بیمه بیکاری و پس انداز احتیاطی

Models of precautionary saving imply that households will hold more assets when faced with greater income uncertainty. However, previous empirical studies of income uncertainty have produced somewhat mixed support for the precautionary saving hypothesis. In this paper, we note that differences in the state-contingent income stream available to workers through the unemployment insurance (UI) program provides an excellent source of variation for testing the presence of a precautionary savings motive. Simulations of a stochastic life cycle model suggest that a UI system similar to the type currently in place in the U.S. can lead to a significant reduction in the assets accumulated by a median worker. Moreover, there is considerable variation in the UI benefit schedules for workers living in different states in the U.S., which provides an exogenous source of variation for empirically testing the precautionary saving hypothesis. We carry out this test using data on expected UI benefit replacement rates and financial assets held by households in the Survey of Income and Program Participation. Our empirical results are consistent with the predictions of the model and suggest that reducing the UI benefit replacement rate by 50 percent would increase gross financial asset holdings by 14 percent, or $241, for the average worker. We also find empirical evidence that this “crowd out” effect of UI on household saving is stronger for those facing higher unemployment risk and weaker for older workers, both of which are implications from our precautionary saving model.

The concept that some household saving may be undertaken as a precaution for a “rainy day” has long been recognized in the savings literature. Recent simulation studies suggest that precautionary saving is a significant, and perhaps the most important, determinant of individual wealth accumulation. Moreover, in the 1995 Survey of Consumer Finances more households report precautionary saving as their most important motive for saving than any other reason.1 Empirical studies of precautionary saving, however, have produced somewhat mixed conclusions. This empirical ambiguity may stem, at least in part, from the difficulty in identifying and measuring exogenous indicators of the income uncertainty facing an individual. A key element of the uncertainty in future income for working households, and thus a potential determinant of precautionary saving, is the risk of lost wages stemming from unemployment. This uncertainty is mitigated in the U.S. by the presence of unemployment insurance (UI), which on average replaces 45 percent of a covered worker's lost earnings for up to 26 weeks after a qualifying loss of a job. A testable prediction of a precautionary saving model is that this type of income insurance should reduce households’ asset accumulation. Furthermore, the extent of the income insurance available to unemployed workers varies exogenously with the benefit schedule of the UI system in their state of residence. This paper therefore uses differences in workers’ expected UI benefits to provide a source of variation for testing the presence of a precautionary savings motive. We begin by developing a model of household savings decisions which allows us to present testable implications to help guide the subsequent empirical work. In particular, we address two questions. First, given the low risk of unemployment faced by many households and the limited time of eligibility for unemployment insurance, should we expect UI to have a noticeable effect on household wealth accumulation? Second, how will these effects vary according to observable characteristics of workers, providing us with testable implications in a cross-section of wealth data? Our framework for addressing these questions is a stochastic life-cycle model in which risk-averse, prudent individuals face uncertainty over both their employment status and their wage, conditional on employment. Income uncertainty stemming from temporary unemployment risk is mitigated by an unemployment insurance system which is completely financed, in the aggregate, by a mandatory payroll tax. Thus, the model captures the insurance effects of a UI program but there are no direct aggregate wealth effects of the UI program; i.e., because the program is self-financing, on average, the only aggregare wealth effects are caused by the program's affect on precautionary saving.2 It is necessary to simulate solutions to the model in order to provide empirically testable implications because analytical solutions for asset holdings in the model are not tractable. Model simulations suggests that government provision of insurance for idiosyncratic unemployment risk that is not insurable in the private market can cause a significant decline in savings, although UI does increase the welfare of risk-averse, prudent savers. The negative effect of UI on asset accumulation is shown to increase, in percentage terms, with unemployment risk and decrease with age. In the second part of our analysis, we test these implications of the precautionary saving model by estimating the “crowdout” effect of UI on financial assets using household-level data from the Survey of Income and Program Participation (SIPP).3 We employ a reduced-form empirical specification for households’ financial assets which controls for many potential determinants of savings behavior that might be spuriously correlated with UI benefit replacement rates, such as wages and fixed location effects. We find that there is a significant crowd-out effect of UI generosity on households’ financial asset holdings: our estimates imply that reducing the benefit replacement rate of UI by 50 percent would raise the average household's financial asset-to-income ratio by 14 percent, or $241. This effect tends to be strongest, in percentage terms, for those facing the greatest risk of unemployment and for younger workers, which is consistent with the predictions of our precautionary-saving model. Overall, our empirical results support the premise that the precautionary motive is an important determinant of individual savings behavior. Moreover, our results imply that the unemployment insurance system, while potentially raising welfare, diminishes household asset accumulation. The paper proceeds as follows. In Section 1, we provide a brief overview of the related literature on precautionary savings behavior. Section 2 discusses the structure of our stochastic life-cycle simulation model and presents simulation results analyzing the effects of UI on asset accumulation in the model. Section 3 outlines our empirical procedures for testing some of the implications of the precautionary saving model. Section 4 presents the basic empirical results, while Section 5 presents some extensions. Section 6 concludes.

Low rates of savings among individuals in the U.S. has stimulated both concern among policy-makers over means of raising the savings rate and interest among academics in modelling how individuals make savings decisions. One view which has gained significant theoretical attention in recent years is that much of individual wealth accumulation is driven by precautionary motives. But empirical tests of the precautionary motive have produced quite mixed results. Our approach in this paper has been to use the exogenous variation in income risk across individuals that arises from the unemployment insurance system to test for the precautionary motive. Our results suggest that individuals do save less when UI is more generous, providing empirical confirmation that the precautionary motive is an important one in practice. The results are large in percentage terms, suggesting that a halving of the generosity of the UI system (a policy change within the range of our data) would raise individual savings by one-third. However, the findings are quite small in dollar terms, reflecting the small average asset holdings of non-elderly families in the U.S. While our finding is consistent in direction with the predictions of our model, it is quite different in its magnitude. In percentage terms, our empirical estimates are somewhat larger than those of the model; we find that a 10 percent rise in replacement rate would lower savings by 2.8 percent, while the model predicts only a 1.5 percent effect. But this larger effect in percentage terms reflects a much smaller base of savings in the data than in the model; in absolute terms, we estimate a change in savings of 0.16 percent of income from a 10 percent replacement rate rise, as opposed to 4 percent of income in the model. The difference in these findings arise from several sources. The first is that we only have one type of asset in the model, and it is fully liquid. In the empirical work, however, we have only modeled the effect of UI on financial assets, which represent a small share of total asset holdings; as noted above, we are unable to draw firm conclusions as to the effect of UI on broader measures of assets. Thus, our small estimate dollar effects may misrepresent the aggregate savings impact of this program. Second, we do not allow for any other forms of insurance (other than savings) in the model. In reality, there still may be partial insurance which mitigates the need for precautionary savings for the event of unemployment, through the labor supply of family members, transfers from extended family and friends, or other social insurance programs. Finally, and perhaps most importantly, we have not incorporated in the model the important underlying heterogeneity in savings behavior that we see in the data. The median worker that we model in this paper only does a small share of aggregate savings. Most savings are done by higher income individuals whose income does not depend in an important way on the generosity of the UI system. The challenge is to find a theoretical basis for these differences, rather than to build them in ad-hoc; the means-tested welfare programs in Hubbard et al (1994) and Hubbard et al (1995) provide one rationale for the wider distribution of accumulated wealth. Nevertheless, despite these limitations in comparing the results, the basic prediction of our model is borne out by the data: more generous unemployment insurance induces a significant reduction in asset holdings. At the same time that it lowers savings, unemployment insurance can raise welfare through completing the missing market for state-contingent insurance. Furthermore, this program has a host of other costs and benefits which must be considered in optimal program design, as discussed in Gruber (1994). The important point that this paper raises is that a critical input to such optimal benefit calculations is the potential welfare costs from reduced savings. Future research on this topic could usefully extend our findings to consider measurement of these welfare costs.