1
Introduction

Eric A. Hanushek and Nancy L. Maritato

Federal, state, and local governments are continually adopting or revising policies and programs that affect the economic well-being of the nation's citizens. Unfortunately, it is all too often true that the process of such policy development and implementation is not closely linked to scientific and analytic policy research. This situation became clear during the health care reform debates of 1993–1994. As task forces and policy makers attempted to develop legislation that would improve the provision of health care services in the United States, they found that key information was missing. The dearth of relevant information was most apparent when analysts attempted to estimate the cost of reform proposals: for each proposal, different analysts arrived at cost estimates that differed by a factor of two or three. Similar problems affect analyses of proposals in the area of retirement income security.

With the aging of the population, with increasing uncertainty about the solvency of the Social Security system, and with growing concern about the availability of adequate private pensions, retirement income policy is likely to be the focus of increasing attention. In some respects, the information necessary to formulate sound retirement income policy is more difficult to obtain than the information needed for health care reform. Economic, social, and demographic trends make the level of income security that future retirees can expect highly uncertain. In addition, long delays between the implementation of policies and the full realization of their effects add to the challenge of making informed decisions in this area. Without the appropriate information, decision makers run the risk that the proposals they adopt may be ineffective, or worse, counterproductive.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 1
Assessing Knowledge of Retirement Behavior 1 Introduction Eric A. Hanushek and Nancy L. Maritato Federal, state, and local governments are continually adopting or revising policies and programs that affect the economic well-being of the nation's citizens. Unfortunately, it is all too often true that the process of such policy development and implementation is not closely linked to scientific and analytic policy research. This situation became clear during the health care reform debates of 1993–1994. As task forces and policy makers attempted to develop legislation that would improve the provision of health care services in the United States, they found that key information was missing. The dearth of relevant information was most apparent when analysts attempted to estimate the cost of reform proposals: for each proposal, different analysts arrived at cost estimates that differed by a factor of two or three. Similar problems affect analyses of proposals in the area of retirement income security. With the aging of the population, with increasing uncertainty about the solvency of the Social Security system, and with growing concern about the availability of adequate private pensions, retirement income policy is likely to be the focus of increasing attention. In some respects, the information necessary to formulate sound retirement income policy is more difficult to obtain than the information needed for health care reform. Economic, social, and demographic trends make the level of income security that future retirees can expect highly uncertain. In addition, long delays between the implementation of policies and the full realization of their effects add to the challenge of making informed decisions in this area. Without the appropriate information, decision makers run the risk that the proposals they adopt may be ineffective, or worse, counterproductive.

OCR for page 1
Assessing Knowledge of Retirement Behavior In order to improve this situation, the Pension and Welfare Benefits Administration in the U.S. Department of Labor, with support from the National Institute on Aging, the Pension Benefit Guaranty Corporation, the Social Security Administration, and TIAA-CREF, asked the Committee on National Statistics at the National Research Council to establish a Panel on Retirement Income Modeling. The charge to the panel was not to recommend changes in public or private policies that affect retirement income security. Instead, the charge was to consider a critical adjunct to the policy process—namely, the projection models, databases, and research findings that are needed to provide reliable information about the likely short- and long-term costs and benefits of today's retirement-income-related policies and proposed changes to them. Modeling the likely future effects of any current policy or policy proposal is a difficult task at best. This is especially true when the policy pertains to retirement income security because of the need to project the effects of both current policy provisions and proposed changes to them over long periods of time. As just one example, to assess adequately the implications that tax law provisions to encourage contributions to 401(k) employer pension plans or Individual Retirement Accounts will have for post-retirement (and pre-retirement) living standards, the effects of those provisions on savings behavior must be projected over at least a generation. The long time horizon for many retirement-income-related policy projections greatly increases the uncertainty of the estimates compared with, say, the 5-year projections that are typically prepared for tax law and welfare policy proposals. Further increasing the difficulty of the modeling task and the uncertainty of the estimates is the need to take account of all, or at least the main, sources of retirement income support (and the potentially major drain on income represented by health care costs). These income sources include Social Security, employer-provided pensions, personal savings, other transfers (e.g., public assistance and public and private disability payments), post-retirement earnings from part-time employment, and bequests and other transfers among family members. For some purposes, it may be appropriate—and challenging enough—to model one of the components in isolation, such as the effects of changes in Social Security payroll taxes or benefits on the solvency of the Social Security Trust Fund and on likely rates at which Social Security benefits will replace earnings. However, such an analysis cannot provide an adequate picture of overall retirement income security if the Social Security changes in turn affect other components of retirement income (e.g., levels of employer pension coverage and other savings). Moreover, to the extent that there is interaction among the components, such an analysis may not provide an adequate picture of the effects of the changes on the Social Security system itself. Particularly when an analysis considers sources of retirement income other than Social Security, there is an added difficulty due to the heterogeneity among workers and employers. While almost all workers are covered by Social Secu-

OCR for page 1
Assessing Knowledge of Retirement Behavior rity, there is wide variation in the extent to which workers are covered by employer-provided pensions and in the provisions of such coverage. There is also wide variation in personal savings behavior. For policy deliberations, this heterogeneity becomes very important. Much of the focus of government retirement policy is on ensuring some minimum living standard for the subset of elderly people who may have been unlucky or unwise in preparing for their retirement. It is thus important that projection models produce estimates of the likely numbers of such people, as well as estimates of the average experience. Priorities for improving data, research knowledge, and models with which to inform debates about retirement income policies need to be motivated by a consideration of what is known and not known about relevant behaviors and the state of the art with respect to analytical and projection modeling. To learn from experts in the field, in September 1994 the panel sponsored the Conference on Modeling the Impact of Public and Private Policies on Retirement Behavior and Income: What Do We Know and What Do We Need to Know? Six papers were commissioned from knowledgeable researchers and policy analysts for presentation and discussion at the conference. This volume presents those papers as they were revised by the authors to incorporate the comments of discussants, panel members, and reviewers. The paper authors were asked to review the issues, literature, and state of knowledge in six topic areas:  projecting the distribution of income and wealth of retired workers and their families from all sources (Alan L. Gustman and F. Thomas Juster);  labor supply behavior that is relevant to retirement decisions and, ultimately, retirement income (Robin L. Lumsdaine);  microlevel savings behavior and substitution effects between public and personal savings vehicles and the implications for retirement income (James M. Poterba);  the behavior of firms that is relevant to workers' retirement decisions and their retirement income (Donald O. Parsons);  projecting the older population, the likely health care costs they face, and the implications for retirement income (Ronald D. Lee and Jonathan Skinner); and  estimating the overall effects of policy changes on future retirement income security in a context that includes the macroeconomic effects of government tax and transfer policies (Gary Burtless). The authors of the first five papers were asked to include the following: (1) an introduction about the relevant policy issues in the area; (2) a review of the current literature, focusing on why some questions have not yet been satisfactorily answered; (3) modeling approaches, both for improving research knowledge and for providing answers to policy questions; (4) an indication of the desired

OCR for page 1
Assessing Knowledge of Retirement Behavior state of the art with regard to data, research, and modeling strategies; and (5) recommendations of research, modeling, and data collection priorities. The author of the final paper was asked to take a broader look at retirement income modeling, concentrating on an overall framework for policy analysis in this area. The panel anticipates that the papers in this volume will serve as a useful reference for the research and policy communities as to what is known about retirement-income-related behaviors and the gaps and deficiencies in knowledge, data, and modeling techniques. The panel has drawn on the papers in its deliberations about priorities for developing an improved capability for assessing the short-term and long-term costs and benefits of retirement-income-related policy changes. The findings and conclusions of the panel will be provided in its final report. INCOME AND WEALTH OF OLDER AMERICAN HOUSEHOLDS The paper by Gustman and Juster focuses on the distribution and sources of income and wealth of households with retired workers. They report that income and wealth data show substantial disparities among elderly households, and the disparities appear to he greater for households whose members are age 70 and older than for other households. The available data also suggest that favorable economic circumstances tend to coexist: for example, people with higher levels of wealth tend also to have income from private pensions. Although there are some problems with income data, Gustman and Juster conclude that wealth data are much more problematic, especially for assets other than housing. In order to more closely examine income and wealth at retirement, Gustman and Juster review separate behavioral models that address several relevant issues: earnings, lifetime savings, and pensions. Beginning with labor supply and retirement decision models, they argue that current models of labor force participation and earnings could be improved to better account for the trend toward early retirement. Although the literature suggests that the preferences of workers have shifted, the models attribute the shift to changing incentives. The primary savings models (e.g., the life-cycle model, the precautionary savings model, the bequest motive model) have not been successful in explaining why so many elderly reach retirement with little or no savings. This theme underscores the analysis of household savings behavior in the paper by Poterba. A majority of research on private pensions approaches the issue in terms of individual behavior. On the employer side, Gustman and Juster do not find a model with sufficient structure that can be used to predict the effects of government policies on employers' decisions about pension benefit levels, other plan characteristics, insurance features, or other outcomes that would be useful in understanding how pension policies affect retirement incomes and wealth. Although Gustman and Juster attribute the lack of models largely to a lack of databases on employers that include detailed pension offering information along

OCR for page 1
Assessing Knowledge of Retirement Behavior with employer characteristics, Parsons argues that more fundamental conceptual issues may also be important. By focusing on the economic well-being of the elderly. Gustman and Juster are led to a key missing element in existing thinking and modeling: the extraordinarily important interaction of labor supply, savings, and pension decisions. Knowledge of the separate components is insufficient unless one makes the unrealistic assumption that they are independent or that part of the decisions is exogenously determined. This interaction issue clearly taxes available modeling abilities, but it cannot be ignored. Gustman and Juster conclude that the new Health and Retirement Survey (HRS) and the Asset and Health Dynamics Among the Oldest Old (AHEAD) survey provide essential data for much of the modeling of individual behavior that is needed. As just one example, HRS will allow researchers to sort out the effects of earlier savings decisions on the retirement decision by providing panel data on detailed asset accumulation before retirement. Similarly, data from AHEAD will enable researchers to determine to what extent older households decumulate assets, particularly housing assets, to finance consumption after retirement. LABOR SUPPLY BEHAVIOR The paper by Lumsdaine considers the choices that workers make, and the factors influencing their choices, about when to retire. She reports trend data showing that the average retirement age is falling while life expectancies are increasing. Both of these trends suggest that retirement savings will need to be stretched over more years of life. However, a large proportion of the retirementage population is reaching retirement with relatively low levels of savings. Because most of the elderly population is affected by Social Security, Lumsdaine suggests that it is the obvious place to start in discussing broad-based policy changes that can influence retirement behavior and income. Although some changes to the program have been legislated over the past decade, much of the research on Social Security suggests that the effects of these changes will not be substantial. However, debate still continues as to the relative importance of Social Security on retirement decisions and retirement income. Because Social Security may not provide enough income for many retirees, pensions are a source of additional retirement income that government policy has encouraged. Lumsdaine reports evidence that suggests, for those people who will receive pension benefits, the magnitude of the expected benefits can strongly influence their retirement behavior. However, it is difficult to make inferences about labor supply behavior with nationally representative data due to the lack of detailed pension plan information in most data sets. There has been increasing discussion about the relationship between Social Security Disability Insurance and retirement and the possibility that a liberaliza-

OCR for page 1
Assessing Knowledge of Retirement Behavior tion of disability benefits is partly responsible for earlier retirement decisions and declines in labor force participation generally. Lumsdaine reports that there is still considerable disagreement in the literature about the causal relationship between increased generosity of disability benefits and declining labor force participation, although there is a correlation. Understanding the link between disability and retirement is the first step to being able to predict how changes in Social Security policy may affect disability application. Another important issue to address is workers' flexibility to choose hours of work. Although evidence suggests that workers would prefer to gradually reduce their amount of work hours rather than make an abrupt transition from full-time work to complete retirement, evidence also suggests that workers are in fact constrained in their choice of hours. One option they have is to retire from a career job but take a new job with fewer hours of work. Overall, Lumsdaine provides a picture of a rich set of analyses addressing retirement decisions of individual workers, although she notes that the appropriate approach to best understand these decisions is not completely determined. Nonetheless, plentiful examples provide a sample of different approaches. By far the most important issue in Lumsdaine's view is better understanding of how individuals form expectations. Because expectations about retirement income and support are crucial to individual choices, fundamental research is needed. She also provides insight into alternative modeling approaches. While seeing the conceptual appeal of dynamic optimization models, Lumsdaine raises a note of caution about the feasibility of developing these very complex and computationally burdensome kinds of behavioral models much further in the near term. Lumsdaine concludes that many of the shortcomings of the literature, as well as the failure to model adequately the multiple factors influencing the retirement decision, stem from insufficient data. HRS and AHEAD, according to Lumsdaine, show great promise for use by future researchers to model the complexities of the retirement decision and retirement income. For example, information on pension plan provisions collected from sample members' employers will make it possible to develop more sophisticated models of the retirement decision. Also, repeated measures of health status and employment and earnings for middle-aged workers as they age should help answer the question of whether poor health status reduces employment opportunities and earnings or whether poor job prospects and earnings impair health. Similarly, such data should help answer the question of whether poor health leads to early retirement or whether a report of poor health is a rationalization of a decision to retire early that would have been made in any case. PERSONAL SAVING BEHAVIOR The paper by Poterba examines the choices that workers make, at various stages in their work life, between types of public and private savings vehicles and

OCR for page 1
Assessing Knowledge of Retirement Behavior about the level of their contributions to private pension plans (including decisions to withdraw contributions). Beginning with an examination of income and wealth holdings of the elderly, Poterba reports (as do Gustman and Juster) that there are wide disparities in wealth among the elderly and that the majority of them reach retirement age holding relatively low levels of liquid assets. The predominant model of savings behavior—the life-cycle/permanent income hypothesis model—fails to explain this phenomenon. Other savings models, such as the precautionary motive model and the bequest motive model, have the same shortcoming. The failure to understand individual motives behind savings decisions is important. For example, individuals' reactions to different policies will differ due to their underlying motivation for saving. Moreover, while improved data will help in understanding savings, the underlying modeling questions cannot be ignored. An important issue, which is a recurring theme throughout the papers, is that individual heterogeneity is central to much of the analysis and policy debate, but appropriately incorporating such heterogeneity is a very difficult analytical task. For those individuals managing to reach retirement age with a significant level of wealth, a lingering question in the literature is the rate of accumulation or decumulation of that wealth. Poterba reports that evidence in the literature is mixed and that this is an important area for future research. Poterba also examines the research related to the question of whether individuals who participate in private pension plans adjust other aspects of their saving to offset their pensions. He suggests that this question is difficult to answer because of data problems. For example, few data sets combine information on the structure of private defined benefit pension plans and other components of household wealth. For many of the important questions of interest with respect to private saving and the future financial status of the elderly, concludes Poterba, the existing research base does not provide detailed and convincing information on crucial parameter values and behavioral estimates. This problem will be partially remedied by HRS and AHEAD, but these surveys alone will not resolve all of the outstanding research questions. EMPLOYER BEHAVIOR The paper by Parsons examines the choices that employers make about their work forces, pension benefits, and financing of pension plans. An understanding of employers' decisions about hiring and retention of older workers and about whether to offer pension benefits (and, if so, what type to offer) is crucial to predicting the future well-being of the retired population. Parsons notes that as employers deal with a heterogeneous population of aging workers, various attributes decline nonuniformly, making it difficult to adjust compensation to new productivity levels. In addition, workers are resistant to downward adjustments

OCR for page 1
Assessing Knowledge of Retirement Behavior in pay, although they seem to more readily accept compensation cuts in the form of actuarially unfair adjustments in pension accruals. Employers vary greatly in the types of pension plans they offer, if they offer one at all. Historically, employers most often offered defined benefit plans. The reasons for offering this type of plan range from forestalling union activity to decreasing voluntary job turnover and enhancing job performance. These types of plans can also be used to induce exit from a company or to provide the company a graceful way to reduce employee compensation. Lately, defined contribution plans, such as 401(k) plans, have been on the rise. These plans are advantageous to workers because they provide a vehicle for tax-deferred savings. At the same time, employers benefit from them because the plans permit the targeting of pension contributions to workers who value them. The questions about employers' behavior are very similar to those about individuals. How do employers' decisions about retirement options and benefits interact with Social Security and other public programs and with individual retirement decisions? These questions, however, go to a deeper level that involves the fundamental underlying motivations of employers. Employers have incentives to encourage or discourage particular workers to stay because of productivity concerns, and these incentives interact with law, regulations, and individual contracts, but little is known about the strength of employer motivations and there are no well-developed models on this issue. Indeed, the contrast in knowledge, compared with modeling individual retirement decisions, is stark. Parsons describes some of the larger questions but without the means to evaluate alternative modeling approaches. Moreover, a basic message from Parsons' discussion is that some key prior questions, such as the pattern of age-productivity relationships, must be addressed before one can move to questions of individual employer or worker motivations and behavior. The central issues in understanding employer behavior thus come down to defining better the key questions to be addressed and to developing basic research programs. Because this area is so underdeveloped, there are also fundamental questions about what kinds of data would be most useful to collect and how one should obtain those data. An implication is that the design of cost-effective data collection is much more important in this area than in others covered in this volume. MORTALITY, HEALTH STATUS, AND HEALTH CARE COSTS The paper by Lee and Skinner looks at the relationship of life expectancy and health issues to retirement behavior and income. To begin, Lee and Skinner examine mortality rates because it is important to have good forecasts of mortality to determine how long people are likely to draw Social Security or private pension benefits or how long their personal savings are likely to last. Lee and Skinner find that there are large differentials in mortality forecasts, which lead to

OCR for page 1
Assessing Knowledge of Retirement Behavior very different forecasts in the number of elderly. Also problematic is that mortality data by race and ethnic group are of suspect quality. Another issue involves the projection of health and disability status. Although current evidence suggests a trend toward lower levels of disability among the elderly population, Lee and Skinner note that accurate projection is difficult because no unique definition of disability exists in historical data. This problem will largely be corrected in the future by the continued collection of a variety of health and disability measures in HRS and AHEAD. Improving forecasts of the general price of health care costs, in contrast, is not something that can be done simply by attaining more historical data on prices. The problem, according to Lee and Skinner, is that there is no good model of how health care costs are determined. Much work remains to be done to understand what factors are important in determining health care costs and how they might be expected to evolve over the next 30 years. Ultimately, because of shortcomings of existing modeling, progress in this area is very uncertain in the near term. How will changing mortality, health, and health care costs affect retirement income security? Lee and Skinner suggest that projections about retirement income security often focus on average levels of health care expenditures within quite broad demographic groups. However, the design of appropriate public policy is more likely to be concerned with the retirement outcomes of a specific group of people, such as those at the bottom portion of the income distribution. Hence, the authors suggest that developing methods for predicting the future financial security of specified demographic groups may be beneficial for policy purposes. A FRAMEWORK FOR ANALYZING RETIREMENT INCOME SECURITY The paper by Burtless outlines a broad framework for conducting analyses of retirement income security and projecting likely future effects of policy changes. His proposed analytical strategy rests on a microsimulation model that would project labor force status, job tenure and turnover, private pension and Social Security accrual, and household saving. Aggregate predictions from this model could be calibrated to predictions of the Social Security actuarial model. A macroeconomic model should also be created and linked to the microsimulation model to explain how savings are divided across alternative uses; to account for growth in capital stock, worker productivity, and wages; and to explain market rates of return on different classes of assets. Burtless admits that the initial implementation of this modeling approach would probably produce relatively unrealistic results. However, he suggests that analysts could offer a range of possible outcomes, from very optimistic to very pessimistic. He also suggests that continuing work be done on improving the

OCR for page 1
Assessing Knowledge of Retirement Behavior reliability of the linked micro and macro models through additional empirical research on key retirement-income-related topics. Pursuing the use of an overall microsimulation strategy for modeling retirement income has two important justifications. First, such a formulation provides structure for separate analyses of behavioral components. An overall integrated model requires continual consideration of what behavioral interactions are most important, of which parameters are crucial for analysis and which are less important, of how concepts and variables might be consistently defined and measured. Second, while various analytical models could play this integrative role, microsimulation models have distinct advantages in the context of retirement modeling because they are designed to explicitly incorporate the heterogeneity of the population. As noted above the central nature of individual heterogeneity is a recurring theme in all the papers. At the same time, the demands of microsimulation models for data and knowledge of individual behavior are truly great, suggesting that when and how to build a new microsimulation model are strategic questions for which there are no obvious or easy answers. CONCLUSION A clear message from the papers in this volume is that a number of questions about retirement income policy could be clarified with new data from HRS and AHEAD. The authors stress the importance of continuing to collect these data so that they provide sufficient observations to explain behavior patterns over time. In some cases, however, additional data are needed, not only through additional questions on current surveys and through links among existing surveys, but through the establishment of new surveys. It seems especially important to collect more employer-level information. Additional work is also needed in modeling. The availability of new data will certainly help, but in some cases, much more conceptual thinking is needed. Currently, there is no consensus about such absolutely key questions as why individuals save. It is certainly true that good data and good models must work in conjunction to provide useful analysis.