Wages and the Value of Nonemployment
Social Science Research Network
Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen
... von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Abstract Nonemployment is often posited as a worker's outside option in wage setting models such as bargaining and wage posting. The value of this state is therefore a fundamental determinant of wages and, in turn, labor supply and job creation. We measure the effect of changes in the value of nonemployment on wages in existing jobs and among job switchers. Our quasi-experimental variation in nonemployment values arises from four large reforms of unemployment insurance (UI) benefit levels in Austria. We document that wages are insensitive to UI benefit levels: point estimates imply a wage response of less than $0.01 per $1.00 UI benefit increase, and we can reject sensitivities larger than 0.03. In contrast, a calibrated Nash bargaining model predicts a sensitivity of 0.39 -more than ten times larger. The empirical insensitivity holds even among workers with a priori low bargaining power, with low labor force attachment, with high predicted unemployment duration, among job switchers and recently unemployed workers, in areas of high unemployment, in firms with flexible pay policies, and when considering firmlevel bargaining. The insensitivity of wages to the nonemployment value we document presents a puzzle to widely used wage setting protocols, and implies that nonemployment may not constitute workers' relevant threat point. Our evidence supports wage-setting mechanisms that insulate wages from the value of nonemployment. JEL-Codes : J310, J600, J650. examine the wage pressure channel of UI. 4 Burdett and Mortensen (1998) and Manning (2011) are references for wage posting models. 5 Shimer (2005), Hall and Milgrom (2008), Chodorow-Reich and Karabarbounis (2016) and Hall (2017) discuss the role of the sensitivity of the Nash wage to outside options for employment fluctuations in matching models. 1 that wages increase by $0.39 whenever UIBs increase by $1.00. This effect comes from two margins: first, while the worker is unemployed, she mechanically receives a higher UIB payment. Second, her re-employment wages also respond to the outside option boost, generating a built-in feedback effect. This logic and thus the high theoretical sensitivity hold across a broad set of model refinements of bargaining models, and are robust to incorporating various features such as finite benefit duration, to micro re-optimization such as job search choices (due to the envelope theorem) and even to market-level adjustment (netted out by a control group in the same market). 6 We document that empirical are insensitive to increases in workers' UI benefit levels. Point estimates for wage-benefit sensitivity are less than 0.01 after one and two years. Our confidence intervals reject that a $1.00 increase in UI benefits increases wages by more $0.03 after two years. The insensitivity holds across various subsets of workers, even those with higher unemployment risk and new hires. These estimates are an order of magnitude smaller than the calibrated model predicts and thus present a puzzle to the Nash bargaining model with nonemployment as the outside option, unless one is willing to believe that workers have close to full bargaining power. Such an interpretation is however empirically rejected by existing small estimates of firm-level rent sharing (inside value shifts from productivity), which we argue imply worker bargaining power of around 0.1 and that we use to calibrate our bargaining model. 7 We juxtapose our implied bargaining power estimates with these values in Figure 1 . Our findings imply that nonemployment is not the relevant outside option in wage bargaining, providing causally identified micro-empirical support for models that insulate wage setting from the nonemployment value or perhaps outside options more generally. Our empirical setting is a unique set of four reforms that generated large, quasi-experimental variation in UI benefit levels in Austria, in 1976Austria, in , 1985Austria, in , 1989Austria, in and 2001 Reforms to UI benefit levels are rarer than potential benefit duration reforms, whose effects on the nonemployment value would be harder to theoretically price. The reforms raised benefits differentially for workers based on their previous salaries (the UI reference wage that determines benefits in the event of a UI claim), for example by as much as 28% in 1985. Our difference-in-differences design compares wage growth between those treated workers and their unaffected peers (control group). We use administrative data on workers and firms covering 1972 through today, with daily information on UI claims and benefit receipts, as well as wage earnings and other labor force statuses. Additional features of the Austrian UI system permit particularly clean variation in nonemployment values. First, Austrian UI affects the nonemployment value for most workers. Conditional on a separation, most workers receive UI due to broad eligibility and high take-up, due to 6 We also show that the model makes the same predictions for the sensitivity if treatment and control workers occupy segmented markets. 7 Manning (2011) and Card et al. (2018) review rent sharing estimates. In Appendix D, we translate the reduced-form elasticities into an upper bound for worker bargaining power. 8 Only the 1989 reform has been studied, with a focus on unemployment spell duration (Lalive et al., 2006). 2 mandatory registration and because of long unemployment spells. Second, Austrian workers who quit are eligible for benefits -crucial for UI to indeed shift workers' threat points. 9 Third, Austrian UI does not feature experience rating. Fourth, even after UI benefit exhaustion, a significant share of Austrian workers receives means-tested benefits that, while lower in levels, move nearly one-to-one with a worker's previous UIB levels and thus our variation. We first visually analyze each reform nonparametrically by plotting raw data. We sort workers by their UI reference wage, and then plot their wage growth before and after each reform, along with the between-worker reform-induced variation in UI benefit levels. The raw data do not reveal any wage responses even after two years -in stark contrast to the large predicted wage effects from our calibrated bargaining model that we also plot. Next, we estimate the wage-benefit sensitivity in a regression-based difference-in-differences analysis. The point estimates are below 0.01, with confidence intervals that rule out even a $0.03 wage increase to a $1.00 benefit increase. We also formally test our identification assumptions with placebo tests and can include rich controls. By including firm-by-year effects, for instance, we leverage variation in nonemployment value shifts between workers within the same firm. We also test a central cross-sectional prediction of the model: UI benefits raise outside options, and thus wages, by more for workers whose separation will entail longer unemployment spells. Yet, when we split up workers by the predicted time on UI post-separation, even workers in the top quintile of UI risk -whose model-predicted sensitivity would have reached nearly 0.6 -exhibit the zero wage effect. Relatedly, we also find find little evidence for larger sensitivity among workers with plausibly lower bargaining power, e.g., blue-collar or female workers, for whom a given outside option shift should have triggered larger wage effects.