UNEMPLOYMENT INSURANCE SAVINGS ACCOUNTS: SIMULATION RESULTS FOR ESTONIA*

Transcription

1 Comments welcome! UNEMPLOYMENT INSURANCE SAVINGS ACCOUNTS: SIMULATION RESULTS FOR ESTONIA* Milan Vodopivec The World Bank and Tomaz Rejec GEA College of Entrepreneurship, Portoroz July 21 Revised September 22 *The authors are grateful to the Statistical Office of Estonia for providing the data used in the study, to Hugo Hopenhayn for many useful insights and comments, and to Ülle Pettai for excellent research assistance. The paper is part of a project Providing Unemployment Benefits Through Unemployment Insurance Savings Accounts, financed by the World Bank s Research Support Budget.

2 Abstract By simulating the working of the unemployment insurance savings accounts (UISAs) in Estonia using a methodology similar to Feldstein and Altman (1998), the paper examines two questions: how viable this system is as an alternative to traditional unemployment insurance (UI) system, and how the redistribution of income would change if UISAs replaced the UI system. The simulations are performed on a representative sample of labor force participants, for which lifetime labor market histories are produced from the panel data collected by consecutive Estonian Labor Force Surveys. The simulations show that UISAs have passed the test of viability: 8 to 27 percent of workers end their active life with negative cumulative balance on their UISA account, and 35 to 54 percent of workers experience negative balance on their UISA account at least once during their working life (the lower numbers refer to the lowunemployment, and the higher to the high-unemployment scenarios). Moreover, although the version of the UISA system studied in the paper permits borrowing from such accounts and hence allows for redistribution, the paper shows that the level of redistribution lags substantially behind the level which occurs under the UI system. ii

3 Table of contents Abstract... ii 1. INTRODUCTION UNEMPLOYMENT INSURANCE SAVINGS ACCOUNTS: WHAT DO WE KNOW ABOUT THEM? METHODOLOGY AND DATA SIMULATION RESULTS (A) VIABILITY OF THE UISA SYSTEM...16 (B) REDISTRIBUTION UNDER UISA AND UI CONCLUDING REMARKS REFERENCES APPENDIX 1: OBTAINING COMPLETE WORK HIS TORIES FROM THE ESTONIAN LABOR FORCE SURVEYS APPENDIX 2: CALCULATION OF THE SIMULATED VARIABLES iii

4 1. Introduction While the traditional system of unemployment insurance (UI) provides protection against the hardship of job loss, there is growing persuasion and new evidence that, in the words of Meyer (1995), unemployment insurance is not a completely benign transfer; it affects claimant s behavior (p. 127). Indeed, evidence shows that unemployment insurance even under the best program design and strict monitoring creates adverse incentives which reduce job-search efforts and increase overall unemployment. Similarly, severance pay offers income protection for dismissed workers, but reduces the job creation capacity and the employment rate of the economy, and adversely affects marginal workers (Mortensen, 1994, Blanchard 1998). 1 Spurred by adverse incentives created by traditional income support systems, new approaches to improve these systems have been embarked upon. The system of unemployment insurance savings accounts (UISAs) is among the most radical and perhaps also promising ones. Under the UISA system, each worker is required to save a fraction of earnings in his or her account, and draw unemployment compensation from it. By internalizing the costs of unemployment benefits, the UISA system is expected to radically change workers incentives and thus to avoid the moral hazard inherent in traditional unemployment insurance schemes while, under some proposals, providing same protection to the unemployed as the traditional UI system does. Too little, however, is known about the working of the UISA system to know for which groups of workers, and under what conditions, this favorable assessment of the system actually holds true. 1 For an extensive evaluation of the efficiency effects of unemployment insurance as well as other income support programs, see Vodopivec and Raju (22). 1

5 This paper is an attempt to provide insights into the working of the UISA system by simulating its introduction in one of transition economies Estonia. In particular, it addresses the following two issues: How viable is the UISA system? Unemployment insurance savings accounts eliminate pooling across individuals. If a significant proportion of workers cannot generate sufficient savings to draw upon during their unemployment spells, such a system may be non-viable. What redistributive effects are produced by the replacement of the UI system with the UISA system? According to Feldstein and Altman (1998), such redistributive effects for the U.S. are likely to be small what can be expected for developing countries? Our simulation results suggest that in Estonia, UISAs pass the test of viability: 8 to 27 percent of workers end their active life with negative cumulative balance on their UISA account, and that 35 to 54 percent of them experience negative balance on their UISA account at least once during their working life (the lower numbers refer to the low-unemployment, and the higher to the high-unemployment scenarios). However, although the version of the system studied in the paper permits borrowing from UISAs and hence allows for redistribution, the paper shows that the level of redistribution under this system lags substantially behind the level of redistribution under the UI system. The paper is organized as follows. To familiarize the reader with this rather new and not so well-known system, section 2 presents the UISA system: it describes its functioning, points out the potential viability problem that UISA may suffer from, and discusses the redistributive effects which would be brought about by the substitution of the traditional unemployment insurance system by the UISA system. Section 3 describes the methodology 2

6 and data sources, and section 4 presents the simulation results. Section 5 summarizes the results and derives policy implications. 2. Unemployment Insurance Savings Accounts: What Do We Know About Them? Partly in response to the shortcomings of the traditional unemployment insurance, the idea of unemployment insurance savings accounts (UISA) has gained popularity in recent years. The main rationale and the key appeal of the UISA system as an alternative to the traditional unemployment insurance system is its potential of improving the incentives of employed workers and job-seekers while simultaneously providing the same protection as traditional insurance schemes. Unemployment insurance savings accounts, however, is still very much a novel idea. Much less empirical evidence exists about this system than about other systems of income support, and there has been no rigorous analysis of existing UISA programs. It is thus premature to give a reliable evaluation of this system as an alternative to traditional unemployment support schemes. Nonetheless, this idea has attracted a lot of attention in Latin America, where variants of this system have recently been introduced by Argentina, Chile, Colombia, Ecuador, Peru, and Uruguay. In Brazil, such a variant has been in place for several decades. Below we describe the proposed system of the unemployment insurance savings accounts and review the theoretical and empirical literature on the effects of such system. Description. The system functions as follows. Employers deposit for each worker some specified fraction of his or her earnings into a special individual savings account on a regular basis (see Table 1). In some countries (Chile), workers are also required to make regular contributions into their accounts. Upon separation and regardless of the reason for 3

7 separation, workers can make withdrawals from their savings accounts as they deem fit. However, in Brazil, workers can only access their accounts in the case of involuntary separation. Furthermore, employers are required to make an additional payment of 4 percent of the account balance (plus interest) to the individual as penalty. In Panama and Venezuela, the penalty is set as a multiple of previous wages, and offered regardless of the reason for separation. In all countries, at retirement, positive account balances are added to old-age pensions. Some programs allow workers to access their savings accounts for reasons other than unemployment, such health and education. According to some proposals (see, for example, Cortazar, 1996, and Feldstein and Altman, 1998), unemployed workers would be able draw benefits monthly as under the traditional unemployment insurance, and the government would lend money to accounts where the balance falls below zero. A close variant of this arrangement has recently been introduced in Chile. In the Chilean system, employers and workers make contributions into individual savings accounts. At the same time, workers and the government make contributions into a separate fund called the Solidarity Fund. After separation, if the unemployed worker s account balance falls below a stipulated minimum, the difference is made up via transfers from the Solidarity Fund (Heckman and Pages, 2). To improve both the welfare and efficiency effects, the unemployment insurance savings accounts could be combined to provide protection against other risks as well. 2 Such recent proposals include the Integrated Unemployment Insurance System (Yun, 21). Under this system, unemployment insurance is provided via integrating unemployment 2 Such is the current dual social/private arrangement in Uruguay, where the individual savings account covers not only unemployment benefits, but also old-age pension, disability, death, sickness and maternity benefits, as well as family allowances (Lipsett, 1999). 4

8 insurance with the pension system. Benefits are financed via a combination of withdrawals from an individual savings account on which a worker accumulates his/her contributions for unemployment as well as for old-age pensions and, under certain circumstances, also from a public unemployment insurance (which operates on a pay-as-you-go basis). Such a program thus combines inter-temporal pooling of risk of an individual with wide-base pooling under the traditional unemployment insurance system, and therefore offers a combination of selfinsurance through savings and public insurance. In addition, it combines several risks under one program, thus pooling the self-insurance component and reducing the amount of savings necessary for providing the same insurance under separate programs (indeed, there are also proposals to include other social insurance systems, such as disability and health-care, under the same roof, which under certain conditions is again welfare improving see Orszag et al, 1999). As Orszag and Snower (1997) point out, such a system could either be fully funded or financed on the pay-as-you-go (PAYG) basis. The main rationale for the PAYG approach, according to the authors, is the reluctance of most OECD countries to quickly embark on the transition to a funded system. In developing countries, another argument in favor of the PAYG approach are insufficiently developed financial markets. However, by opening a host of familiar problems plaguing PAYG pension systems (among others, re-introduction of intra- as well as inter-generational transfers, lack of transferability, susceptibility to political risk, avoidance of raising the national savings), this option needs to be carefully examined. Incentives under unemployment insurance saving accounts. According to theoretical modeling, the main rationale and key advantage of the UISA system as an alternative to the traditional unemployment insurance system is its potential of improving the 5

9 incentives of employed workers and job seekers while conceivably providing the same protection as traditional unemployment insurance. As shown by several theoretical papers, unemployment insurance savings accounts would radically change workers incentives (Orszag and Snower, 1997; Orszag et al, 1999). By internalizing the costs of unemployment benefits, the UISA system avoids the moral hazard inherent in traditional unemployment insurance. The system is thus credited with a potential to substantially decrease overall unemployment and, by lowering payroll taxes, increase wages. In particular, Orszag and Snower (1997) show that unemployment insurance savings accounts reduce unemployment by both increasing on-the-job effort of employed workers as well as job-search effort of unemployed workers. 3 Orszag et al (1999) also recommend a comprehensive vs. a piecemeal approach when introducing savings accounts. They warn that a potential complementarity problem exists if the savings account is not set up for multiple uses: under the traditional unemployment system, workers who have built up substantial resources in their pension accounts have the incentive to withdraw from the labor force and claim unemployment benefits until they retire. Setting up an integrated savings account reduces such incentives. There are also other advantages of the Integrated Unemployment Insurance System. By combining several risks under one program, the system is expected to offer not only superior provision of insurance and thus consumption smoothing, but also to significantly reduce disincentives as compared to the traditional unemployment insurance system. In addition, the government could subsidize low wage workers, which would improve the distributive properties of the system. Moreover, because of the direct link between 3 But note that some workers may stay unemployed so as to be able to draw benefits from their UISA accounts rather than exit to inactivity, and thus there are also some factors which might contribute to higher unemployment under the UISA system. 6

10 contributions and benefits, the system has the potential to attract informal sector workers. While details of the system still need to be determined, theoretical modeling suggests that the more risk averse is the individual and the lower is the job-search elasticity (that is, the less sensitive is the reemployment probability to job search), the higher is the level of optimal borrowing from the public part of the system (Yun, 21). Empirical evidence. Unemployment insurance savings accounts are still largely an "uncharted territory." Much less empirical evidence exists about this system than about other systems of income support, and apart from Kugler s (2) evaluation of the Colombian program there has been no rigorous analysis of existing UISA programs. In her pioneering study, Kugler (2) examines the effects of a 199 conversion of the severance pay program into an unemployment insurance savings accounts program in Colombia. She finds that the lion s share of the costs of the transfer that firms make to individual workers accounts (75-87 percent) show up as a reduction of wages; that implies that the likely effects of the new program on the reduction of labor demand and employment are small. She also finds that, in accordance with the theoretical predictions, the conversion increased both firing and hiring by firms, in comparison with the previous system of severance pay. Her work, however, does not shed light on the interesting question of the effects of UISAs on the reemployment probability, that is, whether or not the system improves job search incentives. There has been no other rigorous empirical work about the effects of real world UISAlike systems. Some researchers, however, have reported that the Brazilian FGTS system, while avoiding the problem of disincentives in job search found under unemployment insurance, creates incentive and other problems of its own (Gill et al, 2). First, the system creates perverse incentives on the part of the worker to precipitate a firing so as to be able to access the 7

11 funds in the savings account. It is estimated that the system increases the labor turnover rates by 3 percent. Second, it also creates additional litigation costs incurred in deciding whether or not the cause for dismissal is "just." More research as well as piloting is needed to learn whether problems of the Brazilian program can be avoided. The above discussion on the incentives under the UISA system is not complete, however. The elimination of the problem of adverse incentives applies only for individuals who save enough to maintain positive balances and who expect that they will end their working life with a positive balance. In contrast, workers who expect to end their working life with negative balances face the same incentives as under the traditional unemployment system. This consideration which obviously has important consequences for the viability of the UISA system is considered next. Viability of unemployment insurance savings accounts. Unemployment insurance savings accounts eliminate pooling of resources across individuals and, instead, rely on incomparably more restrictive intertemporal pooling of resources of one individual only. This raises an important viability question: if a significant proportion of workers cannot save enough via modest contributions from their earnings -- during their productive life to draw upon their accumulated savings during their unemployment spells, then such a system is nonviable. In other words, if unemployment is concentrated among a group of workers, these workers may not be able to finance their unemployment benefits by their own savings. In contrast, there may be a large group of workers who would never use their savings account for unemployment purpose. Under such circumstances, the UISA system would be irrelevant as an alternative to the traditional UI system. 8

12 To investigate the viability of the system, Feldstein and Altman (1998) simulated the working of the UISA system for the U.S. In their simulations, the protection provided by unemployment benefits is completely the same as under the current system, but it is financed from individual s unemployment insurance savings account, to which individuals are required to contribute 4 percent of their wages. Their simulations show that over a 25 year period, only a small proportion of workers (5-7 percent) end their working life with negative balances (these estimates are conservative in the sense that they do not account for any behavioral responses to changes in incentives), and that the cost to taxpayers is reduced by more than 6 percent. Feldstein and Altman thus conclude that the UISA system is a viable alternative to the standard unemployment insurance system. Of course, their conclusion is valid for the U.S. economy. Since in other countries the probabilities of entry into and exit from unemployment differ substantially from those of the U.S., the conclusion of the viability of the UISA system cannot be extrapolated to other countries, particularly not to developing ones. Distributive issues and unemployment insurance savings accounts. The UISA system can in principle provide the same income protection as the traditional unemployment insurance system does (with less adverse incentives, as claimed above). Switching to the UISA system, however, does have distributional consequences, because the benefits are financed in a different way. According to Feldstein and Altman (1998), however, the distributional effects for the U.S. are likely to be small albeit they work in the wrong direction, that is, they tend to hurt the poor. They find that individuals in all quintiles except the bottom one slightly gain, and individuals in the bottom quintile slightly lose (distributive effects appear worse, because the simulation does not take account of behavioral responses to the changed incentives 9

13 following the introduction of the UISA system). It is hard to predict what distributional effects the switch to UISA will have in the context of developing countries. To summarize, the UISA system and its variant Integrated Unemployment Insurance System appear to be promising options, particularly for countries where initial conditions are especially suitable (for example, for countries where the existence of severance pay programs may ease the transition to an UISA system). By internalizing the costs of unemployment benefits, the program avoids the moral hazard inherent in the traditional unemployment insurance program and thus improves reemployment incentives which is, given the weak monitoring capacity, an important advantage particularly for developing and transition countries. In its integrated version with public insurance thus avoiding its main weakness of the absence of risk-pooling among individuals the program promises to yield both superior protection and improved incentives, and has also the potential to attract informal sector workers. There is a need, however, for further investigation and piloting of the program. Little is known about the working of the UISA system to know for which groups of workers, and under what conditions, the above favorable evaluation of the system actually holds true (and important design parameters of the system regarding contribution rates and rules for withdrawal, for example, also need to be examined). And as explained above, the current paper by examining the consequences of the absence of cross-pooling for the viability and the distributive properties of the system seeks to help bridging this gap. 3. Methodology and Data We have adapted and extended the simulation methodology of Feldstein and Altman (1998), as dictated by data availability. The simulation thus consists of applying UISA rules to labor force participants over their entire working life. Such an accounting allows to determine the 1

14 histories of contributions to and withdrawals from their UISA of workers, from which one can derive characteristics of the UISA system which are suggestive of the viability and desirability of the system. In particular, the following characteristics are focused upon: the fraction of workers which end their active lives with negative cumulative balances on their UISAs; the fraction of workers who ever have had negative balances; and the share of public financing needed to finance unemployment benefits. Moreover, to gauge the distributional effects of the replacement of UI with the UISA system, we compared the subsidization/taxation rates by different lifetime wage quintiles implied by each system. We also performed an extensive test of robustness of the derived estimates with the respect of different parameter values of the UISA system. It has to be emphasized that in our calculations, we ignore behavioral consequences which the new system is likely to bring. We refrained from such an exercise because in the absence of any data on the behavioral changes provoked by the changes of UI parameters, assumptions about the likely magnitude of such changes would be too speculative. Below we present the simulation algorithm and discuss the most important data issues arising in the simulation. The simulation algorithm. We assume that each worker s UISA is empty at the time of entering the labor force. Thereafter, each worker contributes to his/her account during employment spells, and draws unemployment benefits during unemployment spells. Each month, the worker puts a fraction of his wage to his account, as set by the contribution rate. The contributions are waived when the balance on the account is above a predetermined maximum accumulation amount. Monthly withdrawals from the account are made when the worker is unemployed. The duration of the benefit receipt is limited by the maximum duration 11

15 of benefits, and the level of the benefit is determined by the replacement rate (a given percentage of the person's previous wage); in testing the robustness of our results, a flat rate is also considered (indeed, in the 199s the Estonian unemployment benefit system paid benefits at a flat rate). Eligibility is further restricted according to minimum prior employment requirement. In addition, we allow the UISA system to have an initial grace period during which the individual s benefit is not financed by withdrawals from the worker s UISA, but instead by the government. Moreover and most importantly, if the account is in the red (that is, if its balance is negative), the benefit is nonetheless paid to the worker and the negative balance of his account is further increased. To finance accounts with negative balances, the government taxes the employed workers in a similar way as under the standard unemployment insurance system (note that we are referring to the contribution rate when resources are put into UISAs, and to the taxation rate when resources are used to finance benefits of workers with negative UISA balances). All UISA accounts also accrue (negative or positive) monthly interest payments, depending on their balances. Data sources. The study s main data source are the Estonian 1995 Labor Force Survey (retrospectively covering the period of ), and the subsequent 1997 and 1998 surveys. The universe for the sampling was the 1989 census of the Estonian population, for the 1995 survey, and the population database of the Statistical Office of Estonia, for the later two surveys. In the 1995 survey, 9,68 individuals were interviewed; the coverage of the later two surveys was about half of the 1995 survey, with 2955 of the same individuals included in both surveys. Respondents were asked about their labor market status (employment, unemployment, and inactivity) as of beginning of the period covered by the survey and all subsequent changes 12

16 of the status. For each spell of employment, they also reported industry of employment, type of employment, and a number of employer attributes. The survey also elicited information on human capital attributes including education, work experience and job tenure, and demographic information on age, ethnicity and gender. Beside standard questions about the current labor market activity, the 1995 survey also asked retrospective questions on wages and employment from the period before transition up to 1995, and the subsequent surveys covered the gaps between the two consecutive surveys. In the 1995 survey, this required recollection of labor activities up to six years before the time of interview, which makes the collected data suspect to recall bias. To minimize this bias, enumerators were carefully trained to cross check answers for employment and unemployment spells to insure consistency. Moreover, research indicates that individuals recall traumatic events more readily, and changes in labor market status are likely to have been particularly memorable in an economy transiting from a system with many years of constant steady employment. Indeed, data validation checks show that the recall bias has been very limited. For example, the data on economic activity from the 1989 census corresponded quite well with the survey responses from the 1995 survey, and the majority of the discrepancies are attributable to changes in labor force definitions. 4 Similarly, the estimates of the number of registered unemployed obtained from the surveys quite closely match the data from the registers of Employment Offices (for example, for the second quarter of 1997 the survey estimate is 36,4 and the Employment Office number 35,7 see Statistical Office of Estonia, 1998). 4 In 5.4 percent of the cases, the recall data indicated labor force participation when the census indicated inactivity. The opposite disagreement occurred in 3.2 percent of the cases. The former cases were concentrated 13

17 The major challenge of this paper was obtaining a complete work and unemployment history for a representative sample of Estonian labor force, based on a panel data that spans only over 2 and 6 years, respectively. To obtain such histories, we produced a population of synthetic labor force participants ( persons ) by connecting the segments of working histories of workers with similar characteristics ( parents ) for whom the survey data existed. We describe this procedure in detail in Appendix 1. In doing so, we took advantage of having two sets of panel data, one for the early and one for the late 199s, to generate two sets of samples, corresponding to a low- and high-unemployment scenarios, respectively (we used the early 199s panel data to obtain the lo w-unemployment, and the late 199s panel data the highunemployment scenario). 4. Simulation Results In this section we describe the results of the simulation of the UISA system in Estonia, obtained by applying the above-described simulation algorithm to the sample of synthetic persons for both low- and high-unemployment scenarios (see Appendix 2 for the details of calculation of simulated variables). In general, the results show that the UISA system is a viable alternative to the traditional unemployment insurance system, but that it entails less redistribution from the rich to the poor than the traditional system. We report separately the results about the viability of the UISA system and separately changes in the redistribution of income brought about by the substitution of the traditional UI by the UISA system. Beside presenting the results obtained under the baseline values for the parameters of the UISA system, we also report sensitivity analysis, produced by modifications of baseline values. Because the results are sensitive to the among women in their twenties, and such mismatches are attributable to a change in labor force definition (see 14

18 selection of parents used in the procedure to produce complete synthetic work histories, we repeated each type of simulation 3 times and generated average values for the quantities we wanted to evaluate, together with their standard errors. simulation: The following values of the UISA simulation parameters were used for the baseline contribution rate: 3 percent of person's current wage, maximum accumulation amount: 3.6 average wages (that is, the amount that would cover a 6-month unemployment spell paying a flat benefit of 6 percent of the average wage), replacement rate: 6 percent of person's last wage, minimum prior employment requirement: 6 months in the last 12 months, maximum duration of benefits: 6 months, no grace period, and interest rate: 2 percent. 5 Our baseline case thus assumes the level and duration of benefits as typically offered by UB systems. By doing so, our simulations are based on more generous benefits than those offered by the Estonian system in the 199s. The main difference is that in our baseline case, we assume a 6 percent replacement rate, while in the 199s, the level of benefits in Estonia Noorkoiv et al, 1998, for details). 5 Calculations show that under the baseline high-unemployment simulation obtained from data, unemployment benefits amount to 1.55 percent of wage earnings. This matches closely the same ratio calculated from the 1995 Estonian household income and expenditure survey (source: online HEIDE databank, World Bank), according to which the ratio was 1.16 percent (note that in the 199s, the Estonian unemployment benefit system was more parsimonious as we assumed under the baseline scenario). 15

19 was set as a flat-fee at a rather low level). 6 Moreover, higher generosity of simulated benefits results also from granting benefits to all workers with sufficient contribution period, regardless of the reason for separation (so we do not disqualify job quitters, although many systems do). It also has to be remembered that whenever we compare the performance of the UISA and UI systems, we assume that under both systems the individual is offered exactly the same benefits (the same level and maximum duration of benefits), and that all behavioral changes are thus assumed away. (a) Viability of the UISA system Overall, the simulation results suggest that the UISA system could offer a viable alternative to the traditional UI system. Under the baseline parameter values of the UISA system and under the low-unemployment scenario, only 8 percent of workers end their working life with negative cumulative balance; the comparable number under the high-unemployment scenario is 27 percent (see Table 2). The number of workers with a negative account balance at least once during their working life is substantially larger, 35 percent and 54 percent for a low- and highunemployment scenarios, respectively. Unsurprisingly, the higher the education of workers, the lesser is the likelihood that they end their working careers with negative balances, as well as that these balances are ever negative. Perhaps more surprisingly, the likelihood of men to experience negative balances is substantially larger than the likelihood of women. Viability of the UISA system is confirmed also by the large share of self-financing of unemployment benefits under this system. As much as 95 percent of the benefits are paid by 6 In 21, Estonia passed a new Unemployment Insurance law which is offers more generous benefits: the replacement rate is set initially at 5, dropping to 4 percent after 1 days (which is likely to exceed the current flat level for most workers), and the potential duration is 6 to 12 months (currently 6months, extended to 9 months on individual basis). First benefits under this law will be paid out in

20 workers from their UISAs, under the low-unemployment scenario, and somewhat less, 84 percent, under the high-unemployment scenario (Table 3); recall that benefits are allowed to be paid also when the UISA balance is negative, and that the government pays the debt of the workers who retire with negative balances on their UISA accounts. The lower percentage of self-financing under the high-unemployment scenario is understandable, because under such conditions, more workers in need are using unemployment benefits for longer time. Note that the average value of lifetime unemployment benefits under the high-unemployment scenario, 7.2 average monthly wages, exceeds the one under the low-unemployment scenario by 64 percent (Table 3). In line with the results in Table 2, women and more educated receive lower lifetime unemployment benefits, and finance higher fractions of their benefits from their UISAs. Additional insights are obtained by examining the accumulation of negative balances on the UISA by age of workers (Figure 1). Under the low-unemployment scenario, the percent of workers with negative balances rises with age, reaches its peak at 15 percent in the year bracket, and gradually diminishes thereafter; there is little difference among genders. Under the high-unemployment scenario, the percent of workers with negative balances rises much steeper and longer, reaching its peak at about 3 percent in the 4-44-year bracket, and modestly decreases for the oldest age cohorts. Interestingly, under the high-unemployment scenario the outcomes at lower ages are very similar for both genders, but after the age of 35 the outcomes start to deviate there is an increase of the percent of men with negative balances, and a reduction of the percent of women. 7 7 This result may be produced by specific circumstances of early transition which favored women: relative demand shifting toward predominantly women industries, low-wage women having strong incentives to withdraw 17

21 Could the incentives to withdraw from the system once the individual incurs maximum indebtedness be improved if retirement savings are used as collateral? Figure 2 shows that, combined with a three-year grace period, contribution rates of 15-2 percent which are quite usual for European pension systems would indeed reduce the number of individuals with negative balances on their UISA below 1 percent, in the low-unemployment, and 5 percent, in the high-unemployment scenario. This means that if pension contributions are used as collateral, even in the high-unemployment scenario very few individuals would have the incentive not to return to the formal sector work, thus removing a potentially important obstacle to the viability of the UISA system (a similar conclusion for Korea is reached by Yun, 22). Sensitivity analysis. Here we report how deviations from the baseline values of the UISA parameters affect selected measures of the viability of the UISA system, under both lowand high-unemployment scenarios. For each parameter, we present two plots. One shows the fraction of workers with negative balances on their UISA at the end of their working life, as well as the fraction with negative balances at least once during their working life. In the other plot, we show the taxation rate needed to finance the benefits of the workers who end their working lives with negative balances. To bring the reduction of taxes under the UISAs into a sharp relief, in the same plot we also present the taxation rate needed under the traditional UI system to guarantee the same level of benefits (the comparison of the two taxation rates thus allows to gauge the extent of self-financing occurring under the UISA system). The results are as follows: from the labor force, and women being more educated than man benefiting from rising returns to education (see Orazem and Vodopivec, 2), and Vodopivec (22). 18

22 Contribution rate. As expected, the larger the contribution rate, the lower the fraction of negative balances and the taxation rate (Figure 3a), with the relationship being non-linear (the marginal effects of increasing the contribution rate are decreasing). Because the UISA system is very sensitive to such changes at the low end of the contribution range, our simulations suggest that even the introduction of a modest UISA system with one percent contribution rate reduces the taxation rate needed to finance unemployment benefits by half, from.85 to.4 percent under the low-unemployment scenario, and from 1.6 to.85 under the highunemployment scenario. (Note that at contribution rate, the UISA system converts to the traditional UI system, and that the taxation rates under the two systems are identical.) By the same token, reducing the contribution rate from 3 percent (the baseline case) to 1 percent substantially increases the percentages of workers who end their working lives with negative balances from 8 to 4 percent, and from 27 to 6 percent, under the low- and highunemployment scenarios, respectively. Replacement rate. Lower replacement rates decrease both the fraction of negative balances and the taxation rate (Figure 3b). At the replacement rate of 4 percent, there are only 4 percent of workers with negative terminal UISA balances under the low-unemployment, and 15 percent under the high-unemployment scenario, and the taxation rate is kept below.1 percent under both scenarios. In contrast to the effects of contribution rate, the replacement rate has much more linear effect on both the fraction of workers with negative account balances and the taxation rate. Similar conclusions apply about changes in the level of the benefit under the flat-fee regime, that is, when each individual is receiving the same amount of the benefit (to save space, the results are not shown). 19

23 Minimum prior employment duration. As expected, large values of this parameter reduce the fractions of workers with negative UISA balances (Figure 3c). For example, under the high-unemployment scenario, raising the minimum employment duration requirement from three to nine months reduces the fraction of workers who end their working careers with negative UISA balances from 3 to less than 2 percent, and the taxation rate is also reduced. Obviously, making the eligibility more stringent eliminates bad risk workers, that is, individuals with long expected duration of unemployment, from receiving the benefits. Maximum duration of benefits. The higher the maximum duration, the higher is the percentage of workers with negative terminal balances and the higher is the taxation rate (Figure 3d). Interestingly, under the high- unemployment scenario but not under the lowunemployment one the increase of workers with negative terminal balances for durations longer than seven months is less than proportional. This result suggests that those who become unemployed during particularly hard times are better able to self-finance their unemployment benefits, presumably because their workforce attachment is stronger. Contribution waiver amount. The UISA system is highly sensitive to the limit on the maximum balance of UISAs necessary to waive contributions to the account (Figure 3e). If this limit is set low, the fraction of workers with negative terminal account balances increases dramatically (the increase of the taxation rate is less pronounced). But there seems to be a ceiling at approximately 4 monthly average wages above which increasing the limit does not affect the performance of the UISA system (the saturation effect). The length of the grace period. The possibility that, at the start of the working career, the government takes over completely the financing of unemployment benefits importantly reduces the number of workers with negative UISA balances. For example, under the high- 2

24 unemployment scenario where these effects are more pronounced, the percent of workers with negative terminal UISA balances is reduced from 27 percent (no grace period) to 2 percent (five years grace period, see Figure 3f) but understandably, the taxation rate is also higher. Let us also mention an interesting asymmetry in the effects of interest rates (to save space, not presented below): under the high-unemployment scenario, a higher interest rate leaves the percentage of workers with final negative balances unchanged (and slightly reduces this percentage under the low-unemployment scenario), but strongly reduces the number of workers who ever sink in the red. (b) Redistribution under UISA and UI If the traditional UI is replaced by the UISA system, how does this substitution change the redistribution of income? It is the reduction of redistribution leading to self-policing! which is central to the idea of UISA, and so under the extreme version with no negative balances, the system allows no redistribution. In our simulations, however, we allowed individuals to borrow from their UISAs. As a consequence, as we showed above, some individuals ended their working lives with negative balances which were financed by general taxes hence under the UISAs cum borrowing, (some) redistribution occurs, too. Below we compare the redistribution under the UISAs to the one under the UI by simulating both systems on our sample of hypothetical individuals. The type of UISA system studied here entails some redistribution from the rich to the poor, but much less than the traditional UI system. Unsurprisingly, the UISAs allows for more redistribution under the high-unemployment scenario (when the demand for benefits is higher), as well as under the specifications of the system which make the benefits more generous or require less savings to 21

25 be accumulated on UISAs. In such cases, the taxation rate is higher, as the government has to repay larger debt arising from financing benefits with negative terminal balances. In Tables 4 and 5 we present redistribution occurring under UISAs and UI for our baseline simulation cases, for both low- and high-unemployment scenarios. 8 Beside lifetime wages, taxes, and unemployment benefits, we also present, for the UISA system, lifetime contributions to and terminal balance on UISAs, by wage quintiles (all variables are expressed in average monthly wages). Note that by assumption, unemployment benefits under both systems are identical. We focus on two measures of redistribution: net unemployment benefits and subsidization/taxation rate. Net unemployment benefits are calculated as the difference between lifetime unemployment benefits and lifetime taxes (the latter are augmented, under the UISAs, by the difference between lifetime contributions and terminal balances). Subsidization/taxation rate is defined as the ratio between net lifetime unemployment benefits and lifetime taxes (the latter are again augmented, under the UISAs, by the difference between lifetime contributions and terminal balances). As shown in the tables, the level of redistribution (as measured by net unemployment benefits and the subsidization/taxation rate, by different quintiles) under the traditional UI system exceeds the ones under the UISA system, particularly under the low-unemployment scenario (see Tables 4 and 5, and Figure 4). Under both UISAs and UI, the rich (particularly those in the two highest quintiles of their lifetime wages) are subsidizing individuals in the bottom quintiles, but the amount of redistribution and corresponding subsidization/taxation rates are much larger under the UI system. For example, under the UI system and low- 8 The baseline case for UI retains the same structure end eligibility conditions of benefits as they are under the UISA (6 percent replacement rate, the maximum duration of benefits of 6 months, minimum prior work requirement of 6 months in the last 12 months) and computes the implied taxation rate. 22

26 unemployment scenario, total unemployment benefits received by the poorest quintile exceed their own payments into the system by 52 percent, and the taxation rate of the richest quintile is 28 percent (that is, 28 percent of their payments into the UI system is used for payments of UI benefits outside their quintile); in contrast, the corresponding numbers under UISAs are much lower, 12 and 5 percent, respectively. Under the high-unemployment scenario, redistribution intensifies under both UISAs and UI system, relatively more under the former. Still, even under high unemployment scenario, the level of redistribution under the UISAs lags substantially behind the level of redistribution under the low-unemployment scenario in the UI case. Some other features of the UISA system are to be pointed out. As expected, lifetime unemployment benefits, and also lifetime taxes and contributions are higher under the highunemployment than under the low-unemployment scenarios. However, for the poorest quintile, combined lifetime taxes and contributions are smaller under the high-unemployment than under the low-unemployment scenario. Of course, lifetime benefits of the poorest quintile are still larger under the high-unemployment than under the low-unemployment scenario, the difference being financed by taxes and contributions from other quintiles, as well as by a larger depletion of the UISA balances of the poorest quintile (see Table 5). Sensitivity analysis. To analyze how redistribution is affected by changes in the parameters of the UISA system, we show in Figures 5.a 5.f the results of varying the parameters of the baseline simulation, one at the time. The results consistently show that redistribution under the high-unemployment is higher than under the low-unemployment scenario. Moreover, the results show that the lower the contribution rate, the minimum employment duration, and the limit on the maximum balance, 23

27 the more redistribution occurs under the UISA system. This relationship is driven by the fact that lowering contribution rate or the limit of the maximum balance reduces the amount accumulated at individual UISA, thus calling for a larger share of benefits to be financed by taxes. Lowering minimum employment duration, on the other hand, widens the pool of recipients for those with less stable employment histories, thus also increasing the need to finance unemployment benefits via taxes. The results also show that the higher the replacement rate, flat-fee benefit (not shown in a figure), maximum duration of benefits, and the length of the grace period, the more redistribution is implied by the UISA system. In these cases, the effect is produced by the increase in the generosity of the benefits; under unchanged UISA financing parameters, this increase calls for higher taxes and hence more redistribution. 5. Concluding Remarks By simulating the working of the UISA system in Estonia, the paper examined how viable this system is as an alternative to traditional unemployment insurance, and how the redistribution of income would change if the UISA replaced the UI system. Assuming a 3 percent contribution rate and a 6 percent replacement rate, our calculations show that 8 to 27 percent of workers end their active life with negative cumulative balance on their UISA account (somewhat more than Felstein and Altman (1998) find for the US), and that 35 to 54 percent of them experience negative balance on their UISA account at least once during their working life (the lower numbers refer to the low-unemployment, and the higher to the high-unemployment scenarios). These simulation results thus suggest that in Estonia, a large majority of workers would be well-protected against the risk of unemployment by the UISA system, and that the system has 24

28 therefore passed this test of viability. Moreover, using pension contributions as collateral would strongly improve the incentives not to withdraw from the system once the individual incurs maximum indebtedness, thus removing a potentially important obstacle to the viability of the UISA system (and given the small share of unemployment benefits in lifetime earnings, such an arrangement would not undermine a pension system). It has to be emphasized that the above simulations do not account for any behavioral changes and that by doing so, they underestimate the viability of the system after all, improved reemployment incentives are the main rationale for introducing the UISA system. The paper also found that the redistributive effects of the substitution of the traditional UI with the UISA system in Estonia hurt the poor. Although the version of the UISA system studied in the paper, which permits borrowing from such accounts, allows for redistribution of income from individuals in the top to those in the bottom quintiles of lifetime wage distribution, the level of redistribution lags substantially behind the level which occurs under the UI system. Interestingly, the UISA system allows for more redistribution when unemployment is high, and when the unemployment benefits are more generous. That redistribution is reduced obviously does not speak in favor of UISAs, but note that this reduction notwithstanding, the level and duration of benefits of all recipients under the UI system was exactly matched by our simulated UISA system. Moreover, by enhancing their access to jobs, behavioral changes induced by UISA may favor low income groups in particular. Our other results suggest that by allowing workers to draw benefits even when the balance on their account falls below zero, the consumption smoothing properties of the UISA system are significantly improved, and that a many of these workers subsequently repay their 25