Labor Markets and Inflation: The International Wage Flexibility Project Presentation at National Bank of Belgium Conference on Wage and Price Rigidities in an Open Economy 13 October,
Download ReportTranscript Labor Markets and Inflation: The International Wage Flexibility Project Presentation at National Bank of Belgium Conference on Wage and Price Rigidities in an Open Economy 13 October,
Slide 1
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 2
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 3
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 4
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 5
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 6
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 7
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 8
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 9
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 10
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 11
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 12
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 13
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 14
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 15
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 16
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 17
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 18
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 19
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 20
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 21
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 22
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 23
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 24
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 25
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 26
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 27
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 28
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 29
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 30
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 31
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 32
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 33
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 34
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 35
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 36
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 37
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 2
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 3
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 4
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 5
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 6
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 7
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 8
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 9
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 10
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 11
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 12
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 13
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 14
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 15
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 16
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 17
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 18
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 19
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 20
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 21
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 22
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 23
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 24
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 25
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 26
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 27
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 28
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 29
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 30
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 31
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 32
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 33
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 34
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 35
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 36
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.
Slide 37
Labor Markets and Inflation:
The International Wage
Flexibility Project
Presentation at
National Bank of Belgium Conference
on Wage and Price Rigidities in an
Open Economy
13 October, 2006
What is the IWFP?
• 13 Country study of wage inflation Sponsored by
the ECB and directed by Erica Groshen (NY Fed.)
and me
• using micro data on individual and occupational
wages analyzed by teams in each country familiar
with the data to be used
• meta-analysis of country level results by team
including directors plus Lorenz Goette, Steinar
Holden, Julian Messina, Mark E. Schweitzer,
Jarkko Turunen, and Melanie E. Ward
• broader than just analyzing wage rigidity (sand
and grease), but that is the part that I’m going to
talk about today
Where to Find Full Paper
• Results: http://brookings.edu/es/research/projects/iwfp_jep.pdf
• Methodology:
http://brookings.edu/es/research/projects/200509_iwfp.pdf
• Or navigate to brookings.edu
– then go to list of scholars
– then go to my page
– then look under current projects for “13 country study
of wage rigidity” and click through to that page
– all the papers for this project are linked to that page
Country Teams
•
•
•
•
•
•
•
Austria
Belgium
Denmark
Finland
France
Germany
Italy
•
•
•
•
•
•
Norway
Portugal
Sweden
Switzerland
United Kingdom
United States
Disclaimer!
Opinions expressed in this
presentation are mine and mine alone.
They should not be attributed to any
other individuals in the IWFP, their
employers, or associates, nor should
they be attributed to the organizations
sponsoring the IWFP.
My Motivation
• Today many central banks have chosen to explicitly target
inflation (and the ones that don’t often do it implicitly).
• Many of these targets are very low (2% or less).
• In 1996 article with Akerlof and Perry (ADP) we showed
that low inflation in the presence of downward nominal
rigidity could lead to substantial unemployment in the
long run.
• Is overly low inflation causing considerable
unemployment (particularly in Europe where
unemployment rates have been stubbornly high)?
• Although ADP model fits well for US and Canada, it fits
very poorly for Britain and continental European
countries. Is “real rigidity” a confounding problem that
makes ADP model inappropriate?
Motivation
(Continued)
• Most previous (pre IWFP) studies of European wage
rigidity use macro data to determine extent (and concept of
rigidity measured is slow adjustment to economic
circumstances rather than downward rigidity)
• Early exceptions include Smith (2000) and Nickell and
Quintini (2001) who both analyze British micro wage data
and find much less evidence of DNWR than in US data
• Biscourp et al. (2004) find mixed results for France
• Knoppik and Thomas Beissinger (2003) find substantial
DNWR in Germany
• Fehr and Gotte (2004) find substantial DNWR in
Switzerland
• Is there really substantial variation across countries or do
differences reflect methodological differences?
How To Measure Rigidity?
• Initially we were unsure about how to get at
presence of different types of rigidity.
• I was mainly interested in looking at the
extent of DNWR across countries using
consistent methodology.
• At first meeting some very interesting
results emerged examining wage change
histograms.
What We Are Going to See
• Histograms of wage percentage wage changes
• We are looking only at job stayers
• In some cases we are looking at reported hourly
wages
• In other cases a measure of income divided by
hours of work
• Some are from surveys, some from social security
data, some from other types of administrative
records
Figure 1: Alternative Wage Change Distributions
.1
0
.05
0
.05
.1
.15
Finland 1988
.15
United States 1987
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
5
10
15
20
25
15
20
25
.1
0
.05
0
.05
.1
.15
Ireland 1996
.15
United Kingdom 1984
0
-15
-10
-5
0
5
10
15
20
25
-15
-10
-5
0
Black lines indicate contemporaneous and prior year's inflation rates
Red-Normal, Green-Weibull, Blue Bars-Data, Yellow-Zero Spike
5
10
Can we use wage change
histograms to diagnose the nature
and extent of wage rigidity?
Wage Change Distribution
(For workers with same wage facing same minimum wage)
No Nominal
Wage Change
Frequency
Observatuions Swept
Down to Zero by
Menu Costs
Observations Swept Up to Rate of
Price Inflation By Downward Real
Wage Rigidity
Observations Swept Up to Zero By
Downward Nominal Wage Rigidity
and Menu Costs
No Real Wage
Change
0.26
0.23
0.20
0.17
0.14
0.11
0.08
0.05
0.02
-0.01
-0.04
-0.07
-0.10
-0.13
-0.16
-0.19
-0.22
Change in Log Wage
Big Problem is Measurement Error
• If people make mistakes reporting their wages then we
see wage changes where there are none.
• If we compute wage=income/hours we see “wage”
changes due to overtime, bonuses, or mistakes in
reporting hours.
• If we use SS data we have similar problems since almost
no country has data that allow us to accurately identify
base wage.
• All evidence suggests that for most data sets frequency
and extent of errors make this a very serious problem
(evidence suggests that in many data sets most reported
wage cuts are actually errors of these sorts).
Correcting for Errors
Use information in correlation between years
• Abowd and Card (1989) suggest that wages have two
components:
– permanent changes
– transient (one period) changes (which result in negative
serial correlation of wage changes)
• New method identifies transient changes as errors and uses
auto-covariance and frequency of sign switching in
changes to identify error rate and error variance.
• This information is used to identify semi-non-parametric
estimate of true wage distribution (that is we estimate the
histogram of wage changes we would see if there were no
errors).
How Do We Know Frequency and
Severity of Errors?
• First crucial assumption is that errors are only
important source of covariance in wage changes
from one year to next.
• With that assumption we can identify the
frequency of errors, and the variance of those
errors when they are made, by looking at the autocovariance of wage changes and the number of
people who have sequential large wage changes of
opposite signs.
• We estimate a statistical model of the wage change
distribution and the error process using method of
moments.
Validating Primary Assumption
• Results applying new method to US largely fit
with those of other studies (a very high degree or
downward nominal rigidity).
• Finnish and German data has virtually no errors
and estimated covariances are tiny and sometimes
positive.
• Portuguese have good and bad data and (as we
will see) the correction doesn’t change the good
data, but makes the bad data look like the good
data.
Portugese Data
Base wage - 1994
Base wage+other labor earnings -1994
0.12
0.12
0.1
0.1
0.08
0.08
0.06
0.06
0.04
0.04
0.02
0.02
0
0
-25 -23 -21 -19 -17 -15 -13 -11 -9
-0.02
-7
-5
-3 -1
1
3
5
7
9
-25 -23 -21 -19 -17 -15 -13 -11 -9 -7 -5 -3 -1
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49
-0.02
Base wage+other labor earnings-1992
Base wage-1992
0.15
0.15
0.13
0.13
0.11
0.11
0.09
0.09
0.07
0.07
0.05
0.05
0.03
0.03
0.01
0.01
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
-25
-0.01
-22
-19
-16
-13
-10
-7
-4
-1
2
5
8
11
14
17
20
23
26
29
32
35
38
41
44
47
50
More Validation
Analysis of Gottschalk Data
• Gottschalk uses regression discontinuity
analysis of individual micro data to
discriminate between true wage changes and
errors in SIPP quarterly data (finds almost no
negative wage changes) .
• We analyzed his data and found that all auto-
covariance in wage changes due to errors
(validating our identifying assumption).
Three Ways to Estimate
Three Types of Rigidity
• Problem is to generate counterfactual distribution
(or notional wage change distribution) with which
to compare actual wage change distribution.
• Three possibilities
– Assume an ideal form for the notional distribution
– Assume that the notional distribution is symmetric
– Assume that the notional distribution has constant
shape over time
We use the ideal form method (2-sided Weibull)
0.
03
5
0.
07
5
0.
11
5
0.
15
5
0.
19
5
0.
23
5
0.
27
5
0
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
-0
.1
95
-0
.1
55
-0
.1
15
-0
.0
75
-0
.0
35
Frequency
Figure 6
2-Sided Weibull vs. Actual for US 97
Percentage Wage Change
Actual
Weibull
Figure 7
2-Sided Weibull vs. Actual for UK 1988
0.1
0.06
0.04
0.02
Percentage Wage Change
Actual
Weibull
0.28
0.25
0.22
0.19
0.16
0.13
0.10
0.07
0.04
0.00
-0.02
-0.05
-0.08
-0.11
-0.14
-0.17
0
-0.20
Frequency
0.08
Ideal Distribution
(only method I’m going to discuss today)
• In general, two-sided Weibull fits upper tail of nearly all
distributions very well and Gottshalk’s true wage
changes (upper tail) have two-sided Weibull shape.
• So we will assume that notional wage change
distribution is 2-sided Weibull and estimate its
parameters along with
– A fraction r of workers are subject to downward real wage
rigidity and if their notional wage change < expected inflation
they get expected inflation
– A fraction n of workers are subject to downward nominal wage
rigidity and if their notional wage change <0 they get wage
freeze.
– A fraction of workers are subject to symmetric nominal rigidity
and get no wage change if their notional wage change is within
2% of zero on either side
Aside: What Sort of Process Gives
Rise to Weibull?
• Wage increases (above some average level) result
from tournament with winners at each level
getting an increase of a fixed size and then
competing only with winners of first round for a
larger increase in next round.
• Size of increase from winning a round grows
exponentially.
• This is notably different from normal distribution
that would result if workers were evaluated on
many independent criteria and then given raises
depending on how many they satisfied.
Real Rigidity
la
nd
K
ay
U
or
w
Ire
N
Ita
ly
U
et
he S
rla
nd
s
G
er
m
an
y
D
en
m
ar
k
re
ec
e
Au
st
ri
Be a
lg
iu
m
Po
rt
Sw uga
l
itz
er
la
nd
Fr
an
ce
Fi
nl
an
Sw d
ed
en
N
G
0
.2
.4
.6
.8
Real and Nominal Ridigity by Country
Nominal Rigidity
International Comparisons
• Considerable variation across countries in the extent of
downward nominal and downward real wage rigidity
despite correction for differences in data quality
• Some tendency for countries to have either DNWR or
DRWR but not both (tendency is stronger when we restrict
comparison to the 90s)
• How do our estimates compare to those of others and
between different data sources in our own sample?
– ECHP results with other data sets correlate .53 for both real and
nominal rigidity
– Wouldn’t expect perfect correlation since time periods don’t
overlap and we do see some changes over time
– Country averages for nominal rigidity correlate .55 with simple
measures constructed from uncorrected data
– Country averages for real rigidity correlate .25 with simple
measure used in JEP paper constructed from uncorrected data
.6
Comparing DNWR from different datasets
Portugal
.4
.5
Italy
.3
France
Austria
Denmark
UK
.2
Finland
.1
Germany
0
.2
.4
ECHP Estimates
.6
.8
Holden and Wulfsberg r=.45(.08)
1
Italy
Portugal
.8
.8
1
Knoppik and Beissinger r=.74(.09)
.6
Belgium
Austria
PortugalGreece
Denmark
Sweden
Austria
Netherlands
Greece
.2
UK
Ireland
Switzerland
Finland France
0
0
.2
.4
.6
IWFP n ECHP only
Denmark
Germany
Belgium
0
.2
Germany
France
Ireland
UK
Norway
.4
.4
Finland
HW DNWR
.6
Italy
.8
0
.2
.4
IWFP n
Source: HW (2005) FWCP from Table B1, page 38; KB (2005) Table 4, page 29 and IWFP.
.6
.8
What is Correlated with
Downward Nominal Rigidity?
1.0
1.0
0.9
NL
0.8
0.7
NL
0.8
0.7
IT
US
US
PT
0.6
n
GR
0.9
GR
0.5
n
SZ
0.4
AT
DK
0.5
SZ
AT
0.4
FR
SE
0.3
SE
FI
UK
DK
BE
0.1
FI
UK
0.2
IE
NO
NO IE
BE
0.1
DE
0.0
DE
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
3.0
3.5
Corporatism
1.0
1.0
0.9
0.9
GR
NL
0.8
0.7
US PT
0.7
IT
GR
NL
0.8
IT
US
PT
0.6
0.6
0.5
n
SE
SZ
0.4
FR
UK
0.2
0.4
FR
AT
0.3
DK
NO
IE
FI
0.1
BE
DE
0.0
SZ
SE
FI UK
0.2
IE
0.1
0.5
DK
AT
0.3
NO
BE
DE
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.6
0.8
1.0
1.2
Wage Indexation
1.0
1.0
0.9
GR
0.8
0.7
US
0.9
NL
NL
0.8
0.7
IT
PT
0.6
n
FR
0.3
0.2
n
IT
PT
0.6
IT
GR
US
PT
0.6
n
0.5
SZ
SE
0.4
AT
FR
DK
0.3
0.5
0.4
0.3
FI
UK
0.2
IE
0.1
NO
0.2
BE
SE
SZ
AT
DK
FI
UK
IENO
BE
0.1
DE
0.0
FR
DE
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with
Downward Real Wage Rigidity?
0.6
0.6
SE
FI
0.5
0.4
r
PT
0.3
AT
UK
r
BE
IE
PT
0.3
AT
UK
NO
0.2
FR
0.4
BE
IE
FI
SE
0.5
FR
NO
0.2
SZ
SZ
DE
0.1
US
IT
0.1
IT
DK
US
GR
NL
0.0
GR
0.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
DE
DK
3.5
4.0
0.0
0.5
1.0
1.5
Aggregate EPL
2.0
2.5
NL
3.0
3.5
Corporatism
0.6
0.6
SE
0.5
FI
0.4
0.4
BE
IE
r
FI
FR
0.5
FR
IE
PT
0.3
r
UK
SZ
0.2
AT
0.3
UK
NO
0.0
0.1
DE
DK
US
NL
0.1
DK
NL
GR
0.0
NO
0.2
IT
US
BE
PT
AT
DE
0.1
SE
SZ
IT
GR
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
0.2
0.4
Union Density
0.8
1.0
1.2
Wage Indexation
0.6
0.6
SE
SE
FI
0.5
0.5
FR
0.4
PT
0.3
UK
0.2
0.4
AT
NO
DK
GR
r
PT
0.3
AT
0.2
DE
0.1
IT
US
BE
IE
SZ
0.1
FR
FI
BE
IE
r
0.6
UK
NO
SZ
DE
DK
IT
US
NL
0.0
NL
GR
0.0
0.0
20.0
40.0
60.0
Bargaining Coverage
80.0
100.0
0.0
10.0
20.0
30.0
40.0
50.0
Minimum Wage / Average Wage
60.0
70.0
What is Correlated with Rigidity?
• Nothing consistently statistically significant
at conventional levels.
• Union membership and bargaining coverage
variables statistically significant at .1 level
(we are getting better results now using time
series variation).
• Notable case is US where real rigidity was
notable in 1970s but disappears in the 1980s
after the breakdown of pattern bargaining.
Does Rigidity Affect Unemployment?
• According to Akerlof, Dickens and Perry
1996 a 1 percent increase in in nominal
wages due to rigidity should create
somewhere between a 1 and 5 percentage
point increase in unemployment
• We compute estimates of the impact of each
type of rigidity on wages and estimate a
series of models of the impact of rigidity.
Table 3
Effects of Rigidity on Unemployment
Dependent Variable/Specification
Unemployment
Change in Inflation
(Phillips Curve)
Unemployment Effect
(b/a)
1.26
.90
2.99
2.90
Standard Error
(.35)
(.32)
(1.28)
(2.09)
.00
.01
.01
.08
no
yes
no
yes
p for null hypothesis
0 b/a
(one tail test)
Controls for year
(all contain controls for
dataset (country))
Conclusions on Unemployment
• Unemployment effects generally
statistically significant.
• Can’t reject the hypothesis that
unemployment effects are in the range
predicted by ADP model.
• Can’t reject the hypothesis that effects of
real and nominal rigidity are equal (as
theory predicts).
(Tentative) Policy Implications
• Most Euro-zone countries characterized by downward
real wage rigidity rather than downward nominal
wage rigidity. Benefits of increased inflation may not
be so great in those countries compared to US and
Canada (recent nominal wage cuts in Germany are
example of how corporatist countries can overcome
nominal wage rigidity).
• Several Euro-zone countries have substantial
downward nominal wage rigidity so allowing inflation
to move to bottom end of ECB target zone could be
very costly. Should this happen these countries might
be better off if they left EMU. Though gains would be
minor given ECB policy to date.
More Policy Conclusions
• No evident effect of EMU on rigidity in these
data (though most data don’t go back far
enough to tell if there are effects of
Maastricht treaty).
• No Evidence that persistent low inflation
reduces the incidence of downward nominal
wage rigidity. Thus no evidence that
institutions adapt to low inflation.
• For US new method confirms results of
previous studies – significant DNWR. New
finding is that there is no evidence of DRWR
by 1990s. Thus Fed should avoid very low
inflation.