The Retirement Wealth of the Baby Boom Generation. Edward N. Wolff* New York University March 2006

Transcription

1 The Retirement Wealth of the Baby Boom Generation Edward N. Wolff* New York University March 2006 *Edward N. Wolff Department of Economics 269 Mercer Street Room 700 New York University New York, NY USA tel fax Introduction The Baby Boom generation (individuals born between 1946 and 1961) had a rocky start economically as a result of its large cohort size. They experienced depressed earnings when entering the labor market in the 1970s (see Berger, 1983, 1984, and 1985), as well as low rates of return to schooling (see Freeman, 1976). Indeed, Berger (1985) predicted that for the Baby Boom cohort that the negative effect of its large cohort size would continue to depress earnings as the cohort aged. In contrast, Easterlin et. al. (1990) found that during the 1980s the economic well-being of the baby boomers was higher than that of their predecessors. Moreover, the Baby Boom generation shared an increase in income inequality during the 1980s that was common to all cohorts during this period. Sabelhaus and Manchester (1996) also found evidence that the baby-boomers in 1989 were doing better than their parent s generation at a similar point in their life-cycle. In particular, by 1989 baby boomers accumulated more wealth relative to income than their parents had. However, this generation also lived through the economic booms of the 1980s and 1990s. The latter half of the 1990s, in particular, was characterized by the stock market boom and a large accretion in share values. This period is also of particular interest because it has seen one of the most dramatic changes in the retirement income system since the end of World War II. In particular, the last two decades have witnessed a 1

2 dramatic decline in traditional Defined Benefit (DB) pension plans and a corresponding rise in Defined Contribution (DC) pensions. The paper will analyze the wealth holdings of the Baby Boom generation (ages 40-55) in 2001 and compare their wealth holdings with those of the same age group in 1983 and The key contribution is to use a measure of augmented wealth, which includes not only standard net worth but also estimates of both pension and Social Security wealth. The paper will analyze changes in both mean and median augmented wealth of this age group over this period, as well as its various components, particularly pension and social security wealth. It will also investigate changes in the inequality of augmented wealth of this age group over this period. The primary interest of the paper is whether this generation has made up for lost ground by the early 2000s. In particular, how does this generation stack up in comparison to the same age group twenty years earlier. The results of the paper can also shed light on the importance of Social Security as a source of retirement income for the Baby Boom generation. It can also highlight the likely deleterious effects of the transformation of the pension system from DB to DC pensions on the well-being of the average worker. This may have important implications for pension policy in this country. The principal data sources used for this study are the 1983, 1989, and 2001 Survey of Consumer Finances (SCF) conducted by the Federal Reserve Board. Each survey consists of a core representative sample combined with a high-income supplement. The SCF provides considerable detail on both pension plans and Social Security for both husband and wife The next section of the paper (Section 2) provides a review of the pertinent literature on retirement wealth. Section 3 describes the data sources and develops the accounting framework used in the analysis. Section 4 shows time trends in standard measures of household wealth over the period. Sections 5 and 6 investigate changes in pension wealth and Social Security wealth, respectively, over this period. Section 7 presents summary measures on total (augmented) household wealth. Concluding remarks are made in Section 8. 2

3 alone. 1 Kathleen McGarry and Andrew Davenport (1997), using the 1992 wave of the 2. Literature Review Previous work on the overall distributional effects of pension and Social Security wealth has been rather limited. The seminal paper on this subject is Martin Feldstein (1974), who introduced the concept of Social Security wealth and developed its methodology. His main interest was in the aggregate level of Social Security wealth and its effect on aggregate savings and retirement patterns. In a follow-up paper, Feldstein (1976) considered the effects of Social Security wealth on the overall distribution of wealth on the basis of the 1962 Survey of Financial Characteristics of Consumers (SFCC), a survey performed by the Federal Reserve Board of Washington. The paper found that the inclusion of Social Security wealth had a major effect on lowering the inequality of total household wealth (including Social Security wealth). The Gini coefficient for the sum of net worth and Social Security wealth among families in age class 35 to 64 was 0.51, compared to a Gini coefficient of 0.72 for net worth. Edward Wolff followed up by examining the distributional implications of both Social Security and private pension wealth. Wolff (1987), which used the 1969 Measurement of Economic and Social Performance (MESP) database, was one of the first papers to add estimates of private pension wealth and examine their effects on the overall distribution of wealth. The paper showed that while Social Security wealth has a pronounced equalizing effect on the distribution of augmented wealth (the sum of marketable wealth and retirement wealth), pension wealth has a disequalizing effect on augmented wealth. In particular, the addition of Social Security wealth to net worth reduced the overall Gini coefficient from 0.73 to 0.48 but the addition of pension wealth to the sum of net worth and Social Security wealth raised the Gini coefficient back to The sum of Social Security and pension wealth has, on net, an equalizing effect on the distribution of augmented wealth but substantially less than Social Security wealth HRS, found that pension wealth is only slightly more equally distributed than net worth and that adding pension wealth to net worth has a modest effect on inequality (with the 1 Also, see Wolff (1992) for a discussion of some of the methodological issues involved in estimating both Social Security and pension wealth. 3

4 wealth share of the top decile declining from 52.9 to 44.8 percent with the addition of pension wealth). 2 Arthur Kennickell and Annika Sunden (1999), using the 1989 and 1992 Survey of Consumer Finances (SCF), found a net equalizing effect from the inclusion of these two forms of retirement wealth. In particular, the inclusion of pension and net Social Security wealth reduced the share of total wealth held by the top one percent with head less than the age of 65 in 1992 from 31.3 percent to 16.2 percent. 3 Moreover, neither paper presents calculations over a long time period as is done here (18 years). 4 There are several other topics of relevance to this paper. First, several papers have used the Health and Retirement Survey (HRS) to examine what proportion of total (augmented) household wealth is comprised of pension and social security wealth. Alan Gustman, Olivia Mitchell, Andrew Samwick, and Thomas Steinmeier (1997) found that in 1992, pensions, Social Security, and health insurance accounted for half of the wealth held by all households aged 51 to 61 in the HRS; for 60% of total wealth of HRS households who are in wealth percentiles 45 to 55; and for 48% of health for those in the 90th to 95th wealth percentiles. In a follow-up study, Gustman and Steinmeier (1998) used data from the HRS to examine the composition and distribution of total wealth for a group of 51 to 61 years olds. They focused on the role of pensions in forming retirement wealth. They found that pension coverage is widespread, covering two thirds of households and accounting for one quarter of accumulated wealth on average. Social Security benefits accounted for another quarter of total wealth. They also reported that the ratio of wealth (excluding pensions) to lifetime earnings was the same for those individuals with pensions and for those without pensions. They concluded that pensions cause very limited displacement of other forms of wealth. 2 Differences in survey years and age coverage prevent any direct comparisons with my results in this paper. 3 Net social security wealth is defined as the discounted present value of future social security benefits less future taxes paid into the social security (OASI) system. Estimates are not provided separately for pensions and social security. As a result, it is not possible to provide direct comparisons with these results and my own (which uses gross social security wealth). 4 In contrast, there are a host of studies that examine the intra-cohort redistributional effects of Social Security benefits relative to contributions into the Social Security system. They consider which groups are net gainers and which net losers from the Social Security System as a whole. These papers include Wolff (1993); Coronado, Julia Lynn, Don Fullerton, and Thomas Glass (2000); Smith, Toder, and Iams (2001); Liebman (2002); and Leimer (2003, 2004). 4

5 Second, several studies have documented changes in pension coverage in the United States, particularly the decline in DB pension coverage among workers over the last two decades. Before this, Laurence Kotlikoff and Daniel Smith (1983) provided one of the most comprehensive treatments of pension coverage and showed that the proportion of U.S. private-wage and salary workers covered by (traditional DB) pensions more than doubled between 1950 and David Bloom and Richard Freeman (1992), using Current Population Surveys (CPS) for 1979 and 1988, were among the first to call attention to the decline in DB pension coverage. They reported that the percentage of all workers in age group covered by these plans fell from 63 to 57% over this period. Among male workers in this age group, the share covered dropped from 70 to 61%, while among females in the same age group, the share remained almost constant, at 53%. Alan Gustman and Thomas Steinmeier (1992) documented the change-over from DB plans to DC plans between 1977 and 1985 on the basis of IRS 5500 filings. They estimated that only about half of the switch was due a decline in DB coverage conditional on industry, size, and union status and the other half was due to a shift in employment mix toward firms with industry, size, and union status historically associated with low DB coverage rates. Other studies include William Even and David Macpherson (1994a, 1994b, 1994c, and 1994d). The 1994c study showed a particularly pronounced drop in DB pension coverage among workers with low levels of education; and Even and Macpherson (1994d), showed a convergence in pension coverage rates among female and male workers between 1979 and Department of Labor (2000) found that a large proportion of workers, especially low wage, part-time, and minority workers, were not covered by private pensions. The coverage rate of all private sector wage and salary workers was 44 percent in Coverage of part-time, temporary and low-wage workers was especially low. This appears to be ascribable to the proliferation of 401(k) plans and the frequent requirement of employee contributions to such plans. It also found important racial differences, with 47 percent of white workers participating but only 27 percent of Hispanics. Another important finding is that 70 percent of unionized workers were covered by a pension plan, compared to only 41 percent of non-unionized workers. Pension participation was found to be highly correlated with wages. While only 6 percent of workers earnings less 5

6 than $200 per week had a pension plan, 76 percent of workers earning $1,000 per week participated. Third, a related topic of interest is whether DC pension plans have substituted for DB-type plans. Leslie Popke (1999), using employer data (5500 filings) for 1992, found that, indeed, 401(k) and other DC plans have substituted for terminated DB plans and that offering a DC plan raises the chance of a termination in DB coverage. Fourth, several papers looked at the issue of whether DC plans have substituted for other forms of wealth and whether there was any net savings derived from DC plans. James Poterba, Steven Venti, and David Wise (1992, 1993, 1995), using SIPP data for 1984 and 1991, Poterba, Venti, and Wise (1998), using HRS data for 1993, and Poterba, Venti, and Wise (2001), using both macro national accounting data and micro HRS data, concluded that the growth of IRAs and 401(k) plans did not substitute for other forms of household wealth and, in fact, raised household net worth relative to what it would have been without these plans. They found no substitution of DC wealth for either DB wealth or other components of household wealth. In contrast, William Gale in a series of papers both by himself and with colleagues, found very little net savings emanating from DC plans. Gale (1995) concluded that when biases in estimation procedures in the previous literature on the subject are corrected, the offset of pension wealth on other forms of wealth can be very high. Using data from the 1984, 1987, and 1991 SIPP, Eric Engen and William Gale (1997) estimated that at best only a small proportion of 401(k) contributions represent net increments to household savings. In later work, Engen and Gale (2000) refined their analysis to look at the substitution effect by earnings groups. Using data from the 1987 and 1991 SIPP, they found that 401(k)s held by low earners may more likely represent additions to net worth than 401(k)s held by high earners, who hold the bulk of this asset. Overall, only between 0 and 30 percent of the value of 401(k)s represent net additions to private savings. Kennickell and Sunden (1999) found a significant negative effect of both defined benefit plan coverage and Social Security wealth on non-pension net worth but concluded that the effects of defined contribution plans, such as 401(k) plans, on other forms of wealth are statistically insignificant. 6

7 3. Data Sources and Accounting Framework The principal data sources used for this study are the 1983, 1989, and 2001 Survey of Consumer Finances (SCF) conducted by the Federal Reserve Board. Each survey consists of a core representative sample combined with a high-income supplement. 5 The SCF provides considerable detail on both pension plans and Social Security contributions. The SCF also gives detailed information on expected pension and Social Security benefits for both husband and wife. For 1983, the Federal Reserve Board also made its own calculations of the wealth equivalent value of both expected pension benefits and Social Security benefits. I use these estimates in this paper. However, this has not been done for other years. 6 The principal wealth concept used here is marketable wealth (or net worth), which is defined as the current value of all marketable or fungible assets less the current value of debts. Total assets are the sum of: (1) the gross value of owner-occupied housing; (2) other real estate owned by the household; (3) cash and demand deposits; (4) time and savings deposits, certificates of deposit, and money market accounts; (5) government bonds, corporate bonds, foreign bonds, and other financial securities; (6) the cash surrender value of life insurance plans; (7) the current market value of Defined Contribution pension plans, including IRAs, Keogh, and 401(k) plans; (8) corporate stock and mutual funds; (9) net equity in unincorporated businesses; and (10) equity in trust funds. Total liabilities are the sum of: (1) mortgage debt, (2) consumer debt, including auto loans, and (3) other debt. I use the symbol NW to refer to standard net worth. It 5 See, for example, Kennickell and Woodburn (1992), Kennickell, McManus, and Woodburn (1996), and Kennickell and Woodburn (1999) for details on the construction of the weights used in the SCF files. 6 The underlying data are not available for the 1983 SCF to re-do these estimates in exactly the same form as for 1989 and 2001, though I try to follow their method as much as possible for these two years (see below and the Appendix). The difference in methodology may introduce compatibility problems between the 1983 estimates and those of the other two years. Moreover, pension and social security wealth imputations in the 1983 data are rather limited for households under the age of 46. Partly for this reason, I focus mainly on age group 46 to 64. Estimates for all households in 1983 have to be interpreted cautiously, though I do show alternative estimates where I adjust the 1989 and 2001 samples to match the coverage of the 1983 sample (see below). 7

8 should be stressed that the standard definition of net worth includes the market value of DC pension plans. (We shall return to this point later on in the paper). 7 A word should be said on why I use the SCF instead of the newer Health and Retirement Survey (HRS), which has much more complete data on earnings histories and has employer-provided information on individual DB pension plans of each employee covered by these plans. There are three reasons. First, the SCF provides much better data on the assets and liabilities that constitute marketable net worth. Second, the SCF data date from 1983, whereas the HRS data start in Since the most important transformation of the pension system occurred beginning in the late 1980s, the SCF data allow us to better track this change over the transition period. Third, the age coverage of the HRS is limited whereas the SCF covers the whole population. The imputation of both pension and Social Security wealth involves a large number of steps, which is summarized below. Greater details can be found in the Appendix. A. Pension Wealth: For retirees (r) the procedure is straightforward. Let PB be the pension benefit currently being received by the retiree. The SCF questionnaire indicates how many pension plans each spouse is involved in and what the expected (or current) pension benefit is. The SCF questionnaire also indicates whether the pension benefits remain fixed in nominal terms over time for a particular beneficiary or is indexed for inflation. In the case of the former, the (gross) Defined Benefit pension wealth is given by: (1) DB r = 0 PB(1 - m t )e -δt dt where m t is the mortality rate at time t conditional on age, gender, and race; 8 δ the discount rate which is set at 5 percent (2 percent when PB is indexed for inflation), and the integration runs from the person s current age to age Only assets that can be readily converted to cash (that is, fungible ones) are included. As a result, consumer durables, such as automobiles, televisions, furniture, household appliances, and the like, are excluded here since these items are not easily marketed or their resale value typically far understates the value of their consumption services to the household. 8 The mortality rate data are from the U.S. Bureau of the Census, Statistical Abstract of the United States,1985, and 2003, Washington, D.C., U.S. Government Printing Office. 8

9 It should be noted that the calculations of DB pension wealth for current workers Among current workers (w) the procedure is somewhat more complex. The SCF provides detailed information on pension coverage among current workers, including the type of plan, the expected benefit at retirement or the formula used to determine the benefit amount (for example, a fixed percentage of the average of the last five year s earnings), the expected retirement age when the benefits are effective, the likely retirement age of the worker, and vesting requirements. Information is provided not only for the current job (or jobs) of each spouse but for up to five past jobs as well. On the basis of the information provided in the SCF and on projected future earnings (see the Appendix for details), future expected pension benefits (EPB w ) are then projected to the year of retirement or the first year of eligibility for the pension. Then the present value of pension wealth for current workers (w) is given by: (2) DB w = LR EPB(1 - m t )e -δt dt where RA is the expected age of retirement and LR = A - RA is the number of years to retirement. As above, and the integration runs from the expected age of retirement to age are based on employee response, including his or her stated expected age of retirement (see Appendix D),. not on employer-provided pension plans. A couple of studies have looked at the reliability of employee-provided estimates of pension wealth by comparing self-reported pension benefits with estimates based on provider data. Using, data from the 1992 wave of the HRS, both Gustman and Steinmeier (1999) and Johnson, Sambamoorthi, and Crystal (2000) found that individual reports of pension benefits varied widely from those based on provider information. However, the latter also calculated that the median values of DB plans from the two sources were quite close (about a 6 percent difference). 9 I also used as alternatives real discount rates of 1.5, 2.5, and 3.0 percent. The results of the analysis are not materially altered (and not shown in the paper). 10 Technically speaking, the mortality rate m t associated with the year of retirement is the probability of surviving from the current age to the age of retirement. The discount rate is again set at 5 percent (2 percent if PB is indexed). 9

10 Here, too, it should be noted that estimates of Social Security wealth are based on B. Social Security Wealth: For current Social Security beneficiaries (r), the procedure is again straightforward. Let SSB be the Social Security benefit currently being received by the retiree. Again, the SCF provides information for both husband and wife. Since Social Security benefits are indexed for inflation, (gross) Social Security wealth is given by: (3) SSW r = 0 SSB(1 - m t )e -δt dt where it is assumed that the current social security rules remain in effect indefinitely. 11 The imputation of Social Security wealth among current workers is based on the worker's projected earnings history estimated by regression equation (see the Appendix for details). The steps are briefly as follows, First, coverage is assigned based on whether the individual expects to receive Social Security benefits and on whether the individual was salaried or self-employed. Second, on the basis of the person's earnings history, the person's Average Indexed Monthly Earnings (AIME) is computed. Third, on the basis of existing rules, the person's Primary Insurance Amount (PIA) is derived from AIME. Then, (4) SSW w = LR PIA(1 - m t )e -δ*t dt. As with pension wealth, the integration runs from the expected age of retirement to age reported earnings at a single point in time. These estimates are likely to be inferior to those based on longitudinal work histories of individual workers (see, for example, Smith, Toder, and Iams, 2001, whose estimates are based on actual Social Security work histories.) In fact, actual work histories do show much more variance in earnings over time than one based on a human capital earnings function projection. Moreover, they also 11 Separate imputations are performed for husband and wife and an adjustment in the Social Security benefit is made for the surviving spouse. See the Appendix for details. The discount rate is again set at 2 percent. 12 As with pension wealth, the mortality rate m t associated with the year of retirement is the probability of surviving from the current age to the age of retirement and the discount rate is set at 2 percent. 10

11 show many periods of work disruption that I cannot adequately capture here. However, I do have some retrospective information on work history provided by the respondent (see Appendix D for details). In particular, each individual is asked to provide data on the total number of years worked full-time since age 18, the number of years worked parttime since age 18, and the expected age of retirement (both from full-time and part-time work). On the basis of this information, it is possible to approximate the total number of full-time and part-time years worked over the individual s lifetime and use these figures in the estimate of the individual s AIME. Nonetheless, since my estimates of SSW assume a continuous work life, I am likely to be overstating the value of SSW for many workers. This is likely to bias upward my estimates of mean and median SSW, as well as a downward bias in the variability of Social Security wealth. It may also lead to an understatement of the correlation between net worth and SSW. For all three reasons, this estimation procedure will likely lead to an over-estimate of the degree to which SSW equalizes total wealth inequality. Estimates will be provided for the following components of household wealth: (5) NW = NWX + DC where DC is the current market value of Defined Contribution pension plans and HDWX is marketable household wealth excluding DC and NW corresponds to marketable wealth or net worth. Total pension wealth, PW, is given by: (6) PW = DC + DB Retirement wealth RW is then given as the sum of pension and social security wealth: (7) RW = PW + SSW Finally, augmented household wealth, AW, is given by (8) AW = NWX + RW. C. Alternative Pension Wealth Imputations. So far I have treated DB and DC pension wealth as well as SSW as comparable concepts. However, there are important differences in their estimation. Most notably, the calculation of DB wealth is estimated on the basis of the future stream of pension benefits on the assumption that the employee remains at his or her firm of employment until the person s expected retirement date. The computation of SSW is also based on the assumption that the worker remains at work 11

12 until the person s expected retirement date. On the other hand, the DC valuation is based on the current market value of DC plans. There are two ways of putting DB (and SSW) and DC on an equal footing. First, one can compute DB and SSW on a separation value basis, assuming that the worker stopped working as of the year of the survey, say It is possible to do this for SSW by computing the SS benefits that would be received if the worker stopped working in For DB, we can roughly do this for responses that indicate DB benefits as a percentage of last year s salary by using current earnings (as opposed to projected earnings) to calculate the expected DB benefit. However, most respondents simply indicate what they expect their DB benefit to be at year of retirement. Since we do not know the actual formula used, it is impossible to compute the expected benefit on the basis of current earnings. Moreover, in the case of the 1983 SCF, we do not have the underlying data to perform the necessary calculations for the separation value of SSW or DB wealth. 13 The second (and only feasible technique with the data at hand) is to project forward the employer contribution to DC plans, like 401(k)s (see Part E of the Appendix). If we assume, as in the case of DB pensions, that workers remain at their company until retirement and that the terms of their DC contract with their employer stays the same, then it is possible to do this. In most cases, the employer contribution is a fixed percentage of the employee s salary. On the basis of the estimated human capital earnings functions for each worker, it is possible to calculate the annual stream of future employer contributions to the DC plan until retirement (which I call DCEMP). 14 The addition of DCEMP to household wealth puts the treatment of DC pension wealth 13 This comparison might actually overstate the case. Some future DB entitlements are based on past jobs (as opposed to current jobs), which, of course, do not require any additional work time with the company to secure future DB benefits. However, the portion of future DB benefits that accrue from past jobs is relatively small I estimate only 9.65 percent of total DB benefits in Moreover, the 1983 data do not present a problem, since DC wealth was a trivial amount, so that we can again safely ignore this in the wealth comparison between 1983 on the one hand and 1989 and 2001 on the other hand. 12

13 roughly on a par with that of DB pension wealth since both represent future additions to household wealth from the employer. 15 The SCF questionnaire indicates how many DC pension plans each spouse has (up to three per spouse). 16 Information on the employer contribution to DC pensions plans is recorded in two ways. First, in some cases, the contribution is given as a flat dollar amount. Though it is not indicated in the survey data whether the dollar contribution is indexed to inflation over time, I assume that it is indexed to the CPI, which seems the more likely arrangement. 17 Let EMPAMT be the dollar amount of the employer contribution to the DC plan. Then, the present value of the stream of future employer contributions, DCEMP a, is given by: (9) DCEMP a = 0 LR EMPAMT (1 - m t )e -δt dt where m t is the mortality rate at time t conditional on age, gender, and race; and the discount rate δ is set at 2 percent. 18 The integration runs from the current year to LR, 15 If, indeed, employee contributions to DC plans are a net addition to household savings, then this treatment might understate the contribution of DC plans to future household wealth. However, as noted above in Section 2, the evidence on this issue is mixed. Moreover, if both the employee and employer contributions to DC plans substitute for other forms of saving, then DC* may actually overstate the net addition to wealth from the DC plans. DCEMP and PW also differ in terms of the risk associated with future benefits. The benefit levels in DB plans are already set by the terms of the plans, and DB wealth depends mainly on future tenure in the company and future earnings. If the employee leaves the firm, the employee is still entitled to the benefits accumulated up to the point of departure. The establishment of the Pension Benefit Guarantee Corporation in 1974 does, at least, insure the pension benefits (up to a fixed amount) in the event of the bankruptcy of a company. However, there is still a risk that a company s DB plan may end, which is not trivial in light of the recent terminations of DB plans in my private businesses. In comparison, DCEMP depends not only on future tenure with the firm and future earnings but also on future employee and employer contributions and future rates of return. There is no statutory guarantee that a DC plan will be maintained over time. Termination with a company ends future employer contributions into the DC plan. All in all, DB benefits are very likely less risky than DC benefits. 16 The SCF records DC plans only for the main job of each respondent. No information on DC plans is provided for secondary employment. This does not appear to be a significant problem because in 2001, 99.4 percent of the total labor earnings of the head and 98.8 percent of that of the spouse came from the person s primary job. 17 This will, if anything, bias upward the estimated employer contribution to the DC pension plan 18 This calculation assumes that the real rate of return on DC assets equals the discount rate δ. It should also be noted that past employer contributions to DC plans are already included in the current market value of DC wealth. 13

14 where RA is the expected age of retirement and LR = A - RA is the number of years to retirement. Second, in most cases, the employer contribution is given as a percent of earnings. If we assume that the proportion, EMPPER, is fixed over time, then DCEMP b, is given by: (10) DCEMP b = 0 LR EMPPER E* t (1 - m t ) e -δt dt where E* t is the predicted earnings of the worker at time t in constant dollars (see Appendix C for details). I then modify the basic accounting variables as follows: DC* = DC + DCEMP (6 ) PW* = DB + DC* (7 ) RW* = PW* + SSW (8 ) AW* = NWX + RW* D. Future Tax Liability on Pension Wealth. A related issue concerns the tax liability attached to pension wealth. DC contributions are tax sheltered when they are made and subject to income tax on withdrawal either in part or as a lump-sum pension distribution. DB pension benefits and lump-sum distributions are likewise taxable on receipt. In the case of DC accounts, their current market value is greater than their posttax value. In principle, the post-tax value of both DC and DB wealth should be used when computing net worth, PA, and AW in order to put pension wealth on an equal footing with marketable assets, which are not subject to income tax on withdrawal. 19 It is possible to make a relatively crude adjustment to the values of DB and DC wealth for future taxes on receipt. 20 In principle, to make a proper calculation we would 19 This treatment might actually overstate the case, since some assets such as stocks are subject to capital gains tax on sale, whereas DC and DB assets are not subject to capital gains tax. The main difference is that the non-pension assets are purchased with post-tax income and therefore their base value is not subject to any additional income tax. For simplicity, I ignore capital gains taxes here. A third tax on wealth, the estate tax, would apply to all asset components of net worth, including DC wealth, though not generally to DB wealth. (The exception would be lump-sum distributions from DB plans. This is a relatively small amount. In 2001, lump-sum distributions from DB plans comprised only 2.7 percent of total DB wealth.) For simplicity, I also ignore estate tax liability here.. 20 See Poterba (2004) for further discussion of the tax treatment of retirement savings. 14

15 have to predict future income (and its composition), future tax deductions and exemptions, and the future tax schedule as well at retirement. For simplicity, I assume that for current workers income at retirement equals 80 percent of the pre-retirement income. 21 In the case of current beneficiaries, I assume that their (post-retirement) income remains fixed over their remaining life. I also assume that marital status remains unchanged, that couples file joint returns, and that the tax schedule remains fixed over the remaining lifetime of the individual. In addition, it is assumed that the full value of the accumulations in the DC plans is withdrawn over the lifetime of the individual. I also net out taxes on social security benefits according to the tax code current at the time of the survey. 22 For convenience, all estimates of pension and social security wealth shown in the subsequent tables are pre-tax unless otherwise specified. 4. Trends in Standard Measures of Household Wealth. Table 1 documents a robust growth in wealth during the 1990s. After rising by 7 percent between 1983 and 1989, real median wealth among all households was 16 percent greater in 2001 than in As a result, median wealth grew slightly faster between 1989 and 2001, 1.32 percent per year, than between 1983 and 1989, at 1.13 percent per year. 24 Mean net worth was 65 percent higher in 2001 than in 1983 and 44 percent larger than in Mean wealth grew quite a bit faster between 1989 and 2001, at 3.02 percent per year, than from 1983 to 1989, at 2.27 percent per year. Moreover, 21 The 80 percent figure is a typical replacement rate. I also use Adjusted Gross Income for the income concept, which is provided in the SCF data. 22 In 2001, social security benefits are tax-free if Adjusted Gross Income excluding Social Security benefits was less than $32,000 for a married couple filing jointly and $25,000 for singles or couples filing separately. 23 I have elected to use the CPI-U as the deflator for both income and wealth. The CPI-U has been criticized for overstating the rate of inflation. As a result, the U.S. Bureau of Labor Statistics (BLS) has been providing a new consumer price series, called the CPI-U-RS, which makes quality adjustments for housing units and consumer durables such as automobiles and personal computers and employs a geometric mean formula to account for consumer substitution within CPI item categories. Indeed, the U.S. Bureau of the Census uses the CPI-U-RS price index to deflate income data from the Current Population Survey (CPS) to constant dollars. The use of the CPI-U-RS series instead of the CPI-U series will not alter relative trends in mean versus median wealth, as well as the various components of augmented wealth. I use the CPI-U series since it is standard for wealth data. 24 The time trend is similar when the value of vehicles is included in net worth. 15

16 mean wealth grew almost three times as fast as the median, suggesting widening inequality of wealth over these years. 25 The robust performance of median net worth over the 1990s contrasts sharply to trends in median income. Median household income, based on Current Population Survey data, after gaining 11 percent between 1983 and 1989, grew by only 2.3 percent from 1989 to 2001 much slower than median net worth. The net change over the whole period was 14 percent. In contrast, mean income rose by 16 percent from 1983 to 1989 and by another 12 percent from 1989 to 2001, for a total change of 30 percent. The SCF income data show a slightly lower growth in median household income than the CPS data from 1983 to 2001 but a somewhat higher growth in mean income (36 versus 29 percent). However, the SCF data show a much slower growth in both mean and median income than the CPS from 1983 to 1989 and much higher growth between 1989 and The median wealth of the baby-boomers was 41 percent above the overall average in 2001, while their mean wealth was only 11 percent above average. However, the relative net worth position slipped in comparison to the same age group in While the overall mean wealth level grew by 65 percent between 1983 and 2001, that of age group 40 to 55 gained only 48 percent, so that the average wealth of this age group fell from 23 percent above average to 11 percent above average. The median wealth of the age group declined from 49 to 41 percent above average. Both mean and median (SCF) income of the baby-boomers were about a third higher than the corresponding overall mean and median in In terms of median income, there was a small dip in their relative position from 1983 to 2001 (from 40 to 36 percent above the overall median), while in terms of mean income, there was a small relative gain (from 28 to 32 percent above the overall mean). Another interesting measure is the ratio of wealth to income. Age group showed gains in the ratio of median wealth to median income over the period, 25 The choice of endpoints, 1983 and 2001, is dictated by data availability was a post-recession year while 2001 was at the end of the boom of the 1990s, one characterized by sharply rising stock prices. The results may be sensitive to the choice of years. On the other hand, 1989 was also a close to peak year, so that a comparison of 1989 and 2001 would reflect the use of two years at similar points in the business cycle. 16

17 from 1.82 to 1.96, as well as the ratio of mean wealth to mean income, from 4.5 to 4.7. The gains in the ratio of median wealth to income were of the same order of magnitude as for all households, but gains in the ratio of mean wealth to income were somewhat lower than that for all households. Table 2 shows trends in both wealth and income inequality. It is most useful to begin with the income trends. Household income inequality, based on CPS data, increased between 1983 and 1989, with the Gini coefficient rising from to over this period. Between 1989 and 2001, the Gini coefficient again increased, from to All told, according to the CPS figures, the Gini coefficient was up by points from 1983 to The SCF data show an even greater level of income inequality and a greater increase between 1983 and 2001, particularly over the years 1983 to The trends are different for wealth. The Gini coefficient for wealth, after rising by points between 1983 and 1989, dipped slightly between 1989 and Over the full period, the Gini index for wealth rose by points, compared to a increase in that for CPS income and a gain for SCF income. Wealth inequality among the Baby Boom generation was somewhat lower in 2001 than for the full population (a Gini index of versus 0.826). However, the Gini index advanced about the same amount for age group as for all households over the period (though it did show a larger jump among this age group in the period than overall and a greater dip in the period than for all households). Income inequality among the baby-boomers was only slightly below the overall figure in Income inequality among age group 40 to 55 showed an even larger surge from 1983 to 2001 than among all households (0.092 versus Gini points). The timing was also different with a much steeper rise between 1989 and 2001 for this age group compared to overall and a smaller gain from 1983 to 1989 than overall. Table 3 shows the overall portfolio composition of household assets in 1983, 1989, and In 2001, owner-occupied housing was the most important household asset in the breakdown shown in the table, accounting for 28 percent of total assets. However, net home equity -- the value of the house minus any outstanding mortgage -- amounted to only 19 percent of total assets. Real estate, other than owner-occupied housing, comprised 10 percent, and business equity another 17 percent. 17

18 Demand deposits, time deposits, money market funds, CDs, and the cash surrender value of life insurance made up 9 percent and pension accounts 12 percent. Bonds and other financial securities amounted to 2 percent; corporate stock, including mutual funds, to 15 percent; and trust equity to a little less than 5 percent. Debt as a proportion of gross assets was 13 percent, and the debt-equity ratio (the ratio of total household debt to net worth) was There have been some notable trends in the composition of household wealth over the period between 1983 and The first is that pension accounts rose from 1.5 to 12.3 percent of total assets. This increase largely offset the decline in total liquid assets, from 17.4 to 8.8 percent, so that it is reasonable to conclude that households have substituted tax-free pension accounts for taxable savings deposits. The second is that gross housing wealth remained almost constant as a share of total assets over this period. Moreover, according to the SCF data, the homeownership rate, after falling from 63.4 percent in 1983 to 62.8 percent in 1989, picked up to 67.7 percent in However, net equity in owner-occupied housing has fallen almost continuously, from 23.8 percent in 1983 to 18.2 percent in 1998, though it did pick up to 18.8 percent in The difference between the two series is attributable to the changing magnitude of mortgage debt on homeowner's property, which increased from 21 percent in 1983 to 37 percent in 1998 but then fell back to 33 percent in Overall indebtedness first increased, with the debt-equity ratio leaping from 15.1 percent in 1983 to 19.4 percent in 1995, before falling off to 17.6 percent in 1998 and 14.3 percent in Likewise, the ratio of debt to total income first surged from 68 percent in 1983 to 91 percent in 1995, leveled off in 1998, and then declined to 81 percent in Moreover, as we saw above, the fraction of household recording zero or negative net worth jumped from 15.5 percent in 1983 to 18.0 percent in 1998 and then fell slightly to 17.6 percent in However, if mortgage debt on principal residence is excluded, then the ratio of other debt to total assets fell off even more, from 6.8 percent in 1983 to 3.1 percent in One implication is that over the 1990s families have been using tax-sheltered mortgages and home equity loans to finance normal consumption rather than consumer loans and other forms of consumer debt. 18

19 The proportion of total assets in the form of other (non-home) real estate fell off sharply, from 15 percent in 1983 to 10 percent in 2001, as did financial securities, from 4.2 to 2.3 percent. Unincorporated business equity fell slightly as a share of gross wealth over this period. These declines were largely offset by a rise in the share of corporate stock in total assets, from 9.0 in 1983 to 14.8 percent in 2001, reflecting the bull market in corporate equities. However, still in 2001, direct stock ownership ranked only third in total value in this breakdown, behind housing and business equity. However, if we include the value of stocks indirectly owned through mutual funds, trusts, IRAs, 401(k) plans, and other retirement accounts, then the share of total stocks owned shoots up to 25 percent of total assets in 2001 more than double the share in This corresponds to an increase in the ownership rate (both direct and indirect) of stocks from 32 to 52 percent between 1989 and In terms of investment assets, 39 percent of households in 2001 held corporate stock, mutual funds, financial securities, a business or a trust fund, in comparison to 34 percent in 1983 and 43 percent in The portfolio composition of age group 40-55, as well as their ownership patterns, are remarkably similar to the overall figures. Business equity made up a slightly larger share of the baby-boomers portfolio in 2001 than that of all households (20.7 versus 17.2 percent) and pension accounts were also a slightly higher share (14.2 versus 12.3 percent), while corporate stock and mutual funds comprised a smaller percentage (12.8 versus 14.8 percent). The baby-boomers also had a somewhat higher debt-equity ratio than overall (18.0 versus 14.3 percent), debt to income ratio than overall (86 versus 81 percent) and a higher ratio of mortgage debt to house value than overall in 2001 (40 versus 33 percent). The home ownership rate among baby-boomers was 75 percent in 2001 (compared to 68 percent overall), and almost 60 percent of baby-boomers owned stock directly or indirectly (compared to 51 percent of all households).. 5. Pension Wealth Table 5 highlights trends in pension coverage over the period. In this and the subsequent tables, it should be noted that the unit of observation is the household, 26 It is not possible to compute this statistic for

20 not the individual worker. I focus on age group 40 to 55 but for comparisons also show corresponding statistics for age group 65 and over. 27 The share of households in age group 40 to 55 with DC pension accounts skyrocketed over the period, from 16 to 64 percent, or by 47 percentage points, with most of the gains occurring after The results are very similar for all households over the years 1989 to 2001, with the proportion holding pension accounts advancing by 28 percentage points. Among the elderly, there is also a dramatic increase in the share with DC plans (from 2 to 35 percent). Opposite trends are apparent for DB pension wealth, with the share of households in age group 40 to 55 with DB plans plummeting by 29 percentage points between 1983 and Most of the loss in DB coverage again occurred after Among all households, the share with DB plans slipped by 11 percentage points between 1989 and 2001, about the same as for age group 40 to 55. Among age group 65 and over, the proportion with DB plans plummeted by 20 percentage points between 1983 and The percentage of households in age group 40 to 55 covered by either a DC or a DB plan increased modestly, from 68 percent in 1983 to 74 percent in Among all households, the share increased by almost 10 percentage points over the 1990s, a little more than among middle-aged households, while among the elderly the share was down by 4 percentage points from 1983 to There were huge increases in the average holdings of DC pension accounts (see Table 6). Among age group 40 to 55, the average value of these accounts increased by a factor of 8 between 1983 and 2001, to $71,000 in Most of the growth occurred after Among all households the increase was a factor of 5 from 1989 to 2001, a bit more than for age group over this period. Opposite trends are again evident for the average value of DB pension wealth, which declined by 21 percent among age group 40 to 55 over the period. Most of the decline occurred between 1989 and It decreased by 15 percent among all households from 1989 to 2001 (about triple the rate of decline of age group 40-55). However, mean pension wealth PW still showed strong 27 As indicated in the Appendix, calculations of pension and social security wealth were not performed for individuals under the age of 40 in the 1983 SCF. As a consequence, for 1983, I show results only for age group 40 to 55 and age group 65 and over. 20

21 gains, rising by 67 percent among age group between 1983 and 2001, with all of the gains occurring after Mean PW gained 66 percent among all households from 1989 to 2001, somewhat less than among households in age group Among elderly households, average PW grew by 70 percent from 1983 to 2001, about the same as among age group Gains look even stronger when DCEMP is included. In 2001, the average value of DCEMP among age group was $39,800, or 56 percent of DC, and $29,300 among all households, or 55 percent of mean DC. In 1989, the corresponding ratios are greater 2,78 and 1.78, respectively. The higher ratios in 1989 reflect the lower accumulations of DC in that year compared to 2001 (the absolute values of DCEMP are actually greater in 2001 than in 1989). The addition of DCEMP augments the mean value of PW by 32 percent among ages 40 to 55 in 2001 and by 31 percent among all households. The corresponding figures in 1989 are 38 and 28 percent, respectively. The addition of DCEMP, not surprisingly, greatly enhances the rate of growth of the mean value of pensions between 1983 and The mean value of DC* (the sum of DC and DCEMP) rose by a factor of 13 between 1983 and 2001 among age group and by a factor of 3.3 among all households from 1989 to 2001, and mean PW* (the sum of DB and DC*) more than doubled over the period among age group (compared to a 67 percent gain in PW) and rose by 69 percent from 1989 to 2001 among all households (compared to a 66 percent rise in PW).The marginal tax rate on pension wealth averaged 12.4 percent in 1983, 12.0 percent in 1989, and 13.4 percent in 2001, so that post-tax PW* grew much slower than pre-tax PW* between 1983 and 2001 among age group (48 versus 122 percent) but slightly slower among all households over the period. The last panel of Table 6 shows the ratios of pension wealth by component of age group to age group 65 and over. There was virtually no change in the relative value of pension wealth PW over the period between age group and 65 and over, though the relative value of PW* shot up from 118 to 154 percent. However, the relative value of post-tax PW* was almost identical in 2001 as it was in So far, in 28 Median pension values are strongly affected by the share of households with pension wealth and, as a result, are not shown here. 21