Manufacturing Busts, Housing Booms, and Declining Employment

Manufacturing Busts, Housing Booms, and Declining 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 October 2012

Motivation / Introduction Sharp increase in non-employment rate between 2007-2011. We estimate how much of the sharp increase in non-employment can be traced back to the (ongoing) decline of manufacturing sector. A key part of our explanation: the decline of manufacturing was temporarily masked by the national housing boom. We test this explanation by exploiting cross-city variation in manufacturing declines and housing booms. Goal is to account for changes in non-employment rate during the 2000s (both prior to and during the Great Recession) by focusing on (1) massive decline in manufacturing employment and (2) the housing boom-bust cycle. Assess how each of these forces separately affected employment (and other labor market outcomes). Run counterfactuals shutting off each force.

Sharp increase in non-employment rate in U.S. (21-55 year olds)

Sharp increase in non-employment rate in U.S. (21-55 year olds) Great Recession (+7pp)

21,000 Total U.S. Manufacturing Employment (in 1,000s) 19,000 17,000 15,000 13,000 11,000 9,000

21,000 19,000 Total U.S. Manufacturing Employment (in 1,000s) ~1.5 Million Jobs Lost During 1980s and 1990s 17,000 15,000 13,000 11,000 9,000

21,000 19,000 Total U.S. Manufacturing Employment (in 1,000s) ~1.5 Million Jobs Lost During 1980s and 1990s 17,000 15,000 ~3.8 Million Jobs Lost During 2000-2007 13,000 11,000 9,000

21,000 19,000 Total U.S. Manufacturing Employment (in 1,000s) ~1.5 Million Jobs Lost During 1980s and 1990s 17,000 15,000 ~3.8 Million Jobs Lost During 2000-2007 13,000 11,000 Even More Jobs Lost After 2007 9,000

0.40 0.35 Employment Trends for Non-College Men (age 21-55) Manufacturing + Construction Share 0.30 0.25 Manufacturing Share 0.20 0.15 Construction Share 0.10 0.05 0.00

Employment Trends for Non-College Women (age 21-55) 0.140 Manufacturing + Construction Share 0.120 0.100 Manufacturing Share 0.080 0.060 0.040 0.020 Construction Share 0.000 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

Motivation / Introduction Sharp increase in non-employment rate between 2007-2011. We estimate how much of the sharp increase in non-employment can be traced back to the (ongoing) decline of manufacturing sector. A key part of our explanation: the decline of manufacturing was temporarily masked by the housing boom. We test this explanation by exploiting cross-city variation in manufacturing declines and housing booms. Goal is to account for changes in non-employment rate during the 2000s (both prior to and during the Great Recession) by focusing on (1) massive decline in manufacturing employment and (2) the housing boom-bust cycle. Assess how each of these forces separately affected employment (and other labor market outcomes). Run counterfactuals shutting off each force.

Research Questions What is the effect of declining manufacturing sector on non-employment (overall and by age/skill/gender)? To what extent did the housing boom/bust mask and then subsequently unmask the effect of declining manufacturing sector? [ How did the housing boom alter college enrollment during 2000-2007 period? ] Research Design Local Labor Markets strategy (Metropolitan Areas = MSAs) Exploit variation across MSAs in severity of manufacturing decline and size of housing boom Jointly estimate effect of manufacturing decline and (local) housing boom on local area employment and wages

Summary of Findings We find that manufacturing declines significantly affect employment and wages (for non-college men in particular). Housing booms significantly mask manufacturing declines: Non-college men: most of this masking comes from construction employment. Non-college women: almost none of the masking comes from construction. We calculate that roughly 30-40% of increase in non-employment during 2000s can be attributed to decline in manufacturing Most of this is occurring along out of labor force margin rather than unemployment margin Manufacturing and Housing together can explain 30-40% of non-employment growth during the recession (the housing bust unmasked this non-employment growth that would have occurred earlier). We can account for roughly 50-60% of non-employment growth for older men and 20-30% for younger men. Important caveat: we are extrapolating local estimates to national labor market

Summary of Findings We find that manufacturing declines significantly affect employment and wages (for non-college men in particular). Housing booms significantly mask manufacturing declines: Non-college men: most of this masking comes from construction employment. Non-college women: almost none of the masking comes from construction. We calculate that roughly 30-40% of increase in non-employment during 2000s can be attributed to decline in manufacturing Most of this is occurring along out of labor force margin rather than unemployment margin Manufacturing and Housing together can explain 30-40% of non-employment growth during the recession (the housing bust unmasked non-employment growth that would have occurred earlier). We can account for roughly 50-60% of non-employment growth for older men and 20-30% for younger men. Important caveat: we are extrapolating local estimates to national labor market

Getting on same page: What do we mean by masking? Masking occurred both across and within individuals Housing booms typically occurred in places that were not experiencing severe declines in manufacturing. [across-city masking] The types of worker affected by manufacturing and housing boom differed slightly (by age, skill, country of birth) [across-individual masking] We find evidence of within-individual masking using evidence from Displaced Workers Survey (CPS)

Outline 1. Conceptual model 2. Empirical model 3. Main results 4. Counterfactual estimates 5. Conclusion

Conceptual model PURPOSE: To provide a simple model which highlights the following: The interplay between shocks in different sectors When those shocks will result in changes in non-employment Reasons why the response to non-employment resulting from a shock may change over time.

Conceptual model Mass of workers have skill endowment s and reservation wage r, distributed according to F(s,r) Workers can either choose to be employed in either sector A or sector B (which pay w A and w B per efficiency unit, respectively), or they can choose to work in home sector H. Worker of type (s,r) can either supply s efficiency units in A or (1-s) in B. Therefore, worker chooses employment in A iff sw A > (1-s)w B and sw A > r To simplify exposition, assume aggregate production function given by the following: Y = αl A + βl B so that w A = α and w B = β

r β L H α L B L A s* given by αs*=β(1-s*) s

r β L H α L B A H L A A B s* s' given by αs*=β(1-s*) s

r β L B H B L H A H α A H B L A A B A A B s* s' s'' given by αs*=β(1-s*) s

Empirical model Changes in Labor Market Outcomes at Local Level ( L k ) M H O L = f( D, D, D, θ ) k k k k k (1) (2) (3) (4) Definitions: (1) Effect of Manufacturing Labor Demand Change (through all channels) (2) Effect of Housing Related Labor Demand Change (3) Effect of Other Labor Demand Change (not proxied by first two) (4) Effect of Labor Supply Change NOTE: k denotes a local labor market (e.g., MSA) L k could be employment rate, wages, employment in a sector, etc.

Empirical model Changes in Labor Market Outcomes at Local Level ( L k ) M H O L = f( D, D, D, θ ) k k k k k Goal: Estimate: M L / D and L / D H k k k k Problems: We do not observe D M k and D H k We will use proxies for both Ideally, the proxies will be orthogonal to the labor supply shock NOTE: We will estimate a causal effect of manufacturing decline on local labor market outcomes. Housing will be more of a catch all.

Creating IV for Local Change in Manufacturing Labor Demand Instrument for the local declines in manufacturing. Construct predicted change in manufacturing employment following Bartik (1991) ( D ) M k We interact pre-existing cross-sectional variation in industry employment with national industry employment trends. Key assumption: Initial industry variation across MSAs uncorrelated with changes in local labor supply Instrument is strongly predictive of actual changes in manufacturing employment.

Predicted Change vs. Actual Change in Manufacturing

Creating a Housing Related Labor Demand Change H Use housing price growth in local area ( P k ) as our measure of housing related demand change. Intuition We have two direct housing related labor demand channels o Wealth Effect Channel: W k H ( P ) k (+) o Construction Demand Channel: C k H ( P ) k (?) The relationship between construction effect on labor demand and house prices will be positive if variation in house prices is primarily due to variation in housing demand.

Relationship Between Housing Price Growth and Change in Construction Share (Non-College Men, 21-55)

Empirical Model β β M H O L = + D + β P + αx + D + θ + ε k 0 1 k 2 k k k k k P H = γ + δ 1 D M + δg( D H ; Z ) + ωx + D O + θ + ν k k k k k k k k Notes: We treat housing price changes as endogenous Z is a variable which is correlated with housing demand shock but uncorrelated with other unobserved labor demand and labor supply shocks H D k is unobserved shock to housing demand

Comments: Empirical Model β β M H O L = + D + β P + αx + D + θ + ε k 0 1 k 2 k k k k k P H = γ + δ 1 D M + δg( D H ; Z ) + ωx + D O + θ + ν k k k k k k k k When we use house price variation in OLS, we do not take a stand on source of house prices changed during the 2000s; i.e., β 2 d L = = d P k H k direct housing effects (wealth effect and construction) + indirect effects ( D M O, D, θ ) k k k When we use housing price IV, then we are exploiting variation in housing prices that is orthogonal to ( D M O, D, θ ) k k k

Comments: What We Actually Estimate (2) = β + β M H L D + β P + αx + ε k 0 1 k 2 k k k H (1) M P = γ + δ D + [ δz ] + ωx + ν k 1 k k k k We estimate this as two-step estimator (either OLS/OLS or OLS/IV). We report the effect of manufacturing decline to be β 1 +δ 1 β 2. In OLS/OLS, we identify β 2 from residual variation in housing prices (and we do not take a stand on where this variation comes from). In OLS/IV, we identify β 2 from instrument Z which we assume to be orthogonal to ε and ν.

Data For Main Results 2000 Census and 2005-2007 and 2009-2010 ACS o Most of our analysis comes using Census/ACS data. o All of our analysis starts in 2000 as a result. o Focus on individuals aged 21-55. FHFA metro house price indexes

Time Periods Base estimation: 2000-2007 period o Start in 2000 because of data limitations. o Want to focus on pre-recessionary period to get estimated responses. o Interesting to focus on the boom period (highlights masking). Follow up with estimation during the 2007-2010 o Can see if the responses change in different periods. Discuss longer run changes in outcomes: 2000-2010 o Highlights the role of the temporary effects of housing booms.

Estimates from the Empirical Model: Some Graphical Results

Change in non-employment rate for non-college men, 2000-2007

Change in average wage for non-college men, 2000-2007

Change in construction employment share, 2000-2007

Change in manufacturing employment share, 2000-2007

Estimates from the Empirical Model: Formal Estimates

Instrumental Variables Estimates

Identification Using Smoothness Instrument

House Price Growth and Smoothness Instrument

Masking, Between and Within

How Much of the Masking Comes from Within Individuals? Spatial correlation of shocks o Shocks were in Different Places

Correlation Between Manufacturing and Housing Instruments

How Much of the Masking Comes from Within Individuals? Spatial correlation of shocks o Shocks were in Different Places Sub-groups of the populations o o Look at masking across broad demographic groups. Focus on age. o o Manufacturing shock explains 3.3 pp increase in non-employment for young non-college men (21-35) <<25%>> and 5.3 pp increase for old (36-55) <<60%>> during 2000-2010 period. Housing shock explains 1.1 pp decline in non-employment for young non-college men and 1.1 pp decline for old during 2000-2007.

How Much of the Masking Comes from Within Individuals? Spatial correlation of shocks o Shocks were in Different Places Sub-groups of the populations o o Look at masking across broad demographic groups. Focus on age. Within Individual Results (Displaced Worker Survey) o o o Construction does not absorb lots of displaced manufacturing workers. Increased some in the 2000-2007 period. Exploit variation in housing market conditions.

Document Within Worker Effects: Displaced Worker Survey Focus on non-college men displaced from manufacturing and look at: o o Fraction who remained non-employed at time of survey and Fraction who were re-employed in construction Divide sample into Housing Boom MSAs and All Other MSAs based on smoothness IV

Document Within Worker Effects: Displaced Worker Survey

Counterfactuals

Extrapolating Local Estimates to National Labor Market We try to address to several concerns with this exercise: o Migration o Housing Boom Manufacturing demand o Construction Boom o Other National GE effects (e.g., interest rates) To the extent we can address these concerns, they seem to indicate our results are conservative.

Education

Share of pop. with at least one year of college (Age: 18-29)

Did Housing Boom Affect College Enrollment? Use same local labor market design to answer this question (IPEDS data) Places that had large housing booms had a large reduction in the propensity to attend at least one year of college. o o Nearly all the action was on two year colleges (community colleges, technical schools, trade schools, etc.). Found effects for both men and women. During the bust, this trend reversed (but, not completely). Estimates can explain roughly 60-80% of the time series change.

Conclusion We find that manufacturing declines significantly affect employment and wages (for non-college men in particular). Housing booms significantly mask manufacturing declines. We calculate that roughly 30-40% of increase in non-employment during 2000s can be attributed to decline in manufacturing. Our results suggest that the housing bust unmasked non-employment growth that would have occurred earlier in the absence of the housing boom.