NBER Reporter: Summer 2000
Pensions and Social Security are fundamental parts of saving. They each account for about a quarter of the $500,000 in total wealth held on average by families approaching retirement age. (1) Pensions have a large effect on retirement incentives as well. A man with a defined benefit (DB) pension plan who works in the year before qualifying for early retirement on average will find his benefits increased by about 60 percent of one year's pay. For a woman, the increase amounts to about one-third of a year's pay. (2) Accordingly, it is necessary to have accurate measures of pensions and Social Security in order to measure the wealth of those entering retirement, to understand saving and retirement behavior, and to determine the true impact of policies meant to influence saving and retirement.
The Health and Retirement Study
To further our understanding of pensions, Social Security, and their effects, we recently helped to develop and analyze data from the Health and Retirement Study (HRS). (3) The HRS, originally fielded in 1992 as a panel survey of 12,652 individuals from households with at least one member born in 1931-41, now has added additional cohorts (age groups) so that it is representative of the U.S. population over the age of 50. Crucial to our work, the HRS has collected pension Summary Plan Descriptions (SPDs) from the employers of two-thirds of those in the original survey who were covered by a pension in 1992. The HRS also collected detailed descriptions of pensions from this group's employers on previous jobs. Also central to our work, 80 percent of the survey respondents granted permission to the HRS to obtain their earnings histories from the Social Security Administration. Social Security records were linked successfully for 95 percent of those granting permission, or 75 percent of the HRS sample.
The HRS cohort is an interesting group to study: they are the group closest to retirement age and the first covered by Social Security to learn that the present value of their benefits will fall below the present value of their taxes paid. (4) Thus, the HRS data can teach us a great deal about: retirement incentives; the relationship of pensions and Social Security to total saving and retirement outcomes; complexities in behavior beyond those incorporated in the simple life-cycle model; and the relationship between pension and Social Security policies and the distribution of benefits plus behavioral outcomes.
Distributions of Pensions, Social Security, Wealth, and Lifetime Earnings
From the HRS data, we have learned a great deal about the distribution of total wealth and its components -- including pensions and Social Security -- and about how they vary with lifetime earnings, for households and for individuals. Contrary to the general impression, pensions are distributed widely among households. Although only half of the employed individuals in the HRS have a pension, three-fourths of HRS households were covered by a pension at one time, and two-thirds of HRS households own the rights to a pension or pension income. In 1992, pension wealth was worth $191,000 per HRS household with a claim on a pension. Moreover, less than 10 percent of pension wealth has been lost because covered respondents have cashed out their benefits after leaving a pension-covered job.
The share of family wealth held as pensions increases with family lifetime earnings, rising from less than 5 percent of total wealth for those in the bottom 10 percent of lifetime earners, to 30 percent of total wealth for those in the 75th to 95th percentiles of lifetime earners. (5) Although men hold pensions that are much more valuable than the pensions held by women -- for example, at age 55, men hold DB pensions worth $200,000, while women's holdings are $108,000 -- the differences are explained largely by differences in earnings. Benefit-earnings ratios are actually higher for women with DB plans than men: for example, at age 60, benefit-earnings ratios are 20 percent for women and 16 percent for men. From age 55 on, the ratio of pension accrual-to-earnings is higher for women than men: for example, at age 60, it is 10 percent higher. (6)
Contrary to the impression left by the progressive Social Security benefit formula, Social Security does not do a very good job of redistributing benefits among families. When families are arrayed by their lifetime earnings, redistribution of benefits amounts to less than half of what is suggested by the progressive benefit formula as applied to individuals. Low-wage earners are primarily women who have worked for fewer years than men have but are married to a spouse with higher earnings; thus they enjoy disproportionately high spouse and survivor benefits under Social Security. Most of the redistribution among families, therefore, is from families in which both spouses work much of their lives to families in which only one spouse had a long-term commitment to the labor market.
When individuals are ranked according to their earnings, almost 11 percent of total Old-Age and Survivors Insurance (OASI) benefits are redistributed from those with high lifetime earnings to those with low lifetime earnings. When families are ranked according to lifetime earnings, only 5 percent of total benefits paid are redistributed. Further, when households are ranked according to their earnings in years when they are fully committed to the labor market, a little more than 2 percent of total Social Security benefits are redistributed. So, if those currently approaching retirement had a system of privatized accounts instead of benefits determined by the current Social Security formula, there would be little difference in benefit redistribution. (7)
The same mechanism that favors families whose members work under Social Security only part of their lifetimes also favors immigrants. The benefit calculation assumes that in the years before entering this country, the immigrants had no earnings. The mechanism's effect is to lower the Average Indexed Monthly Earnings (AIME) on which their benefits are based; using the progressive segment of the benefit formula generates benefits that are a higher share of their average yearly earnings than someone born in the United States who had the same average yearly earnings would receive. Simply prorating the benefits of immigrants based on time spent in the United States would reduce their benefits by 7 to 15 percent, saving Social Security an additional $50 billion to $100 billion in present value. Nor is the favorable treatment of immigrants under Social Security justified by differences in income or wealth. Indeed, the current formula favors immigrants with high earnings who have been in the country a relatively short period of time. (8)
The same mechanism that favors the treatment of immigrants and families whose members work only part of their lifetimes also will favor those who would opt out of a voluntary Social Security system, especially during the transition to such a system. (9) But the safeguards that could limit benefits for immigrants could also work during the transition to a voluntary privatized system. One possibility would be to average benefits only for the time the individual has spent in covered employment, and then to reduce the basic benefit by the ratio of years-spent-in-covered-employment-to-total-years-used-in-computing-benefits, for example, the 35 years currently used as the base period when counting AIME.
Pensions Do Not Stand Still
Changes in coverage and plan generosity alone have more than doubled the value of pensions held by older households between 1969 and 1992, increasing the value of pensions by 145 percent. (10) There have also been fundamental changes in other pension characteristics. Most notably, the typical covered worker now is more likely to have a defined contribution (DC) plan, that is, a pension held in the form of an account. Looking at a panel of pensions between 1983 and 1989 alone, we observe that the early retirement date declined by a year, while the generosity of early retirement benefits increased. By 1992, three-fourths of those in the HRS with a DB plan were eligible for early retirement benefits by age 55. (11) Although changes in pensions are exaggerated in cross-sectional comparisons, they are large enough that studies matching plan descriptions to panel data should refresh the pension plan matches periodically. (12) That is, in panel data the pension plan descriptions obtained from employers and matched in a particular year should not be treated as if they remain unchanged in all future years. Moreover, pension plans have changed so sharply over time that those who are now retiring could not have projected the characteristics and values of plans that are currently in force.
Pensions, Social Security, and Saving (13)
We next consider the relationship between total wealth and lifetime earnings, and its implications. Except for those in the bottom decile of the lifetime earnings distribution, the ratio of wealth -- including the present value of pensions and Social Security -- to lifetime earnings is relatively constant, averaging about 40 percent from the 25th to 95th percentiles of the lifetime earnings distribution. With these wealth-earnings ratios, the majority of those in the HRS are well prepared for their retirement. Annuitizing their wealth on the assumption of a two-thirds joint and survivor annuity, at their expected retirement age, the median nominal replacement rate for HRS households will be 96 percent of final earnings, while the median real replacement rate will be 62 percent of final earnings. Only those with the lowest quarter of replacement rates appear to be in significant trouble.
Because pensions and Social Security are such a large share of wealth, it is natural to ask whether they affect saving. We use the wide variation in pension plans in the HRS to measure substitution with other forms of saving. We find that pensions add to total wealth by at least half the value of the pension, and in most specifications by a good deal more. Thus, those with higher pensions do not reduce their holdings of other wealth correspondingly. Although our finding of a constant wealth-to-earnings ratio across much of the lifetime earnings distribution also might be construed as evidence in favor of the simple life-cycle hypothesis, the lack of substitutability between pensions and other forms of wealth suggests a more complicated relationship.
Pensions may be a good substitute for other forms of wealth, but this is not apparent in our regressions; because people with pensions are informed by their employers and others about the need for retirement saving, they save more. But even if those with pensions are better informed about the need for retirement saving than those without pensions, they are still not very well informed about their own pensions. Only half the HRS respondents with linked pension data correctly identify their plan type, and fewer than half identify, within one year, the dates of eligibility for early and normal retirement benefits. Benefit reduction rates essentially are not reported. Respondents do better at reporting pension values, but the unexplained variation is still considerable. Only 40 percent of respondents who knew enough to actually report a value for their pension correctly identified the relatively wide bracket in which their plan value actually falls. (14)
Pensions, Social Security, and Retirement
In the HRS panel, we find that 77 percent of the transitions among the different retirement states -- retired, partially retired, and not retired -- involve continuations in the same state between waves; 17 percent of transitions involve a move from greater to lesser participation; and, more difficult to model, 6 percent of transitions involve a move from a state of lesser to greater labor-force participation. (15)
Comparing retirement outcomes in the HRS data with comparable information from the old Retirement History Survey (RHS), we find that both full and partial retirement are occurring at much younger ages. Although more than 75 percent of the males in the RHS were working full time at age 60, only 61 percent of the males in the HRS are. The large spike in the population leaving full-time work at age 65 observed in the RHS is reduced by half in the HRS, while the share leaving full-time work at age 62 has almost doubled over time. Of the HRS sample, 22 percent report that they were partially retired at some time in the first four waves of the survey. Changes in pensions clearly appear to have reduced the early retirement age and to have decreased the age at which early retirement actually takes place. Applying structural models of retirement, it seems that changes in pensions and Social Security are capable of explaining about one-fourth of the trend toward earlier retirement observed over the 1970s and 1980s. (16)
However, the puzzles that we find in estimating savings models do not leave us sanguine about applying simple life-cycle models to retirement behavior. Structural retirement models typically assume that saving behavior consistent with a life-cycle model is taking place in the background as respondents make retirement decisions in accordance with the incentives created by pensions and Social Security. (17) But as we have seen, saving behavior does not appear to be fully consistent with the predictions of the life-cycle model. People are poorly informed about crucial features of their pensions. Nor do they substitute pension wealth for other forms of wealth. We also find that workers who are "liquidity constrained" retire later.
Moreover, when we investigate the relationship between retirement and saving in the HRS data, we find other anomalies. A simple life-cycle model suggests that, unless there is a correlation between the tastes for leisure and time preference, the earlier one wants to retire, the more will be saved for retirement. Accordingly, those characteristics associated with earlier retirement also should be associated with higher saving. Yet we find no evidence of such a relationship in the data. (18)
We also find that reduced-form models of retirement that ignore saving are not capable of explaining the spike in retirement at age 62. There is some general confusion about how to value the cost of funds when calculating the future reward to postponed retirement. For example, when calculating the increase in Social Security benefits from working longer, it is often assumed that people value the expected actuarial increase in Social Security benefits. Yet those family members who would benefit from further delaying receipt of Social Security almost uniformly claim their benefits immediately upon retiring. In contrast, unmarried men who should claim their benefits immediately upon retiring have a greater tendency to postpone benefit acceptance. (19) Reduced-form models that relate retirement to measures of Social Security benefit accrual -- for example, peak or premium value -- predict, counterintuitively, that reducing the Social Security early retirement age will induce more people to retire early.
An Agenda for Further Research
To resolve these puzzles, we are extending our structural models of retirement to also model saving. (20) Similarly we are expanding our analyses of family retirement decisions to incorporate household saving. (21) The HRS is crucial for this analysis because, unlike many other surveys, it provides pension and employment records for both spouses in the household. We also hope to incorporate uncertainty into a dynamic structural model, while integrating the saving and retirement decisions explicitly into the model. Eventually we should be able to explain the complex flows among retirement states observed in the HRS panel. In the process, we also would like to allow for the influence on saving and retirement outcomes of imperfect information and incomplete understanding of the maximization process. This is an ambitious agenda, but one that must be met if we are to understand the effects of pension and Social Security policies into the future. Indeed, we cannot understand the effects of those modifications of Social Security policies that are already in place, such as the extension of the Social Security normal retirement age, or the increase in the delayed retirement credit to 8 percent, without understanding linkages among retirement, saving, and benefit-claiming behavior. Nor can we understand the likely effects of such fundamental policy changes as privatizing Social Security.
1. A. L. Gustman, O. S. Mitchell, A. A. Samwick, and T. L. Steinmeier, "Pension and Social Security Wealth in the Health and Retirement Study," in Wealth, Work and Health, Innovations in Measurement in the Social Sciences, J. Smith and R. Willis, eds. Ann Arbor: University of Michigan Press, 1999.
2. A. L. Gustman and T. L. Steinmeier, "Employer-Provided Pension Data in the NLS Mature Women's Survey and in the Health and Retirement Study," Research in Labor Economics, vol. 19, forthcoming. A defined benefit plan uses a formula to determine the yearly benefit paid to a retired worker. Typically the amount of the benefit is determined by years of service and final pay.
3. We are co-principal investigators of the Health and Retirement Study. F. Thomas Juster and Robert Willis are the principal investigators. Charles Brown and Olivia S. Mitchell are the co-investigators we have worked with on questions pertaining to retirement, pensions, and Social Security. Under the leadership of Richard Suzman, the HRS has been supported by the National Institute on Aging, with additional support from the Social Security Administration and the Pension Welfare Benefits Administration of the Department of Labor. Our own work also has received generous support from these agencies.
4. A. L. Gustman and T. L. Steinmeier, "Effects of Pensions on Savings: Analysis with Data from the Health and Retirement Study," Carnegie-Rochester Conference Series, 50 (July 1999), pp. 271-326.
6. A. L. Gustman and T. L. Steinmeier, "Employer-Provided Pension Data in the NLS Mature Women's Survey and in the Health and Retirement Study," Research in Labor Economics, vol. 19, forthcoming.
8. A. L. Gustman and T. L. Steinmeier, "Social Security Benefits of Immigrants and U.S. Born," in Issues in the Economics of Immigration, G. J. Borjas, ed. Chicago: University of Chicago Press, 2000.
9. A. L. Gustman and T. L. Steinmeier, "Privatizing Social Security: First-Round Effects of a Generic, Voluntary, Privatized U.S. Social Security System," in Privatizing Social Security, M. Feldstein, ed. Chicago: University of Chicago Press, 1998.
10. A. L. Gustman and T. L. Steinmeier, "Pensions and Retiree Health Benefits in Household Wealth: Changes from 1969 to 1992," Journal of Human Resources, 35 (1) (Winter 2000), pp. 30-50.
11. A. L. Gustman and T. L. Steinmeier, "Employer Provided Pension Data in the NLS Mature Women's Survey and in the Health and Retirement Study," Research in Labor Economics, vol. 19, forthcoming.
12. A. L. Gustman and T. L. Steinmeier, "Changing Pensions in Cross-Section and Panel Data: Analysis with Employer-Provided Plan Descriptions," in Proceedings, National Tax Association, November 8-10, 1998, published 1999. The Health and Retirement Study is conducting a new employer pension survey now so that researchers can incorporate changes in pension plans made since 1992, as well as provide basic information on the pensions covering the cohort that has been newly added to the survey.
13. The information in the first four paragraphs of this section is from A. L. Gustman and T. L. Steinmeier, "Effects of Pensions on Savings: Analysis with Data from the Health and Retirement Study," Carnegie-Rochester Conference Series, 50 (July 1999), pp. 271-326.
14. A. L. Gustman and T. L. Steinmeier, "What People Don't Know about Their Pensions and Social Security: An Analysis Using Linked Data from the Health and Retirement Study," NBER Working Paper No. 7368, October 1999.
16. P. M. Anderson, A. L. Gustman, and T. L. Steinmeier, "Trends in Male Labor-Force Participation and Retirement: Some Evidence on the Role of Pensions and Social Security in the 1970s and 1980s," Journal of Labor Economics, 17 (4) (October 1999), pp. 757-83.
17. A. L. Gustman and F. T. Juster, "Income and Wealth of Older American Households: Modeling Issues for Public Policy Analysis," in Assessing Knowledge of Retirement Behavior, E.A. Hanushek and N.L. Maritato, eds. Washington, D.C.: National Academy Press, 1996.
18. A. L. Gustman and T. L. Steinmeier, "Retirement and Wealth," paper presented at Michigan Retirement Research Conference, June 2000. Our findings suggest that those with a stronger taste for leisure may also have higher time preference.
20. A. L. Gustman and T. L. Steinmeier, "A Structural Retirement Model," Econometrica, 54 (3) (May 1986), pp. 555-84 and A. L. Gustman and T. L. Steinmeier, "Changing Social Security Rules for Workers over 65: Proposed Policies and Their Effects," Industrial and Labor Relations Review, 44( 4) (July 1991), pp. 733-45.
21. A. L. Gustman and T. L. Steinmeier, "Retirement in a Family Context: A Structural Model for Husbands and Wives," Journal of Labor Economics, 18 (3) (July 2000), pp. 503-45. We find that coordination of retirement by husbands and wives is not attributable to incentives, but rather to correlation in tastes. Each spouse, and in particular the husband, values retirement more once the other spouse has retired
* Gustman is an NBER Research Associate in the Program on Labor Studies and Loren M. Berry Professor of Economics at Dartmouth College. Steinmeier is Professor of Economics at Texas Tech University.