The Measures Manajemen | Fakultas Ekonomi Universitas Maritim Raja Ali Haji 322.full

Klevmarken, Lupton, and Stafford 325 To initiate work with the matching approach we start by matching solely on the Swedish initial period cross-sectional wealth distribution. That is, we simply match on cross-sectional wealth for Sweden in 1993 and for the United States in 1994. Then for these initially matched cross-sectional samples, we examine wealth dynamics as measured by quintile transitions over five-year intervals. We find that standardization on initial wealth is an important component in explaining quintile wealth mobility differences between the two countries. The process of standardizing through matched sampling compresses the U.S. wealth distribution—that is, decreases the initial cross- section dispersion—thereby making movements across quintiles more likely. Given the relationship between lifecycle stages, that is, age and mobility, we ex- tend the initial exercise in standardization for a single wealth variable and create multivariate matched samples. We match on not only initial wealth and age, but also on other possible characteristic differences: family composition, marital status, and income. Somewhat surprisingly, this multivariate matching does little to change the results from the univariate matched sample on initial wealth. It is probable that initial wealth is a sufficient statistic, capturing many of the other differences between Swe- den and the United States. Nevertheless, certain variables do seem more critical than others and we examine these independently. For instance, in the United States many African American families do not participate significantly in the financial world of stocks, financial accounts, or mortgages Chiteji and Stafford 2000; Charles and Hurst 2000. Does the absence of this group in the Swedish sample ‘‘explain’’ some of the intercountry wealth mobility differences? Do other demographic and economic characteristics explain the country differences in mobility?

II. Cross-National Wealth Distributions Mid-1980s to Late 1990s

A. The Measures

Household wealth data for the United States come from the Panel Study of Income Dynamics PSID. 4 The PSID is a longitudinal survey that tracks the economic and demographic activities of approximately 6,000 households over their life course and spans the years 1968 to the present Hill 1992. As of 1994, more than 60 percent of the original set of sample households remained in the study. Weights have been constructed to account for differential attrition as well as the initial oversampling of poor households and the expansion over time in the number of younger households in the sample. Validation studies have documented that analyses of the PSID yield nationally representative results for the nonimmigrant population when sample weights are applied Becketti et al. 1988; Duncan and Hill 1989. These sample weights are used in all analyses presented in this paper. 5 In 1984, 1989, 1994, and 1999, the PSID collected comprehensive measures of assets and liabilities. In this paper, we focus on total household wealth. This includes net equity in homes and nonhousing assets divided into seven categories: 1 other 4. The PSID is sponsored by the National Science Foundation and the National Institute on Aging. 5. A recent comparison of the PSID family income deciles with those from the Current Population Survey, 1968 to 1999, can be found on the internet at: www.isr.umich.edusrcpsidq_inc_datapsid_vs_cps.pdf. 326 The Journal of Human Resources real estate; 2 vehicles; 3 farm or business; 4 stocks, mutual funds, investment trusts and stocks held in IRAs; 5 checking, savings accounts, CDs, treasury bills, savings bonds and liquid assets in IRAs; 6 bonds, trusts, life insurance; and 6 other expensive collections; and other debts. Wealth data were collected using the unfolding brackets technique pioneered in the PSID in 1984. With unfolding brackets respondents unable or unwilling to provide dollar amounts are routed through a series of upper and lower ranges. Several studies have shown that these estimates substan- tially improve the reporting of wealth Heeringa, Hill, and Howell 1995; Hurd and Mc Fadden 1997; Juster and Smith 1997. In the cross-section, there are 6,915 households in 1984; 7,114 in 1989; 7,415 in 1994; and 5,256 in 1999. 6 The panel used to analyze mobility from 1994 to 1999 includes 4,383 households and includes all 1994 households whose head remained head of household in 1999. In this sense, properly weighted, the panel results are nationally representative for 1994 households over the five-year period. Wealth data for Sweden come from the Household Market and Nonmarket Activi- ties Survey HUS. 7 Household definition, following rules and much of the question- naire content, is similar to those of the PSID. In both panels wealth data for the entire household were collected from one household member, usually the head. The first wave of HUS data was collected in 1984 followed by five more waves of which the latest was obtained in 1998. The sample size has increased from about 1,500 households in 1984 to about 2,400 in 1998. This is the net effect of splitoffs from sample households, refreshment samples and attrition. The initial response rate was 75 percent and in the refreshment samples the response rates have been about 70 percent. One exception is 1986 when it was no more than 60 percent. The panel response rates have stayed above 80 percent. In 1993 a special attempt was made to return previous attriters to the panel. The sample used below to analyze mobility in the period 1993–98 includes a maximum of 1,021 households. The initial 1984 sample and all refreshment samples were probability samples from the residen- tial noninstitutionalized population aged 18–74 years. Some people in the panel have become older than 74 years, but the share of this age group in the sample is smaller than in the Swedish population. The definition of a household in HUS is the same as that of the PSID, namely those who live with a designated head. The head is usually, but not always, a man. Changes in wealth may thus depend on changes in household composition. Household wealth data were collected in the 1984, 1986, 1993, and 1998 waves. The measures include real estate own ‘‘main’’ home, secondary home, rental real estate, and forest and farm property; financial assets stocks and shares, bonds, shares in mutual funds, bank deposits, annuities and life insurance, and private pen- sion policies but not occupational group pensions; consumer durables cars, boats, art and antiques, furniture, electronic equipment, washers, dryers, etc.; and mort- gages, consumer credit, and other loans including student loans. The questions about 6. Note that the 1999 PSID data do not include families formed via splitoffs in the 1997 to 1999 time interval. This is because the 1999 data are in early release format and the family composition changes have not been released. For our panel analysis, the absence of newly formed families is not of relevance. 7. See Klevmarken and Olovsson 1993, Flood et al. 1997, and the internet address http: www.handels.gu.seeconeconometricshushusin.html. Klevmarken, Lupton, and Stafford 327 annuities and private pension policies were introduced first in the 1986 wave and questions about wealth in unincorporated business first in the 1998 wave. The un- folding bracket technique was never introduced in the HUS questionnaires. Re- sponses to questions about financial assets were bracketed and the midpoints used as estimates for each bracket. Market values of real estate and in most cases also consumer durables were estimated by the respondent. As a control on the market values on main home, data on tax-assessed values were collected as well. All data were collected in the beginning of a year and stocks of wealth refer to the last of December the previous year or to the beginning of the year of interview. Survey data on assets usually have a problem with partial nonresponse. Following Rubin 1987, partially missing wealth data in the HUS surveys were compensated by multiple random imputations. In addition to nonmissing wealth items, imputations were a function of the age and schooling of the head, number of adults in the house- hold, if head had market work, and the mean market value of one-family houses in the municipality. Imputations were done at subaggregate level. Of eight subaggregates of total household wealth, on average 1.5 were imputed. The number of observations imputed depended on subaggregate and data wave, but the share of imputed observa- tions was typically about 20 percent. Ten replication sets of wealth data were created, making it possible to estimate the variance of a statistic including the uncertainty originating from imputations. For the aggregate ‘‘net worth,’’ however, the share of the total variance generated by imputations is usually so small that one safely can use only one replication. Measurement errors and imputations create a particular problem when analyzing mobility, because they tend to inflate mobility measures. This will in particular become the case when imputations in HUS were generated cross-sectionally. This problem is further discussed below. HUS wealth data previously have been compared with data assembled by Statistics Sweden using information from the tax assessment process see Bager-Sjo¨gren and Klevmarken 1993, 1998. We can trace the discrepancies, generally, to differences in population coverage and valuation principles; they do not suggest any alarming flaws in HUS data. Major remaining limits are that household surveys do not readily measure the top wealth holders. 8 The PSID data appear to track the household wealth distribution well up to the top one percentile point of the wealth distribution Juster, Smith, and Stafford 1998. We needed to obtain information on wealth holdings at the very top, from external sources such as the Statistics of Income estate tax measures of the Internal Revenue Service and the Forbes 400 lists for selected years because families in the upper two percentiles of the U.S. wealth distribution hold about 40 percent of wealth. The Swedish HUS survey is likely to have the same shortcoming—not covering the very top of the wealth distribution very well. Using a complete enumera- tion of register data for the wealthiest in Sweden, Statistics Sweden estimated that in 1997 the top 1 percent held 20.3 percent of total wealth and the top 5 percent 8. Some very high wealth holders have simple portfolios, consisting primarily of the value of ‘‘their’’ company or the value of the shares they hold, but many others have complex balance sheets with large differences between gross and net assets. In this latter case, reasonable measurements are probably out of the realm of accurate measurement in a household survey. 328 The Journal of Human Resources held 44.1 percent. 9 For the United States, it has been estimated that as of 1989, the share of wealth excluding pensions held by the top 1 percent was 25.6 percent and 47.3 percent for the top 5 percent Hurst, Louh, Stafford 1998. Given the extent to which the large gains in wealth over the past decade are a product of capital gains from corporate equities, it is likely that this share has increased. This can have impor- tant consequences for any study of household wealth since even missing the top 1 percent of U.S. families means that data are missing for about one-third of the overall wealth dollars. Household surveys normally do not include private pension wealth, and, arguably, they should ideally include public pension wealth for example, the present value of expected future Social Security payments, or even the present expected value of publicly provided medical care in both countries. As an extreme example, occupants of public housing in England have allegedly argued that they hold an effective and bequethable asset, the present value of the continuing subsidy. Of course, while private and public pensions in the form of annuity or in-kind type payments affect savings behavior and so should not be neglected, it is incorrect to simply add them to wealth since they are almost completely illiquid and cannot be borrowed against. The effects of this fact are crucial for understanding savings behavior Hurd 1989. On the other hand the differences between the two countries in coverage of the social safety net with the complete population coverage of public pensions, and public health care and old age care in Sweden, are likely to explain some of the country differences in wealth accumulation. The current value of the expected future stream of all these benefits is something we cannot estimate in our surveys. 10 For the purist, one can add to the challenge by considering the addition of human wealth and the issue of family wealth, the latter depending on mortality and divorce and separation rates. Although there is some pension information in the PSID in 1984 and much more in 1999 and 2001, we will not be able to include pension wealth consistently across the years in the two surveys. 11 What we have to work with is a measure of what has been called household wealth, or that which the household has in immediate or cur- rent period control for conversion to other assets or spending without high conversion costs. Some scholars define this measure of household wealth as ‘‘fungible wealth,’’ though there is no agreement on what should be included. The concept we use is the sum of the assets mentioned above less all liabilities. This implies that the HUS estimates of household wealth will have a somewhat broader coverage of consumer durables compared to the PSID and from 1986 also include private pension policies annuities. 12 9. Table 12 in Fo¨rmo¨genshetsfo¨rdelningen i Sverige 1997 med en tillbakablick till 1975. Report 2000:1, Statistics Sweden, O ¨ rebro. Please note that both the concept of wealth and the household definition used by Statistics Sweden differ from those used in this paper. 10. Estimates in Andersson et al. 2002 suggest that in 1999, the median value of pension rights from the public pension system for the age group 45–64 was a little more than 1 million SEK. The value of group pensions was about 200,000 for those who were covered, while the median value of financial assets was about 100,000 and of real estate about 350,000. 11. The 2001 PSID also includes information on pension plan providers. 12. This does not include the value of group pensions and public pensions. The share of private pension annuities in total household wealth is just a few percent. Compare with Table 2. Klevmarken, Lupton, and Stafford 329 A limitation of our study is that with the exception of a subsample for 1998, the HUS wealth data do not include the net value of unincorporated family businesses. For the PSID in 1994 this represented about 15 percent of household wealth. The corresponding 1998 figure for the HUS is only about 3 percent. For comparison purposes we net out wealth in the form of closely held family businesses from the PSID in all of our analysis except for Table 1. In the spirit of viewing wealth measures as a kind of continuum, we can defend ending the operational definition short of including equity in closely held businesses, though the definition is far from ideal. Focusing on household wealth that people could reasonably access within a year to consume or to convert to other assets, we can make a case for excluding equity in a family business. First, a business is often indivisible. It can be sold entirely, but then its value may depend on the complemen- tary input of the entrepreneur. 13 In the same vein it may provide poor collateral for a loan because it may only be liquidated on a fire sale basis. In studying wealth dynamics, we have found that the returns to equity in closely held business were quite high from 1989 to 1994 Hurst, Luoh, and Stafford 1998, p. 314. Yet, the willingness to respond to business equity wealth gains in the form of reduced savings is much lower than for gains in publicly held equities Juster et al. 1999. Less willingness to spend such gains supports the view that they are less accessible. 14

B. The Repeated Cross-Sectional Data on Household Wealth