August 2004
Volume 5, Issue 3

Director’s Corner

The Social Security Administration’s Strategic Plan for Fiscal Years 2003-2008 states the following objective: “through education and research efforts, support reforms to ensure sustainable solvency and more responsive retirement and disability programs.” One of the stated means to achieve this objective is to “maintain an in-house capacity of policy, actuarial, and legislative expertise to respond to decision makers’ questions about the fiscal, distributional, and administrative aspects of proposals to modernize the Social Security programs.” The Retirement Research Consortium was created to enhance this capacity, providing access to a broader network of scholars. The Michigan Retirement Research Center continues to serve this mission, fostering scientific research, communication, and education related to Social Security, pension and retirement policies. This August, the Consortium will meet at the National Press Club for two days to present and discuss findings from current research and policy analysis. The annual meeting is a principal means of communicating important results to the policy community. Sessions this year will feature work that is highly relevant for current policy. In the first panel, Sylvester Schieber and Peter Diamond will provide differing viewpoints on reform. Other panels will consider important aspects of personal accounts; savings adequacy; actual trends in savings; and specific reform measures. The last panel will examine interactions among Social Security programs. The presentations will be compelling and the discussion lively. We look forward to seeing you there.


John Laitner, Director MRRC

Regular Features
Research Brief

Research Brief

Research Brief

Research Brief

Research Brief

Research Brief



Inside This Issue

Version D of RAND HRS

Join Us in D.C.


Join Us in D.C.

Join Us for the 6th Annual Conference of the Social Security Administration Retirement Research Consortium August 12-13, 2004 National Press Club, Washington, DC

Information about this conference, including registration information, will be provided and updated as needed on the MRRC website:

The 6th Annual Conference is being organized by the Michigan Retirement Research Center in cooperation with the Boston Center for Retirement Research, the Retirement Research Center of NBER, and the Social Security Administration.



 Research Brief

How to Evaluate the Effects of Social Security Policies on Retirement and Saving When Firm Policies Affect the Opportunities Facing Older Individuals by Alan L. Gustman and Thomas L. Steinmeier

Executive Summary

As the baby boomers retire, the demand for older workers will increase. As a result, firms will relax the conditions that have encouraged early retirement on long-term jobs held by older workers. This project examines the effects on retirement of non-wage aspects of employment emanating from firm side factors, including minimum hours constraints, layoffs, physical and mental requirements of the job, informal pressures to retire, accommodations made by the employer when a person has a health problem, and retirement windows. Accordingly, if these factors change with changing labor market conditions, we will be in a position to predict the effect of such changes on retirement outcomes. By understanding the effects of these factors on retirement, and how soon-to-be-realized changes in these factors will affect retirement, we can draw direct implications for Social Security. Specifically, we can determine the extent to which changes in these non-wage aspects of employment—that firms might adopt to encourage longer work lives—will lead to postponed retirement, and thus will reduce pressure on the Social Security system.

Six factors are examined: indicators of physical and mental difficulty of work; the role of minimum hours constraints; job accommodations to those with a limitation on their ability to work; age discrimination as perceived by the worker; layoffs; and early retirement windows.

Two lines of analysis are attempted. One line of analysis focuses on proper specification. How important is it that certain factors, normally omitted from structural retirement models, be included in the model? Do the indicated factors significantly affect retirement outcomes? Second, we run a number of simulations asking what would happen to retirement outcomes if we were relax the influence of a number of demand side factors.

Of each of these factors, minimum hours constraints—requirements on many jobs that individuals work full time or not at all—exert the most important influence on retirement outcomes. Perceived discrimination also has an effect on retirement outcomes, but its effect is much smaller. The same is true for the effects of indicators of job difficulty or stress. Reductions in the frequency of layoffs, reduction in the frequency that window plans are offered, and increased availability of accommodations to those who report work limitations would have little effect on retirement outcomes.

Should minimum hours constraints be abolished, the percent of the population ages 62 to 69 who are completely retired will decline by 10 to 15 percentage points. The fraction in this age group who are working in partial retirement jobs will increase by roughly twenty percentage points of the population. Were minimum hours constraints abolished, more than twice as many people will enter partial retirement as will leave full time work, so that total FTE employment would increase were minimum hours constraints eliminated. But the change in FTE employment is much smaller than the increase in partial retirement employment as many individuals will leave full time work and enter partial retirement at younger ages than they would have in the presence of minimum hours constraints. On the other hand, because most will retire at a later age, postponing their complete retirement from their lifetime job as they engage in a period of part time work, the effect is to raise the amount of work effort among older persons.

As individuals enter partial retirement at younger ages than they would have left full time work were minimum hours constraints in place, the liquidity demands on the Social Security system will increase as they begin to draw down their benefits at an earlier age. However, they will draw fewer benefits between the time they would have fully retired from their long term job were there no minimum hours constraints, and the time they actually leave that job given the availability of part time employment. The proportion of earnings subject to the earnings test will also be subject to conflicting forces. The fractions of earnings exempt from the earnings test will increase, but the fraction of earnings beyond the reaches of the earnings test will decline. Lastly, if firms were to relax minimum hours constraints to encourage more partial retirement by older individuals, this will increase the importance of regulations pertaining to withdrawals from personal accounts in those cases where a person continues to work on a part time basis.

back to the top.

Research Brief

Understanding Patterns of Social Security Benefit Receipt, Pensions Incomes, Retirement and Saving by Race, Ethnicity, Gender and Marital Status: A Structural Approach by Alan L. Gustman and Thomas L. Steinmeier Executive Summary

In this paper we use data from the Health and Retirement Study to examine differences in retirement behavior, wealth, Social Security and pension benefits among career workers by race and gender. The differences observed among groups are sometimes substantial. We then estimate models jointly explaining retirement and wealth for males and females, and for white, black and Hispanic married males. We decompose differences in outcomes into those due to differences in preferences and those due to differences in the circumstances of the members of each group. By circumstances we mean both the opportunity set and factors that determine the relative value of continued work, such as health status. We find that differences in circumstances promote later retirement by black and Hispanic men relative to whites, while differences in time preferences have the opposite effect. Differences in outcomes between men and women are primarily due to differences in preferences. The conclusion is that in conducting studies of retirement and saving, parameters reflecting preferences for leisure and goods do not differ by race and ethnicity; but time preference rates do differ by race and ethnicity and should be incorporated in any analysis of retirement. In addition, one should definitely disaggregate when analyzing outcomes for males and females.

Descriptive Analysis

The paper begins with a descriptive analysis. It examines retirement outcomes, Social Security benefits, pensions and other wealth for career workers, those who worked full time after the age of fifty and spent more than half the years since age forty in full time work. Social Security is further decomposed into AIME and PIA. Differences in Retirement by Race, Ethnicity, Gender, and Marital Status In both aggregated and disaggregated data, the sharp spikes in retirement at ages 62 and 65, the Social Security early and normal entitlement ages, are readily apparent. Men are 5.7 percentage points less likely to be retired from full time work than women at age 50; 9.4 percentage points less likely to be retired than women at age 60; 8.5 percent less likely at age 62; and 6.7 percent less likely at age 65. The differences in complete retirement are slightly smaller than for retirement from full time work, except at age 65. When we disaggregate retirement rates for white, black and Hispanic males respectively, each data set shows a sharp spike in retirements at age 62, 14.9 percentage points for white males, 15.8 percentage points for black males, and 12.7 percentage points for Hispanic males. However, the retirement spike is much smaller for white males at age 65 than it is for black and Hispanic males. For whites the age 65 spike in retirements from full time work is 8.4 percentage points, while for black and Hispanic males it is 15.2 and 12 percentage points, about the same size spike at 65 as at 62. With exceptions at a couple of ages, black men are more likely to be retired than white men. The largest difference is 8 percentage points at age 62 in the percent completely retired. Differences between Hispanic and white men are smaller than between blacks and whites. After age 60, Hispanic men are less likely to be retired than are white men. The differences rise to over 11 percentage points for 63 and 64 year old men. For black, white and Hispanic women, there is a sharp spike in retirements at age 62. The sharpest spike seen for any group at 62 is that for Hispanic women, with 20.6 percent retiring exactly at age 62, compared to 13.2 percent and 14.3 percent of black and white women at age 62. The spike in retirements from full time work at age 65 is very small for white women at just under 8 percentage points. For black women the spike is 7.3 percentage points. But for Hispanic women, the spike is 14.5 percentage points. Although there is strong evidence of retirement at the early and normal entitlement ages of Social Security for all groups, the strongest relation among all groups is found for Hispanic women. Married men are much less likely to be retired than single men, while married women are much more likely to be retired than single women. When we focus on the differences in the percent retired from full time work, for those between the ages of 58 and 62, these differences reach double-digit levels.

Wealth, Social Security Benefits, Pension Cove age and Pension Wealth by Gender, Race and Marital Status

In this model, wealth is not a determinant of retirement, but is jointly determined with retirement. Social Security benefits and pension benefits actually received depend on work and retirement choices, but are framed by the rules governing Social Security and pensions. As of the initial wave of the Health and Retirement Survey, 1992, mean wealth is $200,000 for the sample, outside of pensions and Social Security. Median wealth is $98,000. The annualized value of AIME at age 62 averages $20,500 and the annualized value of the PIA averages $8,900. Over 70 percent have a pension. Of those with positive benefits from DB plans, the annual DB benefit averages $12,000, while for those with a DC account, the value at retirement is $94,000. Remembering that wealth is a household concept, wealth held by women is 80 percent of the wealth held by men. AIME for women is less than two thirds the AIME for men among those who fit our definition of career workers. PIA for women is about three quarters of the PIA of men. Five percent fewer women are covered by a pension than men. DB pension benefits at retirement received by women are 60 percent of the value of DB pension benefits received by men, while their DC pension balances are worth 43 percent of the balances for men. Thus the difference in DB pension values are in line with earnings differences while the difference in DC values is wider than the difference in earnings. Overall, the differences in non-pension, non-Social Security wealth holdings between Black and Hispanic men on the one hand, and white men on the other, and in the values of DC accounts, are wider than the difference in covered earnings over their lifetimes, while the differences in Social Security benefits, pension coverage, and the values of payments under DB plans for those receiving positive payments are narrower than the differences in earnings. Differences in wealth levels and DC balances are wider than differences in earnings between black and white women, while differences in Social Security AIME, pension coverage and DB pension values are narrower than differences in earnings – indeed, DB pension values are higher for black women receiving benefits than they are for white women. Differences in wealth levels, pension coverage, and DB pension values are wider than differences in earnings between Hispanic and white women. Although single men have 11 percent lower earnings than married men and single women have seven percent lower earnings (AIME) than married women, the differences in wealth holdings are much greater for married than singles, while singles actually have higher DC balances. Remembering that the wealth figures are not prorated in married households, in the case of males, the differences in wealth holdings are almost two to one, while for women they are bigger than two to one. Otherwise, except for a ten-percentage point difference in pension coverage, other differences between married and singles are not substantial.

Results of Structural Estimates

Because of limited numbers of black and Hispanic married males, we pooled the data for married males of all races, and added four variables for minority status to the preference function for leisure. Two dummy variables indicated the person was black or Hispanic, and two variables reflected an interaction between minority status and age. Jointly, these four variables are statistically not significant. However, the distribution of time preferences is statistically different among the groups, so we report the effects of differences in circumstances vs. differences in time preferences. Differences in circumstances promote later retirement by black and Hispanic men relative to whites, while differences in time preferences have the opposite effect. Next we consider the results of simulations of the effects on retirement outcomes of the different preference function parameters estimated separately for men and women, as well as the effects of the different circumstances unique to men and women. Thus these retirement outcomes are generated with the preference function parameters for men and the circumstances for men and then women, and then the preference function parameters for women, and the circumstances for men and then women. Differences in preferences provide the major explanation for the earlier retirement of women. That is, women retire earlier than men due not to differences in health or the budget set, but primarily due to differences in preferences for market work.


In addition to providing an in depth analysis of differences in wealth, Social Security and pensions by gender and race, these findings also lead to a recommendation on methodology. In conducting studies of retirement and saving, parameters reflecting preferences for leisure and goods do not differ by race and ethnicity; but time preference rates do differ by race and ethnicity and should be incorporated in any analysis of retirement. In addition, one should definitely disaggregate when analyzing outcomes for males and females.
back to the top.

Research Brief

The Impact of the 1996 SSI Childhood Disability Reforms: Evidence from Matched SIPP-SSA Data by Lynn A. Karoly and Paul S. Davies Executive Summary

The Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) of 1996 changed the definition of disability used to determine eligibility for disabled children under the Supplemental Security Income (SSI) program and made other changes to the program. The law required the re-determination of eligibility status for children potentially affected by the new definition of disability. As a result of PRWORA and the SSI childhood disability re-determination process, an estimated 100,000 children were expected to lose SSI benefits. The goal of this paper is to understand the impact of benefit loss due to the 1996 reforms on affected children and their families.

The analysis draws on data from the 1992, 1993 and 1996 panels of the Survey of Income and Program Participation (SIPP) matched with Social Security Administration (SSA) administrative records on SSI program participation for children under age 18. The SIPP, a nationally representative longitudinal survey of U.S. families, is used to examine three key outcomes: family income and poverty status, employment status of the child’s mother and father, and family participation in other social welfare programs. These outcomes are among the most relevant for understanding the impact of the SSI changes on child and family well-being and they can be readily measured in the SIPP data. The SIPP data, matched with SSA administrative records on SSI program participation, provide the best opportunity to track a nationally representative sample of non-institutionalized childhood SSI beneficiaries over time to consider the short- and medium-term impact of the legislative changes.

The choice of these outcomes is also motivated by prior research. For example, previous studies suggest that termination of SSI benefits will lead to an increase in parental labor supply, although the magnitude of this effect is uncertain. The anticipated positive labor supply effect may be attenuated to the extent that former SSI recipients are eligible for other assistance programs (e.g., Temporary Assistance for Needy Families, TANF, or other safety net programs such as food stamps or general assistance) which themselves provide a disincentive for greater work effort. The severity of the child’s disability may also present barriers to increased work effort, particularly in single parent families. The SIPP data can be used to examine the extent to which work effort increases as a result of the 1996 childhood disability reforms and families’ transition to other social safety net programs. Changes in work effort and receipt of public assistance will affect total family income and the family’s poverty status as well. Depending on the family’s response in terms of additional employment income and use of government programs, family income (and hence poverty status) may improve or worsen. The net effect on family income and poverty status must be determined empirically.

The goal of the empirical analysis is to determine the impact of the loss of SSI income as a result of the 1996 PRWORA legislation on family labor supply, welfare program participation, and income and poverty. To accomplish this, we adopt a modeling strategy that is designed to control for other factors besides the legislative change that may affect outcomes for families with children. In particular, our analysis approach considers the impact of the excess benefit loss associated with the 1996 reforms that introduced a more restricted definition of eligibility for childhood SSI benefits. We compare outcomes for families of children who lost SSI benefits in the post-reform period with those who retained benefits, netting out of the effect of benefit loss associated with normal program attrition. We further examine the effects for those affected by the 1996 reforms—those children with disability diagnoses that required a re-determination of their eligibility for SSI benefits—versus those children with diagnoses that remained unaffected by the reforms.

Our results indicate that the excess benefit loss associated with the 1996 reforms was associated with a lower propensity (by 16 percentage points) for the family mother or father to be working, compared with families that lost benefits due to normal attrition. At the same time, the propensity to use food stamps was higher (by 26 percentage points) and family income was lower (by 53 percent of the poverty threshold). The propensity to use AFDC/TANF and to have income below the poverty threshold was also higher (by 19 percentage points and 11 percentage points, respectively) but neither effect was statistically significant at conventional levels. These impacts are all measured as of one month after benefit loss. When a longer horizon is examined, there is some evidence that the employment effect and propensity to use food stamps persist at approximately the same magnitude up to 12 months after benefit loss, while the other effects are attenuated with time, even to the point where the effects on income and poverty change sign. However, most of these results are imprecisely estimated. While the SIPP data have the advantage of a nationally representative sample, the sample sizes are relatively small when the focus is on disabled children receiving SSI benefits. Consequently, the SIPP sample sizes were expected to be large enough to estimate fairly precise effects of the1996 reforms if the impact sizes were large, and possibly even moderate-sized impacts could be precisely estimated as well. But if the policy impacts were smaller, the SIPP would not provide much power to precisely estimate such modest effects.

The analysis has largely born out this expectation. Generally, the results are statistically significant when a one-month observation window is used but the estimates are less precisely estimated as the length of the observation window increases. Outcomes such as employment, with consistently large impact estimates, remain statistically significant across specifications. However, other outcomes, such as the poverty rate, where the impact estimates are smaller, do not always have statistically significant effects. Nevertheless, the relative stability of the results across model specifications and the accord between the different outcome measures provides greater confidence regarding the likely direction and magnitude of the impact of the legislation in terms of family work effort, participation in other social welfare programs, and income and poverty status.

The SIPP analyses leave open the question as to the longer-term impact of the PRWORA policy change on family income, beyond the 12-month horizon examined in this analysis. Families that lose SSI benefits may be vulnerable to income declines due to limited earnings capacity, fluctuations in earnings, or the time limits in other welfare programs. Depending upon the severity of the child’s disability, there are also concerns about health insurance coverage and access to medical care, as well as any longer term impacts on child health. These issues remain to be explored in further research.

back to the top.

 Version D of RAND HRS Released

Version D of RAND HRS Released A new version of the RAND HRS Data File has been released, replacing all prior versions. The RAND HRS Data file is a cleaned and easy-to-use version of the Health and Retirement Study (HRS) with derived variables covering a broad range of measures and named consistently across waves. The file was developed by the RAND Center for the Study of Aging with funding from the National Institute on Aging (NIA) and Social Security Administration (SSA). As of early 2004, eight HRS waves are available for study. The RAND HRS Data file (Version D) is based on 1992, 1993, 1994, 1995, 1996, 1998, and 2000 final releases and the 2002 early release. The complete HRS includes four entry cohorts, as does the RAND file. This version adds the 2002 data as Wave 6 and includes updates of final release data for prior waves. The file incorporates only the core interviews. It does not include exit interview or any restricted data. The data may be accessed through the following link: The data are described in detail in the RAND HRS Data Documentation, which is included in this distribution. It contains complete descriptions of the derived variables, including descriptions of how constructed, notes on cross-wave differences, and all raw HRS variables used. The Office of Research, Evaluation, and Statistics at SSA provided important research direction in the design of this data file.
back to the top.

Research Brief

SSI for the Aged and the Problem of “Take-Up” By Todd E. Elder and Elizabeth T. Powers

Executive Summary

Supplemental Security Income (SSI) is a means-tested program intended to enable elderly households with limited financial resources to live with dignity and independence in retirement. SSI recipients may also be categorically eligible for another valuable benefit in Medicaid, which covers a wide range of health services. Since the program's inception in 1974, low participation rates have been an ongoing concern. At any point in time, a substantial fraction of elderly households who appear financially qualified to receive SSI do not enroll. In the parlance of the welfare literature, they fail to "take up" SSI. Zedlewski and Meyer (1989) estimate that only about 30% of the elderly poor receive SSI benefits. Take-up rates for the eligible population of elderly during the programs' first ten years are estimated at 50 to 55 percent. McGarry (1996), in a major study of take-up, attributes this largely to the fact that many elderly poor expect to receive only a very modest cash payment.

There are other signs that SSI may be under-used by the elderly. For instance, the number of SSI-aged recipients has been falling over most of the program's history. By 1998, 1.4 million elderly people participated in SSI, down from 2.3 million in 1975. To the extent that low participation of the elderly in SSI reflects serious unmet need, this is an issue of general public concern. In an environment in which Social Security reform schemes promoting greater individual responsibility are proposed, it is important to better understand the effectiveness of a "safety net" program for households that are poorly positioned to reap the benefits of reform.

Despite the fact that low SSI take-up by the elderly has been perceived as a serious problem for over a quarter of a century, there remains relatively little research on the aged and SSI. This paper builds upon and extends earlier work in several ways. The time period under consideration is expanded to encompass roughly the past 15 years' experience with SSI. Access to the Social Security Administration's program records permits us to estimate the application decision, a true choice variable, in addition to participation, which is the net result of an application and an administrative process. We use an alternative method of imputing expected SSI benefits for all sample units that demonstrably reduces measurement error in this key variable and is exogenous with respect to application choice.

The methods presented in this paper rely on previously unavailable information in order to generate new estimates of the determinants of SSI take-up and application decisions. While some attention has been paid to the determinants of SSI participation, due largely to low take-up rates since the inception of the SSI program in 1974, previous efforts have been hampered by the lack of administrative data availability and difficulties in constructing a sample of eligible aged individuals. This analysis has extended the previous literature in several substantial ways.

First, we have provided estimates of the SSI take-up decision for a number of years from 1984 to 1997. With one exception, no previous research has studied take-up after 1984, but major changes to Medicare and Medicaid have likely contributed to changes in the take-up decision in the meantime. We find that the influence of the expected SSI benefit has slightly declined over time, while the effect of receiving other welfare has grown dramatically.

Second, we have constructed an alternative definition of eligibility and potential benefits based solely on Social Security earnings and asset levels. Evidence from administrative data suggests that this measure substantially increases the accuracy of eligibility and potential benefit imputations, as well as eliminating endogenous issues resulting from considering other sources of current income in determining potential benefits. Models based on this alternative measure point to smaller effects of potential benefit levels in the take-up decision.

Third, administrative data on the timing of applications and benefit receipt allow us to distinguish between those who applied for the SSI-aged program and those who applied for the disabled component and “aged into” the program. These data also allow for improved accuracy of models of take-up because the dependent variable is subject to less measurement error, which could be systematically related to key independent variables.

Finally, SSA administrative records permit a decomposition of observed participation outcomes into the individual’s decision to apply and SSA’s determination of benefit eligibility. Application represents the “true” behavioral component, and as such comprises the real question of interest to researchers. Results from these models provide the most dramatic departures from previous findings, indicating that the expected SSI benefit does not significantly influence the decision to apply for benefits, with the imputed potential benefit being estimated to have small and insignificant effects on applications in every year but one. In contrast, the imputed benefit significantly positively affects eventual acceptance among SSI-aged applicants in every year of the sample. Taken at face value, these results imply that the prior findings of a positive effect of expected benefits on eventual participation decisions results primarily from those with low imputed benefits being declared ineligible according to Federal guidelines.

The findings presented here suggest substantial value for future work in this area. If expected benefits do not significantly explain the SSI application process, there is a need for determining what factors do affect applications, specifically access to affordable health insurance. We will explore measures of the availability of health insurance to reflect not only state categorical Medicaid eligibility, but also alternatives to SSI such as access to private health insurance and the expansion of Medicaid “buy in” programs such as QMB and SLMB. An analysis of these considerations may greatly advance our understanding of the SSI take-up decision.

back to the top.


Quick Reference 2004 Fact Sheet Available On-line

The Office of the Deputy Commissioner for Legislation and Congressional Affairs produces an annual Fact Sheet that provides a quick reference on a range of information related to program benefits and other statistics, such as numbers of beneficiaries in various categories. For the Fact Sheet, go to: .

back to the top.

Research Brief

Back to Work: Expectations and Realizations of Work After Retirement by Nicole Maestas, RAND

Executive Summary

When we think about retirement, we often have in mind the traditional image of an older worker who one day stops working and goes on to pursue a life of leisure. However, researchers have noted that many people do not follow the traditional retirement path; some retire only partially, and still others retire for a time then go back to work. This latter group is of particular policy interest. Do they “unretire” because they have to or because they want to? If people retire expecting to go back to work, then retirement patterns are more complex than the traditional image would suggest. On the other hand, if people retire expecting not to work, then we must ask whether failures in planning, saving, or unexpected financial losses forced them back into the labor force. In this Research Brief, I report on a study that analyzes expectations and realizations of work after retirement in order to understand whether unretirement is anticipated or comes as an unwelcome surprise after retirement.


The Health and Retirement Study (HRS) is a longitudinal, nationally representative study of older Americans. The survey began in 1992 with an initial cohort of 12,652 persons aged 51-61 and their spouses. I examine retirement patterns over the period of 1992 through 2000 for a subset of respondents in the initial cohort who had not yet retired at the time of their first interview.

Summary of Major Findings

Despite countless news articles to the contrary, the evidence overwhelmingly supports the view that unretirement transitions are mostly anticipated prior to retirement. In comparisons of expectations and realizations, I find that over half (55 percent) of HRS respondents had accurate expectations of work during retirement. Some 36 percent were overly pessimistic—they expected to work but did not in fact do so. Only 9 percent were overly optimistic, expecting not to work but in fact doing so. In other words, unretirement was anticipated for all but nine percent of retirees. If anything, expectations err on the side of excessive pessimism about the future rather than uninformed optimism. Unretirement is not associated with poor retirement planning or inadequate retirement resources. Compared to those who do not go back to work, unretirees have higher earnings prior to retirement, higher total income after retirement, and similar wealth levels before and after retirement. Unretirement is not associated with preferences for reduced hours on the pre-retirement job, worries about having enough income in retirement, worries about being productive in retirement, or feeling less satisfied with retirement. Those with employer pensions are no less likely to go back to work than others, though there is evidence that pension holders formed more accurate expectations about work in retirement. Indeed, the fixed nature of pension income may help individuals form more accurate expectations of their future financial position.

I find little responsiveness to changes in financial variables, which suggests that these changes were largely anticipated before retirement. Specifically, large drops in net worth, capital income and stock values have little effect on unretirement rates. Large drops in total income are counter-intuitively associated with reduced probabilities of unretirement. Losing health insurance after retirement has no effect on the probability of unretirement after work expectations are controlled for, and those with access to retiree health insurance are no more likely to unretire than those without. Finally, I find no evidence that unretirement is systematically related to preference shocks—that is, finding retirement more worrisome than anticipated. For the most part, unretirement is qualitatively similar to partial retirement. There are two notable differences between unretirees and partial retirees.

First, those in worse health are more likely to choose a retirement path involving partial retirement rather than one involving unretirement. This might reflect their need to maintain continuous health insurance coverage during retirement. Second, those who said they did not plan to work in retirement were more likely to unretire than partially retire, which suggests that unretirement remains an option for the small subset of individuals who are overly optimistic about their readiness for retirement. Those who were self-employed prior to retirement were more likely to have been in this subgroup, which may reflect their greater exposure to financial risk through their businesses.


Unretirement is an important phenomenon. Nearly one-quarter of retirees go back to work after retiring. The unretirement rate is even higher for younger retirees—as many as 36 percent of those retiring in their early fifties go back to work. Overall, nearly one-half of retirees follow non-traditional retirement paths involving partial retirement or unretirement. Still, there is little evidence to suggest that unretirement is a bad outcome for most retirees, resulting from inadequate savings or unexpected financial losses. In contrast, the evidence suggests that retirement patterns are much more complex than the traditional image of retirement would suggest. This favors richer specifications of the retirement process in models of retirement. The empirical importance of nontraditional retirement patterns, especially among younger retirees, suggests that policymakers may wish to readdress policies that create disincentives for work after retirement, such as the Social Security earnings test for early retirees.

back to the top.

Research Brief

Economic Adjustment of Recent Retirees to Adverse Wealth Shocks By Gabor Kezdi and Purvi Sevak

Executive Summary

Since the mid-nineties, the stock market has had an unprecedented impact on the wealth of current and future retirees. Through the spread of defined contribution pension plans, an increasing number of retirement age individuals have substantial proportions of their retirement wealth invested in the stock market. While the strong performance of the market from 1994 to 2000 substantially increased the retirement wealth of those invested in stocks, the sharp decline of stock values in 2000 and the following bear market had the opposite effect on retirement wealth. Using data from the Health and Retirement Study and the Current Population Survey, this study quantifies the magnitude of the wealth loss and estimates behavioral responses of retirement age households to these losses.

Understanding the impact of market fluctuations on retirement wealth, and the responses of individuals to these impacts, is of fundamental importance to retirement policy. This knowledge is increasingly important as a greater share of retirement resources is subject to risk. This increase in risk is due to the increase in defined contribution (DC) pension plans, individual retirement accounts (IRAs), individual participation in the stock market and will be further increased if individual accounts are included in Social Security. In the 1998 Health and Retirement Study, the median retired household under age 75 had roughly 10 percent of their non-pension wealth invested in stocks. However, roughly 25 percent of retired households have over 40 percent of their wealth invested in stocks. Thus, a non-trivial number of retired households stood to lose much of their wealth after 2000. We estimate a median wealth loss between the end of 2000 and the end of 2002 of 15 percent, among retired households who held some stock outside of retirement accounts if we assume that the average retired households in our sample had a portfolio that performed as the S&P 500, and that it did not reallocate their portfolio.

Households that retired in the late nineties did so with some assumption about the standard of living they could maintain conditional on that retirement age. Those with a substantial share of resources invested in stocks now face a dilemma: reduce consumption or increase labor supply. To maintain their consumption without increasing labor supply, they must continue to withdraw from their portfolios. However, because its value has declined sharply, this strategy means they will run out of resources earlier, even given reasonable upswings in the market in the future.

Among all retirement age households, we consistently find that individuals reduce consumption in response to adverse wealth shocks: the elasticity of consumption with respect to wealth is estimated to be about five percent. This means that a 1 percent drop in wealth is associated with 0.05 percent decrease in food consumption on average. The consumption elasticity is a somewhat larger seven percent for retired households. Retired households that own risky assets and are observed to have experienced a negative wealth shock have the largest estimated elasticity. These are the households whose observed changes in wealth are most likely to be exogenous (due to the stock market). Existing studies estimate a marginal propensity to consume (MPC) of 3 percent to 15 percent, suggesting that a one-dollar increase in wealth leads to a 3 to 15 cent increase in consumption. Our estimate of a consumption elasticity of 5 percent can be directly compared to these MPCs if we assume a constant elasticity with respect to wealth.

Estimating labor supply responses proves more challenging. Pooled cross sectional data from the 1988-2002 CPS show slight increases in labor force participation rates among men and women in the years following the market peak. Logit estimates controlling for other factors that should affect labor supply demonstrate that that this increase is particularly large among stockholders. This is consistent with the prediction that stockholders were less likely to leave the labor force (and more likely to re-enter the labor force) due to losses in their retirement wealth. However, we are hesitant to interpret this as a causal effect of wealth because the differential time trend for stockowners is present well before the downturn in the stock market. In addition, panel data regressions in HRS do not provide any evidence that wealth losses are associated with delayed retirement or greater reentry.

Taken together, our results suggest that the newly retired are more likely to adjust their consumption than their labor supply in response to adverse wealth shocks. Retirement is more or less an absorbing state, for either supply or demand reasons: once an individual retires, it is very difficult to become employed once again. For the well-being of retirees, it is encouraging to see that households are adjusting their consumption very quickly.

back to the top.


University of Michigan
Institute for Social Research
426 Thompson Street
3026 ISR
Ann Arbor, MI 48106-1248

Phone: 734-615-0422
Fax: 734-615-2180