طرف عرضه پسماند: مورد سیستم بیمه بیکاری در کانادا

This paper presents results from a 1971 natural experiment carried out by the Canadian government on the unemployment insurance system. At that time, the generosity of the UI system was increased dramatically. We find some evidence that the propensity to collect UI increases with a first-time exposure to the new UI system. Hence, as more individuals experience unemployment, their lifetime use of the system increases. This supply side hysteresis effect may explain why unemployment has steadily increased over the 1972–1992 period, even though the generosity of unemployment insurance did not.

A recurrent theme in policy debates regarding social welfare programs is the relationship between benefits and the disincentives to work (Moffitt, 1982). In the case of unemployment insurance, there is a great deal of evidence suggesting that it tends to increase both the duration of unemployment and the probability of becoming unemployed.1 Moreover, work by Katz and Meyer, 1990 and Corak, 1993 and Meyer and Rosenbaum (1996) find evidence that workers adjust their labor supply so that unemployment insurance may subsidize part-year work. Despite this micro-econometric evidence, there does not seem to be a direct relationship between unemployment insurance benefits and the recent secular rise in unemployment in the OECD2. Lindbeck (1995) has pointed to social norms and the sluggish response of individual labor supply to changes in incentives as a potential source of ‘supply side hysteresis’ that may help explain this secular trend. The goal of this paper is to build upon this idea, and see whether recent trends in the use of UI in Canada can be explained using a simple adaptive learning model. In a standard labor supply framework one supposes that changes in worker alternatives result in an immediate behavioral response. Moreover, whether or not the individual has had experience with the alternatives is irrelevant to his or her choice. However, there is a large body of evidence demonstrating that experience does matter for human decision making (e.g. Wickens, 1992). It is well recognized in the economics literature that it takes time for individuals to find an optimal response, and hence short-run supply elasticities are likely to be smaller than long-run elasticities.3 The issue that we wish to address in the study is the importance of this lagged adjustment in the case of labor supply responses to changes in the unemployment insurance (UI) parameters. A number of studies have shown that individuals adjust their labor supply as a function of the parameters of the system in the predicted direction. However, in the case of the Canadian UI system, UI use and unemployment increased steadily from 1971 until the 1990s, though during this period benefit level were constant or falling.The hypothesis we wish to explore is that workers did not immediately respond to the large increase in benefits that occurred in 1971. Rather, when workers experienced unemployment for the first time, due to natural turnover or a recession, this exposed them to UI and caused them to begin exploring ways to use the UI system as a subsidy to part-year work. This was possible due to a number of rule changes that occurred in 1971. First, coverage of the UI system was expanded from 68 to 96% of the work force. The number of weeks of work needed to qualify for benefits was reduced from 30 weeks in a 2-year period to 8 weeks in a single year. The maximum number of weeks during which benefits could be received by a worker having worked the minimum number of weeks required to qualify was increased from 15 to up to 44 weeks, depending on the regional unemployment rate (in high-unemployment regions benefits are more generous). The replacement rate was increased from 57% of previous earnings to 66% (or 75% if claimant had dependents). The generosity of the program is summarized by the subsidy rate (replacement rate×number of weeks of benefits for someone who has worked the minimum number of weeks to qualify/number of weeks needed to qualify) which is illustrated in Fig. 1 for the period 1951–1996. Given the shorter qualification period, the 1971 reform created an incentive for individuals to tailor their behavior to the parameters of the system, resulting in what many Canadians affectionately called the lotto 10/42. Work 10 weeks and win 42 weeks of paid holidays. The immediate impact of these changes are readily visible in Fig. 2, which illustrates the evolution of the use and cost of the UI system. In 1972 there was a dramatic increase in both the number of UI recipients and the total cost of the program. However, both of these indexes continued to increase over the 1972–1992 period, with the fluctuations in outlay explained by the business cycle (for reference the unemployment rate is also illustrated in Fig. 2). The cost of the UI system as a fraction of total labor income increased from about 3% in 1972 to 5% in 1991, a 66% increase. This occurred even though the incentives for use were either constant or decreasing.4 To see how conditioning may help explain this observation, consider a cohort of workers that were working full year in 1971, the date of the large-scale change to UI. Over time more and more workers from this cohort will have experienced unemployment and possibly received some UI. We find that the probability that an individual will receive UI increases when he or she has had experience with the system in the past, implying that the fraction of workers receiving UI should also increase over time, even though the parameters of the system are unchanged.5 This also creates a hysteresis effect during recessions. A recession increases the number of workers who leave full-year employment and experience unemployment and UI. The conditioning or learning effect that we identify implies that at the end of the recession, the equilibrium number of workers who are unemployed should be greater, and hence the economy should not return to its pre-recession equilibrium level of employment. This may account for the rising trend in the unemployment rate illustrated in Fig. 2. To ensure that we are identifying a behavioral change rather than a structural change in the economy, we follow the behavior of individual workers using a large administrative data set. In addition to the usual controls, we are able to control for individual effects, year effects and industry effects. Using a random effects probit model, we find some evidence that first-time treatment with the UI system permanently increases future use. In Section 2 we present a discussion of the model. Section 3 presents the administrative data set used in this study, while Section 4 provides some simple results using a difference-in-differences approach. Section 5 discusses the main estimation strategy. The results are discussed and summarized in 6 and 7.

We find some mixed evidence that first-time use of the unemployment insurance system in Canada increases the probability of future use. As workers are exposed to UI for the first time for a variety of reasons, they learn about the functioning of the system and adjust their behavior accordingly. This behavior is consistent with individuals responding to the incentives provided by the system. The fact that experience is required for a change in behavior is consistent with laboratory studies of learning. It also suggests that studies based on cross-section estimates of supply responses underestimate the long-term impact of the disincentive effects of social welfare programs. In Canada’s case, the increased divergence of the Canadian and American unemployment rates has long been a source of concern. The study of Card and Riddell (1993) is unable to identify the source of the difference based upon a standard supply and demand analysis. The lagged adjustment effects identified in this study may provide a coherent explanation for this effect. Given the size of the 1971 changes to the UI system, even as benefits decreased in the subsequent 20 years, workers unfamiliar with the system may still modify their behavior after a spell of unemployment several years after the change because they had never considered part-year work as an option. The results reported in Fig. 4 are consistent with this explanation. If subsequent research supports this conjecture, this has potentially important implications for the design of social welfare programs. Specifically it would imply that the feedback between a policy change, and its ultimate impact on the economy may be very slow, and hence it suggests that it may be very costly to learn about and correct policy errors. Moreover, once individuals have adjusted their behavior to the system, one is likely to face a similar lagged adjustment in the reverse direction as individuals take time to learn and respond to the new parameters. Canada has recently tightened significantly its UI rules. It will be interesting to see how the Canada–US unemployment rate gap responds as a consequence. Secondly, this highlights the importance of coverage in determining the impact of changes in a program. Rule changes that affect current recipients can be expected to have an immediate impact because the individuals experience the rule change. However, for program rule changes that involve an increase in the target population, our results suggest that it may take some time before the new individuals at risk respond fully to the new incentives.28 In summary these results suggest that great care must be taken if we are to properly interpret the relationship between changes in incentives at the individual level, and the subsequent impact on the economy as a whole.