Wage flexibility of older workers and the role of institutions

Wage flexibility of older workers and the role of institutions Evidence from the German LIAB data set Martin Kerndler Vienna Graduate School of Economics University of Vienna Workshop Arbeitsmarktökonomie Nov 11, 2016

Motivation wages of older workers might be too high relative to their productivity, leading to early job exit (OECD, 2006, 2013; Frimmel et al., 2015; Skirbekk, 2004, 2008) wage profiles however, high wages per se do not promote layoffs as long as they are sufficiently flexible how does wage flexibility with respect to firm-level performance shocks evolve over age? (du Caju et al., 2007, 2012) is this age pattern affected by different institutional settings? 2 / 30

Related literature micro wage flexibility: Guiso et al. (2005), Cardoso and Portela (2009), Gürtzgen (2014), Kátay (2016) downwards wage rigidity: Dickens and Goette (2006), Babecký et al. (2010), Messina et al. (2010), Du Caju et al. (2007, 2012, 2015) impact of works councils: Freeman and Lazear (1995), Addison (2009), Addison et al. (2001, 2006, 2010), Mueller (2011, 2012), Gürtzgen (2014) 3 / 30

Methodology Guiso, Pistaferri, Schivardi (JPE, 2005) estimate wage responses to idiosyncratic permanent and transitory shocks to firm performance dynamic panel regressions for firm performance and individual wages using difference GMM (Arellano-Bond, 1991) identify wage elasticities from residuals via IV application to German LIAB data set allow for age- and institutional heterogeneity in wage responses 4 / 30

The data set Linked Employer-Employee Data of the IAB (LIAB) longitudinal version 1993 2010 employer side: representative annual establishment panel 2-digit sectors, establishment size, structure of workforce annual sales, investment existence of works council, CBA coverage employee side: social insurance records average daily gross wage full time/part time (but no hours!) gender, age, education blue- or white-collar worker, occupation 5 / 30

Sample selection mining and manufacturing sector in West Germany nominal values deflated by sector-specific PPI male full-time workers aged 25 59 exclude workers with top coded wages workers and establishments with complete observations in three consecutive spells 6 / 30

Descriptive statistics: establishment characteristics establishment sample linked sample mean s.d. mean s.d. sales/worker 2.033 3.620 2.127 3.506 employment 461.309 2151.942 370.404 1203.177 capital-labor ratio 0.958 2.096 0.988 2.178 works council 0.581 0.565 sector CBA 0.613 0.603 firm CBA 0.082 0.069 tenure 11.644 4.896 age 41.888 3.784 white collar 0.198 vocational degree 0.766 high school 0.040 voc. + high school 0.025 applied university 0.022 university 0.010 establishments 1826 514 7 / 30

Descriptive statistics: worker characteristics worker sample linked sample mean s.d. mean s.d. sales/worker 3.016 2.943 3.232 3.611 employment 4148.707 5749.233 2955.104 4578.758 capital-labor ratio 1.247 1.662 1.200 1.520 works council 0.958 0.956 sector CBA 0.860 0.836 firm CBA 0.063 0.077 tenure 14.953 8.075 15.694 7.721 age 41.764 8.381 42.547 8.092 white collar 0.144 0.152 vocational degree 0.746 0.746 high school 0.004 0.003 voc. + high school 0.019 0.019 applied university 0.017 0.015 university 0.008 0.008 observations (ind.) 723255 (103497) 406994 (93359) 8 / 30

Part 1: Firm side regression model: y jt = ρy jt 1 + Z jtγ + ϕ j + ε jt Arrelano-Bond s (1991) dynamic panel GMM estimator: first differencing sweeps out FE: y jt = ρ y jt 1 + Z jtγ + ε jt cov( y jt 1, ε jt ) 0 IV approach necessary potential instruments for y jt 1 are y jt 2, y jt 3,... validity depends on autocorrelation structure of ε jt 9 / 30

Explanatory variables y jt log(sales per worker) in establishment j in year t Z jt includes log(capital-labor ratio) year dummies (7) sector dummies (15) regional dummies (10) use y jt 2, y jt 3, y jt 4 as instruments for y jt 1 10 / 30

Results sales regression unweighted weighted variable coef. s.e. coef. s.e. y jt 1 0.290 0.117 0.044 0.140 log(k jt /L jt ) 0.114 0.028 0.210 0.052 dummy set χ 2 -stat. p-val. χ 2 -stat. p-val. year 13.96 0.052 13.29 0.065 sector 37.45 0.001 107.52 0.000 regional 17.43 0.065 16.83 0.078 test statistic p-val. statistic p-val. Hansen J test 28.89 0.090 18.92 0.527 AR(2)-test 0.47 0.638 0.24 0.811 AR(3)-test 0.50 0.615 1.11 0.265 AR(4)-test 0.58 0.564 1.18 0.238 observations 6620 6620 s.e. clustered at the establishment level, italic: p < 0.10, bold: p < 0.05 11 / 30

Residual autocorrelation autocorrelation matrix of first differenced residuals ˆε jt Order (k) E( ε jt ε jt k ) s.e. 0 0.0655 0.0115 1 0.0217 0.0081 2 0.0011 0.0055 3 0.0074 0.0072 cov. structure consistent with error process: ε jt = ζ jt + v jt, ζ jt = ζ jt 1 + u jt, E(u 2 jt ) = σ2 u, E(v 2 jt ) = σ2 v, E(u jt u js ) = E(v jt v js ) = 0 for s t, E(u jt v js ) = 0 for all s, t. estimates: σ 2 u = 0.015, σ 2 v = 0.022 12 / 30

Part 2: Worker side from firm side: y jt = Z jt γ + ϕ j + ζ jt + v jt wage regression model: w ijt = λw ijt 1 + X ijtδ + αζ jt + βv jt + φ ij + ψ ijt where X ijt includes Z jt estimation as before: first differencing sweeps out FE: w ijt = λ w ijt 1 + X ijtδ + αu jt + β v jt + ψ ijt }{{} =: ω ijt don t know u jt and v jt treat them as errors, identify α and β in separate step 13 / 30

Explanatory variables w ijt log(daily gross wage) of worker i in establishment j in year t X ijt includes Z jt : log(capital-labor ratio), dummies for year, sector, region log(employment) dummy for works council dummies for sector CBA / firm CBA (ref.: no CBA) cubic polynomial in age, tenure dummy for white collar worker 5 education groups use w ijt 2, w ijt 3, w ijt 4 as instruments for w ijt 1 14 / 30

Results wage regression (exerpt) unweighted weighted variable coef. s.e. coef. s.e. w ijt 1 0.081 0.016 0.085 0.033 log(l jt ) 0.070 0.020 0.060 0.013 ãge 0.016 0.006 0.017 0.007 ãge 2 /100 0.026 0.003 0.026 0.003 ãge 3 /1000 0.005 0.001 0.004 0.002 white collar 0.044 0.006 0.049 0.008 test statistic p-val. statistic p-val. Hansen J test 25.90 0.169 22.00 0.340 AR(2)-test 1.50 0.134 1.05 0.293 AR(3)-test 0.23 0.820 1.77 0.077 AR(4)-test 3.53 0.000 1.49 0.135 observations 459387 459387 ãge = age 40, s.e. clustered at the establishment level, italic: p < 0.10, bold: p < 0.05 15 / 30

Part 3: Identification of wage responses ω ijt = αu jt + β v jt + ψ ijt ε jt = u jt + v jt Identification using IV: Similarly, projecting ε jt on ε jt+1 isolates transitory shock component, regressing this predictor on ω ijt identifies β. projecting ε jt on ε jt 1 + ε jt + ε jt+1 isolates permanent shock component, regressing this predictor on ω ijt identifies α. 16 / 30

Heterogeneous responses 1 baseline obtain estimates for wage responses to u jt and v jt as illustrated 2 age heterogeneity interact shocks u jt and v jt with polynomial in age: α 0 u jt + α 1 age ijt u jt + + β 0 v jt + β 1 age ijt v jt +... 3 age and institutional heterogeneity interact shocks u jt and v jt with polynomial in age and dummies for sector CBA, firm CBA, works council 17 / 30

Main messages significant response to transitory shocks only above age 57 wage flexibility w.r.t. permanent shocks is 0.061 until age 45, convex increasing thereafter (age 59: 0.15) sector CBA lowers response to permanent shocks the role of works councils: for prime-age workers, lower wage response to transitory shocks, but higher wage response to permanent shocks for older workers, higher wage elasticity w.r.t. both types of shocks consistent with informational frictions at the establishment level that works councils are able to resolve 18 / 30

Age heterogeneity: transitory shocks transitory coef. s.e. coef. s.e. ε jt 0.0110 0.0108 0.0102 0.0074 ε jt ãge 0.0016 0.0003 ε jt ãge 2 /100 0.0126 0.0072 ε jt ãge 3 /1000 0.0144 0.0037 test statistic p-val. statistic p-val. KP underid. test 11.80 0.019 19.88 0.098 KP F statistic 19.41 6.70 Hansen J test 3.33 0.343 9.19 0.687 observations 337772 337772 ãge = age 40, s.e. clustered at the establishment level, italic: p < 0.10, bold: p < 0.05 19 / 30

Age heterogeneity: permanent shocks permanent coef. s.e. coef. s.e. ε jt 0.0810 0.0140 0.0609 0.0122 ε jt ãge 0.0005 0.0013 ε jt ãge 2 /100 0.0146 0.0095 ε jt ãge 3 /1000 0.0043 0.0084 test statistic p-val. statistic p-val. KP underid. test 66.82 0.000 52.03 0.000 KP F statistic 43.01 6.95 Hansen J test 5.39 0.145 12.18 0.431 observations 325846 325846 ãge = age 40, s.e. clustered at the establishment level, italic: p < 0.10, bold: p < 0.05 20 / 30

elasticity Age heterogeneity: transitory and permanent shocks 0,20 0,15 0,10 0,05 0,00 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59-0,05-0,10 age red: permanent shock, blue: transitory shock 95% confidence bands based on robust standard errors clustered at the establishment level 21 / 30

Age and institutional heterogeneity: transitory shocks transitory coef. s.e. coef. s.e. ε jt no CBA 0.0132 0.0188 0.0407 0.0243 ε jt no CBA ãge 0.0029 0.0014 ε jt sector CBA 0.0337 0.0342 0.0410 0.0260 ε jt sector CBA ãge 0.0025 0.0016 ε jt firm CBA 0.0441 0.0238 0.0586 0.0255 ε jt firm CBA ãge 0.0057 0.0014 ε jt works council 0.0395 0.0329 0.0540 0.0256 ε jt works council ãge 0.0056 0.0014 test statistic p-val. statistic p-val. KP underid. test 20.37 0.086 0.106 KP F statistic 11.37 10.46 Hansen J test 14.49 0.271 29.10 0.217 observations 872778 392035 22 / 30

elasticity w.r.t. transitory shock Age and institutional heterogeneity: transitory shocks 0,15 0,10 0,05 0,00 25 30 35 40 45 50 55-0,05-0,10 age no CBA sector CBA works council 23 / 30

elasticity w.r.t. transitory shock Age and institutional heterogeneity: transitory shocks 0,15 0,10 0,05 0,00 25 30 35 40 45 50 55-0,05-0,10 age no CBA sector CBA works council 95% confidence bands based on robust standard errors clustered at the establishment level 24 / 30

Age and institutional heterogeneity: permanent shocks permanent coef. s.e. coef. s.e. ε jt no CBA 0.1325 0.0424 0.0723 0.0260 ε jt no CBA ãge 0.0042 0.0026 ε jt no CBA ãge 2 /100 0.0169 0.0235 ε jt sector CBA 0.0505 0.0380 0.0274 0.0210 ε jt sector CBA ãge 0.0011 0.0026 ε jt sector CBA ãge 2 /100 0.0279 0.0201 ε jt works council 0.0176 0.0407 0.0368 0.0220 ε jt works council ãge 0.0002 0.0027 ε jt works council ãge 2 /100 0.0212 0.0206 test statistic p-val. statistic p-val. KP underid. test 28.18 0.002 49.42 0.008 KP F statistic 10.12 88.82 Hansen J test 9.81 0.366 34.83 0.143 observations 302780 302780 25 / 30

elasticity w.r.t. permanent shock Age and institutional heterogeneity: permanent shocks 0,20 0,15 0,10 0,05 0,00 25 30 35 40 45 50 55-0,05-0,10 age no CBA sector CBA works council 95% confidence bands based on robust standard errors clustered at the establishment level 26 / 30

Recap: the differential effects of works councils transitory shocks decrease wage flexibility of young and prime-age workers significantly increase wage flexibility of older workers permanent shocks (weakly) significantly increase wage flexibility for young and prime-age workers significantly increase wage flexibility of older workers effect on older workers seems stronger in firms not covered by a CBA [but low sample size] 27 / 30

A theoretical explanation works councils may help resolve informational asymmetries between management and workforce (Freeman and Lazear, 1995) information rights mandated by law (BetrVG) far-reaching co-determination and consent rights put works councils in a strong position works councils likely to receive more credible information about economic conditions and the expected persistence of shocks 28 / 30

Indications for the informational channel for prime-age workers, dampened response to transitory shocks but (weakly significant) larger response to permanent shocks for older workers, higher wage responses to both types of shocks low expected remaining employment horizon low match surplus surplus sensible to productivity shocks, especially permanent ones knowledge of size and structure of shocks crucial to avoid layoff higher wage responsiveness for elderly mainly due to firms without CBA the more decentralized bargaining, the more important is firm-specific information in wage setting 29 / 30

Next steps robustness checks extend sample to East Germany, other sectors unweighted regression, evtl. drop very small firms clarify role of institutions vs. firm size extend analysis to separations: adjustment to shocks takes place along two margins wages and employment expectation: particularly tight link between wage and layoff risk in firms without works council, especially for older workers appropriate estimation strategy? 30 / 30

Part II Appendix

daily log wage Cross sectional age-wage profile Average over 2000 2008 4,85 4,80 4,75 4,70 4,65 4,60 4,55 4,50 4,45 4,40 <25 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 age group 32 / 30

annual increase in daily log wage Annual increase in log wage of stayers Average over 2000 2008 0,10 0,08 0,06 0,04 0,02 0,00-0,02 mean median p25 p75-0,04 <25 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 age groups back 33 / 30