Reemployment Bonuses, Unemployment Duration, and Job Match Quality

Transcription

1 Reemployment Bonuses, Unemployment Duration, and Job Match Quality Taehyun Ahn School of Economics, Sogang University Seoul , Korea July 2016 ABSTRACT This study examines the impacts of reemployment bonus, the incentive payments to unemployment insurance recipients who find a job within a specified period of time, using Korean data. A sharp discontinuity in treatment assignment at age 55 identifies the effect of increased reemployment bonus on unemployment duration and on subsequent job duration. The results indicate that increases in reemployment bonus boost the job-finding hazards of the claimants early in unemployment spells by 20.0% to 37.8%. Additionally, employment stability is not significantly affected by increased bonus, which implies no negative influence of the bonus on subsequent job match quality. Keywords: Reemployment bonus; Unemployment insurance; Regression discontinuity JEL classification codes: C41; J64; J65

2 I. Introduction The unemployment insurance (UI) systems in many countries are intended to provide short-run income support for involuntarily unemployed workers while they search for work. The UI benefits enhance the welfare of risk-averse individuals affected by adverse employment shocks by smoothing consumption. However, at the same time, UI may induce moral hazard and create disincentives to rapid reemployment. The payment of UI benefits may cause unemployed individuals to search less intensively for new jobs than they would in the absence of UI, which increases the duration of unemployment. 1 The balance between incentives to job search and consumption smoothing through insurance is an important concern in designing the UI program. In response to search disincentives, many countries have traditionally imposed various job search requirements. More recently, policymakers have considered and introduced additional services for UI claimants, including job search assistance, job search monitoring, and financial incentives (OECD 2013). In this study, I investigate the effects of reemployment bonus, the incentive payments to UI recipients who find a job within a specified period of time, on the job-finding rate and on subsequent job duration, using Korean data. The reemployment bonus in the UI system is often considered as a way to increase job search efforts among unemployed workers and has been tested mostly by field experiments conducted in the United States (Wandner 2010). The results from the US cash bonus experiments are mixed and show no conclusive evidence on the effectiveness of the bonuses (Woodbury and Spiegelman 1987, Decker 1994). On the basis of outcomes from the field experiments, Meyer (1995, 1996) argues that, while monetary bonus affects the speed with which unemployed workers find a job, a permanent 1 See Hopenhayn and Nicolini (1997) and Burtless (1990) for a theoretical analysis and a survey of earlier literature on disincentives and UI. See also Tatsiramos and Van Ours (2014) for a recent overview of theoretical and empirical studies. 1

3 bonus program may increase incentives to file for UI benefits, the extent of which cannot be assessed by experimental studies. Korea is one of the few countries that administers a nationwide reemployment bonus program in the UI system. Recently, the country introduced an age-based payment structure in which a higher bonus rate is applied to individuals aged 55 or older, which provides a rare opportunity to identify the causal effects of reemployment bonus. In the estimation, I examine the differences in the job-finding hazard and reemployment duration between individuals just below and above age 55 before and after the bonus policy change, by combining a regression discontinuity design (RDD) and the difference-in-differences (DID) framework. A key feature of my analysis is that the identification strategy relies on a sharp discontinuity in the reemployment bonus rate, which increases from a half of residual UI benefits to two-thirds of the benefits at the age of 55. Comparing individuals who are unemployed and just below the age 55 with those who are just above this threshold is the idea behind the RDD framework. This strategy is intuitively appealing because both the groups of unemployed workers are virtually of the same age. Another important feature is that the analysis uses comprehensive data from the administrative sources that cover the universe of unemployment spells, including information for the period before the age-based bonus policy was introduced. Using this information is helpful, first, in implementing a convincing RDD strategy. Although the treated individuals are barely older than the non-treated ones, an important concern is that the treated individuals may have an inherently different job-finding rate because they may face a different labor market environment and potentially are eligible for other retirement and benefit arrangements in Korea. Comparing the differences in outcomes of the treated and non-treated individuals 2

4 prior to the policy change and then contrasting them to the corresponding differences after the policy change, which is an idea of the DID framework, is an effective way to make the RDD framework persuasive. Second, the data is useful in assessing a potential entry effect suggested by Meyer (1995, 1996). The possibility of increased incentives to file for UI benefits, which cannot be assessed by experimental studies, can be verified by examining the pattern of the inflow into unemployment around the age threshold before and after the introduction of the age-based bonus policy, using the administrative records. In an attempt to examine the influence of increased reemployment bonus on job match quality, I also estimate a model of transition hazard from employment back to unemployment. I extend the existing research in this dimension because of the concern that the treated individuals are less selective during the period in which reemployment bonus is eligible, which may negatively influence the match quality of the subsequent job. A low match quality potentially leads to a higher probability of leaving a post-unemployment job and ultimately undermines the effectiveness of the reemployment bonus program. The data display a significant discontinuity in the job-finding rate at the age-55 threshold after the bonus policy change: the treated individuals, who are just above age 55, are substantially more likely to find a job before exhausting their UI benefits than non-treated individuals, who are slightly below this threshold. The data, however, reveals no significant discontinuity around the age threshold in the old bonus scheme. In addition, I find no evidence on manipulation of timing of entry into unemployment, which supports the validity of the identification strategy. In light of these patterns found in the data, I estimate discrete-time duration models for the transition from unemployment to employment and for that from employment back to unemployment. The estimates from the hazard models indicate that increases in 3

5 reemployment bonus boost the job-finding hazard of the UI claimants early in unemployment spells during the bonus qualification periods by 20.0% to 37.8%. Additionally, the estimates show that employment stability is not significantly affected by increased bonus, which implies no negative influence of the bonus on subsequent job match quality. These findings provide positive evidence for the reemployment bonus program that can be used as a viable tool to shorten unemployment spells and to promote rapid reemployment. II. Background A. Studies on the effectiveness of reemployment bonus Previous studies on the effects of reemployment bonuses have been mostly based on field experiments conducted in the US. The first pilot experiment was conducted in Illinois in In the experiment, randomly selected UI recipients were offered a $500 bonus if they found a job within 11 weeks (qualification period) and remained in the new job for at least 4 months. Results suggest that this experiment reduced the spell of unemployment benefits, averaged over all assigned recipients, by more than one week and that the state UI offices appears to save money even after accounting for the cost of bonuses (Woodbury and Spiegelman 1987, Spiegelman and Woodbury 1987). These positive findings induced the government to test the employment bonuses further in New Jersey ( ), Pennsylvania ( ), and Washington (1988). Unlike the case of Illinois, where the bonus offered a fixed amount during the qualification period, the New Jersey experiment had a payment structure in which the bonus payment declined over the duration of unemployment, so that the bonus received was greater the earlier that reemployment occurred. In addition, job search assistance (JSA) services were combined with bonus payment for the treated claimants. The findings suggest that a declining 4

6 bonus structure has a greater impact on short-term claimants than on long-term claimants who are likely to exhaust their benefits (Decker 1994, Anderson 1992). The experiments conducted in Pennsylvania and Washington tested several different combinations of bonus amounts and qualification periods. The outcomes from these experiments reveal that while the impacts are the largest for the most generous bonus with the longest qualification period, average reductions in the amount of time spent on UI is modest about half a week per claimant suggesting that the bonus program is not a costeffective method of speeding reemployment (Decker and O'Leary 1995). Evaluating Pennsylvania and Washington experiments, Decker and O'Leary (2005) argue that adopting a low bonus amount with a long qualification period, targeted toward UI recipients who are most likely to exhaust their benefits, can be a cost-effective way for reducing benefit payments. While the results from the US field experiments are mixed regarding the effectiveness of the bonus program, Meyer (1995, 1996) points out that a permanent bonus program may increase incentives to file for UI benefits and foster inflow to unemployment. First, the currently unemployed individuals who do not file could respond to the monetary incentives to file for UI benefits. Second, those currently not unemployed but changing jobs could begin work slightly later and file for UI benefits. Third, firms could respond to the increased compensation for short UI spells and may change their layoff policies. Because the possibility and the degree of the entry effects are difficult to be assessed by field experiments, it is unclear whether the impacts of bonuses estimated from experimental studies appropriately reveal the impacts of a permanent bonus program. To my knowledge, the study by Van der Klaauw and Van Ours (2013) is the only nonexperimental study that examines the effects of reemployment bonus. They use 5

7 administrative data on welfare recipients in Rotterdam, where reemployment bonus schemes as well as financial sticks in the form of benefit sanctions were implemented during They find that the benefit sanction significantly raises the exit rate from welfare, while reemployment bonuses have no significant effects. The bonus program in Rotterdam, however, was targeted for the long-term welfare recipients who were not entitled to any other social insurance benefits such as UI benefits or disability insurance, and was offered only to those who had been on welfare benefits for at least one year. Thus, it is unclear whether the estimated results can be directly applied to a bonus program aimed at speeding the reemployment process for UI recipients before exhausting UI benefits. B. The UI system and reemployment bonus program in Korea The UI system of Korea was introduced in July 1995 after a long debate regarding the issues of increasing financial burden to businesses and the possibility of adverse selection and moral hazard. Initially, the UI system only covered workplaces employing 30 or more workers. The coverage of UI, however, expanded rapidly throughout the 1997 Asian financial crisis to all workplaces since October 1998 (Kim 2010). In the Korean UI system, only those who are involuntarily unemployed can file for the UI benefits. The UI benefit payments usually is a half of the average wage at the previous job. The maximum duration of benefit depends on the age at which job seekers become unemployed and the length of the contribution period. 2 The maximum duration for the age group 50 65, which is the subject of the analysis, is 150 days (5 months) if the period of contribution is 1 to 2 years. The longer the contribution period, the longer the maximum duration; the maximum duration is 180 days (6 months) for the contribution period of 3 to 4 2 The length of the contribution period is close to the tenure with the previous employer because of the fact that the accumulated contribution period is set to zero once an individual leaves the job and files for the UI. 6

8 years, 210 days (7 months) for 5 to 9 years, and 240 days (8 months) for 10 years or longer. The maximum duration of each contribution group decreases by 30 days (1 month) for the younger group aged For the unemployed workers younger than 30 years, the maximum duration further decreases by 30 days. From the beginning of the UI system in Korea, reemployment bonuses have been offered to the UI claimants. The basic structure of the reemployment bonus is to provide certain fractions of residual benefits to the UI recipients who find a job that lasts for a long time. In February 2010, the bonus scheme was revised in some dimensions to enhance the effectiveness of incentive and to simplify the payment structure. First, the condition for receiving a bonus was tightened to encourage speedy reemployment. In the old scheme, the UI recipients who find a job that lasts 6 months or longer can receive reemployment bonus. In the revised bonus scheme, however, only those who find a job 30 days (one month) before exhausting their benefits are eligible for the bonus. Second, the rule that determines the bonus rates, which had been somewhat complicated, was simplified. Previously, the bonus rates were differentiated by the time at which the recipients were reemployed. 3 In the new scheme, the amount of bonus payment is simply a half of the residual UI benefit. Third, importantly to my identification strategy, an exceptional bonus rate two-third of the residual UI benefit is applied to the UI claimants aged 55 or older. III. Model specification In analyzing the effect of the reemployment bonus, I specify discrete-time duration 3 If an individual finds a job within a third of the maximum benefit duration, the person receives two-third of the remaining benefit. Alternatively, an individual who is reemployed between a third and two-third of the maximum benefit duration receives a half of the remaining benefit. Finally, if an individual finds a job before exhausting the UI benefit, but later than two-third of maximum benefit duration, the person receives only a third of the remaining benefit. 7

9 models for the transition from unemployment to employment and for that from employment back to unemployment. I use discrete time spell data with monthly (30 days) intervals to allow heterogeneity in treatment effect depending on the months spent in unemployment. Assuming that the monthly duration of an unemployment spell is represented by a non-negative random variable T u i, which takes on only integer values, the hazard rate of transition from unemployment to employment for individual i in month t, conditional on survival up to t, can be written as where t y t P T t T t 1, y t (1) u u u u u i i i i i k k u u u u u u u u 2 i 0 1 d id 1d id i 2 i 0 3 i 0 d 2 d 1 u u 2 u u u 4Di Agei Age0 5Di Agei Age0 6xi t i y t I t I t D Age Age Age Age The effect of duration dependence of unemployment is modeled by including a set of dummy u variables, I () t, indicating that the current period is in duration intervals denoted by id subscript d. Because there is a sharp discontinuity in reemployment bonus, I include the treatment indicator Di, which is equal to one if the individual is above the age threshold of 55 (Age0) at the time of entering unemployment, and zero otherwise. To allow the effect of treatment to vary over time in unemployment, I interact the treatment indicator Di with the u u duration dummy variable I () t. The parameter 1 captures the causal effect of the increase id in reemployment bonus on the hazard of transition to employment. flexible way, which ensures that the coefficient for the treatment variable identifies the causal 8 d To control for the effects of the assignment variable Agei on the hazard of transition to employment, I include linear and quadratic terms in Age i Age 0 (2) and their interaction terms u u with the treatment indicator. Thus, the coefficients 2 to 5 capture general age effects in a

10 effect of reemployment bonus. The vector u xi t represents the observable characteristics, including gender, years of schooling completed, and the information on the previous job, such as the level of daily earnings and occupational category, plus dummy variables for each year and quarter and dummy variables indicating region of residence. The last term, u i, represents unobserved individual heterogeneity that influences the hazard of transition from unemployment to employment. The basic model described in equation (2) assumes that the effect of age is continuous around the age threshold before the age-based bonus policy was introduced. The sensitivity of the results based on the basic model can be assessed by using data before the policy change. I, therefore, specify an extended model the before-after-rdd (BA-RDD) specification which can be described as k k u u u u u u u i 0Posti 1 d id i 1 d id iposti 2 i 0 i d 2 d 1 u 2 u u 2 3 i 0 i 4 i i 0 i 5 i i 0 k k u u u u u u u 2 6 7dIid t 2dIid t Di 8 Agei Age0 9 Agei Age0 d 2 d 1 u u 2 u u u 10Di Agei Age0 11Di Agei Age0 12xi t i y t I t Post I t D Age Age Post Age Age Post D Age Age Post D Age Age Post where Posti equals one if the individual becomes unemployed after the policy change, and u zero otherwise. The parameter 1 provides the estimate on the causal effect of reemployment bonus on the hazard of transition to employment. d In addition to the transition from unemployment to employment, an important purpose of this study is to examine the effect of increased bonus on the stability of subsequent employment. 4 Again assuming that the monthly duration of an employment spell is described (3) 4 It is also useful to examine the reemployment wage in assessing the match quality of the subsequent job. Unfortunately, reemployment wages are available only for those who leave 9

11 by a non-negative integer random variable T e i, the hazard rate from employment back to unemployment for individual i in month t is given as where e e e e t y t P T t Ti t 1, yi t (4) e i i i k k k e e e e e u u e u u e i 0 1 d id 2d id i 1d id i i 3 i 0 d 2 d 2 d 1 e 2 e e 2 e e e 4 Agei Age0 5Di Agei Age0 6Di Agei Age0 7xi t i y t I t I T I T D Age Age Similar to the models for transition from unemployment to employment, I flexibly model the effect of duration dependence by including month dummy variables, e Iid (5) t which indicate time in the current reemployment spell. To control for the effect of the preceding unemployment duration on the reemployment spell, I add dummy variables for the length of u u unemployment duration denoted as id i I T. The main concern regarding the reemployment spell is that individuals in the treatment group potentially are less selective during the period in which reemployment bonus is eligible, which may negatively influence the match quality of the subsequent job and potentially decrease the duration of the reemployment spell. To capture the effect of treatment on reemployment, I interact the treatment indicator, Di, with the dummy variables, u id u i I T, indicating the time interval in which individuals exit unemployment for a job. The causal effect of increased reemployment bonus on the stability of the subsequent job is represented by 1 ed characteristics, e xi. The model for reemployment spell also includes controls for observable t, and for the differentiated effects of age between the treated and the their reemployment jobs, because wages are recorded when the unemployed file for UI benefits. 10

12 non-treated individuals. Finally, e i represents unobserved heterogeneity that influences the subsequent job stability. To control for potential discontinuity around the age threshold prior to introducing the age-based bonus scheme, similar to the transition process from unemployment to employment, I specify an extended model for the reemployment duration the before-after- RDD (BA-RDD) specification using data before the policy change, which is described as k k e e e u u e u u e i d id i d id i i i d 2 d 1 0 i 1 i 1 i 2 0 y t Post I T Post I T DPost Age Age Post Age Age D Age Age D Age Age e 2 e e 2 3 i 0 i 4 i i 0 i 5 i i 0 Post Post Post k k k 7dIid t 8dIid Ti 2dIid Ti Di Agei Age0 e e e e u u e u u e 6 9 d 2 d 2 d Age Age D Age Age D Age Age x t e e e e e e 10 i 0 11 i i 0 12 i i 0 13 i i i i (6) In this specification, the parameter 1 ed provides the estimate on the causal effect of reemployment bonus on subsequent employment stability for individuals who left unemployment for a job in time interval d. Following a standard approach in the literature (Bover, Arellano and Bentolila 2002, Ham and Lalonde 1996, Steiner 2001), I specify the discrete-time hazard rate of transitions for individual i as follows: j i 1 j j t yi t 1 exp yi t. (7) In equation (7), the subscript j represents the type of transition: j = u denotes unemployment to employment and j = e denotes employment to unemployment. As an alternative specification, I consider a linear version of a discrete-time hazard model by using the identity function as a hazard function j i j t y j i t yi t. I additionally estimate a linear specification because it provides easier interpretations of the treatment effect (although treatment is assumed to be related to the probability of outcomes in a linear fashion) and 11

13 coefficients in a linear model directly measure marginal effects for the probability of transitions. Finally, I assume that unobserved individual heterogeneity, u e i and i, which influence the duration of unemployment spell and reemployment spell, respectively, are independently and normally distributed with zero means and constant variances. 5 Furthermore, I compute standard errors that account for nonindependence within individuals. IV. Data A. Data construction The data used in this study are drawn from the UI database that is maintained by the Korea Employment Information Service. The database contains several administrative files, such as records of employment and the UI administrative files, which enable researchers to construct event history data about employment and unemployment spells. Among the UI administrative files, the UI claim and payment files provide basic demographic information on a given individual, such as age, gender, and level of schooling, as well as job characteristics for the job held before entering unemployment, such as daily earnings and type of occupation, as well as the period of contribution before the current unemployment. For the analysis, from UI administrative files, I extract personal and job information regarding individuals entering unemployment from a job between January 2006 and July 5 By allowing correlation between u e i and i, the random effects specification can also take account of the possibility that the unobserved individual factors that affect the transition from unemployment to employment have an additional influence on the subsequent job stability. However, I do not jointly estimate the two transition equations in this study. Before choosing the current specification, following Lillard (1993) and Lillard and Panis (1996, 1998), I estimated a simultaneous system of equations for hazards, which allows the correlation between two individual-specific random terms, but found that for this specific model, they are uncorrelated. 12

14 2013, and merge it with records of employment to construct individuals event history data on unemployment duration and the subsequent employment spell. I restrict the attention to individuals aged between 54 and 56 years with a particular focus on the period since February 2010, when the discontinuity in reemployment bonus rate based on the age-55 cutoff became effective. I further restrict the sample to individuals who have been continuously employed and contributed to the UI for at least one year, but no longer than ten years prior to entering unemployment. I exclude the unemployed who have been continuously employed for more than ten years because they are also eligible for national pension benefits and can choose to retire and claim early pension benefits from the age of 55, albeit with some penalty in terms of benefit levels. Table 1 presents summary statistics for the key background variables included in the hazard models. The data indicate that among 76,630 individuals aged years, who became unemployed since February 2010, about 47% (35,859) comprise the treated individuals aged 55 or more. Table 1 also shows that the individuals in the (non-treated) control group, who are slightly younger than 55, have demographic characteristics similar to those of the treated individuals. The share of females in the treated group is 42%, and it is close to the corresponding share for the non-treated group (43%). In addition, job losers in both groups have completed, on average, similar years of schooling (12.25 for the non-treated versus for the treated). Furthermore, with respect to previous job characteristics, the unemployed in the treated group are quite comparable to the non-treated group in terms of average daily earnings (61.39 versus 61.63) and occupational distribution. The distribution of contribution period reveals that the treated individuals have a slightly longer period of contribution than the non-treated ones, which reflects the fact that the treated job losers are slightly older and have a slightly longer period of tenure at previous job than the non-treated 13

15 individuals before entering unemployment. To explore selectivity issues in detail, in the next subsection, I examine whether the assignment to treatment is random around the age threshold. B. Descriptive evidence In this subsection, I provide graphical evidence on the effects of increasing reemployment bonus on unemployment spell. In addition, to investigate whether the assignment to treatment is random and well controlled and to explore the possibility of entry effects, I examine the influences of changing reemployment bonus (treatment) on the inflow size and composition of inflow. Figure 1A reports the ratio of UI recipients who find a job before exhausting their benefit after the policy change (since February 2010) by age, at entry into unemployment, for each age week from 54 years and week 1 to a week before 56 years. The figure overlays a fit of fourth-order polynomial regression on both sides of the age-55 threshold, allowing for a discontinuity at the threshold. As seen in the figure, treated individuals are more likely to find a job before UI benefits expire than non-treated individuals. There is a significant discontinuity in the job-finding rate at the age-55 threshold: the difference in the probability between the group below and above the age threshold is (standard error ) and statistically different from zero, which implies that treated individuals have a 21% smaller job-finding probability than control individuals (0.24). Is the discontinuity in job-finding rate at age 55 observed in Figure 1A driven by the difference in reemployment bonus? This is true only if the counterfactual job-finding probability, without differences in reemployment bonus, varies smoothly with age and does not increase at age 55. Figure 1B shows the job-finding rate by age for individuals who became unemployed before the policy change (from January 2006 until January 2010). The 14

16 graph reveals little difference in the job-finding rate between the two age groups. More importantly, there is no significant discontinuity around the age threshold: the difference in the job finding probability between age groups just below and above 55 years is (standard error ) and is not statistically different from zero. Thus, the evidence in Figure 1B suggests that comparing reemployment probability in groups at both sides of the age threshold is a useful identification strategy under the revised bonus scheme. Another important concern with the identification strategy is the possibility of manipulating timing of unemployment entry for those who are close to the age-55 cutoff. While it is difficult to delay the timing of losing one s job by oneself, it is potentially possible that firms and individuals together can manipulate the age at entry into unemployment. Thus, I examine whether the relative size of the inflow into unemployment is influenced by introducing the age-based bonus scheme. Figure 2 shows that the relative inflow the ratio of density of age in the period before the policy change to that after the policy change is close to 1 throughout the age range and indicates no discontinuity around the age threshold. As the final check for validity of the identification strategy, I examine the composition of the inflow into unemployment nearby the age threshold. For using discontinuity based on age cutoff as a convincing strategy, the composition of the inflow pool needs to be balanced around the threshold in terms of background characteristics. Figures 3A to 3C report the inflow composition with respect to key variables included in the hazard analysis: the female share, years of schooling, and daily earnings. As shown in Figure 3A, the share of female is around 0.42 and remains stable in the year age bracket. Importantly, there is no discontinuity at the age 55 threshold. With regard to educational attainment, the average years of schooling appears to slowly decline with age; Figure 3B, however, indicates no discontinuity in schooling years at the age threshold ( with standard error ). 15

17 Furthermore, Figure 3C shows no discontinuity at the age cutoff with respect to the average daily earnings from employment: the difference is with standard error The patterns of inflow into unemployment described in this subsection provide quite convincing evidence that there is no manipulation in the running variable (age), which helps to identify causal effect of reemployment bonus. V. Estimation results A. Hazard model estimates In this section, I present the estimation results from hazard models for the transition from unemployment to employment and for that from employment back to unemployment. Although the models described in Section III are heterogeneous treatment specifications that allow the effect of treatment to vary over time of the unemployment duration, I first report the restrictive specifications that assume that the treatment effects are time-invariant in Table 2. Specifications 1 and 4 in Table 2 use data only of individuals who became unemployed after February 2010 when the bonus scheme was revised. The estimated coefficient for the treatment indicator Di in the transition hazard from unemployment to employment is and is statistically significant at the 1% significance level, which indicates that the job-finding hazard rises as reemployment bonus increases. However, the corresponding coefficient in the hazard model from employment to unemployment is small (0.0422) and is statistically insignificant at any conventional level, which suggests that reemployment bonus has little influence on job match quality. To check whether the treated individuals inherently have different hazards regardless of the bonus scheme, in specifications 2 and 5 in Table 2, I use the information only 16

18 pertaining to the unemployed under the old bonus scheme. The estimated coefficients in unemployment and reemployment hazards are small and statistically insignificant, which indicate that the treated individuals were no different from the non-treated individuals in terms of job-finding rate and employment stability. In specifications 3 and 6 of Table 2, using the information before and after the revision of the bonus policy, I present the estimates from BA-RDD hazard models. The estimate for the unemployment transition model is , which is statistically significant at the 1% level, and very close to the coefficient estimate of specification 1. This result confirms the previous finding that the treatment increases the probability of finding a job. The BA-RDD estimates for the employment transition model also confirms that employment stability is not significantly affected by the increase in bonus, which suggest no influence of reemployment bonus on job match quality. In the second panel of Table 2, I estimate hazard models separately by groups determined by the contribution period (1 2, 3 4, and 5 9 years) and present the coefficient estimates. While the effects of reemployment bonus on the job-finding hazard are positive in all contribution groups, the effect is marginally significant in the group with 3 4 years of contribution. With regard to employment hazards, the estimates reveal that reemployment bonuses have no significant influence in all contribution groups. To allow the effect of reemployment bonus to vary over the time of unemployment duration, I estimate the heterogeneous treatment models that interact the treatment indicator with time dummies of unemployment durations, as discussed in Section III. Table 3 presents these estimation results based in the BA-RDD model specification. In Specification 1, I pool all contribution groups and estimate transition hazards models. The estimated coefficients show that reemployment bonus has a positive influence on job-finding probability during the 17

19 first 5 months (150 days), but has no effect from the 6 th month of unemployment. This result suggests that the positive effect of reemployment bonus on unemployment hazard found in homogenous specifications is mainly driven by the impact on job-finding hazards early in the claimants unemployment spells. Consistent with findings from homogeneous treatment models, reemployment bonuses have no effect on the transition hazard from employment to unemployment. As described in Section II A, the maximum duration of benefit depends on the contribution period. The bonus qualification period, during which the UI claimant could qualify for the bonus, is 30 days (one month) shorter than the maximum duration. Thus, the claimants with 1 2 years of contribution have 120 days (4 months) of bonus qualification period; those with 3 4 years of contributions have 150 days (5 months) of qualification period; and those with 5 9 years of contributions have 180 days (6 months) of qualification period. In specification 2, I estimate transition hazard models by contribution group and report the estimation results. As shown in Table 4, the treated individuals are more likely to find a job during the bonus qualification period than the non-treated ones in all contribution groups. While the job seekers with a longer contribution period (5 9 years) have somewhat higher hazard of transition to reemployment in the 10 th and 11 th month, most treated groups show no higher exit rate from unemployment once the qualification periods expire. With respect to reemployment duration, reemployment bonus has no significant influence on the hazard of exit to unemployment regardless of the time at which an individual finds a job in all contribution groups. One exception is that the treated individuals with 3 4 years of contribution, who become reemployed at the 5 th month in unemployment, have a higher probability to exit from reemployment than the non-treated ones. However, it is unclear whether a higher hazard of exit represents a negative influence generated from the 18

20 reemployment bonus incentive because no higher job-finding hazard is found in the corresponding month in unemployment. Overall, the estimates show no significant evidence of reemployment bonus on subsequent employment stability. In Table 4, for a sensitivity check, I present the coefficient estimates from a linear version of a discrete-time hazard model. The results confirm the findings that reemployment bonuses increase the job-finding hazard, but have no influence on subsequent job stability. According to the estimates from homogeneous treatment models (Panel A), the treated individuals show significantly higher hazards of exiting from unemployment in most groups, although the estimated coefficient is imprecise for the individuals with 3 4 years of contribution. Moreover, the estimates from heterogeneous treatment models (Panel B) reveal that the positive impact of reemployment bonuses on job-finding hazards are effective and pronounced only during the bonus qualification period except for the group with a long contribution period. Concerning reemployment duration, Table 4 confirms the previous finding that increased reemployment bonuses have virtually no effects on exit hazard from employment. B. Predicted conditional probabilities To get a better idea of the effect of reemployment bonus on the job-finding hazard among the UI recipients, in Figure 4, I present the conditional probabilities of reemployment in the next month for a benchmark individual (based on estimates in Table 3). An individual is assigned the modes for all categorical variables, the means for all continuous variables, and various months of unemployment. Pooling all claimant groups, Figure 4A reveals that a nontreated individual has a job-finding probability of within two months of unemployment and of within five months of unemployment. If the individual is slightly older than 55 and receives treatment, these probabilities increase by 29.9% and 20.0% 19

21 to and , respectively. The difference in the job-finding rate, however, becomes insignificant after six months in unemployment. In Figures 4B 4D, I additionally present the predicted probabilities separately by contribution group. The figures also show the pattern that reemployment bonuses boost the job-finding hazards during the bonus qualification period by 21.8% to 37.8% while the difference is less pronounced in the group with 3 4 years of contribution. In all groups, the differences in job-finding hazard are concentrated in the earlier period of unemployment and tend to disappear in most cases after the qualification period. The results on the predicted hazard of reemployment suggest that reemployment bonus plays an important role in promoting early reemployment for UI claimants by creating an incentive to shorten unemployment spells to less than the full duration of the bonus qualification period. VI. Conclusion Korea recently implemented a unique reemployment bonus program by introducing an age-based reemployment bonus payment structure, in which a higher bonus rate is applied to individuals aged 55 or older, which provides a unique opportunity to identify the causal effects of reemployment bonus. Using comprehensive data from the administrative sources that cover unemployment spells, including information of periods before and after introducing the new bonus scheme, I examine the differences in the job-finding hazard and reemployment duration between individuals just below and above age 55, by combining a regression discontinuity design (RDD) and the difference-in-differences (DID) framework. The data reveals a significant discontinuity in the job-finding rate at the age-55 threshold after the bonus policy change: the treated individuals, who are just above age 55, are substantially more likely to find a job before exhausting their UI benefits than non-treated individuals. Consistent with the patterns found in the data, the estimation results show that 20

22 increases in the reemployment bonus boost the job-finding hazards of the UI claimants early in unemployment spells during the bonus qualification periods by 20.0% to 37.8%. Additionally, the estimates show that employment stability is not significantly affected by increased bonus, which implies no negative influence of the bonus on subsequent job match quality. One potential limitation of my study is that I evaluate reemployment bonuses only at the intensive margins by investigating the influences of the changes in bonus amounts, which is due to the inherent nature of policy changes experienced in Korea. Additionally, regarding the extent of the bonus effectiveness, generalizing the estimated results of my analysis to youth unemployment requires caution because the identification of the causal effect is largely based on the relatively older population. Nevertheless, the findings of this study provide some positive evidence for the reemployment bonus program that can be used as a tool to shorten unemployment spells, which may shed light on designing UI systems and public policies aimed at promoting intensive job search and inducing strong labor market attachment for the unemployed. 21

23 References Anderson, P.M. 1992, "Time-Varying Effects of Recall Expectation, a Reemployment Bonus, and Job Counseling on Unemployment Durations", Journal of Labor Economics, vol. 10, no. 1, pp Bover, O., Arellano, M. and Bentolila, S. 2002, "Unemployment Duration, Benefit Duration and the Business Cycle", Economic Journal, vol. 112, no. 479, pp Burtless, G.T. 1990, Unemployment Insurance and Labor Supply: A Survey, Brookings Institution. Decker, P.T. 1994, "The Impact of Reemployment Bonuses on Insured Unemployment in the New Jersey and Illinois Reemployment Bonus Experiments", Journal of Human Resources, vol. 29, no. 3, pp Decker, P.T. and O'Leary, C.J. 1995, "Evaluating Pooled Evidence from the Reemployment Bonus Experiments", Journal of Human Resources, vol. 30, no. 3, pp Ham, J.C. and Lalonde, R.J. 1996, "The Effect of Sample Selection and Initial Conditions in Duration Models: Evidence from Experimental Data on Training", Econometrica, vol. 64, no. 1, pp Hopenhayn, H.A. and Nicolini, J.P. 1997, "Optimal Unemployment Insurance", Journal of Political Economy, vol. 105, no. 2, pp Kim, S.T. 2010, "Korea s Unemployment Insurance in the 1998 Asian Financial Crisis and Adjustments in the 2008 Global Financial Crisis", ADBI Working Paper,. Lillard, L.A. 1993, "Simultaneous Equations for Hazards: Marriage Duration and Fertility Timing", Journal of Econometrics, vol. 56, no. 1-2, pp Lillard, L.A. and Panis, C.W.A. 1996, "Marital Status and Mortality: The Role of Health", Demography, vol. 33, no. 3, pp , "Panel Attrition from the Panel Study of Income Dynamics: Household Income, Marital Status, and Mortality", Journal of Human Resources, vol. 33, no. 2, pp Meyer, B.D. 1995, "Lessons from the US Unemployment Insurance Experiments", Journal of Economic Literature, vol. 33, no. 1, pp , "What have we Learned from the Illinois Reemployment Bonus Experiment?", Journal of Labor Economics, vol. 14, no. 1, pp O Leary, C.J., Decker, P.T. and Wandner, S.A. 2005, "Cost-Effectiveness of Targeted Reemployment Bonuses", Journal of Human Resources, vol. 40, no. 1, pp

24 OECD 2013, OECD Employment Outlook 2013, Organisation for Economic Co-operation and Development. Spiegelman, R.G. and Woodbury, S.A. 1987, The Illinois Unemployment Insurance Incentive Experiments, WE Upjohn Institute for Employment Research. Steiner, V. 2001, "Unemployment Persistence in the West German Labour Market: Negative Duration Dependence Or Sorting?", Oxford Bulletin of Economics and Statistics, vol. 63, no. 1, pp Tatsiramos, K. and van Ours, J.C. 2014, "Labor Market Effects of Unemployment Insurance Design", Journal of Economic Surveys, vol. 28, no. 2, pp Van der Klaauw, B. and Van Ours, J.C. 2013, "Carrot and Stick: How Reemployment Bonuses and Benefit Sanctions Affect Exit Rates from Welfare", Journal of Applied Econometrics, vol. 28, no. 2, pp Wandner, S.A. 2010, Solving the Reemployment Puzzle: From Research to Policy, WE Upjohn Institute. Woodbury, S.A. and Spiegelman, R.G. 1987, "Bonuses to Workers and Employers to Reduce Unemployment: Randomized Trials in Illinois", American Economic Review, vol. 77, no. 4, pp

25 Figure 1: Ratio of UI recipients who find a job before exhausting their benefits A. After the bonus scheme change Reemployment rate before exhaustion Reemployed before benefit exhaustion Age Discontinuity at threshold (standard error): (0.0099) B. Before the bonus scheme change Reemployment rate before exhaustion Reemployed before benefit exhaustion Age Discontinuity at threshold (standard error): (0.0106) 24

26 Figure 2: Inflow effect due to the age-based reemployment bonus scheme relatve inflow (after vs. before) Age Discontinuity at threshold (standard error): (0.0520) 25

27 Figure 3: Composition of the unemployment inflow A. Female (share) female share Age Discontinuity at threshold (standard error): (0.0115) B. Education (years of schooling) education (years of schooling) Age Discontinuity at threshold (standard error): (0.0320) 26

28 C. Daily wage daily wage Age Discontinuity at threshold (standard error): (1.8296) 27

29 Figure 4: Predicted hazard of transitions to employment A. All Predicted hazard of reemployment with 90 % CIs Probability Months control treatment B. Contribution period (1 2 years) Probability Predicted hazard of reemployment with 90 % CIs Months control treatment 28

30 C. Contribution period (3 4 years) Predicted hazard of reemployment with 90 % CIs Probability Months control treatment D. Contribution period (5 9 years) Predicted hazard of reemployment with 90 % CIs Probability Months control treatment 29

31 Table 1: Summary statistics for variables used in the hazard models All Age<55 Age 55 Variable Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Female Years of schooling Daily earnings (KRW 1,000) Occupation Manager Professional Semiprofessional/Craftsmen Agriculture Service, others Contribution period (years) Number of unemployed 76,630 40,771 35,859 Note: All models also include age and age squared interacted with a dummy variable for age 55 or more, dummy variables for each year and quarter, and dummy variables indicating region of residence. 30

32 Table 2: Estimation results of eligibility to higher reemployment bonus (homogenous treatment specification) Unemployment Employment (1) (2) (3) (4) (5) (6) After Before BA-RDD After Before BA-RDD All Treated *** *** (0.0499) (0.0600) (0.0790) (0.0535) (0.0755) (0.1406) By contribution period before the UI claim (years) *** ** (0.0511) (0.0623) (0.0783) (0.0610) (0.1562) (0.2017) * (0.1210) (0.1567) (0.1882) (0.2206) (0.2364) (0.3209) *** ** (0.1538) (0.1457) (0.2077) (0.2203) (0.1821) (0.2785) Note: Robust standard errors adjusted for correlation within individuals are in parentheses. *** p<0.01, ** p<0.05, * p<0.1 31

33 Table 3: Estimation results of eligibility to higher reemployment bonus (heterogeneous treatment specification) (1) (2) (3) (4) All Contribution period (1 2) Contribution period (3 4) Contribution period (5 9) Unemployment Employment Unemployment Employment Unemployment Employment Unemployment Employment Treated*Month *** *** ** (0.0673) (0.1473) (0.0936) (0.1196) (0.1496) (0.1979) (0.1405) (0.2908) Month *** *** *** ** (0.0676) (0.1452) (0.0927) (0.1146) (0.1510) (0.2048) (0.1442) (0.1839) Month *** * * * (0.0718) (0.1532) (0.0979) (0.1190) (0.1628) (0.2168) (0.1520) (0.2215) Month *** ** (0.0757) (0.1644) (0.1031) (0.1289) (0.1720) (0.2244) (0.1599) (0.3103) Month ** * ** (0.0792) (0.1696) (0.1054) (0.1337) (0.1826) (0.2412) (0.1730) (0.2903) Month (0.0710) (0.1518) (0.0900) (0.1148) (0.1830) (0.2490) (0.1802) (0.3689) Month (0.0753) (0.1630) (0.1014) (0.1299) (0.1586) (0.2138) (0.1800) (0.2610) Month (0.0777) (0.1647) (0.1137) (0.1420) (0.1800) (0.2396) (0.1493) (0.1948) Month (0.0850) (0.1928) (0.1201) (0.1646) (0.1912) (0.2627) (0.1682) (0.4401) Month ** (0.0936) (0.2074) (0.1331) (0.1757) (0.2134) (0.2990) (0.1819) (0.2664) Month * (0.0996) (0.2252) (0.1406) (0.1897) (0.2239) (0.3118) (0.1964) (0.2866) Month (0.1047) (0.2328) (0.1468) (0.1950) (0.2323) (0.3160) (0.2104) (0.2828) Month (0.0677) (0.1483) (0.0934) (0.1212) (0.1546) (0.2115) (0.1429) (0.1878) Month * (0.0844) (0.1931) (0.1152) (0.1601) (0.1972) (0.2786) (0.1745) (0.2461) Note: Robust standard errors adjusted for correlation within individuals are in parentheses. *** p<0.01, ** p<0.05, * p<0.1 32