Policy Research Working Paper 10325 Take-Up and Labor Supply Responses to Disability Insurance Earnings Limits Judit Krekó Daniel Prinz Andrea Weber Health, Nutrition and Population Global Practice February 2023 Policy Research Working Paper 10325 Abstract In most disability insurance programs, beneficiaries lose a reduction in the earnings limit in Hungary to examine some or all of their benefits if they earn above an earnings screening and labor supply responses. The findings show threshold. While intended to screen out applicants with that the policy changed selection into the program modestly, high remaining working capacity, earnings limits can also but it reduced labor supply significantly. Viewed through distort the labor supply of beneficiaries. This paper devel- the lens of the model, these findings suggest that the earn- ops a simple framework to evaluate this trade-off. It uses ings threshold should be higher. This paper is a product of the Health, Nutrition and Population Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at dprinz@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Take-Up and Labor Supply Responses to Disability Insurance Earnings Limits∗ Judit Krek´o aniel Prinz D´ Andrea Weber Keywords: disability insurance, policy reform, earnings limit, labor supply JEL codes: H53; H55; I38; J22 ∗ Krek´o: Budapest Institute for Policy Analysis and Centre for Economic and Regional Studies (ju- dit.kreko@budapestinstitute.eu). Prinz: World Bank (dprinz@worldbank.org). Weber: Central European University, CEPR and IZA (webera@ceu.edu). We thank Anik´ ır´ o B´ o, Jonathan Cribb, Marianna Endr´ esz, Sandra Gain, Rafael Lalive, R´obert Lieli, Arieda Mu¸ co, Andr´ ´ as Simonovits, David Sturrock, Almos Telegdy, Istv´ an J´anos T´ ´ am Zawadowski, and seminar participants at Central oth, Ross Warwick, Roula Yazigi, Ad´ European University, the Hungarian Economic Association Annual Conference, the Centre for Economic and Regional Studies, Harvard University, and iHEA for helpful comments. Krek´ o was supported by the “Lend¨ ulet” program of the Hungarian Academy of Sciences (grant number: LP2018-2/2018).The data used in this paper is under the ownership of the Central Administration of National Pension Insurance, the Na- tional Health Insurance Fund Administration, the Educational Authority, the National Tax and Customs Administration, the National Labor Office, and the Pension Payment Directorate of Hungary. The data was processed by the Institute of Economics, Centre for Economic and Regional Studies. We thank the Data- bank of the Centre for Economic and Regional Studies for providing access and help with the administrative database. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. The share of working-age adults receiving long-term disability insurance (DI) benefits has increased rapidly over the last few decades and DI programs account for over 10% of social spending in OECD countries (OECD, 2010). The trend in disability rolls raises concerns about the fiscal sustainability of DI programs and has prompted policy makers to examine program designs that encourage potential beneficiaries to remain employed and those already receiving benefits to return to work (Autor and Duggan, 2010; Autor, 2011; Burkhauser and Daly, 2011; Liebman, 2015). One way that policy makers try to limit DI take-up and incentivize work is setting earnings limits: if a beneficiary earns above a certain level, she loses part or all of her benefits. The rationale behind earnings limits is the presence of asymmetric information: the government cannot observe applicants’ true health status or work capacity, so it must rely on a screening mechanism.1 The screening mechanism is supposed to ensure that only workers who are unable to earn above a certain level will apply for benefits, while potential applicants with higher working capacity will find it advantageous to forego benefits and remain employed instead. In the United States, the earnings limit applicable to beneficiaries in the Social Security Disability Insurance (SSDI) and Supplemental Security Income (SSI) programs is called Substantial Gainful Activity (SGA). It is designed as a “cash-cliff”, which means that if a beneficiary earns even $1 above the SGA, she loses all benefits. The SGA thus creates a notch in the benefit schedule such that a higher total income (wages plus benefits) can be obtained by working less and keeping earnings below the SGA (Maestas and Yin, 2008; Weathers and Hemmeter, 2011). In 2023, the SGA is $1,470 per month for non-blind applicants. Someone working full-time at the federal minimum wage would make approximately $1,260. In the 30 states with minimum wages set above the federal level, a full-time minimum-wage worker would make considerably more than the SGA. The benefit design based on a cash-cliff essentially assumes that if an applicant can earn more than the SGA limit in the labor market, they do not need to receive any DI benefits. Policy makers and researchers have recognized that earnings limits like the SGA create strong work disincentives and have potentially negative welfare impacts. Alternative policy approaches adopted in other countries avoid a notch in the benefit schedule by introducing a gradual phaseout of benefits above an earnings threshold. But even under these policy designs, the implicit tax rate may still inefficiently distort labor supply (Kostøl and Mogstad, 2014; Deuchert and Eugster, 2019; Ruh and Staubli, 2019; Zaresani and Olivo-Villabrille, 2022). 1 Other screening mechanisms include medical screening (de Jong, Lindeboom and van der Klaauw, 2011; Liebert, 2019; Godard, Koning and Lindeboom, 2022), benefit amounts (Mullen and Staubli, 2016), wait times (Kearney, Price and Wilson, 2021), and hassle costs (Deshpande and Li, 2019). 1 The usefulness of earnings limits as a screening mechanism and their distortionary effect on labor supply decisions create a trade-off for policy makers. When setting earnings limits, they need to take into account the impact on who takes up benefits and how much bene- ficiaries work. In this paper, we develop a framework to understand this trade-off in the context of a “cash cliff” design where beneficiaries lose all benefits if they earn above the SGA. When the government increases the earnings limit, the set of workers who apply for benefits widens: receiving benefits while working becomes appealing to higher-productivity workers. On the flipside, a decrease in the earnings limit means that the set of workers applying for benefits shrinks as only less-productive workers will prefer benefit receipt and limited work. At the same time, another effect is at play: conditional on receiving benefits, a higher earnings limit means that beneficiaries who can work will increase their labor supply as they still remain eligible for benefits. And a lower earnings limit has the opposite effect: conditional on receiving benefits, beneficiaries will set their labor supply lower in order to remain eligible for benefits. We call these two effects of changing the earnings limit the se- lection and labor supply effects. At the optimal earnings limit, the marginal selection effect and the marginal labor supply effect of moving the earnings limit will balance each other out. Therefore, to characterize the welfare impact of changing the earnings limit, these two effects should be estimated. To empirically estimate the selection and labor supply effects of changing the earnings limit, we study a policy reform in Hungary that reduced the earnings limit for some ben- eficiaries but not others, while leaving benefit amounts unchanged. In 2008, the cash-cliff style earnings limit in Hungary’s Regular Social Assistance (RSA) program for moderately disabled individuals was reduced from 80% of the individual’s last wage before entering disability to 80% of the monthly minimum wage for new entrants (a level similar to that observed in most states of the U.S.), while it remained the same for beneficiaries who were al- ready approved. We exploit this policy change to understand how selection into the program and labor supply once in the program changed. To this end, we compare the evolution of various extensive and intensive margin measures of labor supply relative to the start of ben- efit receipt among beneficiaries who enter before (“old entrants”) and after (“new entrants”) the reform. We find that the decrease in the earnings limit had a small impact on selection into the program. First, we do not find evidence of decreased program entry rates. Second, consistent with the screening mechanism, we show that individuals who entered the program after the reform had worse pre-entry labor market outcomes than beneficiaries who entered earlier. New entrants were 3 percentage points (4%) less likely to work and earned 8% less on average (conditional on working) pre-entry than old entrants. Old and new entrants were similar 2 on a variety of other dimensions, such as age, occupation, geographical location, and sick leave use prior to entering disability. Examining benefit persistence, we find no evidence that new entrants were more likely to exit the program than old entrants. Overall, the moderate selection effects are consistent with a world where the earnings limit and the benefit level were already sufficiently low to deter most potential entrants who were well-positioned to find higher-paying jobs in the labor market. At the same time, we find that individuals who entered the program after the earnings limit was reduced had meaningfully lower labor supply post-entry. New entrants were as likely to be employed as old entrants, but conditional on being employed, they worked less. On average, new entrants worked 7% fewer hours, and had 18% lower earnings (conditional on working) after taking up benefits. This result is driven by beneficiaries with higher pre-disability earnings, who were most affected by the change in the earnings limit. To examine the impact of the lowered earnings limit on beneficiary health, we consider mortality, an imperfect proxy for health. Our results suggest no change in mortality, which means that the primary effect of the reduction of the earnings limit on beneficiaries was through reduced work. Since we study a change in the DI earnings limit in 2008, it is important to rule out the role of the recession in explaining our results. We address this concern in several ways. We start by showing that the overall labor market impacts of the recession were not really present in Hungary until 2009 when the unemployment rate started rising rapidly. This means that inflow into RSA should not yet be affected by the economic downturn. But labor market outcomes in the years after entry might be differentially affected for old and new entrants. Our first strategy to confront this concern involves comparisons across regions that are more and less severely hit by the recession and showing outcomes relative to their national or regional average. Second, we perform placebo analyses based on entrants around non-reform cutoff dates. Third, we compare outcomes of old and new entrants into the accident insurance program, which did not see a change in the earnings limit. This set of robustness checks confirms that the change in labor market outcomes of new entrants is due to the change in the earnings limit rather than the change in the economic environment. Our work contributes to three strands of the literature. We most directly contribute to the literature on earnings limits in disability insurance (e.g., Maestas and Yin, 2008; Schimmel, Stapleton and Song, 2011; Weathers and Hemmeter, 2011; Kostøl and Mogstad, 2014; Greenberg et al., 2018; Deuchert and Eugster, 2019; Ruh and Staubli, 2019; Zaresani, 2020). This literature finds mixed evidence on labor supply responses to earnings limits. Some papers find that DI beneficiaries are responsive to the financial incentives induced by earnings limits. Others suggest little response to easing the earnings limit. We contribute 3 to this literature in several ways. First, unlike much prior work focusing on existing DI beneficiaries, we are able to study both who takes up DI benefits (the selection effect) and how beneficiaries behave when they start receiving benefits (the labor supply effect). Second, we focus on moderately disabled individuals who have relatively high employment rates after entering the program and should thus be responsive to the change in the earnings limit. More severely disabled and longer-term beneficiaries considered in some previous work (e.g., Greenberg et al., 2018) are unable to significantly change their labor supply. Third, to the best of our knowledge this is the first paper to study a decrease in the earnings limit, which is important as the responses to positive and negative changes in the earnings limit are not necessarily symmetric. Fourth, we develop a simple model that clarifies the role of the two key effects of changing the earnings limit: its impact on selection and on labor supply. More broadly, this work contributes to the literature on the work disincentives of DI programs and the literature that studies the labor supply impacts of DI receipt (e.g., Bound, 1989; Gruber, 2000; Chen and van der Klaauw, 2008; Maestas, Mullen and Strand, 2013; Low and Pistaferri, 2015; Mullen and Staubli, 2016). This literature has focused on understanding the effects of disability programs on labor supply taking into account all features of the programs as implemented. It broadly finds that DI receipt discourages work. For example, Maestas, Mullen and Strand (2013) find that for applicants on the margin of program entry, employment would be on average a third higher if they did not receive benefits. Earnings limits and other features, such as benefit generosity and selection criteria jointly determine the effects of DI programs. Our contribution to this literature is examining one feature of disability programs that policy makers can use to influence the incentive effects of DI programs. Finally, this work also speaks to the academic and policy literature that has tried to address the fiscal sustainability of DI programs, partly by suggesting that work disincentives in these programs should be decreased (e.g., Autor and Duggan, 2006, 2007; Autor, 2011; Bipartisan Policy Center, 2015; Liebman, 2015). For example, several policy proposals in the United States included moving from a “cash cliff” to a gradual phase-out. This study can shed further light on the welfare effects of changing earnings limits among moderately disabled workers. 1 Conceptual Framework In this section, we propose a simple conceptual framework to capture the key trade-offs related to the setting of earnings limits in disability insurance programs. We focus on the 4 case of notches, earnings limits above which beneficiaries lose their benefits completely.2 Individuals are characterized by their productivity types θ ∈ [0, 1]. Types are distributed according to CDF F (θ) (PDF f (θ)). Individuals work h hours and have after-tax income y = hθ − τ (hθ) where τ is the income tax rate. Type θ is unobserved and therefore the government cannot use it to condition benefits. Disability benefits are B and there is an income threshold y ¯ for receiving benefits. Individuals have utility NB VθN B = u yθ − v (hN B θ ) (1) if not receiving benefits, and utility B VθB = u yθ + B − v (hB θ ) (2) if receiving benefits where v (h) is the disutility of work. Individuals decide to participate in the disability program if the value of participation is higher than the value of non- participation: VθB ≥ VθN B . This decision rule determines in turn a cutoff type θ ¯, who is indifferent between participating and not participating. Here we consider the program entry decision in a static framework and disregard any dynamics of repeated entry and exit. In our empirical analysis we show that program participation is highly persistent and even the reform-driven change in the earnings threshold did not lead to increased exits. Social welfare is ¯ θ 1 B + B − v (hB NB − v (hN B ¯ W = u yθ θ )dθ + u yθ θ )dθ − θB (3) 0 ¯ θ Program Cost Receiving Benefit Not Receiving Benefit ¯ θ 1 + GB θ dθ + GN B θ dθ (4) 0 ¯ θ Fiscal Externality, Fiscal Externality, Benefit No Benefit where θ¯ is the highest productivity type worker who receives the benefit. The standard fiscal externality is the tax revenue the government realizes on a type θ worker: GB B θ = τ h θ and GNθ B = τ hN B θ . However, other types of fiscal externalities can also fit into this framework. For example, time out of the labor force while receiving benefits may reduce working capacity (Autor, Maestas, Mullen and Strand, 2015; Garcia-Mandic´ o, Garc´ıa-G´omez, Gielen and 2 Our notation follows the framework in Finkelstein and Notowidigdo (2019) who study the take-up of welfare programs in the presence of potential behavioral biases. See also Nichols and Zeckhauser (1982), Kleven and Kopczuk (2011), and Anders and Rafkin (2021) for more general models of welfare eligibility and take-up. 5 O’Donnell, 2020; B´ o et al., 2022) which imposes an additional negative fiscal externality. ır´ The government can vary the earnings limit y ¯ leaving B fixed. This has an impact on what types of workers choose to receive benefits, as the cutoff type θ ¯ changes. This in turn impacts total program costs (how many workers receive benefits) and the fiscal externality of the program. If the government lowers (increases) y¯ the set of workers opting for benefits shrinks (widens). The labor supply of some workers receiving benefits also decreases because some workers will lower their labor supply in order to remain eligible for benefits. At the same time, the labor supply of some workers not receiving benefits may increase as they can work more once they do not need to meet the earnings limit anymore. In particular, a marginal change in y¯ has the following effect on social welfare: ¯ ¯ θ 1 dW dθ dGBθ dGN θ B = GB NB ¯ − Gθ ¯ −B + dθ + dθ . (5) dy ¯ ¯ θ dy 0 ¯ dy ¯ θ dy ¯ Change in Selection Change in the Fiscal Externality The welfare impact consists of two parts. First, the change in selection into benefit take up ¯ dθ (d y ¯ ) has an impact through the program cost (B ) and the fiscal externality (GB ¯ − Gθ θ NB ¯ ). Second, among beneficiaries (types 0 to θ ¯ to ¯) and potentially non-beneficiaries (types θ dGB dG NB 1), the fiscal externality can change too ( dy θ ¯ and dθ y ¯ respectively). In the standard case, the fiscal externality is the tax revenue the government realizes and it changes because beneficiaries may adjust their labor supply to remain eligible for benefits. Note that assuming that individuals were already optimizing, applying the envelope theorem, there is no welfare effect through individuals’ utilities. Welfare is maximized with respect to the earnings limit if the derivative dW dy ¯ is zero. This is realized when the selection and labor supply responses balance each other out. 2 Background Preceding the 2008 reform, Hungary had the highest disability benefit receipt rate in the OECD at 12%, over twice the OECD average (OECD, 2009). Unlike the U.S. system, but similar to other European countries, Hungary’s disability insurance programs are tiered based on the severity of impairment beneficiaries have. The Regular Social Allowance (RSA) program, the focus of this paper, was available to individuals with sufficient work histories and with an at least 40% health impairment who could not work in their pre-disability job or any other job commensurate with their level of education without rehabilitation or fur- ther education. Health impairments are assessed by a panel of physicians and rehabilitation experts. The most common qualifying diagnoses for RSA recipients were musculoskeletal 6 diseases. Different programs were available to more severely disabled individuals (disability pension), as well as those close to the retirement age when becoming disabled (temporary allowance), those who became disabled before age 25, and blind individuals (disability al- lowance). The benefit level of the RSA was low compared to the disability pension: 36% to 38% of the statutory minimum wage throughout the years of our analysis. RSA recipients are allowed to work up to an earnings limit and at the time of the reform about 26% did work. As a comparison, only 12% of beneficiaries with more severe disabilities (disability pensioners) were employed in 2007. Until December 2007, the earnings limit was linked to the previous earnings of the applicants. A person with an at least 40% of health impairment was allowed to apply for RSA if her average earnings over four consecutive months did not exceed 80% of her pre-disability earnings.3 The same rule applied to benefit continuation: beneficiaries whose average earnings over four consecutive months exceeded 80% of their pre-disability earnings were removed from the program. Starting in January 2008, the earnings limit was lowered: irrespective of prior earnings, new entrants were only allowed to earn up to 80% of the monthly minimum wage while receiving benefits. This effectively meant that new entrants could only work part-time. The decision about the new earnings threshold was made at the end of 2007. The first internal proposal was written in November 2007 and passed on December 23, 2007, becoming effective on January 1, 2008. Hence the legislation was unexpected, making anticipatory effects unlikely. The earnings limit remained unchanged for those already approved for benefits. To understand the bite of the reform, it is useful to consider the distribution of earnings among RSA beneficiaries prior to taking up benefits. Their average pre-disability earnings were 126% of the monthly minimum wage. 60% earned more than the monthly minimum wage and among those who earned more, the average pre-disability earnings were 169% of the monthly minimum wage. This suggests that the policy change affected a substantial share of potential beneficiaries. Because the earnings limit was 80% of pre-disability earnings before the reform, the change in the earnings limit varied across beneficiaries. This created additional variation across beneficiaries in the bite of the policy: those with higher potential earnings were more impacted by the reform than those with pre-disability earnings close to the monthly minimum wage. Due to declining benefit generosity and increased stringency of health requirements (Du- man and Scharle, 2011), the inflow into all types disability programs in Hungary had been 3 The calculation of pre-disability earnings is complex. It takes into account earnings during several years before applying for benefits. Previous earnings are adjusted for economy-wide changes in average earnings. Because of this complex calculation and the four-month rule, earnings can exceed the earnings limit in some months without removal from the program. 7 continuously declining since the early 2000s. This has also been the case for the RSA pro- gram where the monthly inflow declined by a factor of five between 2003 and 2007. This downward trend came to an end in 2008 when the inflow stabilized, see Appendix Figure A1. 3 Data and Empirical Framework We use administrative panel data that brings together information on earnings, occupations, benefit receipt, health care spending, and other domains for half of the Hungarian population over years 2003–2017 (Seb˝ ok, 2019). The data is based on a random 50% sample (for privacy reasons) of the population aged 5–74 in 2003 who are followed until 2017. Since our focus is on the working age population, we restrict the sample to individuals aged 20-60 in 2007. In addition to employment status, wages, and working hours, we can observe disabil- ity benefit take-up (regular social assistance, disability pension, other types of disability benefits), unemployment insurance, and other social program (e.g., maternity leave) partic- ipation. We use monthly data, which allows us to precisely identify the timing of benefit take-up. The annual microregion level unemployment data are from the T-STAR database. To study how the reform impacted the selection of beneficiaries into the RSA program and their labor supply conditional on participation, we compare beneficiaries who enter in 2007, the year before the reform (“old entrants”) and beneficiaries who enter in 2008, the year after the reform (“new entrants”).4 We follow these two groups of beneficiaries for 4 years (48 months) before and 3 years (36 months) after they enter disability insurance. We start our empirical analysis by comparing selection into the RSA program between old and new entrants. In particular, we compare program inflow, observed characteristics of entrants and their labor market outcomes in the years before entry. This should give us a good sense of the overall selection effect due to the reform. Next, we compare labor market outcomes of old and new entrants after disability entry. To interpret these differences as labor supply effects of the change in the earnings limit, we have to control for the selection effects, which we do in regression and reweighting analyses. A general concern with our identification strategy is the role of aggregate labor market trends and, in particular, the onset of the Great Recession in 2008/2009. We apply the following strategies to confront this concern. First, we note that the main impact of the recession on the Hungarian labor market occurred in 2009. This is reflected in the unem- ployment rate, shown in Figure A2, which started its dramatic increase only at the end of 4 Program entry is defined based on the original date of application for RSA. Benefits are dated back to the application date. 8 2008. The inflow into RSA in the treatment and control groups should thus not be affected by the recession.5 But we are still worried that labor market outcomes after entry into RSA are affected by the economic downturn, which would imply different time patterns for old and new entrants. Our second strategy exploits large regional variation in the increase of the unemployment rate during the recession. This variation allows us to test whether the magnitude of the economic shock is related to inflow or selection into RSA, or to labor market outcomes after RSA entry. Finally, we also examine outcomes of entrants into Work Accident Allowance, an alternative disability program which was not subject to a change in the earnings limit. 4 Results Selection and Benefit Take-Up Following our conceptual framework in Section 1, we start by analyzing how selection into regular social assistance (RSA) receipt changes with the reform. The framework predicts two sources of selection effects due to the change in the earnings limit. First, the lower cutoff type θ¯ should lead to a drop in program take-up. Second, due to the change in the average type who takes up benefits, the composition of observable characteristics of beneficiaries might change. Appendix Figure A1 plotting the monthly inflow into the RSA program does not provide evidence of a drop in program entry after the reform date in January 2008. In fact, the figure shows that in 2008 program inflow stabilizes after a long period of decline. Next, we compare observed characteristics of “old entrants” (beneficiaries who enter RSA during 2007, the year before the reform) and “new entrants” (beneficiaries who enter RSA during 2008, the year after the reform) in Table 1. We focus on differences in entrant characteristics three years before benefit take-up, because earnings decline rapidly in the year before entry due to deteriorating health and some restrictions embedded in the design of eligibility criteria further reduce labor supply immediately prior to program entry. It appears that the new and old entrants are similar along many dimensions, including age, occupation, and geographic location. But three characteristics show statistically significant differences: gender, employment and average wage. New entrants are 3 percentage points more likely to be male, 2 percentage points (3%) less likely to have been working three years prior to entry and earned 8% less on average (conditional on working). In particular, differences in labor supply prior to program entry deserve some attention to understand selection into the program. 5 Prior work has suggested that recessions push more individuals into DI (Autor and Duggan, 2003; Maestas, Mullen and Strand, 2015). We find little evidence of such an effect. 9 Figure 1 provides more detail on the evolution of labor market outcomes by year of entry, including share working and conditional on working, hours worked, earnings relative to the monthly minimum wage, and share with earnings above 80% of the monthly minimum wage from four years before benefit take-up to three years after. Looking at the period before entering RSA, the results suggest that old entrants were slightly more attached to the labor market. New entrants were 3 percentage points (4%) less likely to work and conditional on working earned 8% less on average pre-entry than old entrants. Column (5) of Table 2 shows mean differences in outcomes between old and new entrants from year four to year one before RSA entry. We leave out the last year before entry, because of the sharp decline in work during that year in order to be eligible for benefits. As a proxy for health status, we show in Appendix Figure A3 that there is no difference between old and new entrants in sick leave use before taking up disability benefits. Selection could also be driven by benefit persistence. Especially, if there is uncertainty about changes in eligibility rules after the reform, new entrants might be more likely to leave the program after they learn about the restrictions from the new earnings limit. Panel (a) of Figure 2 suggests that program participation is as persistent for new entrants as for old entrants, especially in the first two years after entry. Approximately 94% of initial beneficiaries still receive RSA two years after program entry in both groups followed by a slight divergence in the third year. This result suggests that the lower earnings threshold did not lead to significantly increased program exits. Aggregate data from the Yearly Reports of National Rehabilitation and Social Office show that the ratio of accepted/rejected applications for all DI programs was similar in 2007 and 2008 (around 55%), suggesting that the stringency of the assessment process remained unchanged. These findings suggest that the earnings limit had a moderate impact on selection and benefit take-up. Next, we turn to labor supply responses conditional on program entry. Labor Supply After Program Entry Our conceptual framework in Section 1 predicts that some DI benefit recipients restrict their labor supply to remain eligible for benefits. Figure 1 and column (6) of Table 2 show how labor market outcomes of beneficiaries change with the reform. We focus on average outcomes in years 2 and 3 after RSA entry and report mean differences between old and new entrants in Table 2. Panel (a) of Figure 1 shows that less than 20% of entrants work right after entering RSA. Subsequently, the employment rate increases more quickly for old entrants to about 30% during the first year and it equalizes again by the third year of benefit receipt. Differences in intensive margin labor supply are more substantial. Panel (b) shows that, conditional 10 on working, new entrants supply on average 29 hours per week during the second and third years of benefit receipts, 7% less than old entrants. New entrants earn on average 78% of the monthly minimum wage, 18% less than old entrants (Panel (c)) and there is also a significant gap between beneficiaries with earnings above 80% of the monthly minimum wage (Panel (d)). Because the earnings limit applied to the average of four consecutive months’ earnings, some new entrants can still earn above this limit in some months. Appendix Figure A4 provides further evidence that beneficiaries are indeed responding to the new earnings limit at 80% of the minimum wage by setting their earnings exactly at the threshold. While the distribution of monthly wages among old entrants is smooth through the threshold (Panel (a)), there is visible bunching among new entrants, as 5% of them earn within HUF 5,000 ($15) of the earnings limit (Panel (b)). Heterogeneity by Pre-Disability Earnings In Figure 3 we examine heterogeneity by reform exposure, comparing beneficiaries for whom the decrease in the earnings limit was likely binding and those for whom it was likely not binding, because their earnings were too low to be affected by the new limit. Panel (a) shows earnings relative to the minimum wage for RSA beneficiaries who earned below the minimum wage three years before taking up RSA benefits. Among this lower-earning group we find that the small pre-RSA-entry earnings gap of 6% between old and new entrants persists post-entry at about 9%. Panel (b) shows the same comparison for beneficiaries who earned above the minimum wage three years before taking up RSA benefits. For this group, there is a sharp increase in the earnings gap between old and new entrants from 4% pre entry to 21% after taking up benefits. This confirms our prediction that workers with higher earnings potential reduce their labor supply in order to remain eligible for the disability benefit. Mortality Panel (b) of Figure 2 examines the mortality of beneficiaries. It shows that over a three-year horizon after program entry, old and new beneficiaries have the same cumulative mortality (2%). While this is an imperfect measure of health, the result suggests that the change in the earnings limit primarily impacted beneficiaries through changes in labor supply rather than through worsening health. Over a three-year time horizon, lower income does not worsen beneficiary health sufficiently to result in higher mortality rates. Controlling for Selection on Observables Our conceptual framework predicts that lowering the earnings threshold has an impact on who selects into taking up benefits. In particular, we expect a lower earnings limit to result in increased selection toward lower- productivity or less-employable individuals. Our empirical findings above suggest that the 11 selection effect is relatively small. To further confirm that selection on observed characteris- tics is not driving the gaps in post entry labor supply, we apply propensity score reweighting in Figure A5. Specifically, we estimate the probability of RSA entry in 2008 versus 2007 in a logit model controlling for age, gender, work status and wage relative to the minimum wage 12, 24, and 36 months before taking up benefits and construct inverse probability weights based on the predicted propensity score. The patterns in reweighted labor supply measures after RSA entry shown in Figure A5 are virtually identical to those in Figure 1. This sug- gests that observable differences between old and new entrants prior to taking up benefits (i.e. the change in selection) do not explain the decrease in labor supply after taking up benefits. Accounting for the Recession The national unemployment rate in Hungary rose dra- matically with the start of the recession in 2009. We exploit strong regional variation in this increase to test for responses in RSA entry and labor supply of entrants to macroeconomic conditions. Appendix Figure A6 plots changes the unemployment rate at the beginning of the recession relative to changes RSA inflow rates across microregions. The absence of a clear relationship between the two variables indicates that the severity of the recession did not lead to a change in RSA inflow. To absorb macroeconomic fluctuations in outcomes, we compare labor market outcomes of new and old entrants relative to their national or microregion counterparts. Figure A7 shows labor market outcomes before and after RSA entry relative to their local/national average in the given month. The dynamics of relative labor market outcomes in this figure are similar to the absolute outcomes in Figure 1. Next, we present results for two different subgroups: RSA entrants living in microregions with low (below-median) versus high (above-median) increase in the unemployment rate. Appendix Table A1 shows that reform responses are very similar for areas more and less impacted by the recession. Together this evidence indicates the recession cannot explain our results. Placebo Analyses As a further check for the influence of the macroeconomic environment on RSA beneficiaries’ labor supply, we define placebo reform dates in non-reform years (January 1, 2006/2007/2009) and “placebo old entrants” and “placebo new entrants” taking up RSA in the years around the placebo reform dates. Panels (a) to (c) in Appendix Figures A8 and A9 replicate hours worked and earnings outcomes from Panels (b) and (c) in Figure 1 under the placebo reforms. Mean differences in all outcome variables in the pre- and post- reform periods around each of the placebo reform dates are shown in Columns (1)-(4) and (7)-(8) of Table 2. These results confirm that the main differences in labor supply outcomes can be explained by the change in the earnings limit rather than by time trends. While 12 some outcomes in placebo reform years show statistically significant differences, they are much smaller in magnitude than those in the actual reform period. As a second placebo analysis, we examine an alternative health related benefit, the Work Accident Allowance (WAA), which was not affected by the reform. This program is available for individuals who suffer health impairments of more than 13% resulting from workplace accidents or occupational diseases. Importantly, the accident allowance has no earnings limit, reflecting that the health impairment does not necessary imply a loss in working capacity. Otherwise, the award procedure for WAA is similar to RSA and other disability benefits. Labor market outcomes for WAA entrants in the reform years are shown in Appendix Figures A10. Similar to the other placebo analyses, the figure shows no difference in any outcomes between the two groups. 5 Discussion Disability insurance earnings limits can serve as screening mechanisms, ensuring that dis- ability benefits go to those who truly cannot work. At the same time, they may distort labor supply among workers with significant remaining working capacity. In this paper, we studied take-up and labor supply responses to changing earnings limits. We showed conceptually that with the choice of the earnings limit, policy makers must trade off selection and labor supply effects. Empirically, we exploit a reform that lowers the earnings limit in a disability insurance program for the moderately disabled in Hungary. As the lower earnings limit applied to all new disability entrants but remained unchanged for those already receiving benefits, we compared outcomes of entrants in the year before and after this cutoff date to evaluate the reform effects. Our empirical analysis provided three main findings. First, we documented that program entry and persistence in the program were not affected by the change in the earnings limit. This result is consistent with a system in which DI benefits also act as insurance against labor market income risks. Second, we found that the change in the composition of beneficiaries in response to the policy was small as individuals with slightly lower work capacity selected into the program after the reform date. Third, we show that intensive margin labor supply among beneficiaries entering after the reform date decreased significantly, leading to fewer hours of work and lower earnings. In particular, the reform resulted in a sharp reduction of labor supply among previously higher-earning beneficiaries, who presumably have higher capacity to generate labor income and whose labor supply responded to fulfill the stricter benefit eligibility requirement. Overall, our results suggest that decreasing the earnings limit only led to a moderate 13 improvement in screening efficiency. This evidence is consistent with a scenario where the earnings limit and benefit level before the reform were already sufficiently low to deter potential entrants who are well-positioned to find higher-paying jobs in the labor market. At the same time, the reform substantially distorted the labor supply of program participants. Viewed through the lens of our model, the empirical findings suggest that the overall impact of the reform on efficiency and welfare was negative. The reform failed to yield sizable cost savings from benefit expenditures for the government, but left moderately disabled individuals with lower earnings, resulting in lower tax revenues in turn. At the given benefit level, a higher earnings limit would therefore be optimal. 14 References Anders, Jenna, and Charlie Rafkin. 2021. “The Welfare Effects of Eligibility Expan- sions: Theory and Evidence from SNAP.” Mimeo. Autor, David H. 2011. “The Unsustainable Rise of the Disability Rolls in the United States: Causes, Consequences, and Policy Options.” National Bureau of Economic Re- search Working Paper 17697. Autor, David H., and Mark G. Duggan. 2003. “The Rise in the Disability Rolls and the Decline in Unemployment.” Quarterly Journal of Economics, 118(1): 157–205. Autor, David H., and Mark G. Duggan. 2006. “The Growth in the Social Security Disability Rolls: A Fiscal Crisis Unfolding.” Journal of Economic Perspectives, 20(3): 71– 96. Autor, David H., and Mark G. Duggan. 2007. “Distinguishing Income from Substitu- tion Effects in Disability Insurance.” American Economic Review, 97(2): 119–124. Autor, David H., and Mark G. Duggan. 2010. “Supporting Work: A Proposal for Modernizing the U.S. Disability.” Center for American Progress and The Hamilton Project. Autor, David H., Nicole Maestas, Kathleen J. Mullen, and Alexander Strand. 2015. “Does Delay Cause Decay? The Effect of Administrative Decision Time on the Labor Force Participation and Earnings of Disability Applicants.” National Bureau of Economic Research Working Paper 20840. Bipartisan Policy Center. 2015. “Improve the SSDI Program and Address the Impend- ing Trust Fund Depletion: Consensus Recommendations of BPC’s Disability Insurance Working Group.” Bipartisan Policy Center. Bound, John. 1989. “The Health and Earnings of Rejected Disability Insurance Appli- cants.” American Economic Review, 79(3): 482–503. Burkhauser, Richard V., and Mary C. Daly. 2011. The Declining Work and Welfare of People with Disabilities: What Went Wrong and a Strategy for Change. Washington, D.C.:American Enterprise Institute. B´ ır´ o, Anik´ o, Cec´ ılia Hornok, Judit Krek´ o, D´ ´ aniel Prinz, and Agota Scharle. 2022. “Can Disability Insurance Beneficiaries Be Reactivated?” Mimeo. 15 Chen, Susan, and Wilbert van der Klaauw. 2008. “The Work Disincentive Effects of the Disability Insurance Program in the 1990s.” Journal of Econometrics, 142(2): 757–784. de Jong, Philip, Maarten Lindeboom, and Bas van der Klaauw. 2011. “Screen- ing Disability Insurance Applications.” Journal of the European Economic Association, 9(1): 106–129. Deshpande, Manasi, and Yue Li. 2019. “Who Is Screened Out? Application Costs and the Targeting of Disability Programs.” American Economic Journal: Economic Policy, 11(4): 213–248. Deuchert, Eva, and Beatrix Eugster. 2019. “Income and Substitution Effects of a Dis- ability Insurance Reform.” Journal of Public Economics, 170: 1–14. ´ Duman, Anil, and Agota Scharle. 2011. “Hungary: Fiscal Pressures and a Rising Re- sentment Against the (Idle) Poor.” In Regulating the Risk of Unemployment: National Adaptations to Post-Industrial Labour Markets in Europe. Oxford University Press. Finkelstein, Amy, and Matthew J. Notowidigdo. 2019. “Take-Up and Targeting: Ex- perimental Evidence from SNAP.” Quarterly Journal of Economics, 134(3): 1505–1556. Garcia-Mandic´ o, S´ ıa-G´ ılvia, Pilar Garc´ omez, Anne C. Gielen, and Owen O’Donnell. 2020. “Earnings Responses to Disability Insurance Stringency.” Labour Eco- nomics, 66: 101880. Godard, Mathilde, Pierre Koning, and Maarten Lindeboom. 2022. “Application and Award Responses to Stricter Screening in Disability Insurance.” Journal of Human Resources, Forthcoming. Greenberg, David, Daniel Gubits, David Stapleton, Stephen Bell, Michelle Wood, Denise Hoffman, Sarah Croake, David Mann, Judy Geyer, Austin Nichols, Andrew McGuirk, Meg Carroll, Atsav Kattel, and David Judkins. 2018. BOND Final Evaluation Report, Volume 1. Social Security Administration, Office of Research, Demonstration, and Employment Support. Gruber, Jonathan. 2000. “Disability Insurance Benefits and Labor Supply.” Journal of Political Economy, 108(6): 1162–1183. Kearney, Melissa S., Brendan M. Price, and Riley Wilson. 2021. “Disability Insur- ance in the Great Recession: Ease of Access, Program Enrollment, and Local Hysteresis.” AEA Papers and Proceedings, 111: 486–490. 16 Kleven, Henrik Jacobsen, and Wojciech Kopczuk. 2011. “Transfer Program Complex- ity and the Take-Up of Social Benefits.” American Economic Journal: Economic Policy, 3(1): 54–90. Kostøl, Andreas Ravndal, and Magne Mogstad. 2014. “How Financial Incentives Induce Disability Insurance Recipients to Return to Work.” American Economic Review, 104(2): 624–655. Liebert, Helge. 2019. “Does external medical review reduce disability insurance inflow?” Journal of Health Economics, 64: 108–128. Liebman, Jeffrey B. 2015. “Understanding the Increase in Disability Insurance Benefit Receipt in the United States.” Journal of Economic Perspectives, 29(2): 123–150. Low, Hamish, and Luigi Pistaferri. 2015. “Disability Insurance and the Dynamics of the Incentive Insurance Trade-Off.” American Economic Review, 105(10): 2986–3029. Maestas, Nicole, and Na Yin. 2008. “The Labor Supply Effects of Disability Insurance Work Disincentives: Evidence from the Automatic Conversion to Retirement Benefits at Full Retirement Age.” Michigan, Michigan Retirement Research Center Working Paper 194. Maestas, Nicole, Kathleen J. Mullen, and Alexander Strand. 2013. “Does Disability Insurance Receipt Discourage Work? Using Examiner Assignment to Estimate Causal Effects of SSDI Receipt.” American Economic Review, 103(5): 1797–1829. Maestas, Nicole, Kathleen J. Mullen, and Alexander Strand. 2015. “Disability Insurance and the Great Recession.” American Economic Review, 105(5): 177–182. Mullen, Kathleen J., and Stefan Staubli. 2016. “Disability Benefit Generosity and Labor Force Withdrawal.” Journal of Public Economics, 143(1): 49–63. Nichols, Albert L., and Richard J. Zeckhauser. 1982. “Targeting Transfers through Restrictions on Recipients.” American Economic Review, 72(2): 372–377. OECD. 2009. Employment Outlook 2009: Tackling the Jobs Crisis. Paris:OECD Publishing. OECD. 2010. Sickness, Disability and Work: Breaking the Barriers. Paris:OECD Publish- ing. 17 Ruh, Philippe, and Stefan Staubli. 2019. “Financial Incentives and Earnings of Disabil- ity Insurance Recipients: Evidence from a Notch Design.” American Economic Journal: Economic Policy, 11(2): 269–300. Schimmel, Jody, David Stapleton, and Jae Song. 2011. “How Common is “Parking” Among Social Security Disability Insurance Beneficiaries? Evidence from the 1999 Change in the Earnings Level of Substantial Gainful Activity.” Social Security Bulletin, 71(4): 77– 92. Seb˝ok, Anna. 2019. “The Panel of Linked Administrative Data of CERS Databank.” Centre for Economic and Regional Studies Institute of Economics Budapest Working Papers on the Labour Market 2019/2. Weathers, Robert R., and Jeffrey Hemmeter. 2011. “The Impact of Changing Fi- nancial Work Incentives on the Earnings of Social Security Disability Insurance (SSDI) Beneficiaries.” Journal of Policy Analysis and Management, 30(4): 708–728. Zaresani, Arezou. 2020. “Adjustment Cost and Incentives to Work: Evidence from a Disability Insurance Program.” Journal of Public Economics, 188: 104223. Zaresani, Arezou, and Miguel Olivo-Villabrille. 2022. “Return-to-Work Policies’ Claw- back Regime and Labor Supply in Disability Insurance Programs.” Labour Economics, 78: 102215. 18 Figure 1: Labor Market Outcomes of Regular Social Assistance Entrants (a) Share Working (b) Hours Worked (Conditional on Working) (c) Earnings Relative to Monthly Minimum Wage (Condi-(d) Share Earning Above 80% of Monthly Minimum Wage tional on Working) (Conditional on Working) Notes: Figure shows labor market outcomes for individuals who enter Regular Social As- sistance (RSA) the year before the reform, between January 1 and December 31, 2007 (“old entrants” in blue) and those who enter the year after the reform, between January 1 and December 31, 2008 (“new entrants” in red). The pre-entry labels show the mean of each outcome during the period between four years to one year (months -48 to -13) before entering RSA and the post-entry labels show the mean of each outcome during the period between one and three years (months 13 to 36) after entering RSA. 19 Figure 2: Other Outcomes (a) Benefit Persistence (b) Mortality Notes: Figure shows benefit persistence in Panel (a), and cumulative mortality in Panel (b) for individuals who enter Regular Social Assistance (RSA) the year before the reform, between January 1 and December 31, 2007 (“old entrants” in blue) and those who enter the year after the reform, between January 1 and December 31, 2008 (“new entrants” in red). 20 Figure 3: Heterogeneity By Pre-Disability Earnings (a) Pre-Disability Earnings Below Monthly Minimum(b) Pre-Disability Earnings Above Monthly Minimum Wage Wage Notes: Figure shows earnings relative to the minimum wage for individuals who enter Regu- lar Social Assistance (RSA) the year before the reform, between January 1 and December 31, 2007 (“old entrants” in blue) and those who enter the year after the reform, between January 1 and December 31, 2008 (“new entrants” in red). Panel (a) shows individuals whose average pre-disability wage (three years before entering RSA) was below the minimum wage. Panel (b) shows individuals whose average pre-disability wage (three years before entering RSA) was higher than the minimum wage. The pre-entry labels show the mean of each outcome during the period between four years to one year (months -48 to -13) before entering RSA and the post-entry labels show the mean of each outcome during the period between one and three years (months 13 to 36) after entering RSA. 21 Table 1: Regular Social Assistance Entrants Before and After the Reform New Entrants Old Entrants p-value Gender Male 0.414 0.383 0.034 Age 35-44 years 0.186 0.195 0.429 45-55 years 0.734 0.741 0.597 Best job before disability Managers 0.050 0.040 0.113 Professionals 0.019 0.020 0.718 Technicians 0.082 0.087 0.584 Office and customer service 0.038 0.051 0.042 Commercial and services 0.166 0.157 0.428 Agriculture 0.038 0.045 0.232 Industry and construction 0.277 0.267 0.504 Machine operators and drivers 0.114 0.114 0.942 Elementary occupations 0.215 0.216 0.992 Region Budapest 0.064 0.056 0.293 Central Hungary 0.073 0.084 0.201 Central Transdanubia 0.099 0.099 0.999 Western Transdanubia 0.060 0.061 0.907 Southern Transdanubia 0.158 0.148 0.388 Northern Hungary 0.128 0.141 0.208 Northern Great Plain 0.209 0.205 0.761 Southern Great Plain 0.205 0.200 0.711 Working 3 years earlier 0.697 0.715 0.053 Earnings relative to monthly 1.254 1.364 0.001 minimum wage 3 years earlier (Conditional on Working) Number of observations 1,885 2,791 Notes: Table shows summary statistics for individuals who enter Regular Social Assistance (RSA) the year before the reform, between January 1 and December 31, 2007 (“old entrants”) and those who enter the year after the reform, between January 1 and December 31, 2008 (“new entrants”). The table contains disability beneficiaries aged 20-60 years. Occupation categories refer to the Hungarian standard classification of occupations (HCSO-08/FEOR- 08). The number of persons displays observations in the database that includes about half of the disabled population. 22 Table 2: Labor Market Outcomes of Regular Social Assistance Entrants Working (1) (2) (3) (4) (5) (6) (7) (8) 2006 2007 2008 2009 Pre Post Pre Post Pre Post Pre Post Difference 0.00213 -0.00984 0.0340*** 0.0259*** -0.0314*** -0.0179 0.0431*** 0.0740*** (0.00757) (0.00804) (0.00838) (0.00964) (0.00997) (0.0115) (0.0105) (0.0128) Old Entrants 0.666 0.296 0.672 0.286 0.706 0.312 0.675 0.294 Observations 196,876 227,807 229,240 170,177 168,066 110,397 136,295 89,632 Hours Worked (Conditional on Working) 2006 2007 2008 2009 Pre Post Pre Post Pre Post Pre Post Difference 0.0332 -0.0998 0.0786 -0.425 -0.360** -2.056*** -0.0323 -0.490 (0.105) (0.253) (0.107) (0.294) (0.148) (0.423) (0.167) (0.477) Old Entrants 38.68 31.91 38.75 31.81 38.82 31.39 38.46 29.33 Observations 125,053 61,046 149,845 46,777 107,070 30,034 82,464 25,147 Earnings Relative to Monthly Minimum Wage (Conditional on Working) 2006 2007 2008 2009 Pre Post Pre Post Pre Post Pre Post Difference 0.0707*** 0.0420** 0.0480** -0.0435** -0.101*** -0.167*** 0.00897 -0.00916 (0.0166) (0.0172) (0.0194) (0.0209) (0.0245) (0.0237) (0.0251) (0.0223) Old Entrants 1.201 0.95 1.292 0.992 1.34 0.949 1.239 0.781 Observations 132,503 66,856 158,641 50,726 117,566 33,973 95,801 29,986 Earnings Above 80% of the Monthly Minimum Wage (Conditional on Working) 2006 2007 2008 2009 Pre Post Pre Post Pre Post Pre Post Difference 0.0292*** 0.0215 0.0287*** -0.0372** -0.0618*** -0.237*** 0.0178* -0.0240 (0.00760) (0.0138) (0.00715) (0.0163) (0.00911) (0.0195) (0.0103) (0.0201) Old Entrants 0.731 0.609 0.777 0.631 0.806 0.594 0.744 0.357 Observations 132,503 66,856 158,641 50,726 117,566 33,973 95,801 29,986 ***p < 0.01, **p < 0.05, *p < 0.1 Notes: Table shows labor market outcomes for individuals who enter Regular Social Assis- tance (RSA) the year before and after the reform (2008 in columns (5) and (6)) and Placebo reforms (2006, 2007, and 2009 in columns (1), (2), (3), (4), (7), and (8)). In each column, the third row reports the mean for “old entrants”. “Old entrants” are individuals who enter RSA between January 1 and December 31 of the year before the (Placebo) reform. The first row reports the difference between “old entrants” and “new entrants”. “New entrants” are individuals who enter RSA between January 1 and December 31 of the year of the (Placebo) reform. The pre-entry columns are defined over the period between four years to one year (months -48 to -13) before entering RSA and the post-entry columns are defined over the period between one and three years (months 13 to 36) after entering RSA. Standard errors (in parentheses) are clustered at the individual level. 23 Online Appendix Appendix Figure A1: Number of Regular Social Assistance Entrants Notes: Figure shows the number of beneficiaries entering Regular Social Assistance (RSA) by month between January 2011 and April 2011. The vertical lines mark our main sample period. Entrants between January and December 2007 are considered “old entrants” and entrants between January and December 2008 are considered “new entrants” in our analysis. 1 Appendix Figure A2: Seasonally-Adjusted Unemployment Rate Notes: Figure shows the seasonally adjusted unemployment rate of the 15-64 year old population in Hungary from 2005 to 2011 in percent. Source: Labour Force Survey, Central Statistical Office. 2 Appendix Figure A3: Sick Leave Use Notes: Figure shows sick leave use for individuals who enter Regular Social Assistance (RSA) the year before the reform, between January 1 and December 31, 2007 (“old entrants” in blue) and those who enter the year after the reform, between January 1 and December 31, 2008 (“new entrants” in red). The pre-entry labels show the mean of each outcome during the period between four years to one year (months -48 to -13) before entering RSA and the post-entry labels show the mean of each outcome during the period between one and three years (months 13 to 36) after entering RSA. 3 Appendix Figure A4: Wage Distribution in 2009 (Conditional on Working) (a) Old Entrants (b) New Entrants Notes: Figure shows the distribution of monthly wages observed in 2009. Panel (a) shows wages for individuals who enter Regular Social Assistance (RSA) the year before the reform, between January 1 and December, 2007 (“old entrants”) and panel (b) shows wages for individuals who enter the year after the reform, between January 1 and December 31, 2008 (“new entrants”). The dashed line shows the level of the minimum wage. 4 Appendix Figure A5: Labor Market Outcomes of Regular Social Assistance Entrants: Reweighting for Selection (a) Share Working (b) Hours Worked (Conditional on Working) (c) Wage Relative to Minimum Wage (Conditional(d) Share Earning Above 80% of Minimum Wage on Working) (Conditional on Working) Notes: Figure shows difference-in-differences even study estimates of labor market outcomes relative to Regular Social Assistance (RSA) entry, comparing individuals who enter the year before the reform, between January 1 and December 31, 2007 (“old-entrants”) and those who enter the year after the reform, between January 1 and December 31, 2008 (“new- entrants”). The sample is reweighted by the inverse of the propensity score of selection into the reform treatment (new entrants). The propensity score is based on a logit model, where the dependent variable is an indicator for being a new entrant, and the right hand side variables include age, gender, work status and wage relative to the minimum wage 12, 24, and 36 months before taking up benefits. The pre-entry labels show the mean of each outcome during the period between four years to one year (months -48 to -13) before entering RSA and the post-entry labels show the mean of each outcome during the period between one and three years (months 13 to 36) after entering RSA. 5 Appendix Figure A6: Change in Regular Social Assistance Entry Rate vs Change in the Unemployment Rate (a) 2007-2008 (b) 2007-2009 Notes: Figure shows the relationship between the change in Regular Social Assistance (RSA) entry rates and the change in unemployment rates at the microregion level. Panel (a) shows the changes between 2007 and 2008 and Panel (b) shows the changes between 2008 and 2009 in percentage points.The annual microregion level unemployment data are from the T-STAR database of the Central Statistical Office of Hungary. 6 Appendix Figure A7: Labor Market Outcomes of Regular Social Assistance Entrants Rela- tive to National and Microregion Averages Share Working (a) Relative to National Average (b) Relative to Microregion Average Earnings (c) Relative to National Average (d) Relative to Microregion Average Earnings in Unskilled Jobs (e) Relative to National Average (f) Relative to Microregion Average Notes: Figure shows labor market outcomes relative to national and microregion averages for individuals who enter Regular Social Assistance (RSA) the year before the reform, be- tween January 1 and December 31, 2007 (“old entrants” in blue) and those who enter the year after the reform, between January 1 and December 31, 2008 (“new entrants” in red). Earnings are conditional of being employed. The pre-entry labels show the mean of each outcome during the period between four years to one year (months -48 to -13) before entering RSA and the post-entry labels show the mean of each outcome during the period between one and three years (months 13 to 36) after entering RSA. 7 Appendix Figure A8: Placebo Analyses: Hours Worked (a) 2006 (b) 2007 (c) 2009 Notes: Figure shows hours worked for individuals who enter Regular Social Assistance (RSA) the year before and after three Placebo reforms in 2006, 2007 and 2009. In each of the panels “old entrants” in blue are individuals who enter RSA between January 1 and December 31 of the year before the Placebo reform year (2005, 2006, and 2008) and “new entrants” in red are individuals who enter RSA between January 1 and December 31 of the Placebo reform year (2006, 2007, and 2009). The pre-entry labels show the mean of hours worked outcome during the period between four years to one year (months -48 to -13) before entering RSA and the post-entry labels show the mean of hours worked during the period between one and three years (months 13 to 36) after entering RSA. 8 Appendix Figure A9: Placebo Analyses: Earnings Relative to Monthly Minimum Wage (a) 2006 (b) 2007 (c) 2009 Notes: Figure shows earnings relative to the monthly minimum wage for individuals who enter Regular Social Assistance (RSA) the year before and after three Placebo reforms in 2006, 2007 and 2009. In each of the panels “old entrants” in blue are individuals who enter RSA between January 1 and December 31 of the year before the Placebo reform year (2005, 2006, and 2008) and “new entrants” in red are individuals who enter RSA between January 1 and December 31 of the Placebo reform year (2006, 2007, and 2009). The pre-entry labels show the mean of earnings relative to the monthly minimum wage during the period between four years to one year (months -48 to -13) before entering RSA and the post-entry labels show the mean of earnings relative to the monthly minimum wage during the period between one and three years (months 13 to 36) after entering RSA. 9 Appendix Figure A10: Placebo Analyses: Accident Allowance (a) Share Working (b) Sick Leave (c) Earnings Relative to Monthly Minimum Wage (Condi-(d) Share Earning Above 80% of Monthly Minimum Wage tional on Working) (Conditional on Working) Notes: Figure shows labor market outcomes for individuals who enter Accident Allowance the year before the reform, between January 1 and December 31, 2007 (“old entrants” in blue) and those who enter the year after the reform, between January 1 and December 31, 2008 (“new entrants” in red). 10 Appendix Table A1: Labor Market Outcomes of Regular Social Assistance Entrants with Below and Above Median Change in Microregion Unemployment Rate between 2007-2009 Working (1) (2) (3) (4) (5) (6) High Unemployment Low Unemployment Whole Sample Pre Post Pre Post Pre Post Difference -0.0373*** -0.0116 -0.0317** -0.0138 -0.0314*** -0.0179 (0.0130) (0.0154) (0.0138) (0.0172) (0.00997) (0.0115) Old Entrants 0.651 0.304 0.651 0.302 0.706 0.312 Observations 90,536 59,779 73,858 48,189 168,066 110,397 Hours Worked (Conditional on Working) High Unemployment Low Unemployment Whole Sample Pre Post Pre Post Pre Post Difference -0.598*** -2.193*** -0.195 -2.815*** -0.360** -2.056*** (0.200) (0.489) (0.219) (0.594) (0.148) (0.423) Old Entrants 38.93 31.32 38.66 30.88 38.82 31.39 Observations 57,306 16,060 47,133 13,312 107,070 30,034 Earnings Relative to Monthly Minimum Wage (Conditional on Working) High Unemployment Low Unemployment Whole Sample Pre Post Pre Post Pre Post Difference -0.0874*** -0.152*** -0.111*** -0.185*** -0.101*** -0.167*** (0.0305) (0.0277) (0.0406) (0.0409) (0.0245) (0.0237) Old Entrants 1.31 0.92 1.37 0.97 1.34 0.949 Observations 62,320 18,249 52,539 14,972 117,566 33,973 Earnings Above 80% of the Monthly Minimum Wage (Conditional on Working) High Unemployment Low Unemployment Whole Sample Pre Post Pre Post Pre Post Difference -0.0639*** -0.250*** -0.0452*** -0.227*** -0.0618*** -0.237*** (0.0109) (0.0256) (0.0115) (0.0283) (0.00911) (0.0195) Old Entrants 0.81 0.62 0.79 0.59 0.806 0.594 Observations 62,320 18,249 52,539 14,972 117,566 33,973 ***p < 0.01, **p < 0.05, *p < 0.1 Notes: Table shows labor market outcomes for individuals who enter Regular Social Assis- tance (RSA) the year before and after the reform for the whole sample in columns (5) and (6), for a subsample of persons living in microregions with below median change in the mi- croregion level unemployment rate from 2007 to 2009 ( columns (1), (2)), and above median change in the microregion level unemployment rate (columns (3), (4)). The average increase in unemployment rate in microregions with low and high change is 4.6 and 13 percentage points, respectively. In each column, the third row reports the mean for “old entrants”. “Old entrants” are individuals who enter RSA between January 1 and December 31 in 2007. The first row reports the difference between “old entrants” and “new entrants”. “New entrants” are individuals who enter RSA between January 1 and December 31 of 2008. The pre-entry columns are defined over the period between four years to one year (months -48 to -13) before entering RSA and the post-entry columns are defined over the period between one and three years (months 13 to 36) after entering RSA. Standard errors (in parentheses) are clustered at the individual level. 11