Self-Reported Pension Wealth File Health and Retirement Study Wave 1 ----------------------------------------- The Self-Reported Pension Wealth File contains the present value of pension wealth for respondents in the Wave 1 (1992) of the Health and Retirement Study (HRS). The estimates are based on respondent self-reported descriptions of their employer provided pension plans. Data from the employer survey are not used. The file contains two variables, Wave 1 Case ID (columns 1-5) followed by a 10 digit field (columns 6-15) containing the pension value. These data are available for 12,273 of the 12,652 respondents in Wave 1 of the Health and Retirement Study. Missing respondents belong to households in which one partner refused to complete the interview. A brief overview of how the pension wealth variable was constructed is presented below. It is important to note that the respondent could be covered by as many as three pensions on the current job, a pension promising future benefits from a prior job, and could be receiving pension income from prior job pensions. The pension variable contained in this data file sums together pension wealth from all these sources. The many sources of pension data and the low response rate for some items mean that, for most respondents, at least some component used to construct pension wealth had to be imputed. It is thus not feasible to include a separate variable flagging imputation status. Imputation. ----------- As noted above, many respondents were missing key pieces of data needed to construct pension wealth. For some types of pensions less than half of the respondents provided data complete enough to directly calculate pension wealth. The information required to construct pension wealth comes from three sources: the pension on the current job for persons still working, the pension on the last job for persons no longer working, and pension income by source for persons receiving benefits1. All currently employed workers were asked if they are "included in" a pension plan "through your work" (if self-employed), or if they were "included" in a pension plan "sponsored by your employer or union" (if not self-employed). Each respondent could list up to three plans. About 76 percent of the respondents listed a single plan, 21 percent listed two plans, and the remaining 3 percent of the respondents listed three plans. Respondents were much more likely to cite a defined benefit (DB) plan as their first plan. Of the first plans reported, 61 percent were DB and 34 percent were defined contribution (DC). Of the second plans reported, only 16 percent were DB, but 81 percent were DC. Most of the third plans reported were DC. For each of the three plans, if reported plan type was DB, "both" or "don't know", then the respondent was first asked the expected age of retirement, then asked to give an estimate of the pension benefit at retirement. The benefit could be expressed as a percentage of final salary or as an amount ($) per unit of time (month, quarter, year, etc.) or as a lump-sum at retirement. Most plans (44 percent) gave an amount per unit of time and we have converted these to annual pension benefits. For plans providing a percent of final salary (15 percent) we have also computed an annual pension benefit using an assumed (see below) rate of growth of earnings until the expected date of retirement. Still, there is much data missing; the remaining 41 percent of the plans require imputation. To impute pension benefits we first divide the sample by plan number (three could be listed), type of response (DB, both, DK), and ten wage and salary income deciles. We then perform a hotdeck imputation procedure using these ninety cells.2 If the reported plan type is DC or "both", then the survey asks for the balance accumulated in the plan. Missing data, although still a problem, is not as severe as for DB plans. 71 percent of the DC plans reported provide an account balance. If the plan type is DC then further details on the type of plan are asked. Responses to these detailed questions are used to categorize DC contributions as contributions to either "401(k) plans" or to "traditional DC plans". Our definition of "401(k) plans" broadly includes the HRS response categories "401(k)/403(b)/SRA", "tax shelter", "IRA/SEP", "SEPP", or any response combination that includes these (i.e. some respondents indicate their plan was a combination of a 401(k) and a thrift plan). The category "traditional DC plans" covers the remaining types of DC plans, including thrift plans, ESOPs, and money purchase plans. If a respondent indicated plan type is "both", then no detailed questions about plan type were asked. For these plans the entire balance is assumed to be of the "traditional DC plan" type, thus perhaps underestimating 401(k) balances. For plans known to be DC, but for which the balance is unknown, the hotdeck imputation method is again used, based on plan number, plan type (traditional DC and 401(k)), and ten wage and salary income deciles. Persons not currently employed are asked about their most recent job. As above, they can specify four pension types: DB, DC, both, and DK. However, each respondent can provide information on only one plan. In general, the follow-up questions parallel the questions asked for the current job discussed above. We will only note the differences here. First, persons covered by DB plans are asked about expected future benefits, benefits currently being received, and benefits already distributed as a lump-sum. We disregard all but the former because benefits currently received are picked up elsewhere in the survey (the income section, see below), and benefits already paid out will show up as IRA balances if rolled over (and do not represent pension wealth if not). If covered by a DC plan, the balance is only included if "left to accumulate" with the former employer. DC balances rolled-over into an IRA, converted to an annuity, or withdrawn are not included. Finally, if a person indicated coverage by a "both" type plan they are asked "how much money was in your account when you left that employer?" The survey does not ask how much remains in the account as of the survey data. Based on the proportion of DC balances remaining with the employer to accumulate, we randomly include the balances of one-third of these respondents. Again, missing DB benefits and DC data are imputed by the hotdeck method described above. Unfortunately, there is no way to distinguish between 401(k) balances and traditional DC plan balances for prior jobs. Thus all DC balances are assumed to be from traditional DC plans. The final source of information on pension wealth is from income currently being received. We use income streams from pensions, annuities, and veterans benefits. Constructing Pension Wealth --------------------------- For DC type pensions the reported balance is our measure of pension wealth. For persons expecting DB pension benefits and persons currently receiving pension income, we compute the present value of the benefit stream using the following assumptions: * Mortality data are based on population averages by gender, age, and birth cohort provided by the Social Security Administration. See Mitchell, Olson, and Steinmeier (1996) for a discussion of these data. * For discounting and earnings growth we use the "intermediate" interest rate assumptions used by the Social Security Administration. See Board of Trustees (1995). * Public pensions are assumed to be fully indexed, again using the "intermediate" projections of the Social Security Administration. See Board of Trustees (1995). Private sector pension benefits are not indexed. Respondents were asked to provide the expected pension benefit at their expected date of retirement. If benefits are not currently being received, they are assumed to commence at the expected age of retirement (mean is age 62). The average age of HRS respondents is 55. Thus for the typical HRS respondent, retirement benefits do not begin for another seven years. We have assumed that their response to the expected pension benefit question is denominated in future (date of retirement) dollars. Moreover, we have assumed that the benefit amount is a single survivor benefit. Accordingly, we use individual survival probabilities in the computation of the present value. If instead we assumed the responses represented joint survivor benefits, calculated pension wealth would be somewhat higher. We further assume that when respondents report expected pension benefits they do not anticipate separating from their employer prior to retirement. This assumption allows us to calculate the present value of retirement wealth conditional on continued years of service until retirement. The component of this wealth "earned" as of the survey date is this present value multiplied by the ratio of years of service at the survey date to years of service at the expected date of retirement. This adjustment is necessary to make the present value of DB benefits comparable to the accumulated balance in a DC plan at the date of retirement. References ---------- Mitchell, Olivia, Jan Olson and Thomas Steinmeier. "Construction of the Earnings and Benefit File (EBF) for Use With the Health and Retirement Study." NBER Working Paper no. 5707. August 1996. Board of Trustees of the Federal Old-Age and Survivors Insurance and Disability Insurance Trust Funds. 1995 Annual Report. Footnotes --------- 1 There is also some information on pensions associated with previous jobs (other than the last job), but we jedged these data to be too incomplete to use at this time. 2 About 9 percent of the DB plans were also missing the expected age of retirement. We use the modal response of age 62 in these cases.