Printer Friendly

Responses to social security by men and women: myopic and far-sighted behavior.

I. Introduction

The 1983 amendments to the Social Security Act contain two measures designed to increase the labor supply of older persons in the 1990s and beyond. First, the age of eligibility for full benefits will gradually be increased from 65 to 67 during the years 2000 to 2027. Second, the delayed retirement credit for postponing Social Security benefit receipt beyond age 65 is being increased gradually over the period from 1990 to 2008. In addition, bills to liberalize or eliminate the earnings test, with its potential work disincentives, have been passed by each house of Congress at different times in recent years.

The amendments' change in the delayed retirement credit was based on the premise that individuals behave far-sightedly, that is, they take account of their entire future benefit stream (their Social Security wealth) in making labor supply decisions. If they behave myopically, however, in that they respond only to their current Social Security benefits, this policy change will not have the intended effect. Indeed, in this case, changing or abolishing the earnings limit would be more effective in increasing the labor supply of the elderly.

Previous studies of retirement have made explicit or implicit assumptions concerning the time horizon rather than treating it as an empirical question.(1) Most recent studies have assumed that individuals respond to their Social Security wealth position rather than to current benefits, despite evidence that myopic behavior is in fact quite common in other contexts. For example, a number of studies have found strong empirical evidence against the permanent income/lifecycle hypothesis, which is premised on far-sighted behavior (Flavin 1981; Hansen and Singleton 1983; Mankiw, Rotemberg, and Summers 1985; Wilcox 1989; and Zeldes 1989). Moreover, in a recent study of men's perceptions of the budget constraint under Social Security, Reimers and Honig (1993) found strong evidence of myopic behavior, which could be produced by risk aversion, borrowing constraints, or short life expectancy, as well as by short-sightedness.

With few exceptions, research on the labor supply of older individuals has focused on men.(2) We analyze both men and women because gender differences in the labor supply function at older ages are to be expected, given the differences observed at younger ages. The intermittence of women's lifetime participation in particular suggests that their retirement process might differ from that of men. Bernheim (1988, 1989), for example, found gender differences in the extent to which realized Social Security benefits and retirement age correspond to expectations. Older married men tend to make less accurate forecasts of Social Security benefits, but better predictions of their retirement age, than unmarried women.

Even if their labor supply functions did not differ, one would expect Social Security to have different effects on work for men and women. The earnings test, for example, has a uniform exempt earnings limit, which is more likely to be binding on men than on women. Jobs whose annual earnings are below the earnings limit are more likely to be available to women because of their lower wages and because part-time jobs are concentrated in female-dominated occupations. In addition, the uniform actuarial adjustment for postponed benefits is likely to have a different impact on men and women, because women's longer life expectancy means that they stand to gain more from postponing benefits.

This paper treats the individual's time horizon as an empirical question. Using data from the Longitudinal Retirement History Study (LRHS), we estimate identical hazard models of transitions back to work for men and women. Our evidence on the time horizon is based on a comparison of the labor supply response of older men and women to three key variables - current Social Security benefits, Social Security wealth, and the earnings limit - controlling for age, health, marital status, pensions, occupation, and industry. We focus on labor market reentry after a nonwork spell to eliminate a major source of bias in measuring the effects of Social Security income or wealth on labor supply due to lack of information on employer pension formulas.(3)

We find evidence that men and women behave as if they have different time horizons and therefore respond differently to Social Security. We find that women take account of their Social Security wealth, not their current benefits, when making labor supply decisions. Men, on the other hand, behave myopically in that they respond only to current benefits and do not take changes in future benefits into account. We also find that the earnings limit has no effect on women's participation, while it appears to deter men from working.

We conclude therefore that the 1983 amendments will not have all the intended effects on labor supply. The gradual increases in the delayed retirement credit, by increasing the reward to another year of work, should be effective in increasing women's, but not men's, labor supply in their late sixties. On the other hand, a relaxation of the earnings limit, as is frequently proposed, would encourage men's labor force participation, but is unlikely to have such an effect on women. Finally, the increase in the age of eligibility for full benefits, by reducing the benefit available at ages 62-67, should have the intended effect of increasing labor supply for both genders.

In the next section we discuss the implications of the effective time horizon for individuals' responses to provisions of the Social Security system. Section III specifies the hazard model and the restrictions implied by alternative hypotheses concerning the individual's time horizon. Section IV presents the data and results, followed by a discussion of the implications for Social Security policy.


II. Social Security and the Time Horizon

The individual's effective time horizon determines his or her response to the earnings test for receipt of Social Security benefits and the actuarial adjustments of postponed benefits, including the delayed retirement credit. Under the earnings test, individuals between ages 62 (when benefits can first be received) and 69 have their Social Security benefits reduced by a certain amount per dollar of earnings once earnings exceed a certain limit. Both the benefit reduction rate and the earnings limit depend on age; for example, in 1995, benefits were reduced 50 percent for earnings above $8,160 at ages 62-64, and 33.3 percent for earnings above $11,280 at ages 65-69. Currently, the earnings test does not apply after age 69.

Benefits lost at ages 62-64 due to this test - either explicitly because individuals earn too much on a postretirement job or implicitly because they have not yet retired and begun receiving benefits - are recouped later by means of an actuarial increase of postponed benefits. This actuarial adjustment is the same whether the benefits are postponed because the person has gone back to work or because he or she has not yet retired. The actuarial increase is roughly 8 percent per year (considered to be an actuarially fair adjustment) at ages 62 through 64.

At ages 65-69, benefits foregone due to postponed retirement are actuarially increased later (the delayed retirement credit), but there is no adjustment for those who return to work after starting to receive benefits. Under the 1983 amendments, the delayed retirement credit is gradually being raised between 1990 and 2008, from a less than fair adjustment of 3 percent per year to an actuarially fair 8 percent per year. These provisions are summarized in Table 1.

The earnings test is generally assumed to alter the net wage and therefore to influence labor supply. This assumption may or may not be correct, depending on the size of the actuarial increase and the individual's time horizon, which determines whether he or she takes account of the actuarial adjustments for foregone benefits. If an individual behaves myopically by considering only current income and ignoring the adjustments to future income, or if the actuarial increase is less than fair, the earnings test creates a "kink" in the budget constraint at the exempt earnings limit. In this case, the budget constraint would be ABCD in Figure 1. This may either reduce or increase hours of work, depending on the earnings level, as shown in Hanoch and Honig (1978).

If, on the other hand, an individual is far-sighted, faces a perfect capital market, and is risk-neutral, his or her behavior will be governed by considerations of Social Security wealth rather than current benefits. In this case, the timing of benefit receipt will not affect optimal labor supply because one can borrow freely against future benefits at the same interest rate as is earned on savings. This may be feasible for many older individuals, who are net savers while they are still working full-time and who have other assets to draw upon. In this case, if the actuarial adjustment is fair so that Social Security wealth is unaffected by the earnings test, there will be no kink in the budget constraint at the exempt earnings limit and the budget constraint would be ABE in Figure 1.

If the actuarial adjustment is less than fair, however, so that wealth is reduced (after age 65, for example), the kink at the earnings limit will still exist. Moreover, borrowing constraints may affect individuals who consider retiring before they are eligible to collect benefits, and who have no other assets except perhaps their home. In this case, they may be constrained to act myopically even though they take a long-range view. Diamond and Hausman (1984) and Zeldes (1989) provide evidence that a sizable fraction of the older population face such capital-market constraints. Individuals might also prefer current to future benefits having the same present value if they have reason to expect a shorter than average life, are risk-averse, or simply do not understand the actuarial adjustment. In these cases, the effective budget constraint will have a kink, even if the actuarial adjustment is fair for the average person.

Even if the budget constraint does contain a kink, however, it will not necessarily affect the participation decision (that is, to reenter the work force, as distinct from the choice of hours worked). As long as acceptable jobs at sufficiently short hours are available, earnings may be kept below the exempt amount. Individuals who face a kink in their budget constraint will be more likely to reenter if the number of hours they can work without exceeding the earnings limit is greater than the minimum hours on available jobs and greater than their reservation hours, given the fixed costs of working.

III. Specification of the Reentry Hazard Function

A major obstacle to estimating the effects of Social Security income or wealth on labor supply has been the unavailability of information on employer pension formulas in available nationally representative data sets. The absence of pension information means not only that one cannot estimate the effect of pensions, but also that one cannot obtain unbiased estimates of the effects of other variables. Social Security entitlements and employer pensions are likely to be correlated with each other because they both depend on lifetime earnings. This problem does not arise in analyzing the decision to reenter the labor force because, apart from an income effect, a pension from a previous employer plays no role in the decision to take a new job.

Furthermore, reentry is a surprisingly common feature of the retirement process. In our sample of persons whose job ended in the course of the survey, 18 percent of the women and 16 percent of the men reentered the labor force before the last interview. This increase after a reduction in labor market activity may be a response to unanticipated events or planned in advance. Foreseeable changes in the net wage due to the Social Security earnings test or to fluctuations in aggregate economic activity, for example, may induce planned intermittent labor supply.(4) Increasing preferences for leisure with age may also generate exit and reentry if workers cannot adjust their hours smoothly or if they prefer long consecutive blocks of time away from work.(5)

We analyze the determinants of labor force reentry by estimating a hazard function that generalizes a one-period model of labor supply. This framework has been used in a number of recent studies to analyze spell durations in labor market histories.(6) It assumes that during a short time interval (t, t + dt) an individual who is not working may receive a job offer. Such offers occur with a probability p(t); that is, the offer rate may depend on the time elapsed since the person stopped working. Conditional on receiving an offer, the individual chooses the state that maximizes utility, given the configuration of the opportunity set. The latter may reflect fixed costs of working, a dearth of part-time jobs, or variations in the marginal reward to work due to the Social Security earnings test.

Because only relative utilities matter, we normalize the utility of not working as [U.sub.0] = 0. The relative utility of a reentry job offer may be expressed as [U.sub.1](Z, e; B), where Z is observable and e is unobservable factors affecting the net reward to work and the value of nonmarket time, and B is the parameter/vector. Under these assumptions, the hazard rate for reentry can be expressed as

(1) h(t) = p(t) Pr([U.sub.1](t) [greater than] 0)dt

= h(X(t), t, [Epsilon];[Beta])

where in the reduced form the vector X includes observable and [Epsilon] includes unobservable factors affecting the offer arrival rate as well as the net reward to work and the value of nonmarket time; t is the duration of the nonemployment spell; and [Beta] represents the parameters to be estimated. The vector X includes both constant and time-varying variables, and the hazard rate may depend on spell duration.

The hazard function for reentry that we estimate is specified as

(2) h(t) = exp[[[Gamma].sub.0] + x(t)[Beta] + [[Gamma].sub.1]([t.sup.[[Lambda].sup.1]] - 1) / [[Lambda].sub.1]],

where t is measured in quarters. The term capturing the effect of spell duration on the hazard rate is specified as a Box-Cox transformation with parameters [[Gamma].sub.1] and [[Lambda].sub.1], allowing a flexible functional form for the hazard. Thus, [[Gamma].sub.0], [[Gamma].sub.1], and [[Lambda].sub.1], as well as [Beta], are the parameters that we estimate.(7)

Among the variables in the X vector are five that assess the influence of Social Security on reentry. To estimate the labor supply response to the earnings test - in other words, to test whether a kink exists that affects participation - we include LNHRSE, the natural logarithm of the maximum number of hours a person can work per year without having current Social Security benefits reduced because of the earnings test. This variable is the natural logarithm of the annual exempt amount of earnings currently in effect, minus the natural logarithm of the predicted real reentry wage (PLNRWAGE), which is discussed below. Because the kink may not appear until age 65, we allow for a difference in response to exempt hours before and after this age by interacting LNHRSE with dummy variables for the two age groups.(8)

To test for the relevant time horizon, we include four variables: current Social Security benefits (BENCURR) and three different aspects of Social Security wealth (SSWFORE, SSWSURP, and SSWLOSS), based on the individual's own earnings record. Current real Social Security benefits (BENCURR) at age 62 and older were computed as follows (they are 0 at younger ages): from the Social Security earnings record, we calculated Average Monthly Earnings (AME) as of the date the 1969 job ended, using the rules in effect at that date. For that quarter and each subsequent quarter, the Primary Insurance Amount (PIA) was calculated from the AME according to the formula applicable at each quarter.(9) This is the monthly benefit that a single person would receive if he or she retired at age 65.(10) For those whose jobs ended between ages 62 and 65, the actuarial reduction was applied (0.55 percent for each month that the job-leaving age fell short of age 65). For those whose jobs ended after age 65, the delayed retirement credit was applied (0.083 percent for each month that the job-leaving age exceeded 65).(11) Finally, the current benefit was deflated by the Consumer Price Index (CPI).

SSWFORE and SSWSURP are included to capture the pure income effect of Social Security, and SSWLOSS represents the reduction in the net reward to work due to the earnings test. SSWFORE represents the Social Security wealth that the individual foresaw at the beginning of the survey, based on the benefit formula for 1969, and SSWSURP represents unanticipated increases in wealth due to subsequent legislated changes in benefits. Since a given increase in wealth represents different annual income flows depending on the age at which the individual becomes aware of it, we would expect the coefficients of these two variables to differ.

Foreseen Social Security wealth (SSWFORE) is expressed as a monthly flow proportional to its annuitized value, to make it commensurate with other income flows.(12) To arrive at this monthly value, we assumed that individuals at some "planning age," say age 55, determine their optimal retirement age and that their plans are fulfilled. The retirement age thus chosen, together with their earnings history, determines their AME. The benefit formula in effect in 1969 is used to calculate their foreseen PIA. Foreseen Social Security wealth is the discounted value of the resulting benefit stream, which starts at leaving the 1969 job or age 62, whichever is later.

For retirement at age 65, foreseen Social Security wealth evaluated at the planning age (assumed to be 55) is equal to

(3) W = [summation of] PIA [S.sub.t] / [(1 + [r.sub.t]).sup.t - 55] where t = 65 to [infinity],

where r = interest rate; s = survival probability, conditional on having reached the planning age; and t = age.

For retirement at other ages, foreseen Social Security wealth is equal to [Alpha]W, where [Alpha] represents the degree of fairness of the actuarial adjustment. As long as the actuarial adjustment is fair, foreseen Social Security wealth does not vary with the age of retirement because the higher benefits resulting from postponed retirement offset the shorter remaining lifetime. For retirement ages up to the 65th birthday, therefore, [Alpha] equals 1.0 and foreseen Social Security wealth equals W.

For those who retire after their 65th birthday, when the delayed retirement credit is insufficient to compensate for the shorter remaining lifetime, foreseen Social Security wealth declines monotonically with age of retirement until age 72 (when the earnings test no longer applied during the period of the survey). A delayed retirement credit of 8 percent per year is approximately actuarially fair. The delayed retirement credit of 1 percent per year during the survey period therefore reduces [Alpha] to less than 1.0. For those retiring at age 66, [Alpha] = (1.01/1.08); for those retiring at age 67, [Alpha] = (1.02/1.16), etc.

Social Security wealth can be translated into an annuity beginning at the planning age and ending at death. The value of this annuity is

(4) A = [Alpha]W / [summation of] [s.sub.t] / [(1 + r).sup.t - 55] where t = 55 to [infinity] .

Since our subsamples are restricted to whites and are gender-specific, we assumed the same survival rates (and the same interest rates) for all individuals in a given subsample. Thus the monthly flow from this annuity is proportional to Social Security wealth, which in turn is proportional to PIA adjusted for delayed retirement, as described above. We took advantage of this proportionality rather than explicitly calculate the Social Security wealth or the monthly flow. Instead, the variable reflecting the foreseen monthly flow (SSWFORE) was constructed by multiplying the real PIA forecast for the end of the 1969 job (using the 1969 benefit formula) by [Alpha], the adjustment factor for delayed retirement described above. This captures the variation in the monthly flow across individuals. The factor that translates the benefit stream into an equivalent annuity, being constant across individuals within a subsample, is absorbed into the coefficient of the variable.

Unforeseen increments to Social Security wealth due to benefit increases enacted after 1969 are represented by SSWSURP. This variable measures the difference between the [Alpha]PIA calculated under the currently applicable benefit formula and the formula in effect in 1969.

For a given person, SSWFORE remains constant with age. Among those who leave the 1969 job at the same age, this value is lower for those with lower average lifetime earnings; among those with the same average lifetime earnings, the value is lower for those who leave the 1969 job at an older age after age 65 (because Social Security wealth declines with age of retirement for ages 65-71).(13)

The wealth variables SSWFORE and SSWSURP differ from the current benefit (BENCURR) in two important ways: (1) they are positive before age 62, the age of eligibility to receive benefits; and (2) as long as the increase for postponing benefit receipt is actuarially fair, the values of these wealth variables are not affected by reductions of current benefits due to the earnings test. Moreover, the current benefit and these wealth variables are correlated in opposite ways with age at leaving the 1969 job. For two individuals with the same average lifetime earnings but different ages at job-leaving, the current Social Security benefit is higher the later the age at job-leaving (because of the actuarial reduction for early retirement and the delayed retirement credit for postponing retirement past age 65), while Social Security wealth is lower the later the age at job-leaving after age 65 (because the delayed retirement credit is not actuarially fair).

SSWLOSS measures the potential decline in wealth when benefits are lost under the earnings test and there is not an actuarially fair increase in future benefits. For individuals who retire and then reenter before age 65, benefits are actuarially increased starting at age 65 to make up for lost benefits. As discussed above, this adjustment is actuarially fair and therefore SSWLOSS is equal to 0 up to age 65. However, for reentry after age 65, there is no such adjustment for lost benefits; consequently, the decline in wealth (SSWLOSS) is equal to the value of the current benefit.

As mentioned above, Social Security entitlements are likely to be correlated with employer pensions and other accumulated assets because they all depend on lifetime earnings. To obtain unbiased estimates of the effects of the Social Security variables, as well as to measure the effects of other sources of income, we include a dummy variable indicating current eligibility for an employer pension (DPEN), the amount of pension income in real terms (RPENAMT), and real interest, dividend, and rental income (RINTDIV).

DPEN was constructed from survey answers concerning current receipt or eligibility for full or reduced pension benefits at the time of an interview or, for those not currently eligible, the age of expected receipt or eligibility. Because pension income is reported for alternate years only and the annual total does not indicate the month in which benefits began, we could not directly determine an annual rate for each quarter. To estimate the annual rate (RPENAMT), we assumed that the real pension income reported in the person's last interview represented a full year's benefit, and we used DPEN to determine the quarter when benefits began. Because reported income from interest, dividends, and rents varied considerably across interviews, suggesting that reporting error may be a problem, we averaged the real income from these sources across interviews to construct RINTDIV.

The net reward to work is measured by the natural logarithm of the real reentry wage. Because this cannot be observed for those who do not reenter, we use its predicted value (PLNRWAGE) based on a regression model of reentry wages corrected for selectivity bias (see the appendix).

The effect of a health limitation on reentry is captured by the dummy variable DHEALTH, which measures whether a person's health limits the kind or amount of work he or she can do. Whether a limitation of work (or housework) exists and, if so, how long it has existed, are ascertained at the time of each biennial interview. This information was used to construct a dummy variable that is equal to 1 if there was a health limitation. In cases where there was insufficient information to pinpoint the quarter when this variable changed value, we used a linear interpolation between quarters with known values of the variable.

While the LHRS samples were drawn in 1969 from the populations of married men and unmarried women (including widows, divorcees, and separated women, as well as never-married women), some individuals' marital status changed over the course of the survey. To include the effect of these changes on the value of nonmarket time, we also include the variable DMAR, which is equal to 1 if the person is married. It was constructed similarly to DHEALTH from information on marital status at each biennial interview.

AGERET, the age at leaving the 1969 job, and the duration dependence term together reflect the effects on reentry of three different factors. First, there is the effect of aging itself (apart from health); that is, changes in tastes and abilities associated with getting older. Second, there is the effect of time since the job ended. The longer someone has been out of the labor force, the less likely he or she is to reenter, due to habit formation or loss of contacts and skills. Third, there is the effect of age when the job ended, which is expected to be positive if it measures lifetime tastes for work or lack of financing for retirement.

Finally, reentry job markets may depend on the occupation, industry, and sector of the 1969 job. We therefore include dummy variables (DCWHITCL, DCSERV, and DCGOVT) that reflect these aspects of the job.(14) Definitions and means of all the variables in the model appear in Table 2.

IV. Data and Findings

Our samples of white men and women are drawn from the Longitudinal Retirement History Study (LRHS). These individuals were first interviewed in 1969, when they were 58-63, and were then interviewed every two years until 1979. A person was excluded from our samples if insufficient information was available to construct a complete and consistent work history during the time he or she was in the survey. The LRHS is still the best longitudinal data set available for analyzing the labor market behavior of older people, despite its vintage, its lack of detail about pension benefit formulas, and its limited information on women who were married in 1969. Until additional waves of the Health and Retirement Survey begun in 1992 are collected, the LRHS will remain the only nationally representative longitudinal data set that contains as large a sample of people in their sixties and as complete a work history over as long a period, as well as the Social Security earnings records.

In Tables 3 (for women) and 4 (for men) we show the distribution of employment status in 1969 and of subsequent labor market transitions for our samples. Some 66 percent of the 1,003 women and 86 percent of the 2,522 men were employed in 1969. Of those employed at the 1969 interview, 18 percent of the women and 20 percent of the men remained employed as long as they were in the survey. A total of 542 women (82 percent) were observed to leave their 1969 jobs, which had an average duration of 17 years. Among these female job leavers, 98 (18 percent) reentered the labor force at a later time. Among the men, 1,740 (80 percent) left their 1969 jobs, after an average duration of 22 years. Two hundred seventy-three (16 percent) of these male job-leavers subsequently reentered the labor force. The average duration of completed spells out of the labor force was 20 months for women and 18 months for men (see Table 2).

Because we are interested in estimating a hazard function for reentry, we excluded from our sample persons who remained employed in their 1969 job for as long as they were in the survey.(15) This group includes workers who died or for other reasons left the sample before 1979 and those who remained in the LHRS sample for the entire ten years. To avoid the bias arising from left-censored spells in estimating the hazard function, we further limited the sample to persons who [TABULAR DATA FOR TABLE 2 OMITTED] [TABULAR DATA FOR TABLE 3 OMITTED] were employed at the time of the first interview in 1969 (542 women and 1,740 men).

These sample selection rules raise the possibility of other biases in the estimated parameters of the hazard function. The unmeasured characteristics of those who worked for as long as they were in the survey would make them more likely to reenter after leaving their 1969 job than observationally equivalent members of our sample. Those who were not employed in 1969 are a mixture of two groups: those who were temporarily between jobs (and therefore virtually certain to reenter), and those who had permanently left the labor force (and therefore certain not to). The unmeasured characteristics of the excluded groups thus affect the reentry rate in opposite directions, possibly offsetting each other.

With respect to the time horizon, those who remained in their 1969 job as long as they were in the survey are less likely to be myopic since they continued to work when they might have perceived a kink in the budget constraint. Those who were not employed in 1969 are more likely to be myopic, that is, responding to a perceived kink before the survey began. Thus, the biases from our sample selection rules may offset each other here as well.

Estimates of the reentry hazard function appear in Table 5 for women and in Table 6 for men. We first determine the relevant time horizon, that is, whether current Social Security benefits or Social Security wealth governs the participation [TABULAR DATA FOR TABLE 4 OMITTED] decision. To do this, we use a likelihood ratio test to determine which of the alternative Social Security variables belong in the model. An unrestricted model that includes all four variables (Column 1 in each table) is compared first with a model that restricts to 0 the effect of wealth variables (SSWFORE, SSWSURP, and SSWLOSS, Column 2), and second, with a model that restricts to 0 the effect of current benefits (BENCURR, Column 3). A significant difference in the log likelihoods between the unrestricted model and a given restricted model indicates that the "zero" restrictions can be rejected, in other words, that the variables at issue belong in the model.

The results differ by gender. For women, the zero restrictions on the wealth variables can be rejected but the zero restriction on current benefits cannot. Therefore, the wealth model in Column 3 better fits the data for women. Within the set of wealth variables, the annuitized value of foreseen Social Security wealth (SSWFORE) has a significant effect on the reentry hazard. Unanticipated increases in Social Security wealth (SSWSURP) do not have an additional effect on women's reentry, over and above the effect of foreseen wealth. The lack of significance of SSWLOSS, which measures the potential decline in wealth when benefits are lost under the earnings test after age 65 (that is, when there is not an actuarially fair increase in future benefits), suggests that women's participation [TABULAR DATA FOR TABLE 5 OMITTED] [TABULAR DATA FOR TABLE 6 OMITTED] in the labor force (as opposed to hours of work) is not affected by the earnings test.

For men, the zero restriction on current benefits is rejected while the restrictions on the wealth variables are not, indicating that the current benefits model in Column 2 better fits the data for men. In other words, men's reentry is significantly affected by current benefits, but not by Social Security wealth.

These results suggest that women behave as if they are far-sighted and are not constrained by the illiquidity of Social Security wealth, while men, on the other hand, behave as if they are myopic. Multiplying the coefficients in Table 5, Column 3, and Table 6, Column 2, by the means in Table 2, we find that a 10 percent increase in foreseen Social Security wealth reduces the reentry hazard for women by 17 percent and that a 10 percent increase in current benefits reduces the reentry hazard for men by 7 percent.

If there is either a dearth of short-hour jobs or substantial fixed costs of working, the budget constraint will have a discontinuity. In this case (and in this case only), the participation decision (as opposed to the hours-of-work decision) of those who perceive a kink at the earnings limit would be affected. The results discussed above suggest that a kink will be perceived by men of all ages (because they respond to current benefits, which are reduced for earnings above the exempt amount), but by women only after age 65, when they suffer a reduction in Social Security wealth because of the actuarially unfair delayed retirement credit.

The fact that, for women, the coefficient of our measure of the kink (exempt hours under the earnings test [LNHRSE]) is virtually 0 even after age 65 (even allowing for the bias discussed in footnote 8) indicates that women who wish to work part-time in order to stay under the earnings limit are able to find suitable jobs and that fixed costs of work do not deter them from taking a job. Therefore, while the presence of a kink may affect their hours of work, it will not cause them to stay out of the labor force. This conclusion is reinforced by the lack of significance of SSWLOSS.

The effect of the earnings test on men is not as clear. The point estimates of the elasticity of the hazard rate with respect to exempt hours are close to 1.0 (Table 6, Column 2). This suggests that a 10 percent increase in the number of hours a man could work and still keep his earnings under the limit would lead to a 10 percent increase in the hazard of reentry. These coefficients, however, are not precisely estimated, although they are larger than their standard errors. Thus, there is weak support for the proposition that men have difficulty finding part-time jobs to keep their earnings under the exempt amount (or face high fixed costs of work).(16)

In addition to the different responses to Social Security, the most striking difference between men and women is that the amounts of other income from employer pensions, interest, and dividends significantly affect the reentry of men but not of women. The reentry hazard of men is reduced by about 20 percent per $1,000 of additional outside income. Both men and women, however, are significantly less likely to reenter if they are eligible to receive an employer pension. Current eligibility to receive an employer pension reduces the hazard for women by 49 percent and that for men by 44 percent.(17)

The reentry wage (PLNRWAGE) and health (DHEALTH) are also important factors in the reentry decision for both men and women. A 10 percent increase in the predicted reentry wage (PLNRWAGE) increases the reentry hazard of women by 31 percent and that of men by 24 percent. Not surprisingly, a health limitation reduces the likelihood of returning to work for both genders; but the reduction is larger for women (56 percent) than for men (43 percent).

Women who worked in the public sector in 1969 (DCGOVT) were about 70 percent less likely to return to work. Men who worked in service industries (DCSERV) were about 40 percent more likely to reenter. On the other hand, neither the occupation of the 1969 job nor marital status was significant for either group.

The exponent on the duration term ([[Lambda].sub.1]) has a point estimate close to 2.0 for women, but neither this exponent nor the coefficients of the duration term ([[Gamma].sub.1]) or of age at leaving the 1969 job is significantly different from 0. These results indicate that women's likelihood of reentry does not change with duration of the nonemployment spell. This lack of duration dependence is puzzling. One is tempted to cite the intermittence of labor force participation throughout women's working lives. However, this is inconsistent with the 17-year mean duration of women's 1969 jobs in our sample.

For men, we do find the expected duration dependence. The coefficient of the duration term ([[Gamma].sub.1]) is negative and highly significant, while the exponent ([[Lambda].sub.1]) is close to 2.0 and also highly significant. This implies that the hazard of reentry declines at an increasing rate as the nonwork spell lengthens.

Since age at retirement (AGERET) and the duration dependence term ([[Gamma].sub.1]) represent three effects (aging, time in spell, and age at retirement from career job) with only two variables, it is impossible to identify all three effects without some additional structure. In studies of nonemployment spell durations of younger workers, it may be appropriate to assume that there is no effect of aging itself. In the case of older workers, however, this is not plausible (Hanoch and Honig 1983). In the present study, therefore, each coefficient must be interpreted as reflecting a combination of effects. Thus, holding constant the age at retirement (AGERET), the duration coefficient ([[Gamma].sub.1]) measures the effects of both time in spell and aging. On the other hand, holding duration constant, the coefficient on AGERET measures the effect of aging as well as age at retirement. Therefore, the difference in the two coefficients reflects the difference between the effects of time in spell and age at retirement. For men, the significantly negative coefficient on the duration term, together with the insignificant positive coefficient on AGERET, is consistent with the prediction that, controlling for the effect of aging, time in spell has a negative effect while age at retirement has a positive effect (reflecting tastes for work). For women, as noted above, neither time in spell nor age at retirement has a significant effect on reentry.

V. Conclusions

The labor supply response to critical aspects of the Social Security program - the earnings test and the adjustments of benefits that are postponed because of it - depends on whether behavior is affected by consideration of current benefits only or the entire future benefit stream. In this paper we have treated this question of the individual's time horizon as an empirical one. We find that women behave far-sightedly, in that their Social Security wealth rather than their current benefit affects their labor supply. Men, on the other hand, behave myopically; in other words, they respond to current Social Security benefits.

This myopic behavior on the part of men does not necessarily mean that they are short-sighted. Risk aversion, liquidity constraints, or a life expectancy so short that the increase in postponed benefits is not actuarially fair could produce myopic behavior in far-sighted individuals. Since the actuarial adjustment is at least fair at ages 62-64, both men and women gain from postponing benefits and should, on this ground, behave far-sightedly. Even with a fair actuarial adjustment, however, risk aversion or borrowing constraints could cause both men and women to prefer current rather than future benefits. The observed time horizon for each gender results from the relative strength of these effects - longevity on the one hand, risk aversion and liquidity constraints on the other. Our evidence that women respond to the future stream of benefits indicates that their long life expectancy outweighs their risk aversion and borrowing constraints. Men have generally been found to be less risk-averse than women.(18) Moreover, they have more assets and so are less likely to be liquidity-constrained. That they nevertheless respond to current benefits rather than wealth indicates that the gender gap in risk aversion and borrowing constraints is smaller than the gender gap in longevity.

Our finding that women's participation is not affected by the earnings limit, even after age 65, implies that fixed costs of participation and scarcity of part-time jobs are not impediments to women who wish to work. However, we find some evidence that men are deterred from working by the earnings test. If part-time jobs for men are scarce or their fixed costs of work are substantial, jobs with earnings below the exempt earnings limit may not be a reasonable option. In this case, men may prefer not to work at all.

These findings imply that the effects of recent and proposed changes in the Social Security program will differ by gender. The increase in the delayed retirement credit enacted in the 1983 amendments is likely to have the intended effect of increasing older women's labor supply, because it increases the reward to another year of work for women over age 65. But it will have no effect on men because they do not respond to changes in future benefits. The increase in the age of eligibility for full benefits, also in the 1983 amendments, will increase both men's and women's labor supply, but for different reasons: women's, because it reduces Social Security wealth; men's, because it reduces current benefit income. Recently proposed increases in the earnings limit (or its elimination) may well increase older men's labor force participation. While possibly affecting the hours worked by older women who are in the labor force, this change would not increase the participation rate, and therefore has limited potential for expanding older women's total labor supply.

Appendix 1

Wage-Predicting Equation

The real reentry wage (PLNRWAGE) is predicted from an equation estimated on the subsamples of men and women who reentered the labor force after leaving career jobs. The dependent variable is the natural logarithm of the real wage in the first quarter of the reentry job. Independent variables are average monthly earnings (AME) from the Social Security earnings record; the natural logarithm of the real wage in the last quarter of the 1969 job; tenure on this job and its square; age when it ended; time since it ended; health at reentry; sector, industry, and occupation of the 1969 job; and the inverse Mills ratio. The AME is calculated as of the date the 1969 job ended. It is the best available measure of lifetime earnings potential, although it reflects lifetime labor supply as well as the average lifetime wage. The coefficient estimates are reported in Table A1.
Table A1

Parameter Estimates of Reentry Wage Function, Whites Aged 58-63 in
1969 (dependent variable: ln(real wage) in first quarter of reentry

Variable                                        Females      Males

Average monthly earnings/100 from                0.141       0.051
Social Security record (AME)                    (0.047)     (0.027)

Ln(real wage) at termination of 1969 job         0.128       0.391
                                                (0.168)     (0.072)

Tenure on 1969 job (months/100)                 -0.087      -0.108
                                                (0.148)     (0.070)

Tenure/100 squared                              -0.002       0.006
                                                (0.025)     (0.012)

Age of termination of 1969 job (months/100)     -0.517      -0.506
                                                (0.347)     (0.148)

Months/100 since termination of 1969 job        -0.580      -0.379
                                                (0.283)     (0.199)

Health at reentry (= 1 if current               -0.006      -0.140
health limitation; else = 0)                   (0.152)     (0.069)

1969 job in government sector                    0.107       0.010
(= 1 if yes; else = 0)                          (0.338)     (0.088)

1969 job in service industry                     0.034      -0.110
(= 1 if yes; else = 0)                          (0.153)     (0.078)

1969 job in trade industry                      -0.202       0.043
(= 1 if yes; else = 0)                          (0.168)     (0.091)

1969 job in white-collar occupation              0.286      -0.001
(= 1 if yes; else = 0)                          (0.162)     (0.070)

Inverse Mills ratio                              0.223       0.244
                                                (0.518)     (0.164)

Intercept                                        3.829       3.981
                                                (2.188)     (1.024)

Adjusted [R.sup.2]                               0.567       0.720
Number of observations                          98         273

Note: Standard errors in parentheses.

1. The effect of Social Security on retirement behavior is the subject of a vast literature. Early studies are summarized in Danziger, Haveman, and Plotnick (1981) and Mitchell and Fields (1982). More recent studies, many of them stimulated by the 1983 amendments, include Burtless (1986), Burtless and Moffitt (1984), Fields and Mitchell (1984), Gustman and Steinmeier (1985), Hanoch and Honig (1983), Honig and Hanoch (1985), McCarty (1990), Quinn, Burkhauser, and Myers (1990), Rust (1989), Sueyoshi (1989), and Vroman (1985).

2. The effect of Social Security benefits and a number of other factors on the retirement decisions of unmarried women is examined in Hanoch and Honig (1983) and Honig (1985). A few studies have analyzed the retirement or labor force participation decisions of older married women (Clark, Johnson, and McDermed 1980; Henretta and O'Rand 1980, 1984; McBride 1988; McCarty 1990; Pozzebon and Mitchell 1989; Ruhm 1990).

3. A small number of studies have examined the question of reentry after retirement for men: Burtless and Moffitt (1984); Anderson, Burkhauser, and Butler (1984); Hayward, Hardy, and Chiang (1990); and Quinn, Burkhauser, and Myers (1990). In Reimers and Honig (1993), we use reentry behavior to infer men's perceptions regarding the budget constraint under Social Security.

4. The monthly earnings test in Social Security, in which full monthly benefits were payable in months when earnings were below the maximum, regardless of annual earnings, was eliminated to reduce such planned fluctuations in work activity. Annual earnings may still be reduced by part-year employment, however.

5. For evidence that hours cannot be varied without changing jobs, see Altonji and Paxson (1992).

6. See, for example, Butler, Anderson, and Burkhauser (1989) and Hill and O'Neill (1990). Devine and Kiefer (1991) contains a comprehensive survey of such studies. This approach handles the problems of right-censored spells and time-varying explanatory variables, which an ordinary regression model of duration or a probit or logit model of reentry probability does not (Kiefer 1988).

7. We ignore the possibility of unobserved individual-specific effects. Therefore, our estimate of duration dependence may include a spurious negative component, because those persons most likely to reenter due to unobserved factors have the shortest durations out of the labor force. Thus, as duration increases, those remaining at risk are those less likely to reenter. We are, however, less interested in duration dependence than in the effects of explanatory variables. To the extent that these are correlated with the unobservables, their estimated coefficients may also be biased, of course.

8. We ignore the fact that a small proportion of the sample is either too young to be eligible for Social Security benefits or too old (age 72 and above during the survey period) to face the benefit reduction. To restrict the sample to those who retired at 62 or later might introduce selectivity bias. Our inclusion of people who do not face benefit reduction would be expected to bias the coefficient on LNHRSE toward 0; therefore, the true effect is likely to be somewhat larger than the estimated coefficient.

9. Calculations of the AME and PIA take account of the 1967 old-start formula as well as the special minimum PIA.

10. The Social Security variables may be mismeasured for widows, because widows are eligible to receive their deceased husband's benefit if it is larger than their own. The resulting bias in our estimates of the effects of these variables may be interpreted as resulting from an omitted variable. In the case of BENCURR, this omitted variable is the difference between the husband's and wife's benefits. The sign of the bias depends on the covariance between this difference and the included variable (the wife's own benefit) as well as on the sign of the coefficient of the difference in benefits. Assuming that the income effect of husband's earnings on wife's labor supply is negative, we expect the covariance to be negative. We also expect the effect on the reentry hazard of the difference in benefits, holding the wife's benefits constant, to be negative. Thus, the resulting bias would be positive. The analysis is similar for our measures of Social Security wealth, which are based on benefits. Thus, the estimated effects of all four variables will be biased to the same degree, and estimates of their relative importance should not be biased.

11. A delayed retirement credit of 1 percent per year up to age 72 was in effect during the period of the survey.

12. The annuitized value is strictly proportional to Social Security wealth at the time of retirement, but does not decline with age. Social Security wealth, on the other hand, automatically declines with age due to the shorter remaining life span. Using current Social Security wealth as an alternative to the annuitized value would assume that individuals feel poorer in terms of Social Security - and therefore increase their labor supply - as they age.

13. Throughout the period covered by our data (and until 1983), the earnings test applied to ages 62-71.

14. In preliminary investigation, we found no significant difference between goods-producing industries and trade. We therefore grouped these into a single reference category.

15. Consistent with the LRHS questionnaire, we define a "job" in terms of the employer rather than the kind of work performed; that is, a new job begins only when there is a change of employer.

16. In an earlier study of men who had retired from long-term career jobs, we found that the coefficient of LNHRSE was 2.0 and highly significant (Reimers and Honig 1993). The difference in the coefficients of LNHRSE is consistent with the difference in the samples used in the two studies. The earlier sample was restricted to those who had started their 1969 job before age 55 and who did not define themselves as partially or fully retired. Our current sample, in contrast, also includes men whose 1969 job started after age 55 and who therefore may have more experience in job search for part-time or part-year employment. These men, once they are subject to the earnings test, may find it easier to keep their earnings under the limit. If so, their participation is less likely to be affected by the test.

17. For dummy variables, the percentage change is calculated as [100.sup.*](exp(b) - 1), where b is the estimated coefficient.

18. See, most recently, Barsky, Juster, Kimball, and Shapiro (1993).


Altonji, Joseph G., and Christina H. Paxson. 1992. "Labor Supply, Hours Constraints, and Job Mobility." Journal of Human Resources 27:256-78.

Anderson, Kathryn H., Richard V. Burkhauser, and J. S. Butler. 1984. "Work after Retirement: A Hazard Model of Labor Market Reentry." Institute for Research on Poverty Discussion Paper Number 768-84, Madison, Wis.

Barsky, Robert, Thomas Juster, Miles Kimball, and Matthew Shapiro. 1993. "An Experimental Approach to Preference Parameters and Behavioral Heterogeneity in the Health and Retirement Survey." Institute for Social Research Health and Retirement Study Working Paper, Ann Arbor, Mich.

Bernheim, B. Douglas. 1988. "Social Security Benefits: An Empirical Study of Expectations and Realizations." In Issues in Contemporary Retirement, eds. Rita Ricardo-Campbell and Edward P. Lazear, 312-50. Stanford, Calif.: Hoover Institution.

-----. 1989. "The Timing of Retirement: A Comparison of Expectations and Realizations." In The Economics of Aging, ed. David A. Wise, 335-55. Chicago: The University of Chicago Press.

Burtless, Gary. 1986. "Social Security, Unanticipated Benefit Increases, and the Timing of Retirement." Review of Economic Studies 53:781-805.

Burtless, Gary, and Robert A. Moffitt. 1984. "The Effect of Social Security Benefits on the Labor Supply of the Aged." In Retirement and Economic Behavior, eds. Henry J. Aaron and Gary Burtless, 135-71. Washington, D.C.: The Brookings Institution.

Butler, J. S., Kathryn H. Anderson, and Richard V. Burkhauser. 1989. "Work and Health after Retirement: A Competing Risks Model with Semiparametric Unobserved Heterogeneity." Review of Economics and Statistics 71:46-53.

Clark, Robert L., Thomas Johnson, and Ann Archibald McDermed. 1980. "Allocation of Time and Resources by Married Couples Approaching Retirement." Social Security Bulletin 43 (April):3-16.

Danziger, Sheldon, Robert Haveman, and Robert Plotnick. 1981. "How Income Transfers Affect Work, Savings, and the Income Distribution." Journal of Economic Literature 19:975-1028.

Devine, Theresa J., and Nicholas M. Kiefer. 1991. Empirical Labor Economics: The Search Approach. New York: Oxford University Press.

Diamond, Peter A., and Jerry A. Hausman. 1984. "Individual Retirement and Savings Behavior." Journal of Public Economics 23:81-114.

Fields, Gary S., and Olivia S. Mitchell. 1984. "The Effects of Social Security Reforms on Retirement Ages and Retirement Incomes." Journal of Public Economics 25:143-59.

Flavin, Marjorie A. 1981. "The Adjustment of Consumption to Changing Expectations about Future Income." Journal of Political Economy 89:974-1009.

Gustman, Alan A., and Thomas L. Steinmeier. 1985. "The 1983 Social Security Reforms and Labor Supply Adjustments of Older Individuals in the Long Run." Journal of Labor Economics 3:237-53.

Hanoch, Giora, and Marjorie Honig. 1978. "The Labor Supply Curve under Income Maintenance Programs." Journal of Public Economics 9:1-16.

-----. 1983. "Retirement, Wages, and Labor Supply of the Elderly." Journal of Labor Economics 1:131-51.

Hansen, Lars Peter, and Kenneth J. Singleton. 1983. "Stochastic Consumption, Risk Aversion, and the Temporal Behavior of Asset Returns." Journal of Political Economy 91:249-65.

Hayward, Mark D., Melissa A. Hardy, and Mei-Chun Liu Chiang. 1990. "Work after Retirement: The Experiences of Older Men in the U.S." Presented at the annual meeting of the Population Association of America, Toronto.

Henretta, John C., and Angela M. O'Rand. 1980. "Labor Force Participation of Older Married Women." Social Security Bulletin 43(August): 10-16.

-----. 1984. "Joint Retirement in the Dual Worker Family." Social Forces 62:504-20.

Hill, M. Anne, and June E. O'Neill. 1990. "A Study of Intercohort Change in Women's Work Patterns and Earnings." Baruch College Center for the Study of Business and Government discussion paper, New York.

Honig, Marjorie. 1985. "Partial Retirement among Women." Journal of Human Resources 20:613-21.

Honig, Marjorie, and Giora Hanoch. 1985. "Partial Retirement as a Separate Mode of Retirement Behavior." Journal of Human Resources 20:21-46.

Kiefer, Nicholas M. 1988. "Economic Duration Data and Hazard Functions." Journal of Economic Literature 26:646-79.

Mankiw, N. Gregory, Julio J. Rotemberg, and Lawrence H. Summers. 1985. "Intertemporal Substitution in Macroeconomics." Quarterly Journal of Economics 100:225-51.

McBride, Timothy D. 1988. "Women's Retirement Behavior: Implications for Future Policy." The Urban Institute, mimeo, Washington, D.C.

McCarty, Therese A. 1990. "The Effect of Social Security on Married Women's Labor Force Participation." National Tax Journal 43:95-110.

Mitchell, Olivia S., and Gary S. Fields. 1982. "The Effects of Pensions and Earnings on Retirement: A Review Essay." In Research in Labor Economics, vol. 5, ed. Ronald G. Ehrenberg, 115-55. Greenwich, Conn.: JAI Press.

Pozzebon, Silvana, and Olivia S. Mitchell. 1989. "Married Women's Retirement Behavior." Journal of Population Economics 2:39-53.

Quinn, Joseph F., Richard V. Burkhauser, and Daniel A. Myers. 1990. Passing the Torch: The Influence of Economic Incentives on Work and Retirement. Kalamazoo, Mich.: W. E. Upjohn Institute for Employment Research.

Reimers, Cordelia, and Marjorie Honig. 1993. "The Perceived Budget Constraint under Social Security: Evidence from Re-entry Behavior." Journal of Labor Economics 11:184-204.

Ruhm, Christopher J. 1990. "Bridge Jobs and Partial Retirement." Journal of Labor Economics 8:482-501.

Rust, John. 1989. "A Dynamic Programming Model of Retirement Behavior." In The Economics of Aging, ed. David A. Wise, 359-98. Chicago: The University of Chicago Press.

Sueyoshi, Glenn T. 1989. "Social Security and the Determinants of Full and Partial Retirement: A Competing Risks Analysis." National Bureau of Economic Research Working Paper Number 3113, Cambridge, Mass.

Vroman, Wayne. 1985. "Some Economic Effects of the Social Security Retirement Test." In Research in Labor Economics, vol. 7, ed. Ronald G. Ehrenberg, 31-89. Greenwich, Conn.: JAI Press.

Wilcox, David W. 1989. "Social Security Benefits, Consumption Expenditure, and the Life Cycle Hypothesis." Journal of Political Economy 97:288-304.

Zeldes, Stephen P. 1989. "Consumption and Liquidity Constraints: An Empirical Investigation." Journal of Political Economy 97:305-346.

Cordelaia Reimers and Marjorie Honig are professors of economics at Hunter College and at the Graduate School of the City University of New York. Earlier versions of this paper were presented at the 1993 annual meetings of the Association for Public Policy Analysis and Management in Washington, the American Economic Association in Anaheim, and the Eastern Economic Association in Washington. The authors would like to thank Steven Sandell, members of the Applied Econometrics Seminars at New York University, and referees of the journal for their comments. This research was supported by the U.S. Social Security Administration and the PSC-CUNY research award program. Computing resources were provided by the University Computer Center of CUNY. The data used in this article can be obtained beginning in November 1996 through October 1999 from the authors at Department of Economics, Hunter College, 695 Park Avenue, New York, NY 10021.
COPYRIGHT 1996 University of Wisconsin Press
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1996 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:includes appendix
Author:Reimers, Cordelia; Honig, Marjorie
Publication:Journal of Human Resources
Date:Mar 22, 1996
Previous Article:Factors determining participation of the elderly in supplemental security income.
Next Article:The food stamp participation and labor supply of married couples: an empirical analysis of joint decisions.

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters