THE GREAT RECESSION: UNEMPLOYMENT INSURANCE AND STRUCTURAL ISSUES Jesse Rothstein CLSRN Summer School June 2013 Unemployment Rate Percent of labor force, seasonally adjusted 12 10 Oct. 2009: 10.0% 8 6 Oct. 2012: 7.9% 4 2 Oct. 2007: 4.7% 0 2004 2005 2006 2007 2008 2009 2010 2011 2012 Source: Department of Labor, Bureau of Labor Statistics. 1
Long- term Unemployment Rate Percent of labor force unemployed 27 weeks or more, seasonally adjusted 5 4 3 2 1 0 2004 2005 2006 2007 2008 2009 2010 2011 2012 Source: Department of Labor, Bureau of Labor Statistics. Employment- PopulaCon Rate by Sex Percent, seasonally adjusted 100 80 Female Male 60 40 20 0 1948 1953 1958 1963 1968 1973 1978 1983 1988 1993 1998 2003 2008 Source: Department of Labor, Bureau of Labor Statistics. 2
Unemployment Insurance claims Source: Henry S. Farber and Robert Valletta, Extended Unemployment Insurance and Unemployment Duration in the Great Recession: The U.S. Experience. Working paper, June 2011. Maximum duration of Unemployment Insurance benefits Source: Henry S. Farber and Robert Valletta, Extended Unemployment Insurance and Unemployment Duration in the Great Recession: The U.S. Experience. Working paper, June 2011. 3
The effect of UI on unemployment UI subsidizes unemployment in two ways: Reduces incentive to search for work. Creates incentive to remain in labor force. Optimal UI policy trades off consumption smoothing benefit with moral hazard cost. Might expect benefits to be larger, costs lower in recessions à extend/increase benefits when UE is high. Exits from unemployment, by UI eligibility Source: Henry S. Farber and Robert Valletta, Extended Unemployment Insurance and Unemployment Duration in the Great Recession: The U.S. Experience. Working paper, June 2011. 4
Meyer (1990); Katz-Meyer (1990a, b) Early 1990s estimates for 1980s Look for spike in UI exit hazard as benefit expiration approaches. Mortensen (1977) search model Search intensity s determines offer arrival rate. All offers arrive from stationary wage distribution F(). Accepted if exceed reservation wage w. UI benefits reduce urgency of job-finding, so reduce s & raise w. Exit hazard s(1-f(w)) as benefit exhaustion approaches. Data from administrative records exit hazard = 100% at benefit exhaustion. Important role of temporary layoffs. Predicted unemployment exit hazard, by time until UI exhaustion Source: Bruce Meyer, Unemployment Insurance and Unemployment Spells, Econometrica 58(4), July 1990. 5
Empirical unemployment exit hazard, by time in unemployment Source: Bruce Meyer, Unemployment Insurance and Unemployment Spells, Econometrica 58(4), July 1990. Empirical unemployment exit hazard, by time until UI exhaustion Source: Bruce Meyer, Unemployment Insurance and Unemployment Spells, Econometrica 58(4), July 1990. 6
Card-Levine Examine temporary New Jersey extended benefits program. Argue that extension was exogenous to labor market conditions. Focus on probability of exhausting regular benefits, weekly exit hazards. Look at onset and exit of program. Can examine individuals already drawing UI when new program was announced. Regular UI benefits exhaustion rate, by month Source: David Card & Phillip B. Levine, Extended benefits and the duration of UI spells: Evidence from the New Jersey extended benefit program, Journal of Public Economics 78 (2000). 7
Empirical unemployment exit hazards Source: David Card & Phillip B. Levine, Extended benefits and the duration of UI spells: Evidence from the New Jersey extended benefit program, Journal of Public Economics 78 (2000). UI survival curves Source: David Card & Phillip B. Levine, Extended benefits and the duration of UI spells: Evidence from the New Jersey extended benefit program, Journal of Public Economics 78 (2000). 8
Card-Chetty-Weber Two effects of UI on search effort, with different implications: Moral hazard (substitution deadweight loss). Liquidity (wealth welfare improving). A lump sum severance payment has a liquidity but not a moral hazard effect. No effect if workers can smooth consumption. Regression discontinuity design in Austrian data: At 3 years of service, entitled to 2 months severance payment. UI benefits: 20 weeks if < 36 months of employment in previous 5 years; 30 weeks otherwise. Tenure distribution on pre-displacement job Source: David Card, Raj Chetty, and Andrea Weber, Cash-in-Hand and Competing Models of Intertemporal Behavior: New Evidence from the Labor Market, QJE November 2007. 9
Mean wage on pre-displacement job Source: David Card, Raj Chetty, and Andrea Weber, Cash-in-Hand and Competing Models of Intertemporal Behavior: New Evidence from the Labor Market, QJE November 2007. Predicted UI exit propensity given predetermined characteristics Source: David Card, Raj Chetty, and Andrea Weber, Cash-in-Hand and Competing Models of Intertemporal Behavior: New Evidence from the Labor Market, QJE November 2007. 10
Average nonemployment duration around the severance qualification threshold Source: David Card, Raj Chetty, and Andrea Weber, Cash-in-Hand and Competing Models of Intertemporal Behavior: New Evidence from the Labor Market, QJE November 2007. Average nonemployment duration around the extended benefit qualification threshold Source: David Card, Raj Chetty, and Andrea Weber, Cash-in-Hand and Competing Models of Intertemporal Behavior: New Evidence from the Labor Market, QJE November 2007. 11
Average wage change, pre- to post-displacement, around the severance pay qualification threshold Source: David Card, Raj Chetty, and Andrea Weber, Cash-in-Hand and Competing Models of Intertemporal Behavior: New Evidence from the Labor Market, QJE November 2007. von Wachter-Bender-Schmieder Q: How do effects of UI vary over the business cycle? Effective UI replacement rate à larger effects. Lower job arrival rate may make workers less confident in their ability to choose their exit timing; congestion may reduce attenuate aggregate effects of individual decisions à smaller effects. Note: Consumption smoothing effects almost certainly larger in recessions, b.c. time to reemployment rises. Strategy: Regression discontinuity in Germany. Benefits depend on age: 12 months if < 42, 18 months if 42-43, 22 months if 44-48, 26 months if 49-53, Mapping doesn t change over a fairly long time period. Estimate effect separately by year and correlate with business cycle. 12
Source: Johannes Schmieder, Till von Wachter, and Stefan Bender, The Effects of Extended Unemployment Insurance Over the Business Cycle: Evidence from Regression Discontinuity Estimates over Twenty Years, working paper, July 2011. Source: Johannes Schmieder, Till von Wachter, and Stefan Bender, The Effects of Extended Unemployment Insurance Over the Business Cycle: Evidence from Regression Discontinuity Estimates over Twenty Years, working paper, July 2011. 13
Source: Johannes Schmieder, Till von Wachter, and Stefan Bender, The Effects of Extended Unemployment Insurance Over the Business Cycle: Evidence from Regression Discontinuity Estimates over Twenty Years, working paper, July 2011. Source: Johannes Schmieder, Till von Wachter, and Stefan Bender, The Effects of Extended Unemployment Insurance Over the Business Cycle: Evidence from Regression Discontinuity Estimates over Twenty Years, working paper, July 2011. 14
Source: Johannes Schmieder, Till von Wachter, and Stefan Bender, The Effects of Extended Unemployment Insurance Over the Business Cycle: Evidence from Regression Discontinuity Estimates over Twenty Years, working paper, July 2011. Three ways to estimate the effect of recent UI extensions Extrapolate from 1980s estimates of UI effects. Mazumder (2010): UI extensions raised UE rate by 0.8 1.2 percentage points. Extrapolate from pre-recession trends, attributing any discrepancy to UI. Fujita (2011) 1.2 p.p.; Barro (2010) 2.7 p.p. Use haphazard roll-out of UI extensions to identify UI effect in current data. Rothstein (2011), Farber & Valetta (2011). 15
Rothstein (2011) Four identification strategies for effects of 2007-2011 changes 1. Control flexibly for labor market conditions 2. Job-leavers as control group 3. Focus on variation coming from state decisions about optional triggers in Extended Benefits (EB) program 4. Focus on time-to-exhaustion (with state x month FEs) identify UI effect from changes in duration profile over time & space. à Each isolates a different component of variation. à Assume workers myopic about future legislative extensions (but Farber & Valetta make opposite assumption w/ similar results). Source: Jesse Rothstein, Unemployment Insurance and Job Search in the Great Recession, Brookings Papers on Economic Activity, Fall 2011. The roll-out of extended benefits EB triggered EUC triggered EUC automatic Regular benefits EUC authorizations, scheduled expiration, and (retroactive) reauthorizations 16
Variation in EB benefits due to adoption of optional triggers Data Matched monthly CPS, April 2004-March 2011. 77,000 unemployed job-losers, with duration of unemployment and 2 follow-up interviews. Match to state UI rules. Advantages: Current, large samples. Can follow beyond expiration of UI benefits, and have UIineligibles. Can distinguish reemployment from labor force exit. Disadvantages UI receipt / eligibility not measured. Noise in labor force status & duration. Many U-U-U spells are coded as U-N-U (Poterba & Summers, Abowd & Zellner). 17
35 Empirical specification (strategy1) where: λist ln = D ist θ + P n (n ist ; γ)+p Z (Z st ; δ)+α s + η t. 1 λ ist λist is the log odds of monthly unemployment exit for ln 1 λ ist job loser i from state s at time t D ist is the number of weeks of benefits (including those already used) n ist is the unemployment duration, with P n (n ist; γ) =n istγ 1 + n 2 istγ 2 + n 1 ist γ 3 +1(n ist 1) γ 4 α s, η t are state and time effects P Z (Z st ; δ) is a flexible function of observable measures of economic conditions Also: (1) allow θ to vary with n ist < 26 vs. n ist 26 (2) multinomial logit models for reemployment vs. LF exit Strategies 2 and 3 Strategy 2: Use job-leavers as a control group. Allows state-by-month FEs Separate P n (n; γ) functions for UI-eligible, ineligible. Control for simulated D (common to both groups), individual Xs. Allow economic conditions measures to affect relative hazards. Strategy 3: Zero in on variation coming from state take-up of EB program. 3 optional triggers for states to adopt; not all states do. Add controls for: EUC weeks available EB benefits if don t adopt any optional triggers EB benefits if adopt all optional triggers Status of each of the triggers Remaining variation in D comes only from state take-up decisions (interacted with trigger status); none from legislative forecasts. 18
Strategy 4: Time-to-exhaustion Strategies 1-3 focus on effect of number of weeks of available benefits. Alternatively, can model effect of time-to-exhaustion, d=max(0, D-n): λist 99 ln = f (d ist ; θ)+ 1(n ist = v) γ v + α st 1 λ ist v=0 Allows state-by-month FEs plus unrestricted n controls. f() can be arbitrarily flexible; effects of d are identified from: d/ D = 1 if n D, 0 otherwise. D varies across n within s-t cells due to anticipated EUC expiration. Logit models for unemployment exit Strategy 1 Strategy 2 Strategy 3 Sample is job-losers Sample is Sample is joblosers all unemp. (1) (2) (3) (4) (5) (6) Weeks of benefits (/100) X 0.20 0.13 0.10-0.13 0.07-0.12 unemployed < 26 weeks (0.15) (0.15) (0.14) (0.19) (0.20) (0.22) Weeks of benefits (/100) X -0.30-0.34-0.36-0.23-0.43-0.62 unemployed 26+ weeks (0.10) (0.09) (0.09) (0.11) (0.19) (0.27) Controls Unemp duration controls Y Y Y Y Y Y State unemployment rate linear cubic cubic cubic*elig cubic cubic State insured unemp rate State new UI claims rate State empl. growth rate State-by-month FEs Job loser indicator, alone & in interactions Weeks * sensitivity to expectations EUC weeks EB control function cubic cubic cubic Effect of UI extensions on avg. exit hazard in 2010:Q4-1.0 pp -1.3 pp -1.4 pp -1.3 p.p. -1.8 pp -3.1 pp Y Y Y Y Y 19
Multinomial logit models for reemployment vs. labor force exit Sample is job-losers (1) (2) (3) (4) (5) Reemployment Weeks of benefits (*100) X 0.19 0.24 0.18 0.48 0.01 unemployed < 26 weeks (0.19) (0.19) (0.19) (0.24) (0.33) Weeks of benefits (*100) X -0.44-0.42-0.47-0.29-0.64 unemployed 26+ weeks (0.13) (0.14) (0.14) (0.21) (0.37) Labor force exit Weeks of benefits (*100) X -0.19-0.12-0.11-0.41-0.32 unemployed < 26 weeks (0.21) (0.21) (0.21) (0.45) (0.26) Weeks of benefits (*100) X -0.38-0.34-0.42-0.55-0.58 unemployed 26+ weeks (0.13) (0.13) (0.15) (0.37) (0.34) Effect of extensions on average hazards in 2010:Q4 Reemployment Labor force exit -0.6 p.p. -0.5 p.p. -0.7 p.p. +0.2 p.p. -1.2 p.p. -1.2 p.p. -1.0 p.p. -1.2 p.p. -2.0 p.p. -1.8 p.p. Effects of time until exhaustion on log odds of unemployment exit (strategy 4) Reemployment Labor force exit Parametric Relative log odds -.75 -.5 -.25 0.25 Nonparametric Relative log odds -.75 -.5 -.25 0.25 30 20 10 Weeks until exhaustion of benefits 0 30 20 10 Weeks until exhaustion of benefits 0 20
Simulated unemployment without UI extensions So what is going on? (Rothstein, 2012) Insufficient aggregate demand? Something else ( structural )? Demand is there, but it isn t getting to the workers. Several mechanisms proposed High implicit taxes discourage work or job search real shortage is on the supply side Mismatch there are jobs and there are workers, but they aren t searching in the same markets Declines in search efficiency of unspecified origin One definition: Unemployment that wouldn t be reduced by balanced increases in demand (without inflation). Source: Jesse Rothstein, The Labor Market Four Years Into the Crisis: Evaluating Structural Explanations, Industrial and Labor Relations Review. 21
Claims A substantial part of the increase in unemployment since the beginning of the recession reflected factors other than a shortfall in aggregate demand. (FOMC, 1/12) Labor supply reductions (due in large part to expansions of safety net) explain decline in employment (Mulligan 2009, 2011a, b). Firms have jobs, but can t find appropriate workers. The workers want to work, but can t find appropriate jobs. There are many possible sources of mismatch geography, skills, demography and they are probably all at work (Kocherlakota 2010). Structural UE due to supply shift w D S S w w L L L 22
Structural UE due to supply shift w w S S w D w D L L L Structural UE due to demand shifts LABOR MARKET A LABOR MARKET B S A w A S B w A D A w B w B D A D B D B L A L A L B L B 23
How to distinguish? Structural explanations w/ sufficient aggregate demand imply that employers have jobs but are having trouble finding workers. I.e., that there is a supply shortage (in the markets where the jobs are). à Should expect rising wages (in some markets). Estimating wage changes Current Population Survey monthly measures of hourly wages. Construct 12-month changes. Wage changes over business cycle are confounded by compositional changes, stickiness. Strategy 1: Use repeated observations on individuals employed in both periods. Strategy 2: Use wages on new jobs, reweighted for changes in observables 24
Figure 11. Twelve-month changes in mean wages, various subsamples 12-month growth in average wages (percent) -4-2 0 2 4 6 All workers Composition-adjusted New jobs 2006 2007 2008 2009 2010 2011 Date Change in log real wage at percentile -.04 -.03 -.02 -.01 0.01.02.03.04.05 Figure 12. Change in distribution of starting wages, 2007-8 to 2011-12 All industries Industries with large job openings increases 0 20 40 60 80 100 Percentile of residual log real wage distribution 25