The Labor Market Effects of Disability Benefit Loss

ABSTRACT


Introduction
The rise in disability rolls in developed countries during the 1990s (OECD, 2010) combined with low levels of employment among beneciaries prompted policy makers to examine how the design of disability insurance (DI) can facilitate the labor market reintegration of ben-eciaries.Among other tools, proposals typically include improving nancial incentives for work and the better identication of remaining working capacity (Autor and Duggan, 2010;Burkhauser and Daly, 2011;Maestas, 2019).
Whether low levels of reintegration result from limited working capacity, other barriers to employment or poorly designed nancial incentives is an important question for policy design.To the extent that limited working capacity is the reason behind low levels of reintegration, nancial incentives or periodic reassessments are unlikely to have much success in reintegrating beneciaries into the labor market.Moreover, if they include the removal of benets, they can harm beneciary welfare.If, on the other hand, low levels of reintegration are caused by poorly designed incentives, such as earnings limits set too low (Krekó, Prinz and Weber, 2023), governments can improve the eciency and scal sustainability of DI programs by setting incentives appropriately.But even then, supply-side nancial incentives might not be eective without rehabilitation and personalized support services to mitigate other potential barriers to work, such as human capital depreciation (Edin and Gustavsson, 2008), stigma (Eriksson and Rooth, 2014;Fernández-Blanco and Preugschat, 2018) or psychological distress (Diette, Goldsmith, Hamilton and Darity Jr., 2012) caused by long term unemployment.
In this paper, we study a unique large-scale reassessment reform to investigate the extent to which beneciaries can be reintegrated when their benets are removed or reduced.Starting in 2012, Hungarian DI beneciaries under 57 years of age with health damage below 80% had to undergo a reassessment in order to remain eligible for benets.As a result, about 18,000 beneciaries (9% of the reassessed beneciaries and 5% of all beneciaries) lost their benets while about 12,000 beneciaries (6% of the reassessed beneciaries and 4% of all beneciaries) had their benets reduced.We study the labor market consequences of benet loss or reduction by leveraging age and health cutos in reassessments and focusing on a narrow age cohort around the cut-o age.
Comparing beneciaries just below and just above the age cut-o, we nd that among aected beneciaries the probability of disability insurance receipt decreased by 1.5 percentage points due to the reform.About two-thirds of those who left DI were employed in the primary labor market or participated in public works in the post-reform period: without the concurrent receipt of DI benets, employment increased by 0.9, and public works partici-pation increased by 0.1 percentage point.Roughly one-third of excluded beneciaries were not employed: the probability of having no income from DI or employment increased by 0.5 percentage point.
Unlike in the United States but similarly to other European countries, a meaningful share of DI beneciaries are employed while receiving benets and post-reform labor market outcomes dier greatly according to pre-reform employment status.Individuals who were working in 2011 (and were healthier) were more likely to lose their benet as a result of the review.While only a quarter of the reassessed beneciaries were employed in 2011, half of those who lost their benets came from this group.81% of them were still employed postreform, 6% participated in public works, while 14% had no job or benets.The other half of recipients who lost their benets (with no work recorded in 2011) fared worse in the labor market: only 33% were employed post-reform, 6% participated in public works, while 61% had no job or benets.By comparing pre-and post-reform job quality indicators among those who were employed, we also document the deterioration of job quality of former beneciaries.
Moreover, the deterioration is more striking among those who had no employment in 2011.
Our results suggest that the consequences of DI benet removal depend crucially on whether a beneciary is employed while receiving benets.Those who held jobs while on benets have a good chance of being employed after losing their benets, while those who did not work are likely to remain out of work and without benets.This suggests that improving the labor market attachment of beneciaries is an important direction for policy.
Moreover, our results point to the importance of combining nancial incentives with broader labor market programs that increase employability after benet loss.
Our work contributes to three strands of the literature.We most directly contribute to the literature that has examined the labor market consequences of benet reduction or removal.Borghans, Gielen andLuttmer (2014), García-Gómez andGielen (2018) and Garcia-Mandicó, García-Gómez, Gielen and O'Donnell (2020) study two large-scale DI reassessment initiatives in the Netherlands.García-Gómez and Gielen (2018) and Garcia-Mandicó, García-Gómez, Gielen and O'Donnell (2020) show that following both reforms, recipients who experienced benet reduction or removal increased their labor supply substantially, replacing almost two-thirds of lost benets with earnings in the labor market.As an adverse eect of benet loss, Borghans, Gielen and Luttmer (2014) nd an increase in mortality among low-income women whose benets were reduced following the 1993 reform.
We contribute to this literature by pointing out the importance of employment during DI for post-DI labor market outcomes.Our rich database also allows us to analyze the impact of benet loss on employment prospects and job quality.With this part of the analysis we contribute to the literature showing that long periods of inactivity may lead to a growing distance from the labor market, depreciation in human capital, lower probability of work, and lower wages (Vingård, Alexanderson and Norlund, 2004;Edin and Gustavsson, 2008;Bryngelson, 2009).
Using various quasi-experimental approaches, these papers nd that disability insurance receipt reduces labor supply substantially.Our main contribution to this literature is the examination of the consequences of benet reduction among individuals who were already receiving benets for some time.
Finally, our work also speaks to the academic and policy literature that has focused on the scal sustainability of DI programs (e.g., Autor andDuggan, 2006, 2007;Autor, 2011;Liebman, 2015).We show that, although reassessments may be a way to reduce DI rolls, policy makers need to be aware of the potential negative impact on the welfare of beneciaries whose benets are removed or reduced but who are unable to nd employment.
The remainder of the paper is structured as follows.Section 2 describes the institutional background and the details of the 2012 reform.Section 3 describes our data.Section 4 explains our empirical approach.Section 5 presents our results.Section 6 concludes.

Disability Insurance in Hungary
In 1990, the Hungarian DI system was characterized by lenient eligibility rules and relatively high benet levels (Scharle, 2008).The deep recession following the economic transition from socialism to market economy rapidly increased unemployment in the early 1990s and policy makers allowed (or encouraged) the expansion of benet programs such as DI and early retirement in order to ease social and political tensions (Vanhuysse, 2004).As a result, the number of DI beneciaries doubled between 1990 and 2003 and reached over 700,000 or 12% of the working-age population, the highest rate among OECD countries (OECD, 2016).
Following cautious and largely ineective attempts to tighten the eligibility criteria in the late 1990s, a 2008 reform aimed to curb the inow into the system by prioritizing rehabilitation and encouraging labor market integration instead of focusing solely on health impairment in the assessment of new benet claims (Scharle, 2008).The 2008 reform consisted of three key elements.First, a new assessment system was introduced which put more emphasis on remaining working capacity and the potential for rehabilitation and skill de-velopment.The second element was the introduction of a temporary rehabilitation benet, which was granted for up to three years and thus helped to reduce the take-up of permanent disability benets.New claimants with a health damage of at least 50% and assessed as rehabilitable were eligible for this benet.Third, recipients of the temporary benet were obliged to cooperate with the public employment service and participate in employment rehabilitation programs, which were expanded in terms of range and capacity (Adamecz-Völgyi, Lévay, Bördős and Scharle, 2018).While the employment eect of the expanded rehabilitation programs was positive, their take-up, as well as the impact of the reform on DI spending fell below expectations.
The focus of this paper is a 2012 reform which tightened eligibility and reduced benet levels not only for new claimants but also for existing beneciaries (Nagy, 2015;Kovács, 2019).The aim was to curb inow and to reactivate beneciaries with some remaining working capacity in order to improve the sustainability of the DI system, which was considered overly generous even after the 2008 reform, and was believed to contribute to the low activity rate in Hungary.As a consequence of the two subsequent reforms, as well as favorable demographic and economic trends, the share of beneciaries decreased to 4% of the active population and the cost of DI benets decreased to below 1% of GDP by 2017, one of the lowest values in Europe.While the 2012 reform was successful in reducing the costs of the DI system, its harshness generated debates about whether it would reactivate long-time beneciaries or simply leave them without income.

Details of the 2012 Reform
The 2012 reform obliged approximately 200,000 DI recipients to undergo a health review based on new, stricter rules of entitlement.The obligation applied to all DI recipients below age 57 with a partial disability, whose health impairment was below 80%, as determined by the pre-reform assessment system (Table 1).Two disability benet programs were aected: the Category III Disability Pension for those with a health damage of 50% to 79% and the Regular Social Assistance for those with a health damage above 40%.The reform did not apply to recipients of Category I and Category II Disability Pensions (who had at least 80% health damage).1 Beneciaries aected by the reform had to declare by March 2012 whether they wished to undergo the health reassessment.If they failed to do so, they lost their benet entitlement by May 2012.Otherwise, their health status and degree of employability were reevaluated according to the post-reform rules in a complex assessment process carried out by a team of physicians and rehabilitation experts.Individuals whose health impairment was classied higher than 40% during this review retained eligibility to benets.Mainly due to capacity constraints, it took several years to undertake all the reviews, so the process was completed only in 2016.
About 18,000 beneciaries (9% of the reassessed beneciaries and 5% of all beneciaries) who underwent the review lost benet eligibility.The total number of recipients decreased much more, from 473,000 in January 2012 to 355,000 in January 2017 (Hungarian Central Statistical Oce, 2022), due to a large drop in inows, that started from the early 2000s and gained new momentum after 2012.This drop in the number of beneciaries after the reform suggests that while in principle the eligibility conditions (expressed as percent of health damage) did not change, the assessment process became more stringent.On top of the large drop in the number of beneciaries, the benets of 12,000 beneciaries decreased in ination-adjusted terms.2 The pre-reform disability benet categories were consolidated into two benet programs called Disability Allowance and Rehabilitation Allowance.Beneciaries not recommended for vocational rehabilitation became eligible for the Disability Allowance while those who were deemed able to return to the labor market following rehabilitation became eligible for the Rehabilitation Allowance, which was paid for up to 3 years and set at a much lower rate than the Disability Allowance.3Rehabilitation Allowance recipients were required to cooperate with the rehabilitation authority and fulll obligations set out in the employment rehabilitation plan.At the same time, recipients over 62 years of age were reclassied as old-age pensioners.
Although the comprehensive reevaluation of a large subgroup of DI recipients is uncommon, it is not without precedent: the majority of DI recipients under age 44 were reassessed under more stringent rules in the Netherlands in 2004.Garcia-Mandicó, García-Gómez, Gielen and O'Donnell (2020) estimate that the reassessment removed 17 percent of beneciaries from the program and reduced benet income by 20 percent, on average.However, in contrast to the Netherlands, where support for labor market reintegration was substantially expanded between 1997 and 2002, beneciaries in Hungary who lost part or all of their ben-et received little support in returning to the labor market.The capacity of rehabilitation services at the time was very limited and intensive, personalized services were only provided by a handful of small NGOs, operating mainly in urban centers (Krekó and Scharle, 2020).4

Data
The analysis is based on an individual-level linked employer-employee administrative panel database, covering a randomly selected half of the population of Hungary in 2003, who are then followed up until 2017.The database consists of linked data sets at the monthly frequency of the pension, tax and health care authorities and contains detailed individuallevel information on employment and earnings history, use of the health care system, pension and other social benets, and rm-level indicators.Importantly, it also contains information on the type and amount of dierent disability benets and old-age pensions received.Two important limitations of the data are that the employment status of DI recipients cannot be observed until April 2007 and we do not observe the health condition based on which the disability benet is received.Based on the 2011 census (Appendix Table A1), the majority of DI recipients suer from long-lasting diseases.Among those recipients who have an impairment, mobility impairment is the most prevalent form of disability.
When estimating the eects of the reform, we analyze the following monthly indicators of labor market and DI status.DI status is a binary variable that takes value one if the individual is DI recipient in a given month and zero otherwise. 5The binary variable for employment status equals one if the individual is employed on the 15 th of the given month and zero otherwise.Employment includes self-employment but excludes public work.Importantly, employment was always allowed while receiving DI benets, with restrictions on the maximum possible earnings. 6We analyze public work as a separate outcome.In addition, we look at four quarterly indicators of healthcare use: GP visits, outpatient specialist care visits, hospital days, and total spending (social security plus out-of-pocket spending) on prescription drugs.Indicators of healthcare use are included in our data from 2009.
We extend the analysis with job quality indicators derived from the administrative panel database.We generate a binary indicator of earning above the minimum wage, after adjusting the monthly wage to hours worked.We dene a binary indicator of full-time job, which equals one if the weekly hours of work are at least 40.We generate a binary indicator of working in a skilled job, which includes all occupations except for elementary occupations, with elementary occupations corresponding to International Standard Classication of Occupations (ISCO) code 9. Finally, using the entire sample in the administrative database, we calculate the year-specic median of the total factor productivity (TFP) of rms, weighted by rm size.8Based on this indicator, we dene a binary indicator of above-median employer TFP.

Control and Treatment Groups
In our empirical analysis, we estimate the impact of the 2012 reform on DI recipients subject to the compulsory health reassessment-partially disabled individuals with health impairment below 80%, who were under age 57 at the end of 2011.A limitation of our data is that we do not have information on the reassessment procedure itself; we observe exits from the DI system, but not the reason for leaving the system.Consequently, it is not possible to isolate those who lost their benet as a results of the revision from those who would have exited DI even in the absence of the reform.For this reason, to identify the impact of the reform, we focus on a narrow age group around the age cut-o of the policy, assuming that stated functions: to reintegrate participants into the primary labor market and to exclude people not willing to participate in public works from receiving benets and social assistance (Molnár, Bazsalya, Bódis and Kálmán, 2019).However, the vast majority of Hungarian public workers -especially the unskilled and those in depressed areas -worked in separated public works units (Köllő, 2015) and received very low pay.Between 2011 and 2015, both the net and gross basic public work wage ranged between 70-80% of the statutory minimum wage.
outcomes of individuals in this narrow age group below and above the cut-o age would have evolved similarly in the absence of the reform.
Our sample contains DI recipients belonging to the aected benet categories who were aged 56 or 57 in December 2011.Those who were 56 (just below the cut-o) in December 2011 make up the treatment group, while those who were 57 (just above the cut-o) make up the control group.We restrict the sample to individuals claiming DI throughout 2011 who were alive in January 2012.We restrict the control age group to age 57 at the end of 2011 to exclude individuals close to the old-age retirement age in order to improve comparability across the control and treatment age groups.9We focus on men below 62, the statutory retirement age for the oldest cohorts, allowing us to use data up to 2015.Our focus on men is motivated by the "Women 40" policy which since 2011 gives an early retirement option to women with 40 years of work credits, regardless of age.This policy could aect the control and treatment age groups dierently, potentially confounding our results for women.Finally, those who died during the observed time period are included in the sample until the last year they were alive.
Summary statistics for the control and treatment groups are displayed in Table 2.The two groups are quite similar to each other on most dimensions.They have approximately the same employment rate (24.3% vs 24.9%) while receiving benets in 2011 and each group has been receiving benets for 11 years on average.Despite being a year younger, the 56-yearold treatment group may be slightly less healthy with average prescription drug spending of 533 euros vs 512 euros among the 57-year-old control group.Importantly for labor market outcomes, the two groups live in geographic areas with similar economic environments as evidenced by the average unemployment rate of their micro-regions of residence, 20.0% for the treatment group and 19.4% for the control group.They also work in occupations with similar skill levels: 33.1% of the treatment group and 35.3% of the control group work in skilled occupations, while 18.3% and 17.7%, respectively work in unskilled ones.
Figure 1 plots the evolution of the share of individuals receiving DI benet in our sample separately for the treatment and the control groups.The sample is restricted to individuals who receive benets throughout 2011, but we don't impose any restrictions on DI status before or after 2011.The gure suggests that in 2009 and 2010, the DI status of the control and treatment groups evolved very similarly, which suggests that the two groups are likely to be comparable and that absent the reform their status would have involved similarly.
Following the reform, the control and treatment groups diverge: over the next four years, 2% of the control group but 4% of the treatment group is removed from benets.The bulk of the divergence occurs in May 2012, which suggests that although the review process lasted until 2016, most beneciaries were aected early on.10

Difference-in-Differences
To study the "reduced form" impact of the reassessment on labor market outcomes of the reassessed population, we estimate the following equation: To explore the evolution of the reform's impact over time, we also estimate month-specic treatment eects β t from the following equation: where i indexes individuals, t indexes months, [Date t = t] is an indicator for month t, In order for our estimates to represent the causal impact of being subject to the reassessment on the labor market outcomes of the treatment group, the control group must represent a valid counterfactual for the evolution of the treatment group's labor market outcomes.In particular, we assume that absent the reassessment, the two groups' labor market outcomes would have involved similarly.We present several pieces of evidence consistent with this assumption.First, Table 2 shows that the control and treatment groups are similar on a number of dierent measures of health and employment.Second, Figure 1 suggests that prior to the reassessment the disability status of the control and treatment groups evolved very similarly, suggesting that absent the reassessment they would have moved together as well.Third, the month-specic estimates of the dierence in labor market outcomes between the control and treatment group presented in Figure 2 also show that all outcomes move together in the two groups prior to the reform, which also suggests that the outcomes of the control group post-reform are a good counterfactual for the outcomes of the treatment group.
Fourth, we present results using a placebo approach, comparing the labor market outcomes of disabled individuals who fall into the same age groups but were unaected by the reform as they had health impairments over 80%.There is no evidence of dierential changes by age in this unaected group which suggests that our main results indeed identify the impact of the reassessment for the aected group.Fifth, we also present results for a 2011 placebo reform and nd no evidence of dierential changes by age in labor market outcomes, in line with our main results being driven by the 2012 reform.

Instrumental Variables Approach
To quantify the labor market impact of benet loss, we use being subject to the reassessment as an instrument for benet loss.The rst-stage equation is where exit it is a binary indicator for not receiving DI benets (equals one minus the DI status variable), [AGE i = 56] is an indicator for the treatment group, and the µ t are month xed eects.Using the rst stage to estimate predicted loss of benets, we estimate the second-stage equation: where êxit it denotes predicted benet loss and the µ t are month xed eects.Our coecient of interest is β IV , which captures the impact of benet loss on labor market outcomes after the reassessment among individuals who lost their benets due to the reform.
We estimate the impact of benet loss on three of the previously dened ve outcome In addition to the identifying assumptions described above, the two standard IV assumptions of relevance and exogeneity need to be satised for our estimate to represent the causal impact of benet loss on labor market outcomes.Figure 1 and the rst two columns of Table 3 show the relevance of the instrument.Table 3 suggests that over the four years after the reform, beneciaries under the age cut-o had an approximately 1.5 percentage point higher probability of losing their benets.The exogeneity assumption requires that being subject to reassessment aects labor market outcomes only through the DI exit channel.This assumption cannot be directly tested.Our placebo results provide suggestive evidence that being under the same age cut-o did not aect labor market outcomes among disability recipients not subject to reassessment and in a placebo reform year.However, if for example unobservable health status varies signicantly with being 56 or 57 years old at the end of 2011 (i.e., being subject to reassessment), and health status aects labor market outcomes conditional on DI loss then our estimates could be biased.Table 2 indicates that the treatment and the control groups are similar in terms of major observable characteristics, including pre-reform drug spending.Our assumption is that the two analyzed cohorts are similar in all aspects, apart from being subject to reassessment.

Main Results
In this section, we report our dierence-in-dierences estimates of the overall impact of the reassessment and our instrumental variables estimates of the impact of benet removal on labor market outcomes.Figure 2 3 show that there is little change in the number of individuals who receive DI benets while not working.Panel (b) of Figure 2 shows that by May 2012 aected beneciaries were about 2 percentage points less likely to be receiving benets and working at the same time and the gap increased further to 2.5 percentage points by the end of 2015.This suggests that most benet removals happened early on with additional exits happening gradually over the subsequent years as reassessment progressed.
Pooling over the post-period, column (2) of panel A of Table 3 shows that there was a 1.8 percentage point decline in the probability of receiving benets and working at the same time.
Panel (c) of Figure 2 suggests a concurrent jump in the number of former beneciaries who work without receiving benets, followed by a slow increase over the next four years.
Column (3) of panel A of Table 3 shows that pooling over the post-reform years there was a 0.9 percentage point increase in employment without benets.This suggests that approximately 60% of those removed from benets end up working in the open labor market.
The year-by-year instrumental variables estimates displayed in Panel (a) of Figure 3 3).Panel (a) of Figure 3 shows that the impact of benet loss on employment in public works is especially pronounced in 2013 and 2014.Averaging over the post-reform years, Column (2) of panel A of Table 4 shows that according to our instrumental variables estimates 6% of individuals who lose benets due to the reassessment end up in the public works program during the years after the reform.
Panel (e) of Figure 2 shows an initial jump, followed by a gradual decline in the number of beneciaries who are not employed or receiving any benets.These results suggest that after the initial loss of benets, some beneciaries were able to quickly nd employment (or remain employed if they were already working), while a signicant share initially remained without a job but were able to nd employment later on.Column (5) of Panel A of Table 3 shows that the overall increase in the probability of having no income from DI, employment, or public works increased by 0.5 percentage points or one-third of those removed from benets.
Year-by-year instrumental variables estimates show a decline in the impact of benet loss on the share of those without income for employment, public works or benet from 60% in 2012 to about 20% in 2015, with a post-reassessment average of 35.7% as displayed in column (3) of panel A of Table 4.
Overall, our results suggest that relative to unaected DI recipients just above the age cut-o for reassessment aected beneciaries just below the cut-o lost their benets at substantially higher rates.Outcomes varied signicantly among individuals losing benets: about 60% were employed in the primary labor market following benet loss, while a third were left without a job or any benets, the public works program only accommodating a small share.These results suggest the potential presence of important heterogeneity across types of beneciaries which we turn to in Section 5.2.
Placebo analysis.In order to further probe the validity of our main results, Figures 4  and 5 present two sets of placebo results.Figure 4 replicates our main results presented in Figure 2 for DI categories that were not aected by the reassessment policy.Figure 5 replicates the same results but for a placebo reform in 2011.
Figure 4 shows DI coverage and placebo regression results for individuals who belonged to more severe and hence unaected DI categories in December 2011.The gure shows that while the pre-reform trends deviated slightly between the placebo treatment and control groups (although none of the dierences are signicant at the 5% level), there were no statistically signicant post-reform dierences between the outcomes of the two groups.The patterns indicate that in the unaected DI categories the reform had no impact on the probability of benet receipt, employment, and having no income.11 The placebo results presented in Figure 5 indicate that for a placebo reform in 2011, there were no major pre-reform dierences between the placebo treatment and control groups.
Panel (d) suggests a very small, albeit statistically signicant, increase in employment among the placebo treatment group relative to the placebo control group.This increase is about a tenth of the magnitude of our main eects estimated for the real reform year in Figure 2.
There were no post-reform dierences in other outcomes.
These placebo analyses suggest that our main results are driven by the impact of the 2012 reassessment reforms rather than by spurious dierences that arise between our control and treatment groups or by other events that aect the two groups dierently.

Heterogeneity
To better understand the mechanisms underlying the broad eects of the reform documented so far, we turn to assessing the potential heterogeneous eects of the reassessment.We expect the reassessment and benet loss to aect beneciaries with dierent levels of attachment to the labor market in dierent ways.
We start by examining heterogeneity in outcomes by pre-reform employment.Importantly, approximately a quarter of benet recipients were concurrently employed in 2011, the last pre-reform year.We add terms capturing the interaction of treatment status with 2011 employment status to our reduced form equations ( 1) and ( 2).We also re-estimate the instrumental variables equation ( 4) separately on the previously-employed and non-employed samples.Panel B of Table 3 reports the eect of the reform on labor market outcomes by pre-reform employment averaged over the post-reform period from estimating the modied equation ( 1).Appendix Figure A1 shows year-by-year estimates from estimating the modi-ed equation ( 2).Panels B and C of Table 4 displays our instrumental variables estimates by pre-reform employment status.
The results reported in Panel B of Table 3 reveal that the overall decrease in DI receipt was driven by individuals who already had some employment while receiving DI benets in 2011.Within this group, which makes up approximately one quarter of recipients, DI receipt while employed decreased by 4.9 percentage points (column 2), employment without receiving benets increased by 2.3 percentage points (column 3), while the probabilities of participating in public works (column 4), or remaining without income from work or benets (column 5) increased by smaller magnitudes.Within this group, there was also a statistically insignicant 1.7 percentage points increase in the probability of receiving benets without employment (column 1).
Among the group of beneciaries not working in 2011, the patterns are dierent: DI receipt decreased by 0.9 percentage point, and employment without receiving benets increased by only 0.4 percentage point.The IV regression results reported in Panels B and C of Table 4 show that among individuals who lose their benets as a consequence of the reform, labor market outcomes dier markedly by pre-reform employment.Panel B shows that among those with no pre-reform employment, 33.1% end up working after losing bene-ts, while 60.8% are not working but also not receiving benets.At the same time, as Panel C shows among those with some pre-reform employment 80.8% are working and only 13.5% end up with no employment or benets.Approximately 6% of both groups end up in the public works program.
In a similar vein, we investigate heterogeneity with respect to several other individual-and region-specic characteristics that might moderate the impact of the reform on DI and employment outcomes.Appendix Table A2 and Appendix Table A3 show these results.In both tables, we replicate our baseline results in Panel A.
In both tables, Panel B presents the results for individuals with low versus high prereform spending on prescription drugs, a proxy for health.Here we dene high spending as individuals whose annual spending was at least as high as the sample median in 2011.
Appendix Table A2 shows that the impact of the reassessment on employment outcomes was concentrated in the group of relatively healthy individuals, which is consistent with healthier individuals being more likely to lose their benets.12At the same time, the instrumental variables estimates in Appendix Table A3 suggest that the impact of benet loss on outcomes was similar among healthier and less healthy individuals, with the exception of public work, which increases only in the healthier group.
Panel C of both tables shows results by occupation groups.We group individuals into skilled, unskilled and missing occupation categories based on the last observed pre-reform occupation.Occupation information is missing if no employment history is observed for an individual since January 2003.Close to half (48%) of our sample belong to this category.
The results for skilled and unskilled workers are fairly similar.One exception is public work participation: Appendix Table A3 shows that benet loss increases the probability of public work participation for unskilled workers but not for skilled workers.In sum, these heterogeneity results suggest that employment status while receiving ben-ets and health were the key determinants of being removed from benets.Once removed from benets, prior employment was the main driver of labor market success.Most individuals who were already employed while on benets were able to remain employed, while most of those who were not working while on DI remained out of work while also losing their benets.These ndings point to the importance of improving labor market attachment while on benets.

Additional Results
Job Quality.The sudden loss of income compels expelled beneciaries to promptly search for employment.However, this rush can lead to lower-quality employment, for example, in the form of lower wages (Nekoei and Weber, 2017).The risk of human capital depreciation and a potential stigma eect can also lead to employment in lower quality jobs even in the case of successful job placement.
To investigate the quality of jobs held by individuals who leave DI due to the reform, we re-estimate equation ( 4) with employment at jobs with dierent quality attributes as dependent variables.We estimate the eect of DI exit on the following four outcome variables: ( We then divide the estimated quality-specic employment eects with the total estimated eect of DI exit on employment, to obtain the share of employment eect that falls into the specic employment category.We compare this estimated share with the pre-DI share of treatment group individuals who were employed at the specic employment category (conditional on employment).With this approach, we provide insights on whether people who found employment after leaving DI as a consequence of the reform held worse quality jobs than their typical pre-DI jobs.
Figure 6 shows our results.Panel (a) shows that relative to a pre-DI mean of 77%, on average 72% of the employment eect came from jobs paying above the minimum wage.53% of the employment eect came from full-time jobs according to panel (b), signicantly lower than the pre-DI mean of 78%.Panel (c) shows that 51% of the employment eect came from skilled jobs, well below the pre-DI mean of 73%.Finally, panel (d) shows that 17% of the employment eect came from employers with above-median TFP, half of the pre-DI mean of 33%.The dierences between the quality-specic employment eects and pre-DI means are more striking among those who had no employment in 2011.These results indicate that even individuals who were able to secure employment among the population whose benet was terminated as a result of the reform experienced a deterioration in the quality of their jobs.
Results for women.We exclude women from the analysis of the impact of the reform because due to an early retirement option available for women only, the labor force outcomes of the control and treatment group may evolve dierently, as the early retirement option is more likely to be available in the (older) control group.Despite this concern, the results reported in Appendix Figure A2 indicate qualitatively similar reform eects for women as for men (Figure 2).Similarly, the IV estimates for the eect of DI exit on labor market outcomes for women, reported in Appendix Table A5, are similar to the results for men (Table 4).
Effects of the reform on healthcare use.Appendix Figure A3 shows the time pattern of the impact of the reform on healthcare use.These results suggest that there was a jump in GP visits, outpatient specialist visits, and the number of hospital days among treated individuals when the policy came into eect.This is likely explained by participation in the reassessment process.We do not see a similar jump in prescription drug spending.We also see that by 2013 (i.e., one year after the reform came into eect), the dierences between the treatment and control group disappeared.We observe a small permanent increase in outpatient specialist care use -an increase by around 0.4 visit per quarter.Overall, these results suggest that the reform did not have major permanent eects on healthcare use, suggesting that the reform also did not have major health eects (assuming that health deterioration would be reected in higher healthcare use).

Conclusion
This paper provides evidence on the labor market implications of a major reform that aimed to improve the targeting of disability benet receipt by tightening eligibility conditions and reassessing benet entitlement for a large share of beneciaries.We identied the eects of the reform using the fact that the reassessment only applied to beneciaries under an age cut-o and below a certain level of health impairment.
Our results suggest that while the reform decreased disability insurance receipt in the reassessed population, the resulting increase in employment was modest for those with no pre-reform employment in the age groups close to the age cut-o of the reform.The majority of reassessed beneciaries who were not employed pre-reform were left without any income after losing their benet.Further, those who returned to employment typically worked in lower quality jobs than pre-DI.
Overall, while the stricter disability benet rules proved eective in reducing the number of disability recipients, the reform failed to activate those who were not employed pre-reform and thus had weaker ties to the labor market and were likely to be less employable.These results suggest that nancial incentives for reactivating disability benet recipients may
where i indexes individuals, t indexes months, [Y ear t ≥ 2012] is an indicator for the post reform period, [AGE i = 56] is an indicator for the treatment group, and the µ t are month xed eects.Our coecient of interest is β DiD , the dierence-in-dierences estimator, which captures the dierential change in labor market outcomes for treated relative to control individuals.
56] is an indicator for the treatment group, and the µ t are month xed eects.Our parameters of interest are β t , which capture the dierential change in labor market outcomes for treated relative to control individuals in each month relative to December 2011.In our reduced-form analyses, we focus on ve mutually exclusive and exhaustive binary outcome variables Y it : (1) DI & no employment; (2) DI & employment; (3) employment & no DI; (4) public work & no DI; (5) no DI & no employment & no public work.
shows the month-by-month dierence between control and treatment individuals for each of the labor market outcomes from estimating equation (2).It suggests that there were no signicant dierences in the evolution of labor market outcomes before the 2012 reform.The outcomes of treated individuals start to diverge in 2012, with the biggest change occurring in May, in line with the reform timeline which required benet recipients to declare by March their intention to undergo reassessment or lose benets from May.Panel A of Table 3 reports the eect of the reform on labor market outcomes averaged over the post-reform period from estimating equation (1).The sum of the ve point estimates is zero, reecting the mutually exclusive and exhaustive nature of the ve outcome variables.Similarly, Panel A of Table 4 reports the instrumental variables estimates of the eect of DI exit on labor market outcomes pooled over the post-reform period from estimating equation (4).The sum of the three point estimates is one due to the mutually exclusive and exhaustive nature of the outcome variables.Year-by-year instrumental variables estimates are shown in Panel (a) of Figure 3. Panels (a) and (b) of Figure 2 show the change in DI status, breaking the overall eect displayed in Figure 1 down into two categories by concurrent employment status.Panel (a) and column (1) of panel A of Table suggest that among individuals who exit the DI program due to the reassessment, the share of those employed without receiving benets increased from 40% in 2012 to over 70% in 2015.Consistent with the dierence-in-dierences estimates, over the post-reform years on average 58% of those who exit due to the reassessment are employed in the open labor market without receiving benets as displayed in column (1) of panel A of Table 4. Panels (d) and (e) of Figure 2 show the outcomes of recipients who lost their benets but were not employed in the open labor market.Panel (d) suggests that some of those who lost benets end up in the public works program.Over the 2012-2015 period, the average increase in public works employment is 0.1 percentage points (Column 4 of Panel A of Table Panel D displays results by the length of time spent on DI before the reform.DI length is measured as the time between the individual's rst DI entry and December 2011.We estimate our results separately for individuals who received DI benets for more or less than 10 years.The results are fairly consistent across groups with shorter and longer durations on benets.Finally, Panel E compares individuals in low-and high-unemployment areas.We distinguish between high and low unemployment groups depending on whether the unemployment rate in the individual's micro region was above or below the median in 2011.The results are similar for the two groups. 1) employment earning above the minimum wage & no DI; (2) full-time employment & no DI; (3) employment in a skilled job & no DI; (4) employment at a rm with above median TFP & no DI.

FigureFigure 3 :
Figure 1: DI Status Figure A1: Eect of the Reform Over Time-Heterogeneity by Pre-Reform Employment Status (a) DI & No Employment Figure shows our estimates of the impact of the reassessment policy on the outcomes of treated workers below the age cut-o relative to control workers above the age cut-o.Figure displays the estimated βt coefficients from equation (2) with 95% condence intervals over 2009-2015, with the rst quarter of 2011 as the reference quarter.Sample is restricted to men who received DI throughout 2011, and belonged to the aected DI categories in December 2011.Treated people were aged 56 in December 2011 and control people were aged 57 in December 2011.
Figure shows our estimates of the impact of losing DI benets on the outcomes of aected workers.Figure displays the estimated β IV coefficient from equation (4) with 95% condence intervals estimated separately for each year 2012-2015 and by 2011 employment status.Sample is restricted to men who received DI throughout 2011, and belonged to the aected DI categories in December 2011.Figure shows the share of employment eects of leaving DI by job quality.Gray bars display the β IV coefficient estimates of equation (4), capturing the eect of leaving DI on employment in a specic job category (job paying above the minimum wage, full time job, skilled job, employer having above median TFP), instrumented with being aged 56 versus 57 in December 2011, and divided by the IV estimated eect on overall employment.Red lines indicate 95% condence interval.Sample is restricted to men who received DI throughout 2011, and belonged to the aected DI categories in December 2011.Sample is split by having some employment in 2011, which indicator is set to one for people who had at least one month of employment, including self-employment, in 2011.Blue dots display the pre-DI mean outcome of individuals in the treatment group (age 56 in December 2011), restricting the pre-DI sample to months of employment.

Table 1 :
Health Revision Obligation Cut-OsTable shows the health revision cut-os by health impairment and age. Note:

Table 3 :
Eect of the Reform-Dierence-in-Dierences Estimates Note: *** p<0.01, ** p<0.05, * p<0.1.Cluster-robust standard errors in parentheses.Table displays the β DiD coefficient estimates of equation (1), showing the average treatment eect over 2012-2015.Sample is restricted to men who received DI throughout 2011, and belonged to the aected DI categories in December 2011.Treated people were aged 56 in December 2011, control people were aged 57 in December 2011.In Panel B, the binary heterogeneity indicator of some employment in 2011 is set to one for people who had at least one month of employment, including self-employment, in 2011.

Table 4 :
Eect of DI Benet Loss-Instrumental Variables EstimatesNote: *** p<0.01, ** p<0.05, * p<0.1.Cluster-robust standard errors in parentheses.Table displays the β IV coefficient estimates of equation (4), capturing the eect of leaving DI, instrumented with being aged 56 versus 57 in December 2011.Sample is restricted to men who received DI throughout 2011, and belonged to the aected DI categories in December 2011.In Panels B and C, the sample is split by having some employment in 2011, which indicator is set to one for people who had at least one month of employment, including self-employment, in 2011.
Figure displays the βt coefficient estimates of a yearly version of equation (2) interacted with employment in 2011, showing the treatment eects over 2009-2015, with 2011 as reference year.95% condence intervals are displayed.Sample is restricted to men who received DI throughout 2011, and belonged to the aected DI categories in December 2011.Treated people were aged 56 in December 2011, control people were aged 57 in December 2011.The binary heterogeneity indicator of some employment in 2011 is set to one for people who had at least one month of employment, including self-employment, in 2011.
: Eect of the Reform-Dierence-in-Dierences Estimates, Heterogeneity Table displays the β DiD coefficient estimates of equation (1) extended with heterogeneity indicators, showing the average treatment eect over 2012-2015.Sample is restricted to men who received DI throughout 2011, and belonged to the aected DI categories in December 2011.Treated people were aged 56 in December 2011, control people were aged 57 in December 2011.In Panel B, the binary heterogeneity indicator of low (high) drug spending in 2011 is set to one for people whose spending on medicines in 2011 is below (equal to or above) the sample median in that year.In Panel C, occupation classication is based on the last observed pre-reform employment.34% of the individuals are skilled workers (including both white and skilled blue collars), 18% are unskilled workers.Occupation information is missing for 48% of the sample.In Panel D, at least 10 years on DI is DI length measured up to December 2011 of 10 or more years, where 10 years is the sample median DI length in December 2011.In Panel E, high (low) unemployment is unemployment rate equal to or above (below) the median unemployment rate (16.7%) at the micro-region level in 2011.Appendix Table A3: Eect of DI Benet Loss-Instrumental Variables Estimates, HeterogeneityNote: *** p<0.01, ** p<0.05, * p<0.1.Cluster-robust standard errors in parentheses.Table displays the β IV coefficient estimates of equation (4), capturing the eect of leaving DI, instrumented with being aged 56 versus 57 in December 2011.Sample is restricted to men who received DI throughout 2011, and belonged to the aected DI categories in December 2011.Sample is split by heterogeneity indicators.In Panel B, the binary heterogeneity indicator of low (high) drug spending in 2011 is set to one for people whose spending on medicines in 2011 is below (equal to or above) the sample median in that year.
In Panel C, occupation classication is based on the last observed pre-reform employment.34% of the individuals are skilled workers (including both white and skilled blue collars), 18% are unskilled workers.Occupation information is missing for 48% of the sample.In Panel D, at least 10 years on DI is DI length measured up to December 2011 of 10 or more years, where 10 years is the sample median DI length in December 2011.In Panel E, high (low) unemployment is unemployment rate equal to or above (below) the median unemployment rate (16.7%) at the micro-region level in 2011.