Transcript Leisure Inequality Mark Aguiar and Erik Hurst September 2007

Topic 4:
Regional Economics
Part A:
Measuring House Prices
U.S. Housing Data
•
Housing price movements unconditionally
Census data
Transaction/deed data (provided by government agencies or available
via public records)
Household data (PSID, Survey of Consumer Finances, etc.)
•
Repeat sales indices
FHFA (Google it – government agency)
Case-Shiller
Zillow
CoreLogic
Repeat Sales vs. Unconditional Data
•
House prices can increase either because the value of the land under the
home increases or because the value of the structure increases.
o
Is home more expensive because the underlying land is worth more
or because the home has a fancy kitchen?
•
Often want to know the value of the land separate from the value of the
structure.
•
New homes often are of higher quality than existing homes.
•
Repeat sales indices try to difference out “structure” fixed effects –
isolating the effect of changing land prices.
o
Assumes structure remains constant (hard to deal with home
improvements).
FHFA Repeat Sales Index
•
FHFA – Federal Housing Finance Agency
Government agencies that oversee Fannie Mae and Freddie Mac
•
Uses the stated transaction price from Fannie and Freddie mortgages to
compute a repeat sales index. (The price is the actual transaction price
and comes directly from the mortgage document).
•
Includes all properties which are financed via a conventional mortgage
(single family homes, condos, town homes, etc.)
•
Excludes all properties financed with other types of mortgages (sub
prime, jumbos, etc.)
•
Nationally representative – creates separate indices for all 50 states and
a large amount of metro areas.
Case Shiller Repeat Sales Index
•
Developed by Karl Case and Bob Shiller
•
Uses the transaction price from deed records (obtained from public
records)
•
Includes all properties regardless of type of financing (conventional, sub
primes, jumbos, etc.)
•
Includes only single family homes (excludes condos, town homes, etc.)
•
Limited geographic coverage – detailed coverage from only 30 metro
areas. Not nationally representative (no coverage at all from 13 states –
limited coverage from other states)
•
Tries to account for the home improvements when creating repeat sales
index (by down weighting properties that increase by a lot relative to
others within an area).
-5.00%
Jan-92
Aug-92
Mar-93
Oct-93
May-94
Dec-94
Jul-95
Feb-96
Sep-96
Apr-97
Nov-97
Jun-98
Jan-99
Aug-99
Mar-00
Oct-00
May-01
Dec-01
Jul-02
Feb-03
Sep-03
Apr-04
Nov-04
Jun-05
Jan-06
Aug-06
Mar-07
Oct-07
May-08
Dec-08
Jul-09
OFHEO vs. Case Shiller: National Index
20.00%
15.00%
10.00%
5.00%
0.00%
-10.00%
-15.00%
-20.00%
-25.00%
-30.00%
CS Composite 10
CS Composite 20
OFHEO
OFHEO vs. Case Shiller: L.A. Index
0.3
0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
LA-CS
LA-OFHEO
0.15
OFHEO vs. Case Shiller: Denver Index
0.1
0.05
0
-0.05
-0.1
-0.15
Denver-CS
Denver-OFHEO
0.1
OFHEO vs. Case Shiller: Chicago Index
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25
Chicago-CS
Chicago-OFHEO
OFHEO vs. Case Shiller: New York Index
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
NY-CS
NY-OFHEO
Conclusion: OFHEO vs. Case - Shiller
• Aggregate indices are very different but MSA indices are nearly identical.
• Does not appear to be the result of different coverage of properties included.
• The difference has to do with the geographic coverage.
• If using MSA variation, does not matter much what index is used.
• If calibrating aggregate macro models, I would use OFHEO data instead of
Case-Shiller – I think it is more representative of the U.S.
A Note on Census Data
•
To assess long run trends in house prices (at low frequencies), there is nothing
better than Census data.
•
Very detailed geographic data (national, state, metro area, zip code, census
tract).
•
Goes back at least to the 1940 Census.
•
Have very good details on the structure (age of structure, number of rooms,
etc.).
•
Can link to other Census data (income, demographics, etc.).
•
NOTE: The lower the level of geographic area in which house prices are
measured (in all data sets), the more likely the data is either noisy or
imputed.
Part B:
Some More Data: Housing Cycles
Some Housing Facts
1.
Long run house price appreciation averages only 0-2% per year.
o These patterns are consistent across time
o These patterns are consistent across all levels of geographic
aggregation (e.g., countries, state, cities)
2.
Big booms are always followed by big busts
o These patterns are consistent across time
o These patterns are consistent across all levels of geographic
aggregation (e.g., countries, state, cities)
3.
Supply and demand determine housing prices
o Housing supply is very elastic in the long run (as demand goes
up, we build more houses).
Average Annual Real Price Growth By US State (FHFA Data)
State
AK
AL
AR
AZ
CA
CO
CT
DC
DE
FL
GA
HI
IA
ID
IL
IN
1980-2000
-0.001
0.000
-0.009
-0.002
0.012
0.012
0.012
0.010
0.011
-0.002
0.008
0.004
-0.001
-0.001
0.010
0.002
2000-2007 2000-13
0.041
0.015
0.024
-0.001
0.023
0.001
0.061
0.001
0.066
0.013
0.012
0.001
0.044
0.006
0.081
0.038
0.053
0.009
0.068
0.005
0.019
-0.013
0.074
0.025
0.012
0.001
0.047
0.002
0.030
-0.006
0.020
-0.010
State
MT
NC
ND
NE
NH
NJ
NM
NV
NY
OH
OK
OR
PA
RI
SC
SD
1980-2000
0.003
0.008
-0.010
-0.002
0.014
0.015
-0.002
-0.005
0.020
0.003
-0.019
0.009
0.008
0.017
0.007
0.002
2000-2007 2000-2013
0.049
0.016
0.022
-0.003
0.033
0.021
0.007
-0.003
0.041
0.007
0.058
0.013
0.043
0.004
0.060
-0.016
0.051
0.014
-0.001
-0.016
0.019
0.005
0.051
0.006
0.042
0.010
0.059
0.011
0.025
-0.001
0.025
0.009
16
Average
0.011
0.036
0.005
Average Annual Real Price Growth By US State (FHFA Data)
State
AK
AL
AR
AZ
CA
CO
CT
DC
DE
FL
GA
HI
IA
ID
IL
IN
1980-2000
-0.001
0.000
-0.009
-0.002
0.012
0.012
0.012
0.010
0.011
-0.002
0.008
0.004
-0.001
-0.001
0.010
0.002
2000-2007 2000-13
0.041
0.015
0.024
-0.001
0.023
0.001
0.061
0.001
0.066
0.013
0.012
0.001
0.044
0.006
0.081
0.038
0.053
0.009
0.068
0.005
0.019
-0.013
0.074
0.025
0.012
0.001
0.047
0.002
0.030
-0.006
0.020
-0.010
State
MT
NC
ND
NE
NH
NJ
NM
NV
NY
OH
OK
OR
PA
RI
SC
SD
1980-2000
0.003
0.008
-0.010
-0.002
0.014
0.015
-0.002
-0.005
0.020
0.003
-0.019
0.009
0.008
0.017
0.007
0.002
2000-2007 2000-2013
0.049
0.016
0.022
-0.003
0.033
0.021
0.007
-0.003
0.041
0.007
0.058
0.013
0.043
0.004
0.060
-0.016
0.051
0.014
-0.001
-0.016
0.019
0.005
0.051
0.006
0.042
0.010
0.059
0.011
0.025
-0.001
0.025
0.009
17
Average
0.011
0.036
0.005
Average Annual Real Price Growth By US State (FHFA Data)
State
AK
AL
AR
AZ
CA
CO
CT
DC
DE
FL
GA
HI
IA
ID
IL
IN
1980-2000
-0.001
0.000
-0.009
-0.002
0.012
0.012
0.012
0.010
0.011
-0.002
0.008
0.004
-0.001
-0.001
0.010
0.002
2000-2007 2000-13
0.041
0.015
0.024
-0.001
0.023
0.001
0.061
0.001
0.066
0.013
0.012
0.001
0.044
0.006
0.081
0.038
0.053
0.009
0.068
0.005
0.019
-0.013
0.074
0.025
0.012
0.001
0.047
0.002
0.030
-0.006
0.020
-0.010
State
MT
NC
ND
NE
NH
NJ
NM
NV
NY
OH
OK
OR
PA
RI
SC
SD
1980-2000
0.003
0.008
-0.010
-0.002
0.014
0.015
-0.002
-0.005
0.020
0.003
-0.019
0.009
0.008
0.017
0.007
0.002
2000-2007 2000-2013
0.049
0.016
0.022
-0.003
0.033
0.021
0.007
-0.003
0.041
0.007
0.058
0.013
0.043
0.004
0.060
-0.016
0.051
0.014
-0.001
-0.016
0.019
0.005
0.051
0.006
0.042
0.010
0.059
0.011
0.025
-0.001
0.025
0.009
18
Average
0.011
0.036
0.005
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Inflation Adjusted Housing Price Growth in the U.S.
0.10
0.05
0.00
-0.05
-0.10
-0.15
19
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Housing Market: New York
0.20
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
20
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Typical “Local” Cycle: California
0.30
0.20
0.10
0.00
-0.10
-0.20
-0.30
-0.40
21
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Typical “Local” Cycle: Nevada
0.40
0.30
0.20
0.10
0.00
-0.10
-0.20
-0.30
-0.40
22
Average Annual Real Price Growth By OECD Country
Country
1970-1999
2000-2006
Country
1970-1999
2000-2006
U.S.
Japan
Germany
France
Great Britain
Italy
Canada
Spain
Australia
0.012
0.010
0.001
0.010
0.022
0.012
0.013
0.019
0.015
0.055
-0.045
-0.029
0.075
0.068
0.051
0.060
0.081
0.065
Netherlands
Belgium
Sweden
Switzerland
Denmark
Norway
Finland
New Zealand
Ireland
0.023
0.019
-0.002
0.000
0.011
0.012
0.009
0.014
0.022
0.027
0.064
0.059
0.019
0.065
0.047
0.040
0.080
0.059
1970-1999
2000-2006
0.012
0.046
Average
23
Average Annual Real Price Growth By OECD Country
Country
1970-1999
2000-2006
Country
1970-1999
2000-2006
U.S.
Japan
Germany
France
Great Britain
Italy
Canada
Spain
Australia
0.012
0.010
0.001
0.010
0.022
0.012
0.013
0.019
0.015
0.055
-0.045
-0.029
0.075
0.068
0.051
0.060
0.081
0.065
Netherlands
Belgium
Sweden
Switzerland
Denmark
Norway
Finland
New Zealand
Ireland
0.023
0.019
-0.002
0.000
0.011
0.012
0.009
0.014
0.022
0.027
0.064
0.059
0.019
0.065
0.047
0.040
0.080
0.059
1970-1999
2000-2006
0.012
0.046
Average
24
Country Cycles – The U.S. is Not Alone
Real House Price Growth
UK: 1978 - 2006
0.250
0.200
0.150
0.100
0.050
0.000
-0.050
-0.100
-0.150
25
Country Cycles – The U.S. is Not Alone
Real House Price Growth
Italy: 1978 - 2006
0.250
0.200
0.150
0.100
0.050
0.000
-0.050
-0.100
-0.150
26
Country Cycles – The U.S. is Not Alone
Real House Price Growth
Japan: 1978 - 2006
0.120
0.100
0.080
0.060
0.040
0.020
0.000
-0.020
-0.040
-0.060
-0.080
27
Part C:
Some Models of Spatial Equilibrium
Model Particulars (Baseline Model): The City
• City is populated by N identical individuals.
• City is represented by the real line such that each point on the line (i) is a
different location:
i  (, )
• nt (i) :
• ht (i) :

Measure of agents who live in i.
Size of the house chosen by agents living in i.
•

•
nt (i)ht (i)  1

nt (i)di  N
(market clearing condition)
(maximum space in i is fixed and
normalized to 1)
29
Household Preferences
Static model:
max c(i) h(i) 
 > 0 and  > 0
c(i )  R(i )h(i )  Y
normalize price of consumption to 1
ct , ht ,i
Arbitrage implies:
1
Pt (i )  R(i ) 
Pt 1 (i)
1 r
Construction
A continuum of competitive builders can always build a unit of housing
at constant marginal cost  .
Profit maximization implies builders will build a unit of housing anytime:
Pt  
Demand Side of Economy
max c(i ) h(i )   [Y  c(i)  R(i ) h(i)]


c
(
i
)
h
(
i
)
 c(i ) 1 h(i )   

c(i )
 c(i ) h(i )  1
c(i ) h(i ) 

  R (i )
h(i )
 h(i ) 
h(i )
1


 c(i )  (Y  R(i )h(i )) R(i )
(F.O.C. wrt c)
(F.O.C. wrt h)
Housing and Consumption Demand Functions

 1 
h(i ) 
Y

(   )  R(i ) 
c(i ) 

(   )
Y
An Aside: Use of Cobb Douglas Preferences?
•
Implication of Cobb Douglas Preferences:
  
1
h
Y
  



R


  
Rh  
 Y 





(expenditure on housing)
Implication: Constant expenditure share on housing
Implication: Housing expenditure income elasticity = 1
ln(Rh) = 0  1 ln(Y )  
Estimated 1 should be 1
Use CEX To Estimate Housing Income Elasticity
•
Use individual level data from CEX to estimate “housing service” Engel
curves and to estimate “housing service” (pseudo) demand systems.
Sample:
NBER CEX files 1980 - 2003
Use extracts put together for “Deconstructing Lifecycle
Expenditure” and “Conspicuous Consumption and Race”
Restrict sample to 25 to 55 year olds
Estimate:
(1)
(2)
*
*
*
ln(ck) = α0 + α1 ln(tot. outlays) + β X + η
(Engle Curve)
sharek = δ0 + δ1 ln(tot. outlays) + γ X + λ P + ν (Demand)
Use Individual Level Data
Instrument total outlays with current income, education, and occupation.
Total outlays include spending on durables and nondurables.
35
Engel Curve Results (CEX)
Dependent Variable
log rent (renters)
log rent (owners)
log rent (all)
Coefficient
S.E.
0.93
0.84
0.94
0.014
0.001
0.007
* Note: Rent for owners is “self reported” rental value of home
Selection of renting/home ownership appears to be important
36
Engel Curve Results (CEX)
Dependent Variable
log rent (renters)
log rent (owners)
log rent (all)
Coefficient
S.E.
0.93
0.84
0.94
0.014
0.001
0.007
* Note: Rent for owners is “self reported” rental value of home
Selection of renting/home ownership appears to be important
Other Expenditure Categories
log entertainment (all)
log food (all)
log clothing (all)
1.61
0.64
1.24
0.013
0.005
0.010
X controls include year dummies and one year age dummies
37
Demand System Results (CEX)
Dependent Variable
rent share (renters, mean = 0.242)
rent share (owners, mean = 0.275)
rent share (all, mean = 0.263)
Coefficient
S.E.
-0.030
-0.050
-0.025
0.003
0.002
0.002
* Note: Rent share for owners is “self reported” rental value of home
Selection of renting/home ownership appears to be important
38
Demand System Results (CEX)
Dependent Variable
rent share (renters, mean = 0.242)
rent share (owners, mean = 0.275)
rent share (all, mean = 0.263)
Coefficient
S.E.
-0.030
-0.050
-0.025
0.003
0.002
0.002
* Note: Rent share for owners is “self reported” rental value of home
Selection of renting/home ownership appears to be important
Other Expenditure Categories
entertainment share (all, mean = 0.033) 0.012
food share (all, mean = 0.182)
-0.073
clothing share (all, mean = 0.062)
0.008
0.001
0.001
0.001
X controls include year dummies and one year age dummies
39
Spatial Equilibrium
Households have to be indifferent across locations:
Consider two locations i and %
i.
Spatial indifference implies that:
c(i ) h(i )   c(%
i ) h(%
i )





       1 
       1 
Y
Y








 Y 
 Y  %
   
   
   
   
 R(i ) 
 R(i ) 
R(i )  R(%
i)
for all i and %
i

Equilibrium
r
R (i ) 
P (i )
(1  r )
Housing Demand Curve:
    1 r  1 
h(i )=h = 
 
Y 



r
 P 

 
Housing Supply Curve:
P=
Graphical Equilibrium
ln(P)
hD(Y)
ln(κ) =
ln(P*)
ln(h*)
ln(h)
Shock to Income
hD(Y1)
ln(P)
hD(Y)
ln(κ) =
ln(P*)
ln(h*)
ln(h*1) ln(h)
Shock to Income (with adjustment costs to supply)
hD(Y1)
ln(P)
hD(Y)
ln(κ) =
ln(P*)
ln(h*)
ln(h*1) ln(h)
Some Conclusions (Base Model)
•
If supply is perfectly elastic in the long run (land is available and
construction costs are fixed), then:
Prices will be fixed in the long run
Demand shocks will have no effect on prices in the long run.
Short run amplification of prices could be do to adjustment costs.
Model has “static” optimization. Similar results with dynamic
optimization (and expectations – with some caveats)
•
Notice – location – per se – is not important in this analysis. All locations
are the same.
Equilibrium with Supply Constraints
Suppose city (area broadly) is of fixed size (2*I). For illustration, lets index
the middle of the city as (0).
-I
0
I
Lets pick I such that all space is filled in the city with Y = Y and r = r.
2I = N (h(i)*)
    1 r  1 
2I  N 
 
 Y  



r

 P 


 N     1 r 
P   

 Y  
 2I        r 
Comparative Statics
What happens to equilibrium prices when there is a housing demand shock (Y
increases or r falls).
Focus on income shock. Suppose Y increases from Y to Y1. What happens to
prices?
 N      1 r 
P   

 Y  
 2I        r 
  N     1 r  
ln( P)  ln    
   ln(Y )

  2I        r  
With inelastic housing supply (I fixed), a 1% increase in income leads to a 1%
increase in prices (given Cobb Douglas preferences)
Shock to Income With Supply Constraints
ln(P1)
ln(κ) =
ln(P)
hD(Y1)
hD(Y)
ln(h)=ln(h1)
ln(h)
The percentage change in income = the percentage change in price
Intermediate Case: Upward Sloping Supply
ln(P1)
ln(κ) =
ln(P)
hD(Y1)
hD(Y)
ln(h)=ln(h1)
ln(h)
Cost of building in the city increases as “density” increases
Implication of Supply Constraints (base model)?
•
The correlation between income changes and house price changes should
be smaller (potentially zero) in places where density is low (N h(i)* < 2I).
•
The correlation between income changes and house price changes should
be higher (potentially one) in places where density is high.
•
Similar for any demand shocks (i.e., decline in real interest rates).
Question:
Can supply constraints explain the cross city differences
in prices?
Topel and Rosen (1988)
“Housing Investment in the United States” (JPE)
•
First paper to formally approach housing price dynamics.
•
Uses aggregate data
•
Finds that housing supply is relatively elastic in the long run
Long run elasticity is much higher than short run elasticity.
Long run was about “one year”
•
Implication:
Long run annual aggregate home price appreciation
for the U.S. is small.
Siaz (2010)
“On Local Housing Supply Elasticity (QJE 2010)
• Estimates housing supply elasticities by city.
• Uses a measure of “developable” land in the city.
• What makes land “undevelopable”?
Gradient
Coverage of water
• Differences across cities changes the potential supply responsiveness across
cities to a demand shock (some places are more supply elastic in the short
run).
Can Supply Constraints Explain Cycles?
“Housing Dynamics” by Glaeser et al.
Calibrated spatial equilibrium model
Match data on construction (building permits) and housing prices using time
series and cross MSA variation.
Find that supply constraints cannot explain housing price cycles.
Their explanation:
Negatively serially correlated demand shocks.
What Could Be Missing From Simple Model?
• Add in reasons for agglomeration.
• Long literature looking at housing prices across areas with agglomeration.
• Most of these focus on “production” agglomerations.
• We will lay out one of the simplest models – Muth (1969), Alonzo (1964),
Mills (1967)
• Locations are no longer identical. There is a center business district in the
area where people work (indexed as point (0) for our analysis).
• Households who live (i) distance from center business district must pay
additional transportation cost of τi.
Same Model As Before – Except Add in Transport Costs
Static model:
max c(i ) h(i ) 
ct , ht ,i
 > 0 and  > 0
c(i )  R(i )h(i )  Y   i
Still no supply constraints (unlimited areas)
Demand Side of Economy
max c(i ) h(i )   [Y   i  c(i )  R(i )h(i )]


c
(
i
)
h
(
i
)
 c(i ) 1 h(i )   

c(i )
 c(i ) h(i )  1



c(i ) h(i ) 

  R(i )
h(i )
 h(i )   
h(i )
1
 


 c(i )    (Y   i  R (i ) h(i )) R (i )
(F.O.C. wrt c)
(F.O.C. wrt h)
Housing and Consumption Demand Functions

 1 
h(i ) 
(Y   i ) 

(   )
 R (i ) 
c(i ) 

(   )
(Y   i )
Spatial Equilibrium
Households have to be indifferent across locations:
Consider two locations i and %
i.
Spatial indifference implies that:
c (i ) h(i )   c (%
i ) h(%
i )
 

Y   i 
 
%
i 
R (%
i )  R (i )
Y   
%
When i > %
i, R(i) < R(i)
Equilibrium
Equilibrium Result:
All occuppied neighborhoods i will be contained in [-I,I].
Define R(I) and P(I) as the rent and price, respectively,
at the boundary of the city.
Given arbitrage, we know that:
r
R(I) =

(1  r )
 

Y   i 
 
Y   I  
r
  R (i )
(1  r )
Complete Equilibrium: Size of City (Solve for I)

Remember:
h(i)n(i) = 1
and

n(i ) di  N
i 
 1 
2 
di  N
h(i ) 
i 0 
I
 
h(i )  
  
 
 1  r   1 
  r     Y   I   (Y   i )
 



Some Algebra (if my algebra is correct…)

I 
1
2 


 
   1  r   1 
i 0
 (Y   i ) 
 
Y


I


  r   



 



I


N
(Y   i ) di 

2
i 0


di  N



 
   1  r   1 
      r     Y   I  
 



 N 1  r   1 
  

 1
1





2  r   
1


I    (Y )

 
 N 1  r   1 
  
 2  r      1

 


Prices By Distance (Initial Level of Y = Y0)
P
κ
0
I0
i
Linearized only for graphical illustration
Prices fall with distance. Prices in essentially all locations exceed marginal cost.
Suppose Y increases from Y0 to Y1
P
κ
0
I0
I1
i
Even when supply is completely elastic, prices can rise permanently with a
permanent demand shock.
From Glaeser (2007): Suburb House Prices and
Distance to Boston
From Glaeser (2007): Suburb Density and
Distance to Boston
From Glaeser (2007): Cross City Income vs. House
Prices
A Quick Review of Spatial Equilibrium Models
• Cross city differences?
Long run price differences across cities with no differential
supply constraints.
Strength of the center business district (size of τ) drives some
of differences in long run price appreciations across city.
• Is it big enough?
• Fall in τ will lead to bigger cities (suburbs) and lower prices in
center city (i = 0).
Many Urban Models Have Similar Feature
• In model we just outlined, land is made special because of center
city where travel costs = 0.
• Land could be made special for a variety of reasons:
o
Production agglomeration effects (endogenize center city)
o
Export reasons (proximity to ports)
o
Fixed natural amenities (sunshine, nice weather, beautiful
vistas, etc.)
o
Locally provided public goods (school districts, crime)
o
Consumption agglomeration effects (endogenous provision of
amenities).
Part D:
“Endogenous Gentrification and House
Price Dynamics”
(Guerrieri, Hartley, and Hurst)
Within City House Price Growth Appreciation
2000 – 2006
2000 – 2006
2000 – 2006
Midtown
Manhattan
All
NYC
Harlem
45%
130%
~80%
Lincoln
Park
Hyde
Park
All
Chicago
20%
95%
Zip
28277
Zips
28203-7
8%
40%
~40%
All
Charlotte
~8%
70
Within City House Price Growth Appreciation
Between MSA vs. Within MSA Variation in
House Price Appreciation
Mean
Between S.D.
Within S.D.
2000 – 2006
0.81
0.42
0.18 *
1990 – 1997
-0.07
0.21
0.17
• Data from Case Shiller Zip Code Data
• * Within city variation is 2-3 times larger for cities that experienced
non-trivial property price appreciation.
71
What We Do In This Paper
• Present and empirical evaluate a model of within city house price growth
heterogeneity during city wide housing price booms (and busts).
• Formalize the link between neighborhood gentrification and housing price
dynamics in response to city wide housing demand shocks.
• Key ingredient of our model:
o
Assume individual utility is increasing in the income of one’s
neighbors (e.g., a spatial neighborhood externality).
o
Such preferences have been empirically documented by:
Bayer et al. (2007) ; Rossi-Hansberg et al. (2010)
o
Neighborhood amenities are endogenous
72
Where Do the Preferences Come From
• Our preference structure is a catch all for many potential stories.
• As a result, we do not take a stand on what – in particular – people like
about “rich” neighborhoods.
-
Lower crime (dislike poor neighborhoods)
-
Quality and extent of public goods (like schools) – could be through
expenditures or peer effects.
-
Increasing returns to scale in the provision of local service
amenities (restaurants, entertainment options, etc.).
73
Mechanism for Within City Price Movements
• With the externality, any land occupied by rich people will be of higher
value than land occupied by non-rich people.
– Can explain the within city differences in prices such that rich
neighborhoods have higher land prices (Becker and Murphy (2003)).
• Anything that increases the demand for housing of rich people (i.e., an
influx of new rich people) increases the value of the land onto which they
move.
o
New/expanding rich will migrate to the poor neighborhoods that
directly border the existing rich neighborhoods (to maximize value of
the externality)
o
The poor will get priced out of these border neighborhoods.
o
We refer to this process as “endogenous” gentrification.
74
Document Empirical Support for the Model
• Use variation from Bartik-type shocks across cities (cities that get an
exogenous labor demand shock based on initial industry mix).
• For cities that get larger Bartik shocks:
1.
House prices in the city as a whole appreciate more.
2.
Poor neighborhoods that directly abut rich neighborhoods appreciate
the most (both relative to rich neighborhoods and poor neighborhoods
that are far from rich neighborhoods).
3.
Poor neighborhoods that directly abut rich neighborhoods show much
more signs of gentrification (income growth of residents) relative to
other poor neighborhoods.
4.
These patterns occur in the 1980s, 1990s, and 2000s.
75
Caveat 1: Other Stories For Within City Differences
1.
Commuting costs (production agglomeration)
o
o
Classic Urban Story: Muth (1967), Mills (1969), Alonzo (1962))
Recent Work: Van Nieuwerburgh and Weill (2009), Moretti (2009)).
People pay a cost to commute to jobs.
2.
Different fixed amenities
o
o
Classic Urban Story: Rosen (1979), Roback (1982)
Recent Work: Gyrouko et al. (2009)).
Fixed amenities include weather, beautiful vistas, ocean front property, etc.
Note:
The mechanism we highlight could still go through in the presence of
these other stories (even if neighborhood externality is zero).
Note:
We attempt to distinguish among potential mechanisms in our
empirical work.
76
Fact 1: Within City Dispersion is Almost as Large
as Cross City Dispersion
Between MSA
FHFA
CaseShiller
Time
Period
Cross Zip Code
Cross Tract
Within MSA or City
(Within City)
CaseCaseZillow Census Census Census
Shiller Shiller
Median Median Median
(30+
(CS
Tracts
(MSA) (City)
(City)
(City) Cities) Cities)
2000-2006
0.33
0.42
0.18
0.18
0.24
obs
384
20
1,602
472
472
1990-2000
0.17
0.21
0.16
0.17
-
obs
348
17
1,498
496
1980-1990
obs
-
0.15
0.33
0.54
496
9,684
16,161
0.31
0.24
0.44
158
4,640
8,729
77
Fact 2: “Poor” Neighborhoods Appreciate More
New York Metro Area Zip Codes: 2000-2006
78
Fact 2: “Poor” Neighborhoods Appreciate More
Boston, L.A., San Francisco, and Washington: β: -0.22 to -0.49
79
Fact 2: Patterns are Robust Over Time/Space
Top Quartile
Initial House Price
Bottom Quartile
Initial House Price
Washington, D.C.
1.29
1.61
L.A.
1.21
1.76
San Francisco
0.35
0.61
Portland
0.41
0.69
Denver
0.51
0.89
New York City
0.33
1.06
Boston
0.65
0.84
MSA/Time Period
2000-2006 (Case Shiller)
1990-1997 (Case Shiller)
1984-1989 (Furman/Case Shiller)
80
Fact 3: More Variability Among Poor Neighborhoods
•
Variability among neighborhoods in bottom quartile of 2000 house price
distribution was 0.29.
•
Variability among neighborhoods in bottom quartile of 2000 house price
distribution was 0.05.
81
Fact 3: More Variability Among Poor Neighborhoods
•
Variability difference increases with the size of the city wide property price boom.
82
Summary of Facts
• Tremendous amount of within city house price variation.
• Variation across zip codes/census tracts within a city is of similar magnitude
as the well studied cross city variation.
• Poor neighborhoods within a city appreciate most during city wide housing
booms. The more the city as a whole appreciates, the bigger the differential
between rich and poor neighborhoods within a city.
• There is much greater variation in house price appreciation rates among poor
neighborhoods. The variation increases with the size of the city wide
housing boom.
• All the facts are interesting and should be explored more fully in subsequent
theoretical and empirical work.
• Our subsequent theory and empirical work only focuses on trying to
explain the variation among the poor neighborhoods.
83
Model Particulars (Baseline Model): The City
• City is populated by two types (indexed by s) of infinitely lived households;
NR and NP (rich and poor, respectively)
• City is represented by the real line such that each point on the line (i) is a
different location:
i  (, )
• nts (i) :
• hts (i) :
•
•



Measure of agents of type s who live in i.
Size of the house chosen by agents of type s living in i.
nts (i)di  N s
ntR (i)htR (i)  ntP (i)htP (i)  1
(market clearing condition)
(maximum space in i is fixed and
normalized to 1)
84
Model Particulars: Preferences

• Utility

s
max c h ( A  H t (i))
c , h ,i
 ,  ,  0
i 
• Neighborhood Externality:
H (i)   h R ( j )n R ( j )dj
• Preference Assumptions:
 R   P ; can assume ( R   P )
• Static budget constraint:
cs (i) + hs (i)Rs (i) £ y s
• Income (Exogenous)
i 
 y  yR  yP  y
85
Comments on the Model
1.
No distinction between poor people and farm land (nothing interesting
about the poor except they are not rich).
-
Could include a negative externality from living near the poor. We have not
done that at this time.
2.
No bounds on the city (or mechanisms to bound the city – like transport
costs or location specific amenities).
3.
Only two types of income (rich and poor).
4.
Only one dimension of preference externality.
5.
Neighborhoods are of fixed size (do not allow building up).
6.
Externality is over space occupied by rich people (not amount of rich
people).
7.
No uncertainty (more on this later if time allows).
86
Housing Supply/Intermediaries
• Representative builder who builds poor houses in any location at marginal
cost CP and who builds rich houses in any location at marginal cost CR.
•
pts (i ) the price (per unit) of housing in location i at time t for household type
s.
• Assume houses are owned by risk-neutral intermediaries
• Absence of arbitrage implies:
87
Equilibrium
An equilibrium is a sequence of:
•
rent and price schedules:
•
allocations:
•
feasible locations:
Such that:
1.
2.
3.
4.
households maximize utility
representative firm maximizes profits
intermediaries maximize profits
markets clear
88
Full Segregation
•
Many equilibria (with full segregation)
•
Focus on one of the equilibria.
•
Rich live together at center of line (normalize i = 0 to be center of line).
•
Symmetric city – restrict attention to positive side of line.
•
Implications in other equilibria similar (as long as centers are far enough
89
from each other).
Model Predictions:
Neighborhoods, Externality, and Prices
90
Response to Increasing N keeping NR/NP constant
(similar to lower r or increasing yR)
91
Response to Increasing N keeping NR/NP constant
(similar to lower r or increasing yR)
Poor Neighborhoods
That Appreciate Substantially
92
Response to Increasing N keeping NR/NP constant
(similar to lower r or increasing yR)
Poor Neighborhoods
That Do Not Appreciate
93
Implications of Model: Within City
• Lower priced neighborhoods are more price responsive than high priced
neighborhoods to positive demand shocks.
• It is the low priced neighborhoods in close proximity to the high priced
neighborhoods that appreciate the most when there is a positive
housing demand shock.
• The low priced neighborhoods in close proximity to the high priced
neighborhoods that appreciate the most do so because they gentrify
(rich people move into those neighborhoods).
94
Implications of Model: Cross City
• Mechanism is relevant in that it can also explain differences in price
appreciation across cities.
• Higher income growth (NR increase) within a city leads to higher house
price appreciation (P) at the city level, all else equal.
-
Define P as the weighted average of prices within the city.
The city P just reflects the aggregation of the neighborhood p’s.
• The stronger the externality (δ), the larger the price growth at the city level
(P), all else equal.
95
Rest of Paper
• Test predictions of model using:
o
Within city price movements
o
Exogenous “Bartik” shocks to city as a whole (i.e., manufacturing
declines, finance booms, etc.).
o
Show strong support for the model
Poor neighborhoods on the border of rich neighborhoods are more
likely to appreciate in response to a city wide labor demand shock
(relative to other equally poor neighborhoods).
These neighborhoods also experience a rapid turnover in population
type (i.e., they got richer).
96
Part E:
Local Labor Market Adjustment
(Blanchard and Katz)
How Do Locations Respond to Local Shocks?
•
Continue our theme about thinking about regional economics (house prices
are one part of that).
•
The direct mechanism:
•
What implications do mobility have on the response of labor supply, wages,
and unemployment to local economic shocks?
•
Some work:
Mobility.
Blanchard/Katz “Regional Evolutions” (Brookings, 1992)
Topel “Local Labor Markets” (JPE, 1986)
Consider the Following Labor Market (Inelastic Labor Supply)
Labor Supply
W0i  W
Labor Demand
N0i
Consider the Following Labor Market (Inelastic Labor Supply)
Labor Supply
W0i  W
W1i
Labor Demand
N0i
In short run, adjustment takes place on wages (labor supply is less elastic in short run)
Consider the Following Labor Market (Inelastic Labor Supply)
Labor Supply
W0i  W
Labor Demand
N 2i
N0i
In long run, adjustment takes place on N (labor supply is more elastic in long run)
What is the Mechanism?
•
In/out migration of workers…..
Blanchard/Katz Facts: Persistence of Growth Rates
Blanchard/Katz Facts: Cumulative Declines (relative to trend)
Blanchard/Katz Facts: : Cumulative Declines (relative to trend)
Blanchard/Katz Facts: : Cumulative Declines (relative to trend)
Blanchard/Katz Facts: : Cumulative Declines (relative to trend)
Blanchard/Katz Facts: : Cumulative Declines (relative to trend)
Blanchard/Katz Facts: Persistence of Unemployment Rate?
Blanchard/Katz Facts: Convergence of Wages
Blanchard/Katz Facts: Unemployment vs. Growth
Blanchard/Katz Facts: Growth vs. Wages
Blanchard/Katz Facts: Unemployment vs. Wages
Blanchard/Katz Facts: VAR of Negative Regional Shock
Blanchard/Katz Facts: VAR of Negative Regional Shock
Blanchard/Katz Facts: VAR of Negative Regional Shock
Blanchard/Katz Facts: VAR of Negative Regional Shock
Blanchard/Katz Facts: VAR of Negative Regional Shock
Blanchard/Katz Facts VAR of Negative Regional Shock
Blanchard/Katz Facts: VAR of Negative Regional Shock
Blanchard/Katz Facts: VAR of Negative Regional Shock
Conclusions of Blanchard/Katz
•
Regional Adjustments Take Place
•
In short run, response occurs on unemployment and wage margins.
•
In long run, it occurs on labor supply margin (via migration).
•
Spatial equilibrium model has to make individuals indifferent to move
across regions.
Part F:
Regional Convergence
(Barro and Sali-Martin)
Cross-State Convergence in Y/N (R-squared ~ 0.91)
Historical Trends in Convergence
2
Unadjusted 1940-1980
MS
1.5
AR
AL
1
KY
OK
NC
SC
GA
TNLA
KS
SD
ND
TX
NM
VA
NE
WV
MNCO WY
ID IA
AZFLWI
MO
UT
VT IN NHOR WA
ME MT
MD
PA
OH
MI
IL
RI
MA
NJCA
CT
NYNV
.5
DE
2000
4000
6000
8000
Per Capita Income 1940
Fitted values
10000
gr_ipc_40_80
12000
Part G:
Some Facts (I think) – Based on ongoing
work I am doing with Martin Beraja and
Juan Ospina
Figure 1a:
Percentage Point Change in State Unemployment Rate: 2007-2010
10.0
9.0
8.0
7.0
6.0
5.0
4.0
3.0
2.0
0.0
NV
FL
CA
AZ
RI
NC
AL
ID
GA
MI
SC
IN
UT
OR
IL
WA
NJ
CO
TN
CT
KY
DE
DC
MD
NM
OH
MO
MS
WV
WY
HI
NY
PA
VA
MA
TX
WI
LA
ME
MT
KS
OK
MN
AR
NH
VT
IA
SD
AK
NE
ND
1.0
High Unemployment Change States 07-10
Low Unemployment Change States 07-10
Middle Unemployment Change States
2012m9
2012m5
2012m1
2011m9
2011m5
2011m1
2010m9
2010m5
2010m1
2009m9
2009m5
2009m1
2008m9
2008m5
2008m1
2007m9
2007m5
2007m1
2006m9
2006m5
2006m1
2005m9
2005m5
2005m1
2004m9
2004m5
2004m1
2003m9
2003m5
2003m1
2002m9
2002m5
2002m1
2001m9
2001m5
2001m1
2000m9
2000m5
2000m1
Unemployment Rate (Percent)
Figure 2a
Monthly Unemployment Rate:
High, Middle, and Low Unemployment Rate Change States
14.0
12.0
10.0
8.0
6.0
4.0
2.0
0.0
Question
o
Is the persistence across locations that starts during a recession a feature of
recessions?
o
Yes!
Mar_1976
Dec_1976
Sept_1977
June_1978
Mar_1979
Dec_1979
Sept_1980
June_1981
Mar_1982
Dec_1982
Sept_1983
June_1984
Mar_1985
Dec_1985
Sept_1986
June_1987
Mar_1988
Dec_1988
Sept_1989
June_1990
Mar_1991
Dec_1991
Sept_1992
June_1993
Mar_1994
Dec_1994
Dec_1995
Sept_1996
June_1997
Mar_1998
Dec_1998
Sept_1999
June_2000
Mar_2001
Dec_2001
Sept_2002
June_2003
Mar_2004
Dec_2004
Sept_2005
June_2006
Mar_2007
Dec_2007
Sept_2008
June_2009
Mar_2010
Dec_2010
Sept_2011
June_2012
Mar_2013
Unemployment Rate Changes Across States Based on
Unemployment Increase 79-83
14.0
12.0
10.0
8.0
6.0
4.0
2.0
0.0
High Unemployment Change: 79-83
Low Unemployment Chagne: 79-83
Med Unemployment Change: 79-83
Mar_1976
Dec_1976
Sept_1977
June_1978
Mar_1979
Dec_1979
Sept_1980
June_1981
Mar_1982
Dec_1982
Sept_1983
June_1984
Mar_1985
Dec_1985
Sept_1986
June_1987
Mar_1988
Dec_1988
Sept_1989
June_1990
Mar_1991
Dec_1991
Sept_1992
June_1993
Mar_1994
Dec_1994
Dec_1995
Sept_1996
June_1997
Mar_1998
Dec_1998
Sept_1999
June_2000
Mar_2001
Dec_2001
Sept_2002
June_2003
Mar_2004
Dec_2004
Sept_2005
June_2006
Mar_2007
Dec_2007
Sept_2008
June_2009
Mar_2010
Dec_2010
Sept_2011
June_2012
Mar_2013
Unemployment Rate Changes Across States Based on
Unemployment Increase 89-92
12.0
10.0
8.0
6.0
4.0
2.0
0.0
High Unemployment Change: 89-92
Low Unemployment Chagne: 89-92
Med Unemployment Change: 79-83
Question
o
Is the persistence across locations that starts during a recession a feature of
recessions?
o
Yes!
o
Why does such persistence exist? Is it inconsistent with Blanchard and Katz
adjustments? Does it show up in other labor market outcomes in earlier
recessions? Why does the convergence take place at other recessions?
o
Ripe area for potential research. Something I am pursuing now.
A Word on the Price Index
o
Based on goods in the Nielsen dataset.
-
Mostly food
Some non-food (stuff sold in grocery stores or Target).
o
Have detailed observations on prices and quantities.
o
Have detailed location measures.
o
Dataset is massive!
o
Make a price index akin to BLS for these goods.
o
Does not take into account store effects! (Coibion, Gorodnichenko, and Hong
show store effects could be important).
A Word on Wage Adjustments
o
Can one use regional data to test macro models of wage adjustments?
o
Yes – no one has done this now.
Bottom Line
o
Lot of regional variation during the recession in macro variables.
o
Can we use regional relationships to help discipline macro models.
o
Growing area of research (which we will turn to shortly).
Part H:
Some Facts (I know)
Now it is Time For Me to Bring You a Fact
o
I have had this idea for about 18 years!
o
It was one of the original ideas I was kicking around for my dissertation.
o
Very little work on the topic to this day.
o
Big question:
“Does monetary policy help those regions that need the help the least (i.e.,
increase regional dispersion)”
o
Sub-title
“Does monetary policy disproportionately help Las Vegas (doing relatively
bad) or Dallas (doing relatively well).
Mechanism
o
Monetary policy often works through bank lending.
o
Bank lending is dependent on borrower collateral.
o
Borrower collateral is highly pro-cyclical.
o
I think I finally found the empirical approach to get a handle on this issue.
o
New paper I am working on with Martin Beraja (chicago grad student) and
Andreas Fuester (NY Fed).
o
(As an aside – Martin will be presenting this in the student workshop next
thursday night).
Some Data
Some Data
Experiment
o
Use QE1 as a natural experiment
o
First QE by Fed was designed to target the mortgage market (buy up mortgage
back securities).
o
Lower mortgage rates and stimulate economic activity via refinancing
o
Occurred in December of 2008 (three-four months after Lehman).
o
Examine loan origination activity (primarily refinancing activity) in December
2008 across regions.
-
Control for pre and post December 2008 trends
Loan Volume Growth in 12/08 By
Unemployment Increase (Early 2007 through 11/08)
Loan Volume Growth in 11/08 By
Unemployment Increase (Early 2007 through 11/08)
Loan Volume Growth in 12/08 Relative to 11/08 By
Unemployment Increase (Early 2007 through 11/08)
Loan Volume Growth in 12/08 Relative to 11/08 By
LTV Increase (Early 2007 through 11/08)
Estimated Cross Sectional Differences
Cash-Out Refinance Share
o
Cash out refinancing share only slightly higher in bad regions.
o
Total cash out refinancing is still much higher in good places.
What’s Next
o
Write down a model where monetary policy can differ across space
-
Desire to tap into home equity (collateral) – driven by consumption
smoothing motives.
-
Constraints to tapping into home equity (collateral) – driven by falling
collateral values.
o
Otherwise, model is a pretty standard New Keynesian model with multiple
islands (assuming no labor mobility across islands, sticky wages on islands,
tradable goods across the islands, monetary authority controlling money
supply, island specific productivity shocks).
o
Goal is to calibrate the model to assess regional impacts of monetary policy
shocks. Also discuss potential “optimal” monetary policy decisions.
Part I:
Recent Literature Using Regional Variation for
Macro Questions
Caution: Pitfalls of Regional Studies
•
Often not designed to assess general equilibrium effects!
o
Compares outcomes in some region (region 1) with some other region
(region 2).
o
Any effect on the outcome that is the same for both region 1 and region 2
(i.e., aggregate effect) gets differenced out.
What are some potential candidates:
Future tax rate increases (from government spending shock today),
Interest rate changes (due to changing supply and demand of money),
Mobility of capital and labor across regions (Blanchard and Katz type
adjustments),
Effect of local shocks on tradable demand (which effects goods produced
in other regions).
Mian and Sufi (2012)
o
Huge literature exploring effect of housing market (and increase in leverage in
particular) on local labor markets.
Mian and Sufi (2012) (First slide of their talk)
• The decline in aggregate demand, driven by the household
balance sheet channel, is responsible for 65% of the jobs
lost from 2007 to 2009
• We are confident this represents a separate channel from
the uncertainty channel or the construction-related
structural employment channel
• We provide suggestive evidence on the frictions that would
translate demand shocks into employment losses
The Shock
2005
2006
2007
1
.8
.6
.4
Auto sales
1
.9
.8
.7
.6
House prices
Auto sales
(normalized to 1 in 2006)
House prices
2008
2009
2005
2006
2005
2006
2007
2008
2009
2008
2009
1
.9
Groceries
1.1 1.2 1.3
Groceries
(normalized to 1 in 2006)
1
.9
.8
.7
Other durables
1.1
Other durables
2007
2008
2009
2005
2006
High leverage counties, 2006
Low leverage counties, 2006
2007
The Effect on Employment: First Pass
-.2
-.1
0
.1
(Figure 2)
.5
1
1.5
2
2.5
Debt to Income 2006
3
3.5
4
Motivating Example:
Auto Retail versus Auto Manufacturing
(Figure 3)
Auto Manufacturing
0
-1
-2
-.4
-.2
0
.2
Auto Manufacturing Employment Growth 07Q1-09Q1
1
.4
Auto Retail
.5
1
1.5
2
2.5
Debt to Income 2006
3
3.5
4
.5
1
1.5
2
2.5
Debt to Income 2006
3
3.5
4
Employment Growth: Non-Tradable and Tradable Industries
(Figure 4)
0
-.2
-.4
-.6
-.2
-.1
0
Tradable Employment Growth 07Q1-09Q1
.1
.2
Tradable
.2
Non-tradable (excluding construction)
.5
1
1.5
2
2.5
Debt to Income 2006
3
3.5
4
.5
1
1.5
2
2.5
Debt to Income 2006
3
3.5
4
Employment Growth: Non-Tradable and Tradable Industries:
Herfindahl-Based Definition
(Figure 5)
Tradable
-.2
-.5
0
(based on high geographical concentration)
-.1
0
.1
Tradable Sector Employment Growth 07Q1-09Q1
.5
.2
Non-Tradable
.5
1
1.5
2
2.5
Debt to Income 2006
3
3.5
4
.5
1
1.5
2
2.5
Debt to Income 2006
3
3.5
4
Conclusion
.2
Tradable
.2
Non-tradable (excluding construction)
0
-.2
-.4
-.6
-.2
-.1
0
Tradable Employment Growth 07Q1-09Q1
.1
Household balance sheet channel
explains 65% of jobs lost
.5
1
1.5
2
2.5
Debt to Income 2006
3
3.5
4
.5
1
1.5
2
2.5
Debt to Income 2006
3
3.5
4
Nakamura and Steinsson (2011)
o
Look regional variation in military spending and show how it affects local
economic activity.
o
They refer to this as a “local multiplier”
o
I stress that it has little to do with aggregate multiplier. It is about regional
redistribution.
o
However, as they show, you can use local multiplier to learn about what type
of models match the regional data.
Nakamura and Steinsson (2013)
Nakamura and Steinsson (2013)
Autor, Dorn, and Hanson (2011)
o
Look at the rise of imports to China on U.S. regional activity (wages,
employment, population movements, transfer program response, etc.)
o
Use a “Bartik”-like instrument. Use the initial share of manufacturing
employment in specific industries in which China has grown.
-
Identify within manufacturing variation
o
Find it reduces local manufacturing employment
o
Local unemployment and non-participation rise.
o
Wage reductions in local non-manufacturing
o
Large effect on local transfers!
Charles, Hurst and Notowidigdo (2013)
o
Already looked at this in class.
o
Assess housing price booms and manufacturing declines on local labor markets
during the 2000s.
o
Try to assess the deterioration of U.S. labor market prior to recession due to
declining manufacturing.
o
Show that the housing boom masked the deteriorating labor market during this
period (particularly for low skilled workers).
o
Tries to adjust for migration in estimates.