Click for next page ( 200


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 199
APPENDIX D Major Aspects of DYNASIM2 and PRISM DYNASIM2 and PRISM are dynamic microsimulation models that have been used to analyze a range of retirement-income-related policy proposals.] The two models are much broader than the cell-based Social Security actuarial cost model (see Appendix C): they simulate not only Social Security taxes and benefits, but also employer pension and Individual Retirement Account (IRA) contributions and benefits, Supplemental Security Income (SSI) eligibility and benefits, and federal income taxes. PRISM has a long-term care financing module that simu- lates long-term care utilization and financing for the elderly. The two models take some account of asset income (e.g., dividends and savings interest) but do not simulate wealthy Because they use dynamic microsimulation techniques, both DYNASIM2 and PRISM are able to provide highly disaggregated outputs for each year of a projection period (e.g., for population groups categorized by level of earnings and employment status). They can simulate complex policy provisions and inter- actions. For example, for Social Security, they have been used to simulate options that the cell-based Social Security actuarial cost model could not handle (e.g., provisions to credit homemaker spouses with a share of the employed spouse's earnings). Both models can be obtained by others and are documented to some extent. 1DYNASIM2 has been used for other types of policy analysis as well, such as the effects on welfare program costs of alternative rates of teenage childbearing. 2The PRISM long-term care financing module does treat assets to some extent. 199

OCR for page 199
200 ASSESSING POLICIES FOR RETIREMENT INCOME The documentation for DYNASIM2 is more complete than that for PRISM. (For DYNASIM2, see Johnson and Zedlewski, 1982; Johnson, Wertheimer, and Zedlewski, 1983; Zedlewski, 1990; for PRISM, see Kennell and Shells, 1986, 1990; Rivlin and Wiener, 1988.) The accessibility of the models in practice is limited: they were designed well over a decade ago for mainframe (or minicom- puter), batch-oriented computing environments, and they lack user-friendly de- sign features. However, both models were recently moved to a personal comput- ing platform (the PRISM long-term care financing submodel has operated for some years on personal computers). They have had relatively little formal valida- tion, although some older validation studies for DYNASIM2 are available (e.g., Haveman and Lacker, 1984; Hendricks and Holden, 1976; Wertheimer et al., 1986~. DYNASIM2 and PRISM illustrate the power and richness of the micro- simulation approach to policy projections. They also illustrate the limitations of existing models for addressing many current retirement-income-related policy issues. Their limitations are partly a function of their design, which was opti- mized for an older generation of computing technology. Even with a newer design, however, the lack of key data and behavioral parameters for such impor- tant elements as savings decisions of individuals and benefit offering decisions of employers would limit their utility. As we note throughout this report, significant improvements in microsimulation modeling capabilities require improvements in data and research knowledge. This appendix briefly reviews how DYNASIM2 and PRISM work (drawing on Ross, 1991) and comments on key limitations involving their ability to simu- late behavioral responses to policy changes and to simulate the effects on workers and retirees of employer benefit and labor demand decisions (drawing on Burtless, 1989; Ross, 1991; and a review of the pension simulation components of the two models by panel member Olivia Mitchell). HOW THE MODELS WORK Both DYNASIM2 and PRISM take an initial database and age the records for- ward for every year by means of a dynamic approach. These starting databases are exact-match files that contain records for the members of households from the Current Population Survey (CPS) March income supplement matched with their personal earnings histories from Social Security Administration (SSA) adminis- trative records. For DYNASIM2, the starting database is the March 1973 CPS- SSA exact-match file. For PRISM, the starting database is the March 1978 CPS- SSA exact-match file, which, in turn, is matched with the March 1979 CPS and the May 1979 CPS pension supplement.3 The earnings histories in these files are Because there is only partial overlap between the March 1978, March 1979, and May 1979 CPS samples, the PRISM database has considerably fewer records than the DYNASIM2 database.

OCR for page 199
MAJOR ASPECTS OF DYNASIM2 AND PRISM 201 essential for computing Social Security payroll contributions and benefit entitle- ments; they are also useful for simulating employer pension contributions and benefits. Then, for every year after the base year, DYNASIM2 and PRISM age the characteristics of the records in the file; they not only age each person by 1 year, but also, on the basis of probabilities, simulate whether or not each person will go to school, marry, have a child, divorce, die, change employment status, change jobs, participate in a pension plan, retire, and so on. Both models use outside aggregates, such as the Social Security Office of the Actuary's population projec- tions, to control the simulation results for each projected year. The essence of the models is the creation of longitudinal histories for each person in the original cross-sectional database. These histories are created in DYNASIM2 and PRISM by a set of modules that simulate each event in the following sequence: death, birth, marriage, divorce, labor force participation, unemployment. Starting with the first year, the modules run in turn for each individual. Then year two is simulated, starting with the first individual, from death through unemployment, and ending with the last individual. After the longitudinal histories are created, policy simulations (e.g., of tax liabilities) are run on the records. Given this sequence of events, demographic events can influence labor force events occurring in the same year (and demographic events occurring earlier in the sequence can affect those simulated later in the sequence), while events happening later in the sequence affect events in the following year. Events or characteristics that are simulated after the longitudinal histories are created are affected by those histories (e.g., SSI or Social Security benefits), but they cannot affect the sequence of demographic and labor force events. Therefore, the deci- sion to include certain events in the simulation of longitudinal histories is an important one. Table D-1 lists the events included in the simulation of longitudi- nal histories and the events that are simulated using the synthetic histories for the two models. Tables D-2 and D-3 provide additional information on the determi- nants of major events simulated by DYNASIM2 and PRISM, respectively. Originally, both DYNASIM2 and PRISM modeled birth, death, marriage, divorce, disability, and labor force events, and DYNASIM2 also modeled educa- tion, migration, and leaving home. The objective in developing DYNASIM2 was to be able to simulate a range of demographic and economic events for many different policy purposes. Thus, for example, the model simulates the education of children and young adults, including those born during the simulation and added to the file, because a labor force simulation might find these children as prime-age workers and a simulation with a very long time horizon might find these children as retirees. PRISM, in contrast, was designed to simulate incomes and long-term care utilization of the elderly through the year 2025. Given this time horizon, the model originally did not attempt to simulate life histories for people who were younger than age 20 in the base year (1979) or for those who

OCR for page 199
202 ASSESSING POLICIES FOR RETIREMENT INCOME TABLE D-1 Basic Features of DYNASIM2 and PRISM Feature DYNASIM2 PRISM Input Data Simulation Base Year 1972 Information in Simulation Database Exact match of March 1973 CPS and Social Security Exact match of March 1978 CPS and Social Security earnings records earnings records; also matched with March and May 1979 CPS Retirement Plan Provisions databases 1979 Demographic 1973 1978- 1979 Income 1972 1977- 1978 Employment 1972- 1973 1977- 1979 Quarters of Social Security Coverage 1937- 1972 1937- 1977 Social Security Taxable Earnings 1951 - 1972 1951 - 1977 Pension Characteristics 1979 Events Simulated to Create Longitudinal Histories Demographic Death Death Birth Birth Marriage Marriage Divorce Divorce Disability Disability Education level Education level Leaving home Migration Labor Force Participation Annual hours of work Annual hours of Hourly wage participation Hourly wage Whether unemployed Proportion of labor force hours unemployed Job and Pension Characteristics Job change Industry Pension coverage Pension plan assignment

OCR for page 199
MAJOR ASPECTS OF DYNASIM2 AND PRISM TABLE D-1 Continued 203 Feature DYNASIM2 PRISM Retirement and Benefit Acceptance Individual Retirement Account (IRA) Simulations Based on Longitudinal Histories Job and Pension Characteristics Employer Pension Social Security Individual Retirement Account (IRA) Retirement Supplemental Security Income Taxes Job change Industry Pension coverage Plan participation Pension eligibility Type of plan Benefit formula Plan constants Benefit computation Retirement benefit eligibility Retirement benefit computation Disability benefit Spouse benefit Children's benefit Participation Accumulations Distribution Whether leave job Whether accept new job Eligibility Benefits Participation Federal income tax Social Security payroll tax Pension acceptance Social Security acceptance Adoption Contributions Benefit computation Retirement benefit eligibility Retirement benefit computation Disability benefit Spouse benefit Children's benefit Distribution Eligibility Benefits Participation Federal income tax Social Security payroll tax State income tax aDeveloped by the original contractor for PRISM, ICE Incorporated. NOTE: Table does not include the PRISM Long-Term Care Financing Model. SOURCE: Ross (l991:Table 1). The data for DYNASIM2 come from Johnson, Wertheimer, and Zedlewski (1983) and Johnson and Zedlewski (1982); the data for PRISM come from Kennell and Shells (1986).

OCR for page 199
204 ASSESSING POLICIES FOR RETIREMENT INCOME TABLE D-2 Determinants of Major Events Simulated by DYNASIM2 Event or Characteristic Variables Used to Determine Event Simulation of Longitudinal Histories Demographic Event Death Married Women 45-64 All Others Birth Multiple Birth Sex of Newborn Marriage Age 18-29 Other Ages or Ever Married Mate Matching Leaving Homea Divorce Education Mobility Disability Onset Recovery Labor Force Events Labor Force Participation Hours in the Labor Force Age, race, sex, marital status, education, number of children Age, race, sex, marital status, education Age, marital status, number of children, race, education Race Race Age, race, sex, previous marital status, income, education, region, weeks worked, hourly wage, asset income, welfare, unemployment compensation Age, race, sex, previous marital status Difference in age, difference in education Age, race, sex Distribution over time of expected divorces for this marriage cohort, age at marriage, education, previous marital status, presence of young children, weeks worked, wages Race, sex, age, years at current school level, parents' education Number of years married, size of family, age and sex of head, education of head, race, region, size of metropolitan statistical area (MSA) Age, race, sex, marital status Age, race, sex, marital status, education Age, race, sex, education, South, disability, marital status, student, children, spouse earnings Age, transfer income, expected wage, disability, marital status, children

OCR for page 199
MAJOR ASPECTS OF DYNASIM2 AND PRISM TABLE D-2 Continued 205 Event or Characteristic Variables Used to Determine Event Wage Rate Unemployment Job Characteristics and Pension Plans Job Change Industry of New Job Pension Coverage on New Job Pension Plan Participation Type of Pension Coverage Pension Eligibility and Benefits Retirement Eligibility Vesting Benefit Formula Benefit Plan Constants Individual Retirement Accounts Plan Participation Retirementb Probability of Leaving Job Probability of Taking New Job Race, sex, age, South, disability, marital status, education, student Age, sex, race, education, marital status, South, disability, children Simulation of Jobs, Pensions, and Retirement, Using Longitudinal Histories Age, sex, tenure on current job, industry Sex, education, previous industry Sex, industry, earnings level Age, tenure on job, full- or part- time status, sex Industry Age, industry, years of service, type of pension Industry Industry and type of pension coverage Benefit formula, industry, type of pension coverage Sex, earnings Age, sex, disability, marital status, pension eligibility and amounts, Social Security eligibility and amounts, wage, earnings Age, disability, marital status, pension eligibility and amounts, Social Security eligibility and amounts, imputed wage aLeaving home for reasons other than marriage, birth of a child, divorce, or death. bThe retirement module's choice of retirement age for an individual overrides whatever pattern of labor force participation may have been modeled earlier in the simulation of the individual's longitu- dinal history. SOURCE: Ross (l991:Table 2). The data come from Johnson, Wertheimer, and Zedlewski (1983) and Johnson and Zedlewski (1982).

OCR for page 199
206 ASSESSING POLICIES FOR RETIREMENT INCOME TABLE D-3 Determinants of Major Events Simulated by PRISM Event or Characteristic Variables Used to Determine Event Simulation of Longitudinal Histories Demographic Events Death Disability Onset Recovery Divorce Marriage Mate Matching Birth Labor Force Events Hours Worked per Year Wage Rate Job Change Industry Pension Characteristics Pension Coverage Pension Plan Assignment Retirement Pension Acceptance Social Security Acceptance Individual Retirement Accounts Adoption Contributions Disability status, age, sex, years of disability Age, sex Age, sex, years of disability Age of husband and wife Age, sex, previous marital status Age of male, age of female Marital status, age, number of children, employment status last year Hours last year, age, sex, marital status, education, composite of hours in previous 3 years, female with young children, female divorced or widowed this year, receiving pension or Social Security income Age, sex, whether changed job this year, whether unemployed this year Hours worked, age, years on job Age, education, previous industry, full- or part-time status Age, industry, full- or part-time status, wage rate Industry, multi- or single employer plan in 1979, hourly or salaried status Age, sex (conditional on eligibility) Age, sex (conditional on eligibility) Age, family earnings, pension coverage Sex, marital status, family earnings SOURCE: Ross (l991:Table 3). The data come from Kennell and Shells (1986).

OCR for page 199
MAJOR ASPECTS OF DYNASIM2 AND PRISM 207 were born during the simulations. In 1993, PRISM was modified to simulate educational attainment and life histories for all people alive in the base year. PENSION COMPONENTS DYNASIM2 DYNASIM was designed in the 1970s and updated to DYNASIM2 in the early 1980s as a microdemographic model focused primarily on family formation and dissolution and active worker employment, and only secondarily on retirement. DYNASIM2's labor market segment is contained in the Family and Earnings History Model. It is a recursive system in which demographic characteristics are first simulated for each person for each year. The last of these is a disability probability, which is determined as a function of age, race, sex, marital status, and education. Disability status, in turn, is then used to determine earnings and labor force participation. Earnings, labor force participation, disability, and hours variables are determined for all years of the simulation before pension accruals or benefits (or taxes) are determined and before retirement outcomes are set. The Jobs and Benefits History segment of DYNASIM2 takes as input the results from the first-stage simulation just described and inputs several additional outcomes, including retirement age, Social Security and SSI benefits (if entitled), IRA account accumulations, and job history variables. People are also given a pension coverage probability, which depends on sex, industry, and earnings level; if covered, a plan participation outcome is determined, which depends on age, tenure, sex, and full- or part-time status. These participation and coverage prob- abilities are taken from 1979 CPS data. People assigned positive pension partici- pation are then assigned one of four pension plan types based solely on industry of employment: single-employer defined benefit, single-employer defined con- tribution, multi-employer defined benefit, or multi-employer defined contribu- tion. There appears to be no provision for a worker to have more than one pension plan from a given employer, although multiple mans have become Suite widespread in recent years. The actual determination of pension benefits is done in an Employer Pension Module subroutine attached to the Jobs and Benefits History segment of DYNASIM2. This module is used whenever a termination benefit must be computed for a covered vested worker or when a worker is simulated to retire in order to evaluate retirement benefits. The kernel of the Employer Pension Mod- ule is that it determines the parameters of each worker's pension plan, including the early, normal, and special early retirement ages, probability of full vesting after 10 years of service, defined benefit formula type, and specified constant terms within the defined benefit formula. Maximum service and minimum ben- efit amounts are also set. Each of these parameters is assigned on a randomized basis. Factors used to

OCR for page 199
208 ASSESSING POLICIES FOR RETIREMENT INCOME assign the parameters are age, years of service, and industry, drawing from a number of different data sets where available. Pension eligibility requirements for normal and early retirement are based on a 1974 defined benefit plan database of the Bureau of Labor Statistics (BLS). Vesting assignments are determined solely on the basis of industry, using a 1976 data set. Benefit formulas for defined benefit plan participants are assigned annual contribution rates of 7 per- cent and a nominal rate of interest on their account balances of 7 percent. There are some arbitrary assumptions in the Employer Pension Module. For example, the documentation notes that pre-retirement survivor protection probabilities are arbitrarily set at .02, and the probability of joint and survivor benefits at .75 for men and .25 for women. Pensions are also considered in the Retirement Model. Once the simulated worker is old enough, the probability of leaving a job and taking a new one is computed as a function of age, sex, disability, earnings, pension and Social Security wealth, and changes in pension and Social Security wealth if retirement is delayed a year. This model differs dramatically from the other job change models in DYNASIM2 in that many more economic variables are allowed to enter on the right-hand side as explanatory factors. One problem with this model is that the pension and Social Security wealth terms assume that people die with certainty at age 85, rather than using a more actuarially correct declining survival table. DYNASIM2 is a useful model for many policy purposes, but it is limited in a number of ways for retirement income policy analysis in the 1 990s and beyond. Although DYNASIM2 could be useful in a partial equilibrium simulation of some retirement income policies, such as the effects of Social Security benefit changes on retirement behavior, doing so requires the assumption that nothing else changes a questionable assumption for many policy scenarios. Thus, for instance, simulation of changes in payroll taxes would have to assume no labor demand response, since no employer behavior is included in the model. Changes in the taxability of retirement income could be simulated, but the structure of the model would not permit these tax changes to influence earnings, labor supply, or retirement. Although the model can be used to examine how exogenous changes in pension benefit parameters might influence retirement, it cannot do a good job of examining the effects of changes in nondiscrimination laws or the pension tax treatment of employer contributions, since employers' decisions about pension offerings and the consequences for employees are not included. Similarly, it cannot be used to examine the effect of mandated pensions, since no demand-side behavior is built in there is no provision for job losses that might result or for depressed earnings. DYNASIM2 is primarily a labor supply model, and thus no labor demand consequences can be examined regarding effects of changes in payroll taxes, effects of privatizing a piece of Social Security, etc. Finally, since only a few "hypothetical" pension plans are included in the model and their

OCR for page 199
MAJOR ASPECTS OF DYNASIM2 AND PRISM 209 parameters are not constrained to be internally consistent with each other or for the employees they cover, there is not much room for estimating how changes in specific rules (e.g., nondiscrimination rules) would affect employees through the effects on employers' compensation decisions or their willingness to offer pen- sions at all. PRISM PRISM is a microdemographic model designed in the early 1980s to simulate retirement income from public and private sources. The model simulates earn- ings, employment, retirement, and benefits from employer pensions, Social Secu- rity, and SSI. It also generates retiree health care coverage outcomes. It does not model asset accumulation or savings in general, though it does have an IRA savings segment. The PRISM model was augmented in the mid-1980s to include a long-term-care financing module. This module assigns asset portfolios to people when they reach age 65 and then in each year of the projection makes a simple adjustment of housing asset values by an assumed inflation rate and of nonhousing asset values by an assumed constant rate of dissaving. The structure of the model is recursive. The people in the starting database are "aged" by using background information on such characteristics as earnings histories, pension coverage, and employment. The paths of demographic out- comes are determined first, which in turn feed into the labor force segment. Labor supply outcomes (hours of work, job and industry assignment, pension coverage, date of pension benefit acceptance) are determined next, as a function of such exogenously specified factors as pension plan provisions. Wage growth is also determined at this step, independently of benefit accruals and taxes. (Dis- ability, job change, and entry to the labor force do not depend on benefits or pay.) Next, retiree income levels are specified, including Social Security and pension benefits, using as input the labor force outcomes. Finally, long-term-care out- comes are simulated. The PRISM pension module includes more than 400 different pension plans in comparison to a handful of generic types in DYNASIM2. Pension coverage is determined in the labor force module. Pension plan coverage is allocated on the basis of 1979 and 1983 CPS coverage rates, assigned as a function of industry, age, wage, and full- or part-time status. Plan type is assigned on the basis of very tight assignment criteria: industry, firm size, Social Security coverage, union, multi- or single-employer status, and hourly or salary status. The model also accounts for the person's 1979 CPS pension plan type, vesting, contributions, and participation in a supplemental plan as a way to benchmark starting values. Retirement benefits depend on pension benefit formulas, cost-of-living adjust- ments, and pro- as well as postretirement survivors' options. The probability of accepting pension benefits depends on sex and age, but not on benefit amounts. The probability of participating in a savings or thrift plan depends on wages and

OCR for page 199
210 ASSESSING POLICIES FOR RETIREMENT INCOME employer contributions, and employee contributions are always assumed to be the maximum allowed by the plan. A probability distribution estimated from CPS data is used to determine whether a lump-sum distribution is rolled over. (Most small distributions are not rolled over in the model.) All defined contribu- tion accounts are assumed to be annuitized at the individual annuity rate rather than the group annuity rate. Workers are assumed to receive benefits when they retire from all vested defined benefit plans from all prior jobs (although retirees must actually apply for deferred vested benefits from jobs they left before retire- ment, and some proportion may not do so). Many pension parameters are drawn from a Retirement Plan Provisions data- base. Information on a total of 475 plans was collected for 325 plan sponsors, so that in many cases both a primary and a secondary plan was obtained. Private single-employer and multi-employer plans and nonprofit plans came from a strati- fied random sampling of a 1981 filing of Form 5500 reports. Data for the public employer plans came from a random sample of a 1982 listing of public plans with more than 200 participants. Self-employed plans were also added to the sample, and weights were calculated. The 1981 information was updated in 1983; how- ever, the database does not include 401(k) plans, so, in simulations, savings and thrift plans are assumed to grow to the requisite numbers. Most pension parameters in the model are taken as fixed numbers, rather than being developed endogenously. Thus, industry pension coverage rates are constraints, and pension plan provisions are assumed not to change over time. Normal, early, and disability retirement formulas are given, as are Social Security integration rules and employer and employee contribution amounts. Employees are assumed to contribute 5 percent of their earnings in profit-sharing plans and to contribute the maximum possible amount to defined contribution plans that require employee money. The PRISM model is a second-generation approach to what DYNASIM was seeking to accomplish with regard to pension policy simulation. It is still limited, however, for today's pension and retirement income policy needs. One strength of the model is its richness of detail on pensions, both private and public. The Retirement Plan Provisions database with its several hundred pension plans, including their lengthy vector of plan parameters, builds into this model a more realistic degree of cross-sectional variability than the DYNASIM2 format. How- ever, the database has not been updated since 1983, and it is essentially static. Presumably, the sample weights could be altered to allow for changes in the distribution of types of employers and type of plans, but there is no provision for interaction between labor supply and labor demand behavior that could alter the distribution of plan offerings. Also, there is no feedback of pension plan provi- sions and changes in them to workers' labor supply decisions or wage growth.

OCR for page 199
MAJOR ASPECTS OF DYNASIM2 AND PRISM 2 LIMITATIONS Both DYNASIM2 and PRISM are limited by the old age of their starting data- bases. Neither model has switched to a newer initial database because in the face of confidentiality concerns and resource constraints, government agencies have not been willing to prepare new exact-match files, although this situation may be changing (see discussion in Chapter 4~. The consequence is that both models must simulate more and more years of historical data before they can begin projecting into the future. Aggregate information, such as employment rates and total payroll contributions, can keep the simulations in line with actual historical experience for a handful of key variables for years up to the present, but there is no way to ensure that the models accurately reflect the interrelationships among all of the important variables for those years. In addition, the models have not been revised to incorporate newer research findings about the underlying behaviors. Many of the key transition probabili- ties for example, estimates of labor force participation and retirement are based on old analytical studies (e.g., in DYNASIM2, on the 1969-1979 Retire- ment History Survey and early years of the Panel Study of Income Dynamics) and use overly simplistic functional forms. As an example, the labor force simulation in PRISM is based on a simple Markov probability model of transi- tions between categories of hours of work estimated from matched files of the March 1978 and 1979 CPS. This functional form, over time, will not preserve the characteristics of the distribution of hours worked in the population. Such a formulation captures limited information about previous work patterns, so that it will not properly project the work choices of individuals at the extremes of the distribution people who rarely or always worked. With regard to employer-provided pensions, the models do not incorporate more recent knowledge about trends in pension plan coverage, such as the growth in defined contribution plans. Similarly, neither model reflects well the trend toward increasing heterogeneity of labor force behavior, in which an individual may "retire" from a succession of jobs. Also, many pension-related assumptions in both models allow little or no variation over time or across workers: for example, assuming fixed contribution rates for defined benefit plans, that em- ployees always contribute the maximum to defined contribution plans, or that workers receive all of the benefits from defined benefit plans from all prior jobs that the model's pension rules say they are entitled to when they retire. A reason that neither model has been revised to update key transition prob- abilities or to incorporate more appropriate functional forms that reflect newer understanding of behavior has to do with the cost and time constraints imposed by their mainframe-oriented design. Because the models are not easy to modify, it is not in the modelers' interest to invest in redesign until substantial funding is available and there are clear research findings about appropriate functional forms and parameter estimates. In addition, since the 1983 Social Security amend

OCR for page 199
212 ASSESSING POLICIES FOR RETIREMENT INCOME meets, there has been relatively little demand for retirement-income-related mod- eling until now. Moreover, the decade of the 1980s saw important changes in labor markets and employee benefits that have made it difficult to determine the best professional judgment about key parameters to incorporate in models. Another limitation of the models is their lack of a ready capability to esti- mate the feedback effects of policy changes on behavior, also because of their design. That design separates the construction of longitudinal year-by-year de- mographic and economic "histories" for each person in the database from policy simulations as such. Because creating longitudinal histories was expensive on a mainframe, the practice has been to generate one set of histories (or a few sets under different reasonable assumptions) and then run a large number of program simulations on that longitudinal file. As noted above, events or characteristics that are simulated after the creation of the longitudinal histories are affected by those histories but cannot in turn affect the sequence of demographic and labor force events. For example, in both models, changes in the Social Security system cannot lead to compensating changes in labor supply over an individual's work life unless the longitudinal histories are re-created with a new set of labor supply equations incorporating the expected response. Although it would be possible for other models to simulate the second-round effects of policy changes that are initially simulated by DYNASIM2 or PRISM, neither model has a ready capabil- ity to link with second-round effects models. Finally, the models allow very few employee decisions to be endogenously determined by pension or tax parameters. Thus, the pension parameters are treated as exogenous to the simulation, and retirement plans are more or less "dropped" on workers, with few or no consequences for their decisions to take a job, change jobs, retire, save, or consume. Ideally, models to evaluate retirement income policy would take pensions as endogenously determined by interactions between employees and employers, in the context of regulatory and overall eco- nomic constraints. Because neither DYNASIM2 and PRISM makes pensions endogenous, these two models do not afford the opportunity to investigate how entire classes of policy changes might affect pension sponsorship and participa- tion, pension accumulation and investments, and, finally, pension decumulation. The broader missing piece is employers' demand for labor, including the trade- offs between benefits and wages and employment as well as pay decisions.