Discussion of Liquidity Effects of Benefits Extension

The positive wage effect of the UI extension for potential exhaustees (those with a predicted prob-ability in fourth quartile above) who earned below the median of the income distribution provides evidence suggestive of the existence of a liquidity effect resulting from the benefits extension. In this section, we gauge the magnitude of the liquidity effect of extended benefits by comparing the wage responses to the benefits extension for low-wage and high-wage potential exhaustees in Table A.21, and discuss its limitations.

In the Appendix C, we show the effect of extended benefits on the reservation wage is a

com-bination of a liquidity effect (^{∂ Rt}

Both effects increase the reservation wage. The reemployment bonus counteracts the moral hazard effect by offering θ percent of remaining benefits for workers reemployed before the benefits’

exhaustion point. Nevertheless, extended benefits are predicted to increase the reservation wage because the bonus offer only partly offsets the moral hazard. Equation 2.5 is the reservation wage version of the formulas in Chetty (2008) and Landais (2015a), with an adjustment for the moral hazard effect.

Based on Equation 2.2, the difference between the wage responses for low-wage and high-wage potential exhaustees is the difference in the two group’s liquidity effects if the decline in reservation wage are similar between the two groups and the moral hazard effect of extended benefits does not vary over the wage distribution. Assuming the liquidity effect for high income individuals is zero, we can obtain the ratio of the liquidity effect to the total effect of UI extension via:

0.035 − 0.023

0.035 = 0.34. (2.6)

This means that the liquidity effect of the benefits extension explains 34% of the effect of the ben-efits extension on the reemployment wage for potential exhaustees whose previous wages were below the median. This calculation is not satisfactory for at least four reasons. First, it under-estimates the liquidity effect if the liquidity effect also exists for high-wage workers. Second, it overestimates the liquidity effect if the marginal utility of consumption when employed is lower (smaller substitution effect) for high income individuals. And third, Krueger and Mueller (2016) find that the reservation wage declines significantly over unemployment spells for workers with higher savings, but not for those with lower savings. Hence, if wage and savings are positively correlated, the smaller wage effect of extended benefits for high-wage workers might be attributed to a steeper decline in reservation wage over an unemployment spell, rather than a smaller liq-uidity effect. Fourth, and most importantly, we have imposed a strong assumption that the effect of extended benefits on the reemployment wage is the same as the reservation wage response,

which almost surely does not hold true when the reservation wage or wage offer distribution over unemployment spells exhibit significant non-stationarities.

We are currently investigating a different way of disentangling the liquidity effect from the total effects of extended benefits. We are using the variation brought about by the reach back provision of the reemployment bonus program to estimate the effects of the reemployment bonus as well as the moral hazard effect of extended benefits. This will perhaps be the most important extension of this paper, because it will allow us to recover the liquidity effect and estimate the welfare effect of a benefits extension.

2.8 Conclusion

In this article, we study the unemployment benefits extension for Taiwanese workers aged 45 and over. Taiwan’s benefits extension is different from those in other countries due to its reemployment bonus program. Workers can receive half of their remaining benefits as a bonus after they become reemployed, meaning the benefits extension not only increases the potential duration of benefits but also the bonuses for workers reemployed before exhausting their benefits. Using a search model with borrowing constraints, we find that the bonus offer reduces the moral hazard effect of the benefits extension by 50%, while it did not change the liquidity effects of extended benefits. Since both the liquidity and moral hazard effects increased unemployment duration and the reservation wage, the model predicted that Taiwan’s version of a benefits extension lengthens unemployment duration as well as raising the reservation wage. Furthermore, the model generated a testable prediction that workers who were more likely to exhaust benefits and become liquidity-constrained at the exhaustion point would be more responsive to a benefits extension.

Empirically, we use the administrative records for the population of UI recipients and the RD design to estimate the effects of extended benefits on unemployment duration and reemployment wage. Our RD design yields three sets of findings for UI recipients around 45 years old. First, an increase in potential benefit duration from 180 to 270 days increases nonemployment duration by 39% and insured duration by 15%, both of which are within in the range of previous estimates from

the United States and European countries. Second, the three months’ increase in potential duration decreases the probability of being reemployed within 180 and 270 days of the initial claim by 12%

and 21%, respectively. Moreover, the probability of finding a job within 730 days is still 3% lower for workers eligible for extended benefits. Third, we do not find any job match quality gain due to the benefits extension for overall UI recipients.

Guided by the theoretical predictions of the search model, we examine how the duration and wage effects of the UI extension varied across the probability of exhausting one’s benefits as well as across the degree of liquidity constraint. Our estimates show that workers who are most likely to run out of benefits see a greater increase in duration, and experience a significant wage gain of 2.8%, while other workers did not. The positive wage effect for potential exhaustees was significant at 3.5% for lower-wage workers, but it was insignificant for higher-wage workers. The wage effect for overall UI recipients was small and insignificant, which might be because the wage gains are canceled by the insignificant wage effect for potential non-exhaustees or workers who could sustain consumption after exhausting their benefits. These results suggest that the ratio of exhaustees who are liquidity-constrained at the exhaustion point can be an important determinant of the wage effect of extended benefits. It will be helpful for a meta analysis to be conducted in the future, with the exhaustion rate and some measure of liquidity constraint correlated to the estimated wage effects of a benefits extension.

Since extended benefits are estimated to increase the reemployment earnings only for lower-wage potential exhaustees, one might argue it is welfare enhancing to tie the potential benefit duration to workers’ previous earnings and their predicted likelihood to exhaust benefits. However, it might not be appropriate to do so because the duration response to extended benefits is also larger for these workers. In the future, it will be useful to study the optimal UI design in a theoretical framework that allows potential duration varies over individual characteristics, such as previous earnings or age.

Finally, we still have limited knowledge of the ratio of the liquidity effect to the moral hazard effect of UI extension. To fill this knowledge gap, we propose an approach that compares the

wage response to a benefits extension between high-wage and low-wage workers. Assuming the mean wage response is the same as the reservation wage response and the moral hazard effect was constant over the wage distribution, we estimate the liquidity effect accounts for 34% of the total effect of a benefits extension on the reemployment wage for potential exhaustees who earned below the median wage prior to job loss. The robustness of this result requires further investigation.

CHAPTER 3

ESTIMATING THE EFFECTS OF A TIME-VARYING REEMPLOYMENT BONUS

3.1 Introduction

An inevitable cost of unemployment insurance (UI) is that it reduces unemployed workers’ incen-tives to search and lengthens unemployment duration. To put unemployed workers back to work, countries like Netherlands, Hungary and South Korea provided reemployment bonuses for workers finding jobs early before they run out of unemployment benefits.² In the U.S., field experiments suggest bonuses significantly reduce duration of unemployment (Woodbury and Spiegelman, 1987;

Decker et al., 2001). However, the structure of Taiwan’s bonus program differs from the bonuses tested in the U.S., so it is useful to examine Taiwan’s program separately.

The purpose of this chapter is to estimate the incentive effects of the reemployment bonus program in Taiwan. It offers 50% of the remaining UI entitlements to workers who became reem-ployed before exhausting their unemployment benefits and remain emreem-ployed for at least three months after leaving UI. The Taiwanese bonus program was implemented in 2003, and it reached back to incumbent claimants who entered UI prior to the reform, as discussed below. We con-sider the period before the bonus program took effect as a comparison group for the bonus offer periods. Since only incumbent claimants after 2003 are eligible for a bonus, and the bonus offer declines over duration of unemployment, whether we can identify the effects of the bonus program on reemployment hazard relies on whether business cycle and duration dependence are adequately controlled. We use a discrete-time hazard model to attempt to separate calendar time effects and duration dependence from the effect of the bonus program.

1This chapter is coauthored with Tzu-Ting Yang. Dr. Tzu-Ting Yang is an Assistant Research Fellow at Institute of Economics, Academia Sinica. I am the first author and led the project at every stage of the research.

2Van der Klaauw and Van Ours (2013) estimate the labor supply effect of the reemployment bonus program for welfare recipients in Rotterdam. The bonus programs in Hungary and South Korea are discussed in Lindner and Reizer (2016) and Kim et al. (2012), respectively.

Using administrative UI claim and corresponding earning records, we find that the bonus pro-gram in Taiwan provides unemployed workers strong incentives to accept reemployment—the bonus program is estimated to increase the reemployment hazard rate in the first four months of a nonemployment spell by more than 40 percent. Consistent with the declining bonus offer schedule, the bonus program increases the reemployment hazard by 20-30 percent in the following three months, and it gradually disappears afterwards.

This study connects to the four random experiment conducted in New Jersey, Illinois, Wash-ington and Pennsylvania (Corson and Spielgelman, 2001). Although the designs of bonuses differ from each other, these experiments suggest bonuses significantly reduce insured duration of unem-ployment by about one-half week. For example, the Illinois bonus offered 500 dollars (about four weeks of unemployment benefits) if workers who found a job before the eleventh week of insured unemployment, and if they held the job for at least four months. Woodbury and Spiegelman (1987) found that offering such a bonus reduced insured duration by about 5% without the lowering reem-ployment wage. Davidson and Woodbury (1991)’s estimates also show the Illinois bonus reduces the insured duration by 0.75 weeks for workers eligible for 26 week of benefits (1.75 weeks for those eligible for 38 weeks).

Among bonus experiment, New Jersey’s design is the most similar one to Taiwan’s design. It provided 50% of remaining entitlement, and the amount declines 10% per week. Anderson (1992) finds that the New Jersey bonus increases the reemployment hazard early in the offer period and the effect diminishes over time. However, the bonus offers were made after seven weeks of insured unemployment, and participants did not know an offer would be made before that time. Hence, the New Jersey’s experiment is not externally valid because in a real program, individuals would know that a bonus offer would be made in week seven.

The next section of this chapter describes the reemployment bonus program in Taiwan and explains what the program means for cohorts entering UI at various calendar times. Section 3.3 discuses the bonus impact on labor supply using a search model. Section 3.4 introduces the ad-ministrative UI data and the sample for empirical analysis. We provide descriptive evidence for

how the hazard function evolves over time in Section 3.5 and specify our hazard model in Section 3.6. Section 3.7 contains our estimates for the effects of the bonus program on the reemployment hazard. Section 3.8 concludes.

3.2 Taiwanese Reemployment Bonus Program

Before 2000, the unemployment rate in Taiwan was rarely over 3%, but it began to rise in 2000 and reached over 5% by mid 2001, and 2002 (Figure E.3). The increasing unemployment rate in the beginning of 2000 is consistent with the dramatic increase in the UI caseload during the same period (Figure E.4). To encourage unemployed workers to return to work, the reemployment bonus program was announced in May 2002 and implemented in January 2003. It offers 50% of the remaining UI entitlement to a recipient who becomes reemployed before exhausting his/her six months of benefits, and if they remain employed for at least three months. The three months of reemployment does not have to be continuous, or with a single employer. A person who worked for multiple employers for three months after reemployment would also qualify for the bonus.³

In Taiwan, workers who have one year of work experience in the three years prior to job loss are eligible for six months of unemployment benefits. Unemployment benefits recipients were eligible for six months of benefits regardless of age before 2009. Since 2009, the potential benefit duration has been extended to nine months for workers aged 45 or older. Importantly, the one year of work experience cannot be repeatedly used to initiate a new UI claim. If a worker finds a job and becomes unemployed again, he is eligible for his remaining UI entitlement. For example, if a UI recipient finds a job in the end of the second month of his UI spell, works for one month, and is then laid off, he is eligible for the remaining four months of unemployment benefits, rather than six months of benefits. This worker is not eligible for a bonus because he did not remain employed for at least three months. However, if he is reemployed before he exhausts the four months of remaining benefits and works for at least three months or more, he will be eligible for a bonus.⁴

3The three months reemployment period does not include recalls (the work experience in the firm prior to layoff).

4In order to initiate a new claim for six months of benefits, the worker must accumulate an

The bonus program reached back to UI recipients who were receiving benefits when the pro-gram took effect in 2003. Figure E.5 illustrates the reach back provisions of the bonus propro-gram using three examples. Claimant A started his UI spell on June 1, 2002, and would not have been eligible for the reemployment bonus even if he had found a job before exhausting his benefits, because the bonus program started on January 1, 2003, after this claimant would have exhausted benefits. Claimant B started his UI spell on August 1, 2002 and would have been eligible for half a month of benefits as a bonus if he had found a job on January 1, 2003 and kept the job for three months. If this claimant had found a job before January 1, 2003, however, he would not have been eligible for the bonus because the program had not yet started. If this claimant had found a job after February 1, 2003, he would not have been eligible for a bonus either, because he would have exhausted his benefits by then. Claimant C started a UI spell on January 1, 2003 and would have been eligible for two months of benefits as a bonus if he had found a job on February 1, 2003. The end of his qualification period would have been July 1, 2003. If he had found a job after that date, he would not have been eligible for a bonus.

The introduction of the bonus program creates useful variation in bonus eligibility, allowing us to separate the effect of the bonus program from effects of the business cycle. In Figure E.6, I plot the bonus offer for three cohorts entering UI at different times. Cohorts entering UI before July 2002, which is six months away from January 2003 would not be entitled to a bonus because they would have exhausted their benefits when the bonus program took effect. Cohorts entering UI in October 2002 would be entitled to a bonus if they were reemployed between the 90th and 180th day of their UI spell because the bonus program took effect on the 90th day of their UI spell.

Finally, cohorts entering UI after January 1, 2003 would be entitled to a bonus as long as they were reemployed before exhausting benefits and remained employed for at least three months.

In short, cohorts starting their UI spells before July 2002 were not exposed to the bonus pro-gram. Cohorts starting UI spells between July 1, 2002 and December 31, 2002 were partially exposed to the program due to the reach back provision. Cohorts starting UI spells after January additional year of work experience.

2003 were fully exposed to the program. As we will further discuss in Section 3.7, the variation in bonus entitlement rules across inflow cohorts and within cohorts over calendar times will help us identify the effect of bonus program.

3.3 Theoretical Discussion

In this section, we adopt a discrete-time search model with borrowing constraints from Chetty (2008, 178-179). We make two adjustments to Chetty (2008)’s model. First, we incorporate the reemployment bonus into the model. Second, we assume workers control their reservation wages rather than search intensities to help understand the effects of extended benefits on reemployment wages.⁵ We also consider Landais (2015b, 34-38)’s proposal that uses Euler conditions to simply model derivations. The model introduced below predicts that the bonus will reduce workers’ se-lectivity for jobs, which is capture by a lower reservation wage, and increase job acceptance rate.

We lay out the model and state the main results, leaving the theoretical derivations in Appendix C.

Consider an unemployed worker becomes unemployed at time t and holds an initial asset A_t. She lives for infinite periods of time and draws a wage offer, w, from a known and stationary wage distribution, F(w), in each period of unemployment. If she rejects the offer, she receives an unemployment benefit, b_t, with a potential duration, P, that is

b_t = b, if 0 ≤ t < P 0, if t ≥ P

If she accepts the offer, she earns a wage rate wt, pays a tax rate, τ, and keeps the job forever.⁶

5The model from Chetty (2008) assumes that workers control their search effort, which deter-mines the job finding rate. However, the model assumes a fixed wage rate, so workers’ job search decision does not change their reemployment wages. For search models with reservation wages, see Burdett (1979) for a discrete-time model and Mortensen (1977) for a continuous-time model.

6The timing definition in Chetty (2008) is different from the conventional timing definition in the search model such as Burdett (1979). Burdett (1979) assumes the search intensity and reservation wage at time t determines the probability of reemployment at time t + 1. Chetty (2008) follows Figure 1 of Lentz and Tranæs (2005, 471) and assumes the search intensity at time t determines the probability of finding a job at time t itself. The advantage of doing so is that it is easier to decompose the effect of UI into a liquidity and a moral hazard effect.

Moreover, if she is reemployed before running out of benefits, she receives a reemployment bonus,

Moreover, if she is reemployed before running out of benefits, she receives a reemployment bonus,