آیا تغییرات در انگیزه های مستمری بازنشستگی را تحت تاثیر قرار می دهد؟ مطالعه طولی از انتظارات بازنشستگی ذهنی

This paper investigates the responsiveness of individuals’ retirement decisions to forward-looking measures of pension accumulations. In contrast to previous research, we use within-person variation in retirement incentives and are able to control for unobserved heterogeneity in tastes for retirement by studying a panel of subjective retirement expectations. We confirm that individuals do respond as expected to pension incentives, even when we control for individual fixed effects. However, the magnitude of these responses differs when estimated from models based on within-person versus cross-sectional variation: the inclusion of fixed effects reduces the response by about half.

Understanding the determinants of retirement from the labor force is crucial for designing retirement programs and public policies that affect older individuals. While there has been much research in this area, virtually all of the existing empirical studies that use longitudinal data have relied heavily on cross-sectional variation in pension incentives, and have paid little attention to the possibility of unobserved heterogeneity in tastes for retirement that are correlated with both retirement behavior and pension incentives. This paper makes use of repeated observations of individuals’ retirement expectations to investigate the effects of retirement incentives on those expectations, and to evaluate the robustness of the results to controls for unobserved heterogeneity. We confirm that individuals’ retirement expectations respond to measures of forward-looking pension incentives, even after controlling for individual fixed effects. However, the magnitude of these responses differs when estimated from models based on within-person versus cross-sectional variation: the inclusion of fixed effects reduces the response by about half. The retirement problem does not lend itself directly to approaches based on within-person variation because actual retirement (defined as a permanent exit from the labor force) will be observed only once. Thus, repeated observations of actual behavior will provide limited information about responses to changes in retirement incentives. We are able to estimate the effect of incentives on retirement using within-person variation by examining a different dependent variable: individuals’ subjective probabilities of continuing work beyond age 62, or beyond age 65, from a survey question in the Health and Retirement Study (HRS). The advantage of this measure is that we observe these subjective probabilities at up to four different dates. We combine these expectations measures with self-reported pension data from the first four waves of the HRS. Thus, we can control for any fixed individual characteristics and learn more about the updating of retirement plans in response to new information. While there are certain data limitations associated with relying on the self-reported pension data (that we discuss below), this is a unique opportunity to examine responses to changes in retirement incentives or perceptions of those incentives. Most longitudinal studies of the effect of pension incentives on retirement behavior use cross-sectional variation as the main source of identification and ignore the likely correlation between pension incentives and tastes for retirement. Even studies that are supposedly based on time-varying incentives are often identified from the cross-section since there is little unanticipated within-person variation over time. One exception to this focus on cross-sectional variation is Krueger and Pischke (1992) who investigated the labor supply effect of Social Security using cohort analysis within an aggregate panel data set. Their source of within-person variation comes from a natural experiment: a substantial and unanticipated decline in Social Security benefits for those born between 1917 and 1921 (the so-called notch generation), compared with otherwise identical individuals who were born before 1917. They find an insignificant relationship between Social Security wealth and labor force participation for the notch generation and their results suggest that previous measures of a large negative Social Security wealth effect were overestimated. Although they were not able to investigate the role of private pensions or other sources of wealth that affect the retirement decision, their general conclusion is consistent with findings in this paper. Using more recent longitudinal data from the HRS, Coile and Gruber (2000) have tried to overcome the reliance on cross-sectional variation in retirement incentives by exploiting nonlinearities in the Social Security program rules as their main source of variation in Social Security wealth accumulation. This, they argue, isolates a source of variation in retirement income accumulation that is less subject to bias from omitted unobserved characteristics. However, they do not consider the analogous potential correlation arising from private pensions. If earnings (and thus Social Security benefits) are correlated with tastes for leisure or retirement, it also seems likely that private pension structures will be similarly correlated. Since we are not using a natural experiment in our analysis, it is important to consider what might be driving within-person variation in the incentives to retire. Among workers who are employed by the same firm throughout their 50s and 60s, the rules of their pension eligibility and benefits should be known to them from the very beginning of their job tenure. This assumption is the basis of the optimizing behavior inherent in forward-looking models of retirement. This also suggests, however, that unless new information is revealed from one year to the next, within-person variation in expectations will not be informative. Below we describe possible sources of within-person variation, including information acquisition, early retirement incentives and job separations, and we show that our main results are robust to a sample restriction to individuals who have experienced an identifiable and arguably exogenous change in their retirement incentives. This paper contributes to our understanding of the relationship between retirement incentives and retirement by providing a crucial cross check of estimates based on studying actual retirement. We demonstrate that ignoring the likely correlation between retirement incentives and tastes for retirement leads to a substantial overstatement of the effects of pension incentives on retirement decisions. Since we are using a different dependent variable—retirement expectations rather than actual retirement—our findings will not be directly comparable to the previous literature. However, our evidence on the sensitivity of the incentive estimate to unobserved heterogeneity will have important implications for any work that seeks to understand the timing of retirement. The next section of the paper discusses in more detail the relationship between our approach and that taken in the recent literature on retirement decisions. Section 3 presents our data and econometric strategy. Section 4 discusses the sources of within-person variation that is driving our estimates. Section 5 presents our main results and we consider the possible effects of measurement error in Section 6. The final section concludes.

This paper investigates the relationship between retirement expectations and incentives for retirement provided by employer-sponsored pension plans and Social Security. Because we focus on repeated observations of individuals’ subjective expectations of continuing work, we are able to control for unobserved individual fixed effects that may be correlated with pension incentives. Our main conclusions can be summarized as follows. First, forward-looking measures of pension wealth only, and broader measures that include earnings, Social Security and assets, are significantly related to individuals’ expectations of continuing work into their 60s. This confirms the main finding of the literature based on actual retirement behavior. Second, this relationship remains significant even when we control for unobserved heterogeneity using fixed-effects estimators. Thus, there can be little doubt that individuals consider these forward-looking incentives when making their retirement plans. Third, however, the magnitude of these incentive effects varies dramatically between OLS and differenced estimation strategies. Coefficients on pension incentive measures from fixed-effects regressions are about half the magnitude of similar OLS regressions. Available evidence, making use of alternative differencing strategies, is not consistent with classical measurement error as an explanation for this pattern. Our results suggest that existing research which largely ignores such heterogeneity may be substantially overstating the responsiveness of individuals to pension-related incentives. While most of the literature in this area has focused on actual retirement behavior, our results pertain to individuals’ expectations about continuing work beyond age 62. We show that there are strong connections between these subjective expectations and actual future work behavior, and so our results are also relevant to longitudinal studies of actual retirement. These findings pose a challenge to future research on retirement to better control for heterogeneity in tastes for retirement that may be correlated with observed pension and other retirement wealth.