CHAPTER 11 CONCLUDING COMMENTS

CHAPTER 11 CONCLUDING COMMENTS I. PROJECTIONS FOR POLICY ANALYSIS MINT3 produces a micro dataset suitable for projecting the distributional consequences of current population and economic trends and for analyzing the impact of changes both in Social Security policy and in other important components of the income security system. It incorporates behavioral responses to policy and economic changes by making the decision to retire, claim Social Security benefits and continue to work after retirement endogenous. It also introduces explicit consideration of the living arrangements of the elderly and their claiming of Supplemental Security Income benefits. Table 11-1 explores the role that MINT3 could play in analyzing a number of Social Security policy changes that have been suggested recently. As noted, MINT3 is capable of analyzing both the labor supply and distributional impacts of many of these proposals, particularly those involving changes in the benefit formula and in the structure of dependents benefits. In many other cases, MINT3 is able to inform the analysis of a particular policy impact, even though it may be necessary to supplement the MINT3 model with some assumptions about savings or other impacts not modeled in MINT3. II. POTENTIAL AREAS FOR FUTURE MODEL DEVELOPMENT The validation and sensitivity tests revealed several areas in which the MINT model might be improved. One particularly promising area involves the use of additional information found in the Detailed Earnings file in constructing the matched earnings records, including earnings above the taxable maximum, earnings in employment not covered by Social Security, and the division of earnings into earnings from selfemployment and those from employment. The addition of this information to the MINT matched earnings records should allow improved estimates of many of the income sources and of retirement decisions. Estimates of earnings above the maximum would allow us to generate better estimates of the pension and asset income that will be available to the highest earners in the MINT sample and of their earnings after age 50. In turn, these improvements should lead to better estimates of retirement decisions. Estimates of earnings from employment not covered by Social Security should improve our ability to estimate pension income from noncovered employment and the asset holdings of people who have noncovered earnings. Each of these improvements would also help improve estimates of the retirement decision. Finally, generating separate estimates of income from selfemployment should allow us to take better account of the tendency of workers to shift to self-employment as they approach retirement age and thereby produce better estimates of XI-1

Policy Change Table 11-1 Simulating Policy Changes with MINT 3 Suitability of MINT3 for Analysis Distributional Impacts Behavioral Impacts Financing Changes: Raise payroll tax rates Not particularly suited No built -in behavioral impacts Increase income taxation Could analyze impact on retirement income, but would first have to add income tax module and ability to track taxability of different assets, deductions for itemizers. No built -in behavioral impacts. Could analyze distributional impact of assumptions about savings Rais e taxable maximum Could analyze impact on benefit distribution, but would first have to add earnings above the current maximum. No built -in behavioral impacts. Could analyze distributional impact of assumptions about savings Invest Trust Fund in equities Not particularly suited No obvious behavioral impacts that model could simulate. Coverage Changes: Cover remaining state and local employees Entitlement Changes: Raise early retirement age or age of first eligibility for widow(er)'s benefits. Relax disability definition for certain age groups Individual Accounts Mandatory (carve out or add on) Could analyze impact on retirement income, but would first have to add earnings from employment not now covered by Social Security Can analyze impact of alternatives assumptions about behavioral impact on retirement income, given assumptions about impact on disability incidence. income of different assumptions about changes in disability incidence rates. income given assumptions about net returns to accounts. No obvious behavioral impacts that model could simulate. Predicting change in retirement or savings behavior would require projecting outside of current data observations. Model could simulate SSI claiming and other labor supply effects, given retirement and savings behavior. Current version not set up to predict DI or SSI benefit claiming response to changes in disability rules. responses, given assumptions about savings responses and net returns; can estimate distributional impact of different assumptions about savings XI-2

Table 11-1 Continued Policy Change Suitability of MINT3 for Analysis Distributional Impacts Behavioral Impacts Voluntary (carve out or add on) Government matching contribution for low income workers Worker Benefit Changes: Raise normal retirement age Reduce COLA or increase the DRC Increase actuarial reduction Change benefit formula brackets & marginal rates Increase averaging period Award child care drop out years Add minimum benefit or increase the special minimum income given assumptions about net returns to accounts and participation patterns. income given assumptions about net returns to accounts and participation patterns. incomes, given assumptions about impact on disability incidence. incomes, given assumptions about impact on disability incidence. Depending on specifics of proposal, should be able to analyze impact on retirement (Have information on number and ages of children.) responses, given assumptions about participation rates, savings responses and net returns; can estimate distributional impact of different assumptions about savings claiming, and labor supply responses, given assumptions about participation rates, savings responses and net returns; can estimate distributional impact of different assumptions about savings response of the nondisabled, given assumptions about savings responses. Can analyze impact of different assumptions about disability Can estimate changes in OASI & SSI claiming and the labor supply response of the nondisabled. Can analyze impact of different assumptions about disability response of any particular proposal that model can simulate. XI-3

Table 11-1 Continued Policy Change Suitability of MINT3 for Analysis Distributional Impacts Behavioral Impacts Income test benefits Auxiliary Benefit Changes: Reduce spouses benefit Base widow(er) s benefit on combined earnings Introduce earnings sharing at divorce Eliminate children s benefits for living workers or reduce maximum family benefit Means-test auxiliary benefits SSI Changes Increase national guarantee Eliminate asset test; impute asset income Reduce age of eligibility to 62 Change benefit reduction rates Depending on specifics of proposal, should be able to analyze impact on retirement incomes, given alternative assumptions about impacts on savings behavior. Model does not now estimate children's benefits. Depending on specifics of proposal, should be able to analyze impact on retirement incomes incomes incomes incomes Can estimate impact on OASI & SSI claiming. Impact on labor supply may be complicated, depending on structure of proposal. Would have to make alternative assumptions about savings impacts. Can estimate impact on SSI eligibility and take-up rates. Can estimate impact on SSI eligibility and take-up rates Can estimate impact on SSI eligibility and take-up rates Can estimate impact on SSI eligibility and take-up rates the split between employment and self-employment among future cohorts. Better modeling of this shift will also improve our predictions of pension and asset income by allowing explicit recognition of the different asset accumulation patterns and pension arrangements of the self-employed. Our sensitivity tests suggested that these data elements could be added to the earnings matching process without significant difficulties. Their addition did not appear to make finding good donors for each target record more difficult. Before a new baseline is created using these additional data elements, however, a more careful review is needed of the quality of the data currently found in these additional data elements. Our work did not turn up any serious quality issues, other than those already known, but we also did not XI-4

specifically look for data problems. In addition, more careful analysis will be needed of the implications of different strategies generating matches using earnings above the taxable maximum. Using the current matching algorithm will substantially increase the size of the adjustments needed to equalize five-year average earnings variables among the highest earners, with unknown consequences for the quality of the resulting career earnings records. Changing to an algorithm that matched to the nearest neighbor would reduce this problem, but would not eliminate it, and could introduce other problems, such as finding that one record was being overused as a donor. Introducing any of these additional data elements will also require reestimation of many of the asset, pension, earnings and retirement decision equations. A second potential improvement involves adding future immigrants to the MINT population. Our sensitivity tests suggested that inserting additional records to represent future immigrants was not a particularly difficult computational step. The larger challenge may be the development of the assumptions about the number and source of future immigrants. The addition of future immigrants will probably produce more accurate estimates of the distribution of future benefits and the size of the future population in poverty. It should also provide more realistic estimates of the future SSI caseload, although projecting the impact of the current restrictions on immigrant eligibility will be a challenge. The implications of our test of an alternative treatment of marital status are less clear. Implementing the alternative approach to creating the matched earnings records required a six-fold multiplication in the number of donor pools, undoubtedly reducing the quality of some of the matches. The alternative approach did not increase the correlation between husband s and wife s earnings, but probably did improve the link between changes in marital status and changes in labor force behavior. Determining the net impact of all of these changes on the quality of the resulting earnings records probably requires more analysis than we were able to do under this contract. The pension chapter notes three areas where additional work could improve the MINT model. One is an improved job history module that ties job tenure and job changes more closely to the earnings projections. A second is the development of a more sophisticated mechanism for determining the characteristics of the pension plans in which state and local workers participate. Ideally, an alternative should be developed that uses an approach similar to that now used to assign plan characteristics to private sector workers. Finally, it would be desirable to simulate the formation of future IRAs for current cohort members. The model currently handles all IRAs that exist at the time of the SIPP interview and, indirectly, all IRAs subsequently created through rollovers from employer-sponsored pension plans. It does not, however, allow for new IRAs that are created with new money after the SIPP interview date. Another area of potential improvement is the treatment of defined benefit pensions in divorces. The current model makes no adjustment to either the financial assets or the future defined benefit pension income of parties to a divorce as long as the divorce occurs prior to the receipt of pension benefits. Pensions in payment status are XI-5

divided equally. Although the future value of a defined benefit pension may in fact be ignored in most current divorce proceedings, it is surely an oversimplification of the situation when long-term marriages end in divorce near to the retirement age of one or both partners. By ignoring the adjustments made in these latter cases, we are underestimating the income of divorced women and overestimating the income of divorced men, although we don t know by how much. While improving the treatment of defined benefit pensions in divorce is desirable, it may require additional research on current practices and is likely to involve some moderately challenging changes in the structure of the computer processing. As noted in the discussion of the retirement equations, the model appears to over predict slightly the incidence of Social Security benefit take-up at age 64 and under predict slightly the incidence of retirement at age 65. In the course of estimating the retirement equations, we noticed that self-reports of age at retirement often differed slightly from administrative records. It may be useful at some future date to explore the reasons for this difference. In particular, one might explore whether the administrative records tend to show slightly lower retirement ages than either self-reports or data on the actual age at benefit award because of a tendency for individuals to back date claims and obtain lump-sum retroactive payments, when their situation allows. Several minor categories of adult beneficiaries are not now projected by MINT. Probably, the two most important are disabled widow(er)s and surviving spouses entitled by the presence of entitled children. As part of a future package of MINT modifications, it may be desirable to add these two categories of beneficiaries. In principle, the information needed to simulate each category is currently being generated in the MINT projection process. We are already estimating the number of children ever born and using the results in the simulation of living arrangements. The same estimates ought to provide a basis for projecting entitlement among surviving spouses caring for entitled children. Before adding disabled widow(er)s to the MINT simulations it would be necessary to evaluate the accuracy of the disability assignments growing out of the earnings matches. Assuming these were producing plausible results, the addition of the benefit category would not be difficult. Explicit recognition of disabled widow(er) beneficiaries would allow simulations of the impact of proposals to liberalize the treatment of these beneficiaries. It would also provide a more accurate projection of the benefit after age 62 of persons who had previously received disabled widow(er)s benefits. A final issue meriting attention is the unification of the benefit calculators. At present, MINT employs two benefit calculators. One is used to project alternative measures of Social Security wealth for the purposes of simulating retirement behavior. The other is used to calculate the actual Social Security benefit received. The use of the two separate calculators complicates somewhat the simulation of benefit changes that might affect the retirement decision, since it requires that the change be separately programmed into each calculator. In principle, the calculator used for final benefit calculations could also be used for simulating the retirement decision, although converting to the alternative calculator would require reestimating the retirement equations. XI-6

III. CONCLUDING OBSERVATIONS MINT3 builds on the work previous done under the MINT1 contract by researchers from the Brookings Institution, RAND, and the Urban Institute. In some cases, MINT3 adds new simulation capability that was not present in MINT1. In other cases, it introduces refinements to the projection techniques used to construct MINT1. The need for many of these refinements only became clear after the initial MINT1 results could be carefully analyzed. The major changes introduced in MINT3 include: A new technique for generating individual earnings records that relies on statistical matching to preserve a greater degree of individual variation in earnings experience, produce more realistic distributions of lifetime earnings patterns and better link lifetime earnings with key demographic factors, such as ethnicity and education. A new technique for determining which workers become disabled and tracking the earnings of disabled workers after the onset of disability. The new technique allows the incidence of disability among each birth cohort and age group to be adjusted to match external assumptions, particularly those used in Social Security cost projections. A new technique for determining mortality which links the mortality projections to the actual experience of workers at different education and earnings levels. This technique also allows total mortality in the projected population to be adjusted to external control totals, such as those contained in the Trustees Report. A new procedure for modeling the retirement decision which separates the decision to retire from the career job from the decision to file for Social Security benefits, providing a more satisfactory way of handling bridge jobs and other mechanisms for phasing in retirement. The new model also incorporates current pension entitlements and the impact of continued work on future entitlements into the retirement decision. New pension modules that provide more accurate estimates of both defined benefit and defined contribution pension plans. Individual workers projected to be covered by defined benefit plans are assigned to one of a number of different plans through a new procedure that recognizes the variety of different defined benefit arrangements now in existence and reproduces the approximate distribution of workers among such different plan types. Similarly, those participating in defined contribution plans are assigned different combinations of employer and employee contribution rates and are XI-7

assumed to make different asset allocation decisions in accordance with actual distributions of each variable. A new asset accumulation module that relates better controls for differences in actual and projected initial wealth holdings, better controls for life-time earnings and retirement, and better controls for private pensions. A new procedure to simulate living arrangements of the elderly, SSI eligibility and the SSI take-up decision. This is an entirely new module for the MINT model and allows the analysis of more complex policy proposals that combine changes in the insurance program with enhancements of the safety net provided by the companion assistance program. A new procedure to simulate the characteristics and income of co-resident family members to allow better measures of the well-being of co-residing retirees. Each of these is a valuable addition to the MINT model. Together, they substantially strengthen MINT s ability to project the future status of the aged and simulate the impact that Social Security policy changes will have on the economic status of the aged. XI-8