Transcript Document

Manufacturing Decline, Housing Booms, and Non-Employment

Kerwin Kofi Charles University of Chicago Harris School of Public Policy And NBER Erik Hurst University of Chicago Booth School of Business and NBER Matthew J. Notowidigdo University of Chicago Booth School of Business and NBER April 2013

Motivation

 Large and persistent increase in non-employment rate since 2007, particularly for those without a 4-yr college degree.

o For prime age “non-college” men, non-employment rate increased from 17.2% (2007) to 26.2% (2010) and remained at 24.4% (late 2012).

 Recent research has focused on trying to understand the post-2007 patterns in labor market outcomes.

o Policy uncertainty (Bloom et al. 2012); unemployment benefit extension (Rothstein 2012); expansion of government programs (Mulligan 2012); de leveraging and falling house prices (Mian and Sufi 2012, 2013); Spatial and industry mismatch (Sahin et al. 2012)  Most of this literature ignores the fact that non-employment rates have actually been increasing for last decade (particularly for non-college men), despite the unprecedented housing boom that occurred during this time.

This Paper

 We study how changes in the manufacturing sector and in the housing market affected non-employment since 2000.

 Focus on these two phenomena because since 2000 there have been historically dramatic changes in both sectors Accelerated manufacturing employment decline starting around 2000.

Starting in the late 1990s, massive housing boom with housing prices rising by 37% followed by a near total reversal.

Research Questions

 What were the

separate

effects of the decline in manufacturing and the boom and bust in the housing market on non-employment for less-skilled men and other groups?

 How did these two market wide changes

interact

to affect labor market outcomes?

 Use our estimates to assess how labor market outcomes

would have evolved

had the manufacturing decline and/or housing boom/bust not occurred.  Assess whether the housing boom/bust affected longer run skill acquisition [Preliminary Results] Note: Our analysis focuses on the 2000-2007 period and the 2000-2011 period.

Research Strategy

 Local Labor Market Strategy (use MSAs as unit of analysis)  Exploit variation across MSAs in the severity of the manufacturing decline and size of housing boom.

 We

jointly

estimate effect of predicted manufacturing decline and predicted housing demand change on: o MSA non-employment, wages, construction and manufacturing employment, and population.

Empirical Model

Empirical model

Changes in Labor Market Outcomes at Local Level ( 

L k

)  

k f

( 

D M

, 

D H

, 

D O

,  

k k k

(1) (2) (3) (4)

k

) Definitions : (1) (2) (3) (4) Effect of manufacturing labor demand change (through all channels) Effect of “housing related” labor demand change Effect of other labor demand changes (not captured by first two) Effect of labor supply change/labor supply parameters.

Goal is to estimate and

k

k M

k

/ 

k H

Remember:

k

indexes local variation (MSA in our empirical content)

An Instrument for Manufacturing Changes

Predicted Changes in Local Manufacturing

Construct predicted change in manufacturing employment following Bartik (1991) o o o o Refer to this measure as 

D k M

Widely-used measure (Blanchard and Katz, 1992; Bound and Holzer, 1993; Autor and Duggan 2003; Notowidigdo, 2013).

Interact pre-existing cross-sectional variation in industry employment share with national industry employment trends: 

D k M J

  

j

 1  (      )

j

= three digit manufacturing industries o Strongly associated with changes in manufacturing employment. Key assumption: Initial industry variation uncorrelated with

Δθ k

(local labor supply changes) and other unobserved trends in labor demand.

Pause for a Moment….. What is my “wish” test?

o I am wishing that initial MSA differences in manufacturing propensities are uncorrelated with factors that determine labor supply AND any trends in labor demand that are unrelated to subsequent manufacturing declines.

Is it plausible?

o Seems uncorrelated with subsequent changes in factors that determine labor supply (aside from migration). Hard to test the latent trends in labor demand assumption.

Predicted vs. Actual Change in Manufacturing

A Measure of Housing Demand

Predicted Change in Housing Demand

• To derive this measure, we start with a simple formulation of housing demand and supply: ln(

H k D

) ln(

H k S

)  

k D

 

k S

 

k D

 

k S

ln(

P k

) ln(

P k

) • • 

k D

, 

k S

: factors that drive local housing demand and supply, respectively.

k D

, 

k S

: price elasticities of housing demand and supply, respectively.

• Denoting Δ as the log difference and imposing equilibrium condition that

H k D

H k S

, the shock to housing demand can be expressed as:  

k D

 

k D P k H k S

Housing demand shock affects both prices and quantities.

Predicted Change in Housing Demand

 

k D

 

k D P k H k S

• • • Price change can affect local labor market outcomes via “wealth/liquidity effect” which affects demand for local services.

Quantity change can affect local labor market outcomes via changes in local construction activity.

Assuming no shocks to housing supply, we can express housing demand change in terms of observables.

 

k D

 ( 

k D

 

k S

) 

P k

(Key Expression) o o o Local house price change (from FHFA data) Local housing supply elasticity (from Saiz, 2010).

Local housing demand elasticity: base case = 0.7 (Polinsky and Elwood 1979). (Key assumption is local housing demand elasticity is uncorrelated with Δ

ω k

’s,

η k

’s, and other determinants of Δ

L k

.)

Estimating Equations

L k

 0  

k D

  0   1

D k M

  2

k D

 

X k

 

D k O

k

 (Unobserved)

k

(1)  1 

D k M

 

X k

D O k

  

k k

(2) Effects of interest: o o

β 1

+

δ 1 β 2

(Total effect of predicted manufacturing decline)

β 2

(Effect of predicted housing demand change) Key Assumption: Housing demand change does not affect predicted manufacturing decline (Data strongly support this assumption)

An Instrument for Housing Demand Changes

Estimating Equations

L k

 0  

k D

  0  1

D k M

  2

D k

 

X k

 

D k O

k

 (Unobserved)

k

(1)  1

D k M

k

)]  

X k

D O k

  

k k

(2) • Motivation for using an instrument for housing demand change: o Housing demand change measured with error (e.g., housing supply shocks are possible, measurement error in supply elasticity estimate).

o

O

Housing demand change may be result of or

k Δθ k

(omitted variables bias) • Instrument using sharp, structural break in quarterly house price series that occurred in some MSAs during mid-2000s.

Illustration of Instrument

Illustration of Instrument

Discussion of Instrument

• We construct instrumental variable as follows: o Using quarterly house price data, we run MSA-specific regression of (residualized) house prices on quadratic and structural break term. o Choose location of structural break to maximize R 2 of above regression. The magnitude of this estimated coefficient for each MSA is the instrumental variable.

o Instrument scaled to be an annual growth rate, so a value of 0.1 indicates that annual percentage increase in real house prices jumps discontinuously by 10 pctg pts at the location of the structural break.

• Instrument strongly predicts predicted housing demand change • Uncorrelated with many observable characteristics o Housing supply elasticity, predicted manufacturing decline, lagged non-employment, lagged house price growth, etc.

Relationship Between Instrument and Lagged Housing Change

Relationship Between Instrument and Supply Elasticity

Relationship Between Instrument and Lagged Non-Employment

Relationship Between Instrument and Lagged Wages

Instrument vs. “Out of Town” Buyers (21 MSAs)

Relationship Between Instrument and Housing Demand Change

Table 6: First Stage

Data

Data For Main Results

• 2000 Census and 2005-2007 (pooled) and 2009-2011 (pooled) ACS o Compute share non-employed, working in manufacturing, working in construction, population, wages, etc. for each of our 235 MSAs.

o Compute separately by sex × skill groups.

o Focus on individuals aged 21-55.

o Exclude those living in group quarters.

• Focus on two time periods explicitly : o 2000-2007 (before recession started) o 2000-2011 (over the entirety of the 2000s) • Also explored 2007-2011 period ( Implicit in some of our analysis given that effects during recession reconcile differences between two periods)

Main Results

Predicted Manufacturing Change vs. Change in Non-Employment, Non College Men 2000-2007

Predicted Housing Demand Change vs. Change in Non-Employment, Non College Men 2000-2007

Table 2a: Baseline Non-Employment Results (OLS)

Specification: Dependent Variable: Change in Non-Employment Rate, 2000-2007 Non-College Men College Men Non-College Women College Women  

k D

-0.014

(0.004) -0.003

(0.003) -0.010

(0.003) -0.003

(0.002) All -0.010

(0.002) 

D k M

0.641

(0.246) 0.356

(0.149) 0.833

(0.171) 0.288

(0.162) 0.665

(0.148) Standardized Effects  

k D

(1

σ

) 

D k M

(1

σ

) -0.011

0.007

-0.002

0.001

-0.008

0.009

-0.003

0.003

-0.008

0.007

R 2 0.71

0.17

0.69

0.12

0.77

Table 2a: Baseline Non-Employment Results (OLS)

Specification: Dependent Variable: Change in Non-Employment Rate, 2000-2007 Non-College Men College Men Non-College Women College Women  

k D

-0.014

(0.004) -0.003

(0.003) -0.010

(0.003) -0.003

(0.002) All -0.010

(0.002) 

D k M

0.641

(0.246) 0.356

(0.149) 0.833

(0.171) 0.288

(0.162) 0.665

(0.148) Standardized Effects  

k D

(1

σ

) 

D k M

(1

σ

) -0.011

0.007

-0.002

0.001

-0.008

0.009

-0.003

0.003

-0.008

0.007

R 2 0.71

0.17

0.69

0.12

0.77

Specification:

Table 2b: Baseline Construction Results (OLS)

Dependent Variable: Change in Construction Employment Share, 2000-2007 Non-College Men College Men Non-College Women College Women All  

k D

0.012

(0.003) 0.002

(0.002) 0.002

(0.001) 0.001

(0.001) 0.006

(0.002) 

D k M

-0.333

(0.219) -0.075

(0.096) -0.110

(0.036) -0.008

(0.035) -0.184

(0.107) Standardized Effects  

k D

(1

σ

) 

D k M

(1

σ

) 0.009

(82%) -0.003

0.002

-0.001

0.001

(13%) -0.001

0.001

-0.000

0.004

(50%) -0.002

R 2 0.46

0.08

0.21

0.05

0.43

Table 7: Baseline Non-Employment Results (TSLS)

Specification: Dependent Variable: Change in Non-Employment Rate, 2000-2007 Non-College Men College Men Non-College Women College Women  

k D

-0.018

(0.006) -0.006

(0.002) -0.007

(0.004) 0.001

(0.003) All -0.010

(0.003) 

D k M

0.698

(0.228) 0.393

(0.120) 0.859

(0.155) 0.378

(0.164) 0.706

(0.134) Standardized Effects  

k D

D k M

(1

σ

) (1

σ

) -0.014

0.007

-0.004

0.004

-0.005

0.009

0.001

0.004

-0.008

0.007

R 2 0.71

0.22

0.69

0.13

0.77

Table 3: Non-Employment Results, Sub Groups

Table 3: Non-Employment Results, Sub Groups

Table 3: Non-Employment Results, Sub Groups

Table 4: Non-Employment Results, Long Run (2000-2011)

Specification: Changes Defined 2000-2007 Non-College Men All  

k D

-0.001

(0.009) -0.001

(0.005) 

D k M

0.710

(0.219) 0.726

(0.244) Standardized Effects  

k D

(1

σ

) 

D k M

(1

σ

) -0.000

0.007

-0.001

0.008

R 2 0.54

0.60

Table 4: Non-Employment Results, Long Run (2000-2011)

Specification: Changes Defined 2000-2007 Non-College Men All  

k D

-0.001

(0.009) -0.001

(0.005) 

D k M

0.710

(0.219) 0.726

(0.244) Standardized Effects  

k D

(1

σ

) 

D k M

(1

σ

) -0.000

0.007

-0.001

0.008

R 2 0.54

0.60

Table 5a: Wage Response

Table 5b: Population Response

Conceptual Framework and “Masking”

Interpretation

Did housing boom “mask” the deterioration of U.S. labor markets during the 2000-2007 period?

• Masking could occur within and between MSAs and individuals: o

Between-MSA Masking

: Places that experienced manufacturing decline could be spatially different than places that experienced housing demand boom. (Masks at aggregate level, not local level).

o

Within-MSA / Between-Individual Masking

: People within a given MSA affected by manufacturing shock not the same as those affected by housing demand boom. (Already showed some evidence of this across age groups).

o

Within-Individual Masking

: Some workers who would have been negatively affected by manufacturing decline were able to find work because of temporary increase in housing demand.

Exploring Within-Individual Masking

• Micro data from the 1994-2006 Waves of Displaced Worker Survey.

• Workers asked about displacement over three previous years and some facts about previous job (and standard CPS questions).

• We observe MSAs, so we can identify workers in “housing boom MSAs” (top ½ of MSAs with respect to predicted housing demand change 2000 2007).

• Estimate series of Diff-in-Diff models using whether individual had different labor market outcomes if displaced from manufacturing job and lived in housing boom city (during housing boom).

o o Look at non-employment rate at time of survey.

Look at re-employment in construction industry at time of survey.

Within Individual Masking: Re-Employment Rate

Within Individual Masking

Within Individual Masking

Counterfactuals

Counterfactual Estimates

Counterfactual Estimates

Counterfactual Estimates

Counterfactual Estimates

Housing Booms and Human Capital [PRELIMINARY]

Propensity to Have At Least One Year of College (Age: 18-29)

Did Housing Boom Reduce College Enrollment?

 Use same local labor market design to answer this question  Data from IPEDS  States with MSAs that had large housing booms had a large reduction in first-time, full-year college enrollment o o o Effects are heavily concentrated in two-year colleges (e.g., community colleges, junior colleges, technical schools, trade schools, etc.) Similar magnitudes for both men and women Noticeably larger IV results as compared to OLS (likely due to greater measurement error in housing demand shock in state-level analysis)  During the bust, this trend reversed, but only partially, so effects persist

• • •

Conclusion

Manufacturing declines significantly affected non-employment (and wages) during the 2000s.

We estimate housing boom reduced non-employment growth by roughly 30% between 2000-2007 and that roughly 40% of non-employment growth since 2000 can be explained by declining manufacturing.

Housing boom/bust has had a muted effect on non-employment over the entirety of the 2000s.

Key takeaways

: o Sectoral shocks are important part of understanding weak labor market.

o o The full effects of these sectoral shocks would have shown up earlier in aggregate statistics if not for the temporary housing boom.

Individual workers who would have left labor market earlier were kept in market by housing boom. Important to keep in mind given rising U→N transition rate (especially among long-term unemployed).

Long-Run Increase in Non-Employment?

Results do not imply a permanent increase in non-employment:

o o Workers could choose to acquire skills so as to increase market wage. Workers could choose to move to different labor markets.

We think of this is as more of a medium run increase (as opposed to being just do to cyclical fluctuations) – adjustments take time.

Our force is different from traditional mismatch stories.

o For us, people are just moving up and down labor supply curve in response to labor demand shocks (market wage < reservation wage).

However, our results suggest that temporary government policies to stimulate labor demand will NOT have lasting effects on employment.