Explaining pension dynamics.
Though the full impact of labor force aging has not yet been felt in U.S. labor markets, there is already widespread controversy regarding its likely effects. Some forecaster predict with alarm that the labor market will fail to adapt smoothly to an aging populace, creating severe and long-lasting labor market shortages; other analysts contend that market forces will come into play as they have in the past, making older workers more valuable and easing shortages (Levine and Mitchell 1988). Whether and how the labor market adapts to changes in the work-force age distribution depends on how able companies are induce desired turnover patterns among older and younger employees. In this regard, the form and function of pension plans will play a key role in inducing people to remain on their jobs, or perhaps even more importantly to leave, at later ages.
Analysis of pension plan retirement incentives has been a growth industry in the labor economics profession over the last decade, though surprisingly few studies investigate the dynamic behavior of this important labor market institution. Many researchers assume pension plans are static, probably because there are no nationally representative time-series survey of pensions. The present paper documents changes in pension retirement incentives over time, using a large set of collectively bargained pension plans surveyed periodically by the U.S. Bureau of Labor Statistics since the early 1960s. Analysis of these pension plans' retirement incentives reveals how they changed over time, and some possible explanations for why these changes occurred. We evaluate empirically various reasons for observed changes in pension plan behavior. Our goal is to determine whether and how pension structures can be said to respond to labor and product market conditions, as well as to regulatory policy.
II. UNDERSTANDING PENSION RETIREMENT INCENTIVES
A. The Target Retirement Age
Understanding retirement benefit structures embedded in pension plans and why they change from time is made easier with the concept of a target retirement age. In particular, it is known that defined benefit pensions are not generally actuarially neutral; rather, one important function of such pension is to benefits so as to induce workers to leave the firm at a time consistent with the sponsoring firm's preferences. Specially, the firm is posited to have a target retirement date R*, when the value of compensation paid equals the worker's productivity. In a multiperiod context with pensions, an employee's cumulated compensation value is the sum of his discounted wage stream until retirement PDVE(R), plus the discounted pension benefit from retirement to death PDVP(R). Cumulative total compensation paid must not exceed the value of the worker's lifetime productivity VMP(R) for a competitive firm. Hence at the firm's target retirement date R*, lifetime productivity must be at least as great as cumulative total compensation including the pension discounted value (Lazear 1979):
(1) VMP(R*)[is greater or equal to]PDVE(R*) + PDVP(R*).
This structural implicit function formulation makes clear that the firm's target retirement date R* depends on factors affecting employee compensation and pension profiles.
Turning from the firm to the worker's viewpoint, we posit that workers select their desired retirement dates by maximizing remaining lifetime utility, as a function of consumption (C) and retirement leisure (L), U=U(C,L), subject to time and money constraints (Burkhauser 1979, Fields and Mitchell 1984). The time constraint recognizes the finiteness of the worker's expected remaining lifetime; the money constraint is determined by labor market earnings for as long as employment continues, and by pensions and Social Security after retirement. Define PDVY(R+) as the discounted value of total income available from retiring at age R+:
(2) PDVY(R+) = PDVE(R+) + PDVP(R+) + PDVS(R+).
Here, PDVE is the present discounted value of earnings now defined up to the worker's desired retirement date (R+): PDVP is the expected present value of the pension annuity and PDVS is the expected present value of the Social Security annuity from R+ to death. Here too, the retirement equation is an implicit function in R+ where the worker's target retirement date depends on factors reflecting preferences for work and income versus retirement leisure.
In equilibrium, firms design earnings and pension benefit profiles so as to ensure that workers' target retirement dates coincide with their own. To the extent that older workers' productivities vary across firms and workers differ in their taste for income and leisure, the jointly preferred target retirement ages will vary in the cross-section. As long as nothing else changes, employees select jobs and associated compensation profiles/retirement dates consistent with their preferences, and neither employees nor employers have an incentive to alter the status quo. An exogenous shift in any of the underlying variables may, however, induce workers to reevaluate and perhaps to alter their target retirement date. Thus, for instance, unexpected increase in Social Security retirement benefits could induce unexpectedly (and to the company, undesirably) earlier retirement. Pension plans provide one means to offset unexpected behavioral changes with greater incentives to work longer.
B. Previous Evidence and Hypotheses
Existing empirical pension studies suggest that company-sponsored pension plans do appear to be fairly dynamic, having modified their retirement incentives several times over the years. There are, as yet, no clear-cut explanations for why they behave this way. For instance, Lazear (1983) analyzed a Banker's Trust survey and found that some defined benefit (DB) pension plans (those determine benefits primarily on service) increased their early retirement incentives over the five-year span between 1975 and 1980. He hypothesized that the observed pension changes were attributable to the raising of the legal mandatory retirement age in 1978, though no formal testing of this hypothesis was undertake. A second study, by Bell and Barclay (1987), investigated a handful of large pension plans observed from 1974 to 1983, and again concluded that the payoffs to early retirement increased substantially during this period. (No rationales were given). Most recently, Mitchell and Luzadis (1988) evaluated pension changes over a much longer period than any of the previously mentioned analyses, but did so for only 14 pension plans. Here too, they reported marked improvements in pension benefit early retirement incentives, particularly in cases where mandatory retirement had been in effect previously.
Building on previous literature, we propose to examine here two general explanations for changes in pensions over time: the buyout hypothesis, and the response to regulation hypothesis. Each is developed in turn.
First, the "buyout" hypothesis. This theory holds that pension plans have induced earlier retirement in the United States in response to declines in the relative value of employing older workers. One factor which undoubtedly altered the terms of trade across age groups in the last two decades was the large size of the "baby boom" cohort. Indeed, as Russell (1982) and others have shown, when this group entered the labor market it depressed young workers' earnings relative to those of older workers. Some companies doubtless used their pension plans to enhance early retirement's appeal, permitting them to substitute low-wage younger workers in place of more expensive older employees.(5) If pension plans did perform this buyout function, it would be expected that enhanced early-out pension incentives would become most prevalent for the most expensive, longest-tenured employees.
This labor market rationale for the buyout hypothesis should be complemented with a product-market explanation recognizing sector-specific shifts in profitability over the last two decades. At least some firms seeking to downside but finding themselves unable to adjust older workers' wage profiles altered pension benefit offerings over time (Mutschler and Schulz 1987). Hence the product-market aspect of the buyout hypothesis postulates that a trend toward early retirement in pension incentives was probably most evident in declining industries. Both aspects of the buyout hypothesis are examined empirically below.
A second, but not mutually exclusive, explanation for changes in pensions over time is the "response to regulation" hypothesis. By this we mean that early retirement pension incentives were probably altered in response to myriad changes in the pension regulatory environment between 1960 and 1980. One important legal change pertained to the raising of mandatory retirement ages: in the mid-1960s, Congress outlawed mandatory retirement earlier than age 65, and then in 1978 raised the age to 70.(6) Prior research suggests that there may be something to this hypothesis: our earlier study on a handful of pension plans suggested that firms imposing mandatory retirement during the 1970s were also those who enhanced their early retirement benefits by 1980 when the government circumscribed the practice (Mitchell and Luzadis 1988). Another change in the legal environment came with a series of regulations governing both form and function of private pension plans, beginning with the Employee Retirement Income Security Act (ERISA) in 1974, and continuing at a rapid pace ever since (Clark and McDermed 1990, Turner and Beller 1989). Yet a third important set of regulatory changes affecting the pension environment had to do with frequent revisions in Social Security rules which made earlier retirement more attractive over time. Benefit entitlement levels and incentives for retiring early rose significantly,(7) and at least some firms may have opted to offset Social Security early retirement enhancements by altering pension benefit formulas in the opposite direction over the years. Below we evaluate empirically several tests of the "response to regulation" hypothesis.
III. Empirical Methods
A. Sample of Pension Plans
The pension plans examined below are defined benefit collectively bargained pension plans described in a series published through the years by the U.S. Bureau of Labor Statistics.(8) Using the published pension descriptions, we produced computer algorithms projecting pension benefit amounts for each plan, assuming workers attained age 60 in 1960, 1970, and 1980. Of the 83 pension plans available for analysis, 42 percent were in durable manufacturing, 24 percent were in nondurable manufacturing, 16 percent in transportation, communications and utilities, 10 percent were in services, and 8 percent were in construction and mining. (Data are described more fully in the appendix.)(9)
Since pension plans tend to reward employees differently depending on their age at retirement, marital status, seniority with the firm, and play, benefits are computed in each plan for 60 illustrative workers whose characteristics span a reasonable range of possible combinations of characteristics. Early retirement pension payments were assumed to be determined by rules in effect as of that date (using age 60 as the earliest possible retirement age). Benefit amounts for retirement delayed beyond age 60 build in expectations regarding how each pension plan would have been expected to increase benefits over time (both prior to and after retirement). The BLS data source offered no direct insight on post-retirement increases these pensions plans might have granted, so we were required to make an educated assumption about them in order to value expected future benefit streams. Two alternative inflation adjustments on pension benefits were explored: in the first case, benefits were assumed to be constant in nominal terms, and in the second case nominal benefits were assumed to be fully adjusted for inflation. (Reality lies somewhere between these alternatives though probably closer to the nonindexed case for the majority of pension plans in this study; see Allen, Clark, and Sumner (1986). While the choice of inflation assumption did alter computed pension benefit amounts, it did not affect qualitative conclusions in the multivariate models and had virtually no impact on quantitative estimates, as will be demonstrated below. As a final step, all annual pension benefit amounts were converted to present values relevant to each retirement age using a 2 percent real discount rate as well as mortality probabilities based on life tables specific to workers reaching age 60 in each of the three decades.
One additional factor is important in differentiating among the BLS pension plans in our sample, and his is whether plans were operated by a single employer, or incorporated employees from a number of different employers. The latter type of pension, commonly termed a multiemployer plan, constituted 39 percent of the BLS sample and was concentrated in trucking, construction, and retail trade. Previous studies have shown that multiemployer plans operate under economic constraints and regulatory obligations which differ from those faced by the single employer plans (Mitchell and Andrews 1981). In addition, most multiemployer plans are quite large and permit considerably more portability than do their single employer counterparts, allowing workers to carry their coverage with them from one job to the next (Employee Benefit Research Institute 1985). Because multiemployer plans appear distinct from other plans in these important respects, separate analysis of this group is warranted.
Multivariate regression was used to explore the relationship between pension outcomes and the set of dependent variables of most interest. Explanatory variables reflect the hypothesis developed above.
1. Dependent Variables
There are several ways to parameterize a pension plan's retirement incentives. Some prior empirical studies have examined indicators of pension plan structures such as the early or normal retirement ages, though our preferred approach is to analyze how benefit levels change as retirement is postponed. The key variables describing pension benefit values and provisions can thus be divided into two groups for initial examination: (1) Pension provision variables, including indicators of the ease with which workers can retire early or late compared to the normal retirement age; and (2) Pension value variables, including the age at which pension present value peaks, and the dollar value of incremental pension benefits earned as retirement is deferred.
Eight specific "pension provision" variables, or key structural features of pensions, might be thought to influence whether a worker chooses to retire early, normal, or late. These include the pension plan's early retirement age, if any was permitted; the years of service required to qualify for early retirement; whether a supplement was added to early retirees' benefits; the normal retirement age; the years of service required to qualify for normal retirement; whether the plan had a mandatory retirement policy; and the plan's mandatory retirement age. In addition we include a term indicating whether service was credited by the plan after the worker attained normal retirement age.
A second set of pension indicators we believe to be more economically meaningful: these are "pension value" measures, which directly attest to the plan's built-in financial incentives and penalties associated with retiring early, normal, or late. Three dependent variables are the subject of special attention in this analysis. The first indicates the age at which the worker's pension present value stream attained a maximum under the rules in effect for a given year, PeakAge. The second dependent variable reflects how a worker's pension present value changed if retirement were delayed from 62 to 65, and is termed Pslope 65-62. The third variable, Pslope67-65, refers to the same concept but now measured between age 65 and 67. Previous research suggests that variables constructed in this fashion are useful measures of the incentives to retire early versus late (Burkhauser 1979, Fields and Mitchell 1984).
2. Explanatory Variables
Above we noted that target retirement dates differ from one pension plan to another for various reasons, including differences in product market and labor market conditions faced by sponsoring employers. In order to test the buyout hypothesis more directly, we postulate that firms in declining or unprofitable businesses will structure their pensions so as to lower their target retirement date. The empirical model therefore incorporates a profitability variable, which measures the average return on investment in the industry in the five years preceding the observation (e.g., the 1955-60 two-digit industry average is used for a pension plan observed in 1960). Differences in industry growth are captured using the change in industry employment over five years preceding the year in question. In general, profitable and growing industries are expected to encourage rather than discourage continued work at older ages. Hence, pension slopes are anticipated to be higher and the age higher at which the pension plan benefits peak, in these cases.
We next focus on employee variables needed to test the buyout hypothesis. Ceteris paribus, firms would probably wish more highly paid employees to retire earlier, with the same holding true for workers with more years of service.(10) This is particularly true in defined benefit pension plans because pay and service is usually explicitly rewarded in the benefit formula. For these reasons we include a variable indicating the amount of job tenure attained by each worker at age 60, and a measure indicating whether a worker's earnings were at or above the Social Security maximum taxable wage level (the Social Security threshold). Both variables would be expected to be related to earlier retirement incentives.
Testing the hypothesis regarding responses to regulation is more problematic because of the difficulty of constructing appropriate regulation variables. Anticipated Social Security benefits are computed for each worker assuming he left employment at alternate dates, which should capture Social Security's exogenous effects on the retirement incentives across employees over time. Previous research shows that Social Security benefit rules may be usefully summarized by indicators reflecting an early retiree's expected present value of Social Security benefits, and also by the change in benefit entitlements if the worker deferred retirement (Fields and Mitchell 1984). The present value of Social Security benefits a worker was entitled to if he retired at age 62 is defined as SSvalue 62 (and at age 65, SSvalue65); the change in the present value of the worker's Social Security benefits between the ages of 62 and 65 is SSlope 65-62 (between 65 and 67 it is SSlope67-65). The empirical model also incorporates time effects, to indicate whether the 1960s, an era of relatively little governmental reform and pension regulation, differed from later periods. It is recognized that time dummies cannot prove conclusively whether regulatory changes had an impact, but they should be suggestive about the hypothesis. In addition we estimate separate models for single and multiemployer pensions, in order to exploit the fact that multiemployer plans have been relatively less heavily regulated in the United States than the single employer pensions in the last 30 years.
IV. Empirical Results
A. Descriptive Statistics
Table 1 reports mean value of key plan provisions for our sample of 60 workers in each of 83 pensions over time. Information is provided on the early, normal, and mandatory retirement ages as well as services requirements and other plan provisions. The top panel of the table indicates that pension incentives for early retirement increased over the period under study. Both the average early retirement age and the service requirement for early retirement dropped by roughly one full year between 1960 and 1980. In addition, monetary supplements to induce early retirement became more prevalent over time. The bottom panel reiterates this conclusion, showing that the age at which pension discounted values attained a maximum (PeakAge) declined by more than a full year. Also deferring retirement became more costly over the period, in that pension benefit increments fell as retirement was deferred from age 62 to 65 (Pslope65-62). On all these counts, then, early retirement pension incentives appear to have grown more prevalent and more generous among these collectively bargained pension plans over the period under study.(11)
Inducements to retire at the normal age or beyond changed much less, but nevertheless the average normal retirement age dropped by about six months over the period under study. Several other changes also enhanced the incentives for retiring earlier: marginal gains to deferring retirement beyond age 65 (Pslope67-65) became substantially more negative over the time period examined, and the average age of mandatory retirement dropped by a half year while at the same time mandatory retirement became more prevalent.(12) Each of these increases the payoff to retiring earlier rather than later. The picture is not completely clear cut, however, since between 1960 and 1970 plans began to require more service for normal retirement and were more willing to credit service beyond the normal retirement age. Both of these changes created (other things equal).
(1.) Mitchell and Luzadis (1988) review several studies which make this assumption. (2.) Several previous studies have found conclusive evidence that nonneutral benefit formulas strongly influence older workers' retirement decisions: see Burkhauser (1979); fields and Mitchell (1984); Gustman and Steinmeier (1989); and Stock and Wise (1988). (3.) One prior study has suggested that pension incentives inducing workers to retire and actual retirement dates do vary across firms and employees in the cross-section; however, that analysis evaluated retirement patterns of workers in only 14 pension plans (Fields and Mitchell 1984). (4.) In a fully flexible world firms might also be expected to readjust their wage profiles, but current law makes wage adjustments for older workers difficult if not impossible. In this event, firms will be forced to alter their pension plan offerings. (5.) Evidence on substitution among workers of different ages is offered by Levine and Mitchell (1988). (6.) Mandatory retirement has since been eliminated for most jobs in the private sector. (7.) Prior to 1970, those who postponed benefit acceptance beyond age 65 received no actuarial adjustment to their Social Security benefits (Quinn, Burkhauser, and Myers 1990). The evidence suggests, however, that additional work still paid off for many people, since higher earnings at older ages raised Social Security benefits via the system's benefit recomputation procedure. By the 1980s, workers postponing retirement beyond age 65 received discounted benefits that were much lower than those they would have received by retiring earlier (Fields and Mitchell 1984, Ippolito 1990). (8.) To make benefit calculation comparable across years, all values are computed in 1970 dollars. Of the plans under study, 13 percent were integrated with Social Security in all years, 66 percent were not integrated in any year, and the remainder changed integration status during the period. With respect to plan type, 30 percent could be termed "pattern" plans throughout the period in that benefit formulas depended on service alone, 38 percent had benefit formulas which depended on earnings in some way, 4 percent structured benefits as a flat amount upon attaining eligibility, and the remainder of the plans revised their plan types over the period. (See U.S. Department of Labor 1961, 1970, and 1976-78). (9.) Some plan years in the BLS series were missing; a full list of plans and plan years is available on request from the authors. Plan descriptions were missing when companies went out of business or were bought out; likewise new plans were added to the sample by the BLS over twime when companies (and pension plans) were started. (10.) This is true holding constant productivity, which is difficult to do in practice as is noted in the discussion of the empirical results below. (11.) These findings also compliment evidence form large firms with pension plans covering 100 or more workers recently reported by Ippolito (1990). (12.) This pattern may seem surprising in light of the prohibition of mandatory retirement in late 1978, but is explained by recognizing that some plans had not yet been revised by the time the BLS pension survey was performed.
Rebecca A. Luzadis is an assistant professor of management at Miami University. Olivia S. Mitchell is a professor of labor economics at Cornell University and a researcher at the National Bureau of Economics Research. This research was funded by the Social Security Administration under Grant No.10-P-98289-1, Section 702 of the Social Security Act, in conjunction with the National Bureau of Economic Research programs on Labor Studies and the Economics of Aging. Excellent computer programming was provided by Vivian Fields, and capable research assistanship by Michelle Ciurea, Angela Mikalauskas, Ed Montemayor, Amy Myers, and Silvana Pozzebon. The authors remain solely responsible for opinions expressed herein. an incentive for workers to delay retirement. On the other hand the early rise in service requirements was reversed by the end of the period.
In sum, while changes in the sample means over the years do not tell an entirely consistent story, the overall trend was toward greater rewards for either retirement.
One question of interest to researchers is whether indicators of pension provision structures, such as whether plan had a mandatory retirement provision or an early retirement supplement, are good indicators of plans' financial retirement incentives. Key correlation coefficients between pension provision and pension value variables appear in Table 2, and prove to be remarkably small in this sample of pension plans. These low correlations suggest that it is useful and probably necessary to have access to actual pension benefit formulas when analyzing retirement incentives, rather than just the indicator variables so often available in public use datafiles such as the Current Population Survey. The low correlations also prompt us to focus only on the more economically interesting pension outcome terms, the pension value variables, in the multivariate analysis described next.
Table 2 Correlations Between Pension Provision and Pension Value Variables Pension Value Variables PeakAge Pslope65-62 Pslope67-65 Pension Provisions Early retirement age .10 .001 .23 Early supplement provided -.21 -.06 -.12 Service required for early -.08 .05 .10 Normal retirement age .13 .01 .09 Service required for normal -.11 -.01 .10 Service credited after normal age -.04 -.08 .30 Mandatory retirement in effect -.11 -.05 -.45 Mandatory retirement age .09 .001 .61
B. Regression Results: Single Employer Sample
Estimated determinants of the three pension outcome variables among single employer plans are reported in the left panel of Table 3. (Multiemployer plan results are discussed below.)(13) Here benefits are assumed to be constant in nominal terms; results for models using constant real b;enefits are similar in sign and relative magnitudes. Derivatives using both inflation assumptions appear in Table 4.
The buyout hypothesis implied that firms in industries experiencing profitability and growth should not be those offering early retirement incentives, while companies in declining industries might use their pensions to reduce the fraction of older, more expensive workers. This surmise is partly corroborated as can be seen from the first line of Table 3. Workers appear to be encouraged to remain employed at least to age 65 when profitability is high, as seen by the result that the peak pension age is higher in more prosperous industries. Rewards for continued work between age 62 and 65 are likewise higher where returns on investment are greater, though the coefficient is not statistically significant. Beyond age 65, retirement incentives drop off even in profitable sectors. Contrary to expectations, firms in industries experiencing faster employment growth offer lower payoffs for continued work between the ages of 62 and 65.
[Tabular Data Omitted]
Table 4 Anticipated Change in Peak Pension Age Given a 10 Percent Change in Each Explanatory Variable Single Employer Plans Multiemployer Plans (Nominal) (Real) (Nominal) (Real) Buyout variables Profits 0.11 0.08 ns ns Employment -0.02 -0.01 -0.03 ns Highly paid 0.01 0.01 ns ns Job tenure -0.20 -0.16 -0.04 -0.06 Regulation variables SSvalue62(65) ns 0.06 ns ns Sslope65-62(67-65) ns ns ns ns Other variables Firm size ns ns 0.02 0.03 ns: Not statistically significant at p = 05. Nominal: Dollar amounts not indexed. Real: Dollar amounts indexed to CPI.
The buyout hypothesis also implies that some employers might use pensions to encourage early retirement of more expensive, longer tenured employees. Evidence here is also strong, though again not uniformly supportive. Statistically significant coefficients on the job tenure variable indicate that workers with longer service are consistently encouraged to leave earlier, facing a lower peak pension age, negative payoffs for delaying retirement from age 62 to 65, and even more negative payoffs for work beyond age 65. Nevertheless, the evidence remains mixed since highly paid workers appear to face higher peak pension ages and are offered somewhat larger gains for working up to age 65. After that age, however, even the highly paid face large benefit cuts for continuing to work as would be forecasted by the buyout hypothesis.(14) In general, then, the buyout hypothesis is supported relatively well in the data.
A firm size control is included in recognition of the fact that previous researchers have concluded that workers in large firms are more productive than are smaller firm employees (Brown, Hamilton, and Medoff 1990). Since the data set does not include measures of labor productivity we cannot test this directly, but the positive and significant effect of firm size in the peak pension age equation supports the notion that workers in larger firms are more productive - even at older ages - than are workers in smaller firms. Increments to pension streams at all ages (62 to 65, and after 65) are also positive for workers in larger firms, corroborating this view.
Evidence on the response to regulation hypothesis is best considered under two headings: Social Security effects, and other effects. Many of the Social Security variables have statistically important impacts on pension value variables at conventional significance levels, implying a strong association between retirement incentives in private pensions and the incentives embedded in Social Security benefit structures.
Despite the strength of the pension/Social Security links, the evidence offers only mixed support for this version of the regulation hypothesis. There is a strong positive association between Social Security benefit amounts and the age at which pension benefits peak, as anticipated. This is consistent with the view that employers offset rising Social Security benefits by increasing company pension incentives to delay retirement. No statistically significant relationship emerges, however, between the Social Security variables and changes in pension rewards for deferring retirement between ages 62 and 65. After age 65, both Social Security and pension incentives appear to work in the same direction, rather than offsetting one another. Thus someone with a high level of Social Security wealth at age 65 would be tempted to retire early, a tendency which would be reinforced by the negative pension increment for staying on after that age. Similarly, workers facing a drop in Social Security benefit values after age 65 would also find that their pension wealth would be lower after that age.
Other variables informative on the regulation hypothesis are the time effects, specified as year-specific intercepts for 1970 and 1980 (as compared to 1960). The estimated coefficients are only suggestive as to whether mandatory retirement and other regulation affected pension incentives, since unmeasured factors could certainly have played a role as well. On the other hand the fact that industry and labor market controls are included in all models should increase the chance that the time effects actually do reveal the impact of changes in the regulatory environment. The estimates indicate that peak pension ages fell during the 1960s and 1970s, even after holding constant all other factors included in the model. Also, the incentives for retiring early versus late were strengthened over time, and all estimated time coefficients are quantitatively larger in 1980 than in 1970. Since most important pension reforms were enacted after 1970, this evidence does not contradict the conclusion that changes in regulation induced pension plans to reward earlier retirement over time.
The relative importance of the various pension determinants is explored by means of Table 4, which reflects how the peak pension age would be expected to respond to a 10 percent change in particular explanatory variables for the single and multiemployer subsamples. The largest expected response is associated with job tenure: if this were to rise by 10 percent (two years, in our sample). PeakAge would be expected to fall by about 0.2 years. Strikingly, this finding (and in fact virtually all of the simulated changes) are robust to the inflation assumption used in estimating the underlying model. The next largest response would be induced by a change in industry profitability; if profits rose by 10 percent. PeakAge would be predicted to increase by about 0.1 years. Much smaller changes would be expected from the other variables: a .06 to .08 year rise in PeakAge is predicted in response to a 10 percent increase in Social Security early retirement benefits at age 62, and changes of only about .01 to .02 from equivalent changes in earnings levels and employment growth. In general, these responses are quantitatively fairly small, despite the statistical significance of the underlying coefficient estimates.
[Tabular Data Omitted]
C. Regression Results: Multiemployer Pension Plans
The multiemployer subsample is deserving of attention for several reasons. As noted above, these plans have been subject to every different regulatory influences as compared to single employer plans, and might have behaved differently as a consequence. In addition, the multiemployer plans are among the few in the U.S. economy which permit covered workers to change jobs and carry their pension rights with them. This portability feature makes them interesting as a potential model for increasing retirement income security among an ever-more mobile workforce.(15)
A comparison of the overall sample with the multiemployer pension group in Table 1 suggests that workers in multiemployer plans did face substantially different retirement options. The pension provision variables suggest that multiemployer pensions rewarded later retirement, most noticeably by means of more stringent early retirement criteria. For example, the early retirement age averaged about 60 for the multiemployer group, as compared to about age 58 for the entire sample in 1980. Multiemployer groups also offered greater incentives for continued work, as evidenced by the almost universal willingness to credit service after normal retirement. These patterns are reiterated in the pension value variables: peak pension ages are somewhat higher in the multiemployer subsample, benefit increments for workers wishing to remain employed between age 62 and 65 are greater, and benefit disincentives to remain employed after age 65 are much smaller. It seems safe to conclude that in the BLS dataset, multiemployer pensions rewarded continued work and deferred retirement more powerfully than did other plans. Some of these patterns grew more pronounced over time. For instance, during the 1960s, multiemployer plans had slightly lower peak retirement ages than the rest of the sample, but by the 1980s their peak pension ages were higher than the single employer average.
Factors associated with retirement incentives in the multiemployer sample appear in Table 3. Like their single employer counterparts, multiemployer plans offer some support in favor of the buyout hypothesis: the higher is job tenure, the lower are peak pension ages and dollar inducements to continue on the job. We also find that more profitable industries offer positive benefit enhancements for those remaining on the job between 62 and 65, and also after age 65; these effects are very large as compared to the single employer sample. The Social Security variables play less of a role here than found previously, with no coefficients statistically significant at conventional levels. Estimated year effects indicate that most of the observed change in peak pension ages in this subsample occurred during the earlier period, with no additional effect differentiating the later environment. This probably reflects the fact that pension regulatory activity during the 1970s was primarily focused on single employer plans.
It is somewhat surprising that, even holding other factors constant, multiemployer plans were still to reward longer worklives during the period under study, while many single employer plans took the opposite tack, offering greater incentives toward earlier retirement. One possible explanation of this finding is that multiemployer plans were and are more prevalent in sectors where workers are required to internalize their own productivity differences. This is likely to be true where output is easily measured (e.g., piecework in the garment industry, or in miles driven in trucking) and where work effort can be closely monitored (e.g., in construction). In such cases employers may have less need to use their pension plans to effect early retirement for those whose declining productivity would be more difficult to detect in other types of jobs. If this is so, it suggests that pension plan design reflects underlying differences in employment conditions.
Differences across plan types appear important in that an F-test rejects the hypothesis that explanatory variables play the same role in explaining pension incentives among the multiemployer and single employer subsamples (at the 5 percent level). Nevertheless the coefficient differences translate into relatively similar behavioral responses, as reported in Table 4. Hence the differences across plans should not be overstated.
At the outset we noted that some believe the U.S. labor market will fail to adapt smoothly to an aging workforce, whereas others contend that pensions can and will play an important role in helping companies induce desired turnover patterns. The present research suggests several important conclusions along these lines. First, the defined benefit pension plans under study instituted many changes between 1960 and 1980, several of which enhanced the financial payoffs for early retirement. These changes included increases in benefit levels, reduction in
(13.) The sample size in Tables 3 and 4 is smaller than that in Table 1 and 2 because some explanatory variables are not available for all firms in the sample for all years. The regression analysis utilizes a total of 8.220 observations on workers in each of 137 plan years; 6,540 observations for 109 plan years represent workers in the single employer group with the remainder being in the multiemployer sample. (14.) These findings extend work of Mitchell and Luzadis (1988) who examined a sample of 14 firms. It is also consistent with the contention that worker productivity does not keep pace with wage increases after the age of 65 (Mitchell 1988). (15.) The U.S. Department of Labor has recently suggested that greater pension portability would substantially increase well-being during retirement; see for instance Walker (1988). early, normal, and mandatory retirement ages, and cuts in the age at which pension present values peaked (with retirement after that age being penalized).
A second conclusion pertains to the empirical relationship between pension provision variables and pension value variables which had previously been surmised but not proved by pension analysts: it is quite weak. This suggests that simple indicators of pension plans' structural features (e.g., the plan's early retirement age) do not adequately summarize the complexity of pension incentives as they influence workers' financial retirement constraints.
A third conclusion has to do with explanations for observed pension dynamics. Some evidence supports the buyout hypothesis: that is, workers with high tenure were probably encouraged to retire earlier, thus allowing firms to hire new (and presumably less expensive) employees. This portends that changes in the demographic composition of the labor force may imply future changes in pension retirement incentives. The implications of the buyout hypothesis with respect to product market dimensions received weaker support, but firm in less profitable sectors do appear to use pension incentives to buy out older workers.
Additional information is available regarding two other policy issues. First, the regulatory environment in which pensions operate was found to have an important impact on employer-sponsored pension retirement incentives, particularly wit regard to single employer plans' responses to Social Security rule changes. Less direct evidence was available on other changes in the pension regulatory environment, but the single employer sample offers some indirect support for the view that employers altered pensions so as to offset the legal raising of the mandatory retirement age. Second, multiemployer plans appear to have evolved differently from single employer plans. Specifically, multiemployer plans in this sample were less likely to discourage continued employment at older ages than single employer pensions, proving that even among defined benefit pensions, economic incentives differ widely. The evidence on the multiemployer plans indicates that government efforts to enhance portability may also facilitate another government policy objective - namely, delayed retirement.
This research contributes to knowledge by focusing on the largest set of pension plans ever followed for such a long time period. On the other hand, it remains a small sample, and a collectively bargained sample at that. Nevertheless, with these caveats, it is safe to conclude that retirement incentives inherent in employer-sponsored pensions can and do change markedly over time. Further, pension plan behavior is systematically related to labor force characteristics, firm profitability, and labor market regulation. Our results imply that government as well as corporate retirement policy must recognize more fully the dynamic nature of pension plans.
I. Pension Variables
Pension benefits were computed for 60 illustrative workers obtained by specifying five different earnings levels, six tenure levels, and workers in two marital status groups (single and married). The earnings profiles examined included an average earnings profile reflecting actual pay of older workers covered by pensions taken from the Longitudinal Retirement History Survey and two additional profiles at 20 percent above and below this average level. LRHS earnings streams were moved back to the 1060s and forward to the 1980s using the inflation rate. This approach held constant real earnings across the decades so as to better compare resultant benefit patterns. In addition a low-wage profile (a minimum-wage worker) and high-wage profile (someone at the Social Security taxable maximum) were examined. Tenure profiles were assumed to vary according to the number of years of service accumulated as of age 60; ten, 20, and 30 years were used, as well as 15, 22, and 25 representing low, mean, and high years of seniority found in earlier research (Fields and Mitchell 1984).
Annual pension amounts are computed by evaluating expected benefits for each of the plans described in the series of U.S. Department of Labor publications entitled Digest of One Hundred Selected Pension Plans Under Collective Bargaining (various issues). These amounts were then converted to present values using a 2 percent real discount rate, as well as survival probabilities based on life tables specific to workers reaching age 60 in each of the three decades (see Mitchell and Luzadis 1988). Alternative inflation assumption described in the text were used to inflate pre- and post-retirement benefits. Pension variables were taken directly from the Digest description and coded as indicated in the text.
II. Social Security Benefit Variables
Social Security benefit and present values were constructed by evaluating for each worker in the sample the benefits he would receive if he retired at the specified age. Because Social Security regulations changed over time, we employ the rules in effect at the time the worker turned age 60 (in 1960, 1970, and 1980, respectively). Net present values of real Social Security benefits at age 60 assume retirement at age 60 followed by filing at age 62 when the retiree is first eligible. Other present values are defined assuming that the worker retired at that age and immediately applied for Social Security. Present value computations assume a real discount rate of 2 percent, real and mortality figures applying to the cohort in question. In all cases Social Security present values are computed assuming that nominal benefits are inflated at the same rate as had prevailed in the decade prior to the worker attaining age 60. (For a justification see Fields and Mitchell 1984, Mitchell and Luzadis 1988).
III. Other Explanatory Variables
Time Effects: Time effect is equal to one if the observation was from
1970 (1980), 0 otherwise. Job tenure: Years of service at age 60 (by construction).
Highly paid: Equal to U if earnings at Social Security taxable maximum
in all years, 0 otherwise (by construction). Firm Size: Employees per firm (U.S. Bureaus of the Census, County Business Patterns). Growth: Five year employment change (U.S. Bureau of the Census, County Business Patterns). Profits: Five year (quarterly average) rate of return on investment (U.S. Bureau of the Census, Quarterly Financial Report for Manufacturing Corporations). Table 5 Means of Explanatory Variables (1970 dollars) Pension Plans Observed in 1960 1970 1980 Buyout variables Profits 9.95 11.45 13.87 Employment -125.97 966.80 1,685.55 Highly paid 20.33 20.33 20.33 Job tenure .40 .40 .40 Regulation variables SSvalue62 ($) 24,099 35,857 27,857 SSvalue65 ($) 26,077 40,297 28,535 Sslope65-62 ($) 1,977 4,440 678 Sslope67-65 ($) -2,524 -1,819 -1,799 Other variables Firm size 78.18 77.52 72.99
Allen, S. G., R. L. Clark, and D. A. Summer, 1986. "Postretirement
Adjustments of Pension Benefits." Journal of Human Resources
21(1):118-37. Bell, D.m and W. Barclay. 1987. "Trends in Retirement Eligibility and Pension
Benefits, 1974-83." Monthly Labor Review 110(4):18-25. Brown, c., J. Hamilton, and J. Medoff. 1990. Employers Large and Small.
Cambridge, Mass.: Harvard University Press. Bureau of Labor Statistics. Various years. Digest of One Hundred Selected
Pension Plans Under Collective Bargaining. Washington, D.C.: GPO. Burkhauser, R. V. 1979. "The Pension Acceptable Decision of Older
Workers." Journal of Human Resources 14(1):63-75. Clark, R. L., and A. A. McDermed. 1990. The Choice of Pension Plans in a
Changing Regulatory Environment. Washington, D.C.: American Enterprise
Institute. Employee Benefit Research Institute. 1985. Fundamentals of Employee Benefit
Programs, 2nd Edition. Washington, D.C.: EBRI. Fields, G. S., and O. S. Mitchell. 1984. Retirement, Pensions and Social
Security. Cambridge, Mass.: MIT Press. Gustman, A. L., and T. L. Steinmeier. 1989. "An Analysis of Pension Benefit
Formulas, Pension Wealth, and Incentives From Pensions. " In Research in
Labor Economics, ed. R. Ehrenberg, 53-106. Greenwich, Conn.: JAI Press. Ippolito, R. 1990. "Towards Explaining Earlier Retirement After 1970."
Industrial and Labor Relations Review 43(5):556-69. Kotlikoff, L., and D. E. Smith. 1983. Pension and the American Economy.
Chicago: University of Chicago Press. Lazear, E. 1983. "Pension as Severance Pay". In Financial Aspects of the
United States Pension System, eds. Z. Bodie and J. Shoven, 57-89. NBER.
Chicago: University of Chicago Press. Lazear, E. 1979. "Why Is There Mandatory Retirement?" Journal of Political
Economy 87(6):1261-84. Levine, P. B. and O. S. Mitchell. 1988. "The Baby Boom's Legacy: Relative
Wages in the 21st Century." American Economic Review 78(2):66-69. Mitchell, O. S. 1988. "The Relation of Age to Workplace Injury." Monthly
Labor Review 111(7):8-13. Mitchell, O.S., and E. Andrews. 1981. "Scale Economics in Private
Multiemployer Pension Systems." Industrial and Labor Relations Review
34(4):522-30. Mitchell, O. S., and R. A. Luzadis. 1988. "Changes in Pension Incentives
Through Time." Industrial and Labor Relations Review 42(1):100-108. Mutschler, P. H., and J. H. Schulz. 1987. "Factors in the Retirement
Decision." Brandeis University. Mimeo. Quinn, J., R. V. Burkhauser, and D. A. Myers, 1990. Passing the Torch: The
Influence of Economic Incentives on Work Retirement. Kalamazoo.
Mich.: W. E. Upjohn Institute for Employment Research., Russell, L. B., 1982. The Baby Boom Generation and the Economy. Brookings
Institution: Washington, D. C. Social Security Administration, 1986. Social Security Bulletin, Annual
Statistical Supplement. Washington, D.C.: GPO. Stock, J. H., and D. A. Wise. 1988. "Pensions, The Option Value of Work,
and Retirement. Cambridge, Mass.: NBER Working Paper No. 2686. Turner, J., and D. Beller, eds. 1989. Trends in Pensions. U.S. Department of
Labor, PWBA. Washington, D.C.: GPO. U.S. Bureau of the Census. Various years. County Business Patterns.
Washington, D.C.: GPO. U.S. Bureau of the Census. Various years. Quarterly Financial Report for
Manufacturing Corporation. Washington, D.C.: GPO. U.S., Department of Labor. Various years. Digest of Selected Pension Plans
Under Collective Bargaining. Washington, D.C.: GPO.' Walker, D. M. 1988. Testimony on Pension Portability, House Ways and
Means Oversight Subcommittee. U. S. Congress.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||includes appendix|
|Author:||Luzadis, Rebecca A.; Mitchell, Olivia S.|
|Publication:||Journal of Human Resources|
|Date:||Sep 22, 1991|
|Previous Article:||Do the school nutrition programs supplement household food expenditures?|
|Next Article:||Opportunistic behavior by firms in implicit pension contracts.|