http://www.federalreserve.gov/pubs/oss/oss2/papers/concentration.2004.3.pdf
Currents
and Undercurrents: Changes in the Distribution of Wealth, 1989–2004
Arthur B. Kennickell
Senior Economist and Project Director,
Survey of Consumer Finances
Mail Stop 153
Federal Reserve Board
Washington, DC 20551
Phone: (202) 452-2247 Fax: (202) 452-5295
Email: Arthur.Kennickell@frb.gov
SCF Web Site: http://www.federalreserve.gov/pubs/oss/oss2/scfindex.html
January 30, 2006
Arthur B. Kennickell
Senior Economist and Project Director,
Survey of Consumer Finances
Mail Stop 153
Federal Reserve Board
Washington, DC 20551
Phone: (202) 452-2247 Fax: (202) 452-5295
Email: Arthur.Kennickell@frb.gov
SCF Web Site: http://www.federalreserve.gov/pubs/oss/oss2/scfindex.html
January 30, 2006
Note: as with every document on this site, yellow cells and boxes are Wealthandwant comments and calculations, not part of the original work! See also: Recent Changes in U.S. Family Finances: Evidence from the 2001 and 2004 Survey of Consumer Finances, by Brian K. Bucks, Arthur B. Kennickell, and Kevin B. Moore |
Abstract
This paper considers changes in the distribution of the wealth of U.S. families over the 1989–2004 period using data from the Survey of Consumer Finances (SCF). Real net worth grew broadly over this period. At the same time, there are indications that wealth became more concentrated, but the result does not hold unambiguously across a set of plausible measures. For example, the Gini coefficient shows significant increases in the concentration of wealth from 1989 to 2004, but the wealth share of the wealthiest one percent of families did not change significantly. Graphical analysis suggests that there was a shift in favor of the top of the distribution, while for the broad middle of the distribution increases were about in proportion to earlier wealth. Within this period, there are other interesting patterns. For example, from 1992 to 2004 the wealth share of the least wealthy half of the population fell significantly to 2.5 percent of total wealth. The data show little in the way of significant distributional shifts since the 2001 survey. The paper also presents some information on underlying factors that may explain a part of the distribution of wealth, including capital gains, saving behavior and income, inheritances, and other factors. There are two special topic sections in the paper. The first presents information on the distributions of wealth of African American and Hispanic families. The second presents information on the use of debt across the distribution of wealth.
The views presented in this paper are those of the author alone, and they do not necessarily reflect the views of the Board of Governors of the Federal Reserve System or its staff. The author wishes to thank Michael Neal for assistance with the figures in this paper, staff at NORC for collecting the data, and the SCF respondents for generously sharing their information for research purposes. Thanks to Brian Bucks, Gerhard Fries, Diana Hancock, and Kevin Moore for comments. The author bears sole responsibility for any errors.
This paper considers changes in the distribution of the wealth of U.S. families over the 1989–2004 period, an interval that contains a variety of events that had strong effects on the finances of families. The period includes two recessions, one in 1990–1992 and one in 2001. Leading up to 2001, there was a tech-led boom of the stock market, which deflated in that year and had approximately recovered by the end of 2004. Between 2001 and 2004, real estate prices rose sharply in most areas, while home equity borrowing flourished in a market of relatively low interest rates. Over the whole period strong forces were altering the nature of production, work, and many other aspects of life. For example, at the beginning of the period, the “World Wide Web” was something known to only a relatively small number of technologically sophisticated people, and by the end “www” addresses were commonly seen nearly everywhere. Entirely new markets and jobs were created as older structures faded or transformed themselves to remain competitive. In the underlying demographics, the bulge of baby boomers continued to move through the age distribution, while total population grew about 19 percent over the 15 years, with immigration explaining a non-negligible fraction.
As a consequence of these disparate forces, the distribution of family wealth did shift — most certainly so for individual families. But trends in the overall distribution of wealth are hard to characterize, and often different statistics give different impressions. The data used in this paper, the triennial Survey of Consumer Finances (SCF), supplemented by data from Forbes, offer what is probably the best hope for identifying shifts in the wealth distribution for the whole population.1 But despite the special design of the SCF and the great care taken in processing that data, it is still a relatively small survey, and as such it may lack the statistical power to identify some relatively small changes clearly. That said, the data do identify statistically significant shifts in the wealth distribution over the period considered here. But for the 2001–2004 interval, even while the survey clearly records the surge in real estate values and home-secured borrowing, it shows little in the way of significant overall distributional movements.
1See Kopczuk and Saez [2003] for
an examination of the wealth of
the population sufficiently wealthy to file an estate tax return.
The first section of this paper characterizes the data used. The second section reports a series of different views of the wealth distribution and its dynamics between 1989 and 2004.
Three special topic sections follow. The first traces some of the sources of wealth variation that can be seen in the SCF data. The second examines the relative wealth of African American families and Hispanic families. The third examines the use of debt across the wealth distribution. A final section offers a summary of key findings.
I. Data Used in this Paper
The primary data used in this paper derive from the Survey of Consumer Finances (SCF), a triennial survey sponsored by the Board of Governors of the Federal Reserve System in cooperation with the Statistics of Income Division (SOI) of the Internal Revenue Service. The version of the data used is the full internal data set available only within the SCF group at the Federal Reserve Board. Beginning with the 1989 survey, great efforts have been made to ensure the maximal amount of comparability of the surveys over time. Earlier years of the survey have been used to examine wealth changes (see Kennickell [2003] and references cited therein). This paper focuses on changes relative to the most recently available data at the time this paper was written, the 2004 wave of the survey.2 Data collection for this survey and all the surveys beginning with the 1992 survey was undertaken by NORC, a social science and survey research organization at the University of Chicago.
2See Bucks, Kennickell, and
Moore [2006] for an overview of the 2004
survey and see Kennickell [2000] for a review of the survey methodology.
The SCF collects detailed information on the assets and liabilities of
families, in addition to data on their work history, their use of
financial institutions,
their attitudes and expectations, a variety of demographic
characteristics, and other variables. The asset
and liability data are used to build the measure of net worth used in
this paper.3
This measure
includes the sum of financial assets (checking, savings and money
market accounts, certificates of
deposit, savings bonds, other types of bond, mutual funds, hedge funds,
stocks, annuities,
managed investment accounts, trusts, the cash value of life insurance,
retirement accounts, and
miscellaneous financial assets) and nonfinancial assets (principal
residences, other residential real
estate, net value of nonresidential real estate, businesses, vehicles,
and miscellaneous
nonfinancial assets) net of the sum of all outstanding debts (loans on
a primary residence or other
residential real estate, credit card balances, installment loans,
margin loans, loans against cash
value life insurance and pension accounts, and miscellaneous debts).3For comparability, this measure
is the same as that used in
Bucks, Kennickell, and Moore [2006].
It is important to note that retirement assets are only partly captured
in this measure of wealth. The wealth measure is intended to reflect
only assets where the
family has substantial control or direct interest. Thus, the measure of
retirement assets used
includes Individual Retirement Accounts (IRAs), Keogh Accounts, and
balances in
account-type pensions from which withdrawals could be made, either as a
simple withdrawal or a
loan; other types of employer- or union-based retirement account or
annuity right and
coverage under the Federal Old Age and Survivors’ Insurance (OASI) are
excluded. In 2004, of the
33 percent of families headed by a person with some sort of pension on
a current job, 64
percent had at least one account-type plan of the sort included in the
net worth measure, 20
percent had at least one account-type plan that would not be included
in the wealth measure, and
33 percent had at least one non-account-type plan other than OASI.
Although broadening the net
worth measure to include the omitted account-type plans would be
straightforward,
including an appropriate representation of the other plans would not be
so simple.4
To do
so would require computing an expected present value of annuity
benefits, which would entail
assumptions about the proper framework to use in including or excluding
future employer and employee
contributions to such plans as well as assumptions about how benefits
might be affected by
future employment and wages, the rate of future inflation, and future
interest rates. There
is no consensus about what approach to take in making such assumptions.
Moreover, the additional
effort that would be required is beyond the scope of this paper.4See Gale and Pence [2005]
and
Kennickell and Sundén [1997] for
approaches for including a present value of annuity benefits in the
calculation of net worth.
This is not a trivial matter. See Arthur Kennickell and Annika Sunden's Pensions, Social Security, and the Distribution of Wealth (1997), Tables 4 and 5, for calculations of the magnitude of these contributors to Networth for 1989 and 1992, and Edward Wolff's The Unraveling of the American Pension System, 1983-2001 (2004). Kennickell and Sunden report that, for 1992, Pension Wealth and Social Security Wealth together are approximately equal to the Household Networth reported by the SCF. Social Security Wealth in 1992 was slightly more than half of the combined total. So the increases shown here in RETQLIQ are offsetting an unmeasured loss in Pension Wealth.
|
The data collected in the survey are subjected to an intensive review,
with the aim of detecting serious errors on the part of interviewers or
respondents.
Often comments recorded by interviewers play a key role in this
determination, but computer-driven
searches for common types of problem are equally important. Sometimes
such editing
discovers recorded values that are clearly wrong, but there is not
sufficient information to determine
the correct answer; in such cases, the erroneous value may be set to a
missing value. In other
instances where there are multiple interrelated responses that are
inconsistent, irreconcilable
discrepancies may be allowed to stand if there is no information to
determine the most reliable of
the interrelated values.
Missing data in the survey are imputed using a technique of multiple
imputation.5
Multiple imputation allows an analyst of the imputed data to develop a
measure of the uncertainty associated with the fact that some of the
values of
variables were originally missing. The sample for the SCF is designed
to provide adequate information to
examine a broad range of financial behaviors. Because some assets are
held very
disproportionately by relatively wealthy families, a straightforward
area-probability sample (or other
such sample with equal selection probabilities) would be unlikely to
capture enough wealthy
families for meaningful analysis of such variables, unless the sample
size were quite large.
Short of a huge sample size, an area-probability sample would provide a
very inefficient
representation of wealthy families, and consequently of assets
concentrated among such families. In
addition, there are strong indications that wealthy families are far
less likely to respond to
surveys than other families (see Kennickell [2005]); thus without some
means of identifying relative
wealth a priori, the realized sample for a survey would be very likely
to be biased in terms of its
representation of such families.
5See Rubin [1987] for a
discussion of multiple imputation in general and Kennickell [1998] for
its application to the SCF.
The SCF addresses both these statistical efficiency and bias concerns
through the use of a dual-frame sample design.6 A
national multi-stage area-probability design provides broad coverage of
common economic behavior; this part of the sample provides
about two-thirds of the final interviews. The other part of the sample
employs information from
SOI, under stringent provisions to protect the privacy of taxpayers, to
select a sample with
disproportionate representation of families more likely to be
relatively wealthy; this
sample is stratified by a “wealth index” computed using observed
capital income flows and related
information (see Kennickell [2001]. The two parts of the sample are
adjusted for sample
nonresponse and combined using weights to provide a representation of
families overall.6See Kennickell and Woodburn
[1999] and Kennickell [1999] for a discussion of the construction of
the SCF sample and weighting design.
It is important to note that the SCF excludes one small set of
families by design. People who are listed in the October issue of
Forbes as being among the 400
wealthiest in the U.S are excluded. This exclusion is made for two
reasons.
- First, it is very unlikely that an interviewer could manage to reach a sufficient number of such people to justify the time and effort to attempt to do so.
- Second, it is almost certain that interviews with such people could not be included in the public version of the SCF data without introducing too large a probability that the identity of the respondent might be compromised.
To enable the calculation of statistical hypothesis tests, the SCF uses a replication scheme (Kennickell [2000] and Kennickell and Woodburn [1999]). A set of replicate samples is selected by applying the key dimensions of the original sample stratification to the actual set of completed SCF cases and then applying the full weighting algorithm to each of the replicate samples. To estimate the variability of an estimate from the SCF, independent estimates are made with each replicate and with each of the multiple imputations; a simple rule is used to combine the two sources of variability into a single estimate of the standard error.
II. The Distribution of Wealth
A. Forbes Data
Every October, Forbes publishes a list of what it estimates to be the 400 wealthiest people in the U.S.7 These people probably represent the segment of wealthy families best known to the public in general, though their characteristics may well differ from those of families even a fraction of a percentile lower in the wealth distribution. Because, as noted above, the SCF and Forbes samples do not overlap, these sources are, in principle, natural complements in describing the distribution of wealth. For simplicity, the data from the two sources are treated separately.
7See the October 2004 issue of
Forbes and Canterbury and Nosari [1985] for details on the methodology.
The
Forbes data for recent years are available at www.forbes.com; the
earlier data are only available in the printed version of the magazine.
Unfortunately, on the basis of the very limited documentation
available, it is not clear how consistent the Forbes methodology is
within a given year and across time. From what is known, the estimates
represent an “educated guess,” with a variety of inputs. Probably
the
largest sources of potential error in these estimates are in the
assignment of ownership of assets spread within a family and the
valuation of assets that may not be publicly traded.
According to calculations based on the data reported in Forbes, the wealth held by the 400 wealthiest people grew by widely varying amounts over the period covered in this paper (table 1). Based on the three years of data transcribed for the 1989-1995 period, the annualized growth rate in real terms was 0.5 percent over the first three years and 3.0 percent over the second three years.8 Reflecting in part the rise and decline of high technology stocks over the succeeding five years, the growth rate hit a high of 33.7 percent in 1997 and a low of 12.0 percent in 2000, before turning strongly negative — minus 23.1 percent in 2001 and minus 9.3 percent in 2002. There was growth of 7.0 percent in 2003 and 2.2 in 2004. From 1989 to 2004, total real wealth of the group grew by 6.4 percent at an annual rate, but obviously with considerable variability within that period.
8Except where otherwise noted,
all dollar values reported in this paper have been adjusted to 2004
dollars using the CPI-U-RS, a research series computed by the Bureau of
Labor statistics that is intended to extend methodological improvements
in the current consumer price index back in time to the degree possible.
| Table 1: Wealth of the Forbes 400; 1989–2004. | |||||||||
| Year | Total
wealth ($B) |
Max
wealth ($B) |
Min.
wealth ($M) |
Avg.
top 10 ($B) |
100th value ($M) | Max ÷ min | Avg.
top 10 ÷ min wealth |
100th
value ÷ min value |
Forb.
wealth ÷ (Forb. wealth + SCF wealth) (percent) |
| 1989 | 396.4 | 7.7 | 405 | 4.5 | 1029.8 | 18.9 | 11.1 | 2.5 | 1.54 |
| 1992 | 402.2 | 8.3 | 350 | 6.7 | 1055.5 | 23.8 | 19.1 | 3.0 | 1.68 |
| 1995 | 439.5 | 18.2 | 418 | 8.1 | 1107.8 | 43.5 | 19.3 | 2.6 | 1.67 |
| 1996 | 547.5 | 22.2 | 340 | 9.2 | 1318.6 | 65.1 | 27.0 | 3.9 | NA |
| 1997 | 732.2 | 46.7 | 557 | 15.4 | 1760.5 | 83.8 | 27.6 | 3.2 | NA |
| 1998 | 854.2 | 67.6 | 579 | 22.0 | 1967.5 | 116.8 | 38.0 | 3.4 | 2.49 |
| 1999 | 1,173.1 | 96.3 | 708 | 32.2 | 2720.4 | 136.0 | 45.4 | 3.8 | NA |
| 2000 | 1,313.7 | 69.1 | 795 | 32.6 | 2851.9 | 86.9 | 41.0 | 3.6 | NA |
| 2001 | 1,009.9 | 57.6 | 640 | 25.6 | 2134.2 | 90.0 | 40.0 | 3.3 | 2.20 |
| 2002 | 916.2 | 45.6 | 578 | 23.6 | 1890.7 | 78.2 | 40.2 | 3.3 | NA |
| 2003 | 980.1 | 47.62 | 616 | 24.4 | 2053.8 | 76.7 | 39.6 | 3.3 | NA |
| 2004 | 1,002.1 | 51.0 | 750 | 22.7 | 2200.0 | 68.0 | 30.3 | 2.9 | 1.96 |
| Note: All dollar-related figures are adjusted to 2004 constant dollars. | |||||||||
Within the Forbes group, there were substantial variations in the concentration of wealth held by the group over the 15-year period shown. For example, the ratio of the highest value to the minimum value rose monotonically from 18.9 in 1989 to a peak of 136.0 in 1999 — about seven times the ratio in 1989 — and then declined with slight interruption to 68.0 in 2004. Although the general shape of the ratio of the average of the highest 10 values to the lowest value is similar over this time, the change is much less dramatic — it peaks at 45.4, or about four times the value of the ratio in 1989. The 100th value ranged from 2.5 to 3.8 times the lowest value over the period; at its peak in 1996, this ratio was only about fifty percent higher than its low point in 1989.
For the years where the SCF and the Forbes data overlap, it is possible to see what proportion of wealth is, in principle, missing from the SCF. From 1989 to 1995, the total wealth of the Forbes 400 as a proportion of the sum of that wealth and total wealth measured in the SCF ranged from 1.5 to 1.7 percent; following the pattern of growth in the top rank of the Forbes group, the proportion jumped to 2.5 percent in 1998, before falling off a bit in both 2001 and 2004. In 2004 the fraction was 2.0 percent.
Because membership in the Forbes group is not constant over time, these shifts refer to changes in a slice of the wealth distribution, not the fortunes of individual families. However, since the group members are identified by name, it is possible to trace their dynamics. As shown in Kennickell [2003] for the period from 1989 to 2001, of the 400 people in the 2001 list, 230 were not anywhere in the 1989 list. Persistence in the list was highest for people who were in the wealthiest 100 — of the people in this group, 45 were in the same group in 1989 and 23 others were elsewhere in the list.
| Might some have not yet inherited in 1989 the fortunes that put them on the list by 2001? |
B. SCF Data
Broad growth. Across the 1989 to 2004 period, the inflation-adjusted wealth distribution rose broadly (table 2), though the pattern for individual families over the period might well have been otherwise.9 Although the fraction of families with negative net worth stayed about the same across the fifteen-year period aside from a jump in 1998, the population with non-negative wealth tended overall to shift to higher wealth groups, with some possibly cyclically-influenced deviations within the period.10 For example, in 1989, 26.5 percent of families had net worth of less than $10,000; by 2004 the figure was 22.7 percent. Over the same period, the share of families with at least $500,000 in net worth rose from 10.8 percent to 17.7 percent. Beneath this general trend are many undercurrents affecting the distribution of wealth, some of which are explored in this paper.
9Panel data would be needed
to
address wealth changes for
individual families. There are SCF panel data only for the 1983–1989
period. Because of the notable substantive and methodological
differences between the 1983 data and the cross sectional surveys
beginning with 1989, the 1983 information is not used here. See Avery
and Kennickell [1991] for an analysis of wealth dynamics based on the
1983 and 1989 SCF. In the comparisons reported here, no adjustments are
made for variations in the size and composition of households.
Furthermore, no use is made of the Forbes data in the SCF estimates
reported.
10See Kennickell [2003] for a detailed discussion of families with negative net worth.
10See Kennickell [2003] for a detailed discussion of families with negative net worth.
| Table 2: Percent distribution of net worth in 2004 dollars, 1989–2004. | ||||||
| |
1989 | 1992 | 1995 | 1998 | 2001 | 2004 |
| <0 | 7.2 | 7.2 | 7.1 | 8.0 | 6.9 | 7.1 |
| |
0.6 | 0.4 | 0.4 | 0.4 | 0.3 | 0.3 |
| =0 | 3.9 | 3.2 | 2.6 | 2.5 | 2.6 | 1.7 |
| |
0.5 | 0.2 | 0.3 | 0.2 | 0.2 | 0.2 |
| 1-999 | 4.0 | 3.0 | 2.4 | 3.2 | 2.7 | 2.3 |
| |
0.4 | 0.3 | 0.3 | 0.3 | 0.2 | 0.2 |
| 1K-2.49K | 3.1 | 3.6 | 2.5 | 2.4 | 2.3 | 3.2 |
| |
0.3 | 0.3 | 0.2 | 0.2 | 0.2 | 0.2 |
| 2.5K-4.9K | 4.3 | 3.4 | 3.4 | 3.2 | 3.3 | 4.0 |
| |
0.5 | 0.3 | 0.2 | 0.2 | 0.3 | 0.2 |
| 5K-9.9K | 4.0 | 4.9 | 5.5 | 4.7 | 4.6 | 4.4 |
| |
0.4 | 0.3 | 0.3 | 0.3 | 0.3 | 0.3 |
| 10K-24.9K | 8.0 | 9.2 | 9.2 | 7.9 | 7.9 | 7.8 |
| |
0.6 | 0.4 | 0.5 | 0.5 | 0.3 | 0.4 |
| 25K-49.9K | 9.3 | 10.5 | 10.0 | 9.4 | 9.0 | 8.8 |
| |
0.6 | 0.6 | 0.5 | 0.4 | 0.4 | 0.5 |
| 50K-99.9K | 13.3 | 14.2 | 15.8 | 12.5 | 12.2 | 11.9 |
| |
0.6 | 0.6 | 0.6 | 0.6 | 0.5 | 0.5 |
| 100K-249.9K | 20.8 | 21.5 | 22.2 | 21.9 | 19.1 | 18.6 |
| |
1.0 | 0.9 | 0.7 | 0.9 | 0.6 | 0.7 |
| 250K-499.9K | 11.3 | 10.0 | 9.9 | 12.5 | 13.5 | 12.4 |
| |
0.7 | 0.5 | 0.3 | 0.8 | 0.6 | 0.6 |
| 500K-999.9K | 5.6 | 5.2 | 5.5 | 6.5 | 8.4 | 9.6 |
| |
0.5 | 0.3 | 0.3 | 0.6 | 0.6 | 0.7 |
| >=1M | 5.2 | 4.2 | 3.8 | 5.3 | 7.5 | 8.1 |
| |
1.3 | 0.2 | 0.2 | 0.3 | 0.4 | 0.3 |
| Standard errors are given in italics. | ||||||
| Table 2a: Cumulative percent distribution of net worth in 2004 dollars, 1989–2004. | ||||||
| |
1989 | 1992 | 1995 | 1998 | 2001 | 2004 |
| negative | 7.2 | 7.2 | 7.1 | 8.0 | 6.9 | 7.1 |
| less than or equal to $0 | 11.3 | 10.4 | 9.7 | 10.5 | 9.5 | 8.8 |
| under $1,000 | 15.3 | 13.4 | 12.1 | 13.7 | 12.2 | 11.1 |
| under $2,500 | 18.4 | 17.0 | 14.6 | 16.1 | 14.5 | 14.3 |
| under $5,000 | 22.7 | 20.4 | 18.0 | 19.3 | 17.8 | 18.3 |
| under $10,000 | 26.7 | 25.3 | 23.5 | 24.0 | 22.4 | 22.7 |
| under $25,000 | 34.7 | 34.5 | 32.7 | 31.9 | 30.3 | 30.5 |
| under $50,000 | 44.0 | 45.0 | 42.7 | 41.3 | 39.3 | 39.3 |
| under $100,000 |
57.3 | 59.2 | 58.5 | 53.8 | 51.5 | 51.2 |
| under $250,000 | 78.1 | 80.7 | 80.7 | 75.7 | 70.6 | 69.8 |
| under $500,000 |
89.4 | 90.7 | 90.6 | 88.2 | 84.1 | 82.2 |
| under $1,000,000 | 95.0 | 95.9 | 96.1 | 94.7 | 92.5 | 91.8 |
Means and quantile indicators. The relationship between the mean and the median of net worth is often taken as a simple indicator of changes in distribution. From 1989 to 2004, the mean value of wealth measured in the SCF rose 61.2 percent, while the median rose 35.3 percent (table 3). Although the mean and median both grew, the difference in these growth rates over the 15-year period signals that wealth moved in relative terms to the upper half of the distribution during this time. At the beginning of the period, the mean was 4.0 times the median, and owing the differences in growth rates, the mean was 4.8 times the median at the end. It is noteworthy that the ratio of the mean to the median was relatively little changed from 1989 until 2001, when it rose 0.7 percentage point. Yet, it was in 2001 that the wealth of the Forbes 400 saw the largest percentage decline over the period considered here. This difference suggests that changes for the Forbes group may be relatively loosely coupled with those for other families.
Examination of other key percentiles of the distribution suggests that the overall picture is less straightforward than that shown by the means and medians. Although there was growth from 1989 to 2004 at the 10th, 25th, 75th and 90th percentiles, the ratio of the 75th and 90th percentiles of the wealth distribution to the value of the 25th percentile declined over the period with considerable variation within the period.11 But this decline was not statistically significant, owing in part to the unusually large standard errors for the 1989 estimates. However, if 1992 is taken as the starting point of the period, the ratios increase significantly by 2004 and roughly in parallel with the ratio of the mean to the median. Thus, the data at this level generally support the idea that wealth may have shifted toward the upper part of the distribution at least from 1992 to 2004.
11The table shows the ratio of
the 90th and 25th percentiles,
rather than the ratio of the 90th and 10th percentiles more familiar
from analysis of income distributions, because the 10th percentile is
often zero or a very small positive or (absolute) negative value.
Gini coefficient. Another common indicator of the distribution of wealth is the Gini coefficient, which is defined in terms of the Lorenz curve, a graph of the cumulative percent of wealth against the cumulative percent of families, where the families are sorted by wealth. The wealth Gini coefficient is given as one minus twice the area under the Lorenz curve. In a world of perfect equality, (where the lorenz curve would be a 45 degree line) the value would be zero, and in a world where all wealth was held by one person, the value would be approximately one. Thus, the wealth Gini coefficient gives a measure of the relative size of the deviation of a distribution from perfect equality. Two important and interrelated auxiliary points are that the deviations are weighted equally, independently of location in the distribution, and that two different distributions could generate the same Gini coefficient. Thus, the Gini coefficient does not provide an unambiguous and neutral index of the wealth distribution.
From 1989 to 2004, the wealth Gini coefficient rose from 0.79 to 0.80, a relatively small but statistically significant change (table 4). At the same time, there was a slight increase in the comparable Gini coefficient computed for assets and a slight decrease in the Gini coefficient for debt. In contrast, the coefficient for income began the period at about the same level at which it ended, after having fallen and risen in between; moreover, it is about two-thirds the level of the coefficient for wealth.
| Table 4: Gini coefficients for net worth, assets, debt and income, 1989–2004. | ||||
| |
Net worth | Assets | Debt | Income |
| 1989 | 0.7863* | 0.7481 | 0.7447* | 0.5399 |
| |
0.0055 | 0.0080 | 0.0063 | 0.0086 |
| 1992 | 0.7808* | 0.7379* | 0.7479* | 0.5005* |
| |
0.0061 | 0.0064 | 0.0049 | 0.0049 |
| 1995 | 0.7841* | 0.7323* | 0.7243* | 0.5146* |
| |
0.0043 | 0.0044 | 0.0040 | 0.0042 |
| 1998 | 0.7935 | 0.7444 | 0.7138 | 0.5302 |
| |
0.0051 | 0.0052b | 0.0037 | 0.0040 |
| 2001 | 0.8034 | 0.7603 | 0.7104 | 0.5643* |
| |
0.0041 | 0.0042 | 0.0034 | 0.0037 |
| 2004 | 0.8047 | 0.7540 | 0.7053 | 0.5406 |
| |
0.0049 | 0.0051 | 0.0036 | 0.0040 |
| Standard
errors are given in italics. *=significantly different from the 2004 level at 95 percent confidence. |
||||
Concentration ratios. Because the Gini coefficient attempts to summarize many complex changes in terms of a single number, it may miss important variation for particular parts of a distribution or for particular subpopulations. A more detailed means of summarizing the relative distribution of wealth is the use of concentration ratios, the proportion of total wealth held by specific groups. In 2004, slightly more than one-third of total net worth was held by the wealthiest one percent of families (table 5). Although the estimated level of this share has changed over the surveys since 1989, the differences are not statistically significant. In 2004, the next-wealthiest nine percent of families held 36.1 percent of total wealth, again, a figure not significantly changed over the course of the surveys. This leaves less than a third of the total for the remaining ninety percent of the population. A subset of that group, families in the bottom half of wealth distribution, held only 2.5 percent of total wealth in 2004, and this figure is significantly different from the higher estimates for 1995, 1998, and 2001; of course, those differences reflect movements elsewhere in the distribution, but the statistical power of the tests is not sufficient to identify where among the groups shown the offsetting changes occurred. A possible explanation of the decline for the lowest wealth group might be changes in their use of debt, but a separate examination of gross assets yields a pattern similar to that seen for net worth.
| Table 5: Proportions of total net worth and of gross assets held by various percentile groups, 1989–2004. | |||||||||||
| Proportion of total net worth held by group | Proportion of total gross assets held by group | ||||||||||
| Net worth percentile group | Net worth percentile group | ||||||||||
| 0-50 | 50-90 | 90-95 | 95-99 | 99-100 | 0-50 | 50-90 | 90-95 | 95-99 | 99-100 | ||
| 1989 | 3.0 | 29.9 | 13.0 | 24.1 | 30.1 | 5.4 | 32.5 | 12.6 | 22.3 | 27.1 | |
| |
0.3 | 1.8 | 1.6 | 2.3 | 2.3 | 0.4 | 1.8 | 1.6 | 2.1 | 2.1 | |
| 1992 | 3.3* | 29.6 | 12.5 | 24.4 | 30.2 | 6.6* |
32.1 | 12.0 | 22.6 | 26.7 | |
| |
0.2 | 1.1 | 0.7 | 1.3 | 1.4 | 0.3 |
1.1 |
0.7 | 1.2 |
1.3 | |
| 1995 | 3.6* | 28.6 | 11.9 | 21.3 | 34.6 | 7.5* | 31.2 | 11.4 | 19.5 | 30.4 | |
| |
0.2 | 0.7 | 0.6 | 0.9 | 1.3 | 0.3 | 0.7 | 0.6 | 0.8 | 1.2 | |
| 1998 | 3.0* | 28.4 | 11.4 | 23.3 | 33.9 | 6.7* | 30.8 | 10.9 | 21.7 | 29.9 | |
| |
0.2 | 0.9 | 0.6 | 1.2 | 1.5 | 0.3 | 0.9 | 0.6 | 1.2 | 1.4 | |
| 2001 | 2.8* | 27.4 | 12.1 | 25.0 | 32.7 | 5.6 | 29.9 | 11.7 | 23.4 | 29.5 | |
| |
0.1 | 0.7 | 0.7 | 1.1 | 1.4 | 0.2 |
0.8 | 0.7 | 1.0 | 1.3 | |
| 2004 | 2.5 | 27.9 | 12.0 | 24.1 | 33.4 | 5.8 |
31.0 | 11.4 | 22.2 | 29.5 | |
| |
0.1 | 0.9 | 0.7 | 1.2 | 1.2 | 0.2 | 0.9 | 0.7 | 1.2 | 1.1 | |
| Standard errors are given in italics. *=significantly different from the 2004 level at 95 percent confidence. |
|||||||||||
Graphical analysis. A more direct and comprehensive way of characterizing changes across the wealth distribution is to use a quantile-difference (QD) plot, a graph of differences between the levels of two distribution at each quantile of the distributions. Figure 1a shows a QD plot of the difference between the wealth distributions for 2001 and 2004, where the line plotted represents the 2004 level minus the 2001 level.12 At the bottom of the distribution, the estimate shows that wealth became more negative in that range, though the changes are not significantly different from zero. From there up to about the 50th percentile, there was very little change in levels between the two years. Above that point, the estimates show some substantial gains, but they are significantly different from zero in this point-wise sense only from about the 75th to the 85th percentiles.
12The dots around the central
line indicate the 95 percent
point-wise confidence intervals at selected points across the
distribution. The vertical axis is scaled using the inverse hyperbolic
sine transformation (with a scale factor of 0.0001), which has the
convenient property of being approximately linear around zero and
approximately logarithmic away from zero.
In general, the level changes may be misleading as indicators of shifts of wealth shares across the distribution; for a group to increase its share of wealth, its wealth has to grow at a faster rate than the wealth of other families. A relative quantile-difference (RQD) plot addresses this point by normalizing the change in a QD plot by the level of the base year; that is, the amount shown for each common quantile in the two distributions is the percent change in the level of wealth corresponding to the quantile. Viewed in this way, the changes in the lowest fifteen percent of the distribution tend to explode, largely because the denominator is quite small over much of this range (figure 1b). For the next approximately ten percent, the relatively small dollar changes in the levels are shown to be more substantial as a proportion—on the order of minus ten percent. For those higher in the distribution, the main effect of the normalizing is progressively to flatten the differences. But as with the QD plot, the only changes that were significantly different from zero were those in the range of about the 75th to 85 percentiles.
It is somewhat surprising over a three-year period when there were substantial increases in real estate values and some recovery of earlier stock market losses that there were not more notable changes at this level of distributional analysis, but the data suggest that the implied wealth changes were offset to a substantial degree by borrowing and were also diffused fairly broadly across the wealth distribution. Over longer periods, economic forces may have an opportunity to play out more fully. Figure 2a shows change over the longest period possible with the consistent series of SCF data, from 1989 to 2004.13 Here there are statistically significant increases in wealth almost everywhere across the distribution. Above about the 10th percentile, the plot slopes linearly upward in inverse hyperbolic sine space (an approximately logarithmic transformation over most of this range) until about the 95th percentile, from which point the line spikes sharply upward. In the RQD transformation (figure 2b), the data show large proportional changes below the 25th percentile, but with very wide confidence intervals.
13Both QD and RQD plots are
given in the appendix for 2004 relative to all of the other intervening
survey years.
Between the 25th and 80th percentiles, the graph forms a rough “bowl” shape, where the bowl is flat across the middle at about 35 percent–implying about 2 percent growth on an annual basis over the period. From the 80th percentile, the line drops off again before spiking upward in the top few percentiles. The spike is sufficiently well estimated that it is significantly different from the other changes above the 25th percentile. Thus, this plot does provide some support for the increase in wealth concentration at the very top, as one would expect from the Forbes data over the same period.
Variability of cross-sectional wealth over time. Because only cross sectional data are available from the SCF in the period considered here, it is not possible to examine the movements of families in the wealth distribution over time. Still, it is of macroeconomic interest to know how variable the overall distribution was over the period. To this end, figure 3a shows an estimate of the coefficient of variation (the standard deviation divided by the mean) of the level of wealth associated with each quantile in the six surveys from 1989 to 2004 (a QCV plot). The shape of the figure looks like a somewhat exaggerated version of the RQD plot for 1989–2004 shown in figure 3a. That is, excluding the range of negative and zero wealth, the least variability is in the middle of the distribution with generally increasing variability on either side. Note that because the data are not de-trended, there is a substantial baseline level of variability, and because growth differed notably across the distribution, some of the differences in the figure may be largely a product of the spread induced by variations in the trend in growth rates across the distribution.14 Adjusting the 1989–2004 figures for each quantile to remove the geometric mean of growth over the period yields the graph of de-trended variation shown in figure 3b. Restricting attention to the meaningful range for interpretation, beginning around the 20th percentile, the highest variability by far is for the lowest and highest wealth groups.
14The precision of estimation of
each cross sectional element
may vary because of differences in the degree of sampling error. Thus,
some of the differences observed in cross sectional variability may
reflect the differences in sampling error.
Because wealth for these lowest groups is a relatively small buffer against personal and macroeconomic shocks, it is not surprising to see such high variation across a period that included two recessions as well as important restructuring of employment. A minimum of variation is reached at about the 35th percentile, and above that point, variability increases approximately linearly until the very top of the distribution. The rising variation across the upper 65 percent of the distribution reflects the riskiness of the underlying portfolios, a factor discussed later in this section.
14
Figure 1a: QD plot for net worth, 2004 minus 2001.
Figure 1b: RQD plot for net worth, 2004 minus 2001 as a percent of 2001.
15
Figure 2a: QD plot for net worth, 2004 minus 1989.
Figure 2b: RQD plot for net worth, 2004 minus 1989 as a percent of 1989.
16
Figure 3a: QCV plot for net worth, not de-trended, 1989–2004.
Figure 3b: QCV plot for net worth, de-trended, 1989–2004.
17
III. Sources of Wealth Changes
Changes in wealth overall are due to capital gains and losses — realized and unrealized — on portfolio items and to variations in saving out of varying current income, including returns on assets. Portfolio selections, inheritance or gifts — both those received and those given — and changes in household structure and other demographic factors may also be important for individual families. Given the cross-sectional nature of the SCF data and the lack of relevant retrospective information in the survey, these points can be addressed only obliquely. Here income, saving behavior, capital gains, inheritances, portfolio structure, and demographics are addressed separately as contributors to the shape of the wealth distribution.
The role of income. Two relevant sources of income data are available for this analysis. There are six cross sections of SCF data, and for the period 1999 to 2001 there is information on the variability of income over time for individual observations from the SOI data used in the design of the 2004 SCF sample.
When the SCF cross sections are analyzed with a QCV plot, the result is that the data show very little difference in variability across almost all of the income distribution (figure 4a).15 Only at the extremes of the distribution does the variability increase substantially. The lowest region of variability is the area around the median. De-trending the data by quantiles makes the plot more spikey, but other than shifting the low point to about the 25th percentile, the relative pattern is maintained (figure 4b). These results reflect the volatility of the income distribution, so if families remained at the same relative position over the period, the volatility would have implications for the plausibility of income shocks as a driver of wealth changes. But the small changes compared to wealth shifts over the period suggests that idiosyncratic variations are key or that income is not the primary driver of changes in wealth distribution.
15The appendix provides QD and
RQD plots comparing the
distribution of family income in 2004 with that in the earlier surveys.
Figure 4a: QCV plot for income, not detrended, 1989–2004.
Figure 4b: QCV plot of income, detrended, 1989–2004.
Individual-specific income information is available in the SOI data used to construct the list sample for the SCF. To support the selection of that sample and later nonresponse adjustments, every observation of domestic residents in the file is assigned a value of the wealth index used to create the strata for sample selection. For the great majority of cases, this assignment is made on the basis of three years of data on the components of income, with the intent of smoothing transitory shocks in the estimation of the wealth index.16 In addition to its utility for the subject of this paper, the inspection of the variability of the SOI income data is an important element of the evaluation of the sample design. For each wealth-index stratum in the SOI data, figure 5 reports selected quantiles (25th, 50th, 75th, and 90th) of the distribution of the coefficient of variation of total income, where the underlying coefficients of variation are computed using three observations for each of the individual taxpayer units.17 Although the exact correspondences of the stratum indicators with income and wealth are not disclosable, it can be said that the first stratum encompasses approximately the lowest 75 percent of the wealth index distribution and the strata above the third one encompass about the highest 2 percent of the distribution.
16Cases where more limited
income information is used include
those where there was a change in filing status (typically from married
filing jointly to filing separately, or vice versa) or where a return
was not filed at all.
17Note that the wealth index turns more strongly on capital income flows (e.g., interest, dividend, and business income and capital gains and losses) than on total income. Observations with fewer than three years of income data are not included in the plot.
17Note that the wealth index turns more strongly on capital income flows (e.g., interest, dividend, and business income and capital gains and losses) than on total income. Observations with fewer than three years of income data are not included in the plot.
Figure 5: Distribution of the coefficient of variation of income, 1999–2001, by wealth index stratum; 25th, 50th, 75th, and 90th percentiles.
What is clear in the figure is that there is a longer right-hand tail for the higher strata. Variability of labor income is the most important factor at the bottom end of the distribution and variability of capital income is the most important factor at the other end. According to financial theory, risky assets (those more variable in price) should have a higher expected rate of return. Thus, loosely speaking, one would expect to see a relatively greater density of higher positive returns for risky assets than would be the case for safer assets. For some families in this data set, the spikes in returns could then be taken as directly reflective of one reason they were as wealthy as they were. However, it may also be that other families, particularly very wealthy ones, acted over this time in a deliberate way to time their income receipts, either for tax-related purposes or for more narrow personal purposes. In this case, the variability would have no direct implication for wealth.
For many families, the income measured either in the SCF or in SOI data is, in principle, reasonably close to their true economic income. For others, employer contributions to retirement plans and health insurance may be important elements in a broader concept of income. According to the 2004 SCF, 7.7 percent of families had some sort of employer-provided vehicle that could be used for personal purposes. Some families have access to perquisites such as employer-provided vacation properties. But surely the largest hole in the measurement of the income of U.S. families is unrealized capital gains. Some gains may be realized only very rarely – for example, upon the sale of a house – and others may never be realized as income–for example, an appreciated business passed to heirs. As discussed in more detail later in this section, unrealized gains appear to be a very important factor in the observed distribution of wealth.
Saving out of income. The SCF also contains information on families’ saving practices. Respondents are asked to describe their family’s typical saving practices and then to characterize the level of their expenditures (excluding capital investments) relative to their income. In 2004, the proportion of families that routinely spent at least as much as their income declines across wealth percentile groups — from 42.7 percent in the bottom quartile of the wealth distribution to 5.9 percent in the highest one percent of the distribution (table 6).18 The proportion of families with some type of saving plan rises from 26.5 percent in the bottom quartile to 69.1 percent for the top one percent. A substantial proportion in every wealth group saves whatever is “left over” at the end of the month. A very similar pattern of increasing saving with wealth is seen in the data on the actual saving behavior of families in the preceding year. Thus, it seems reasonable to characterize relatively wealthy families as being ones where it is more likely that at least additional wealth is generated by active saving out of current income.
18Because the general patterns
across wealth groups of both typical saving behavior and saving over
the previous year are very similar across
the 1989–2004
surveys, only the 2004 data are shown.
| Table 6: Usual saving behavior and saving behavior last year, by percentile of the distribution of net worth, percent, 2004. | ||||||||
| All | 0-25 | 25-50 | 50-75 | 75-90 | 90-95 | 95-99 | 99-100 | |
| Usual saving: | ||||||||
| Spend more than income | 7.0 | 13.5 | 7.6 | 4.2 | 3.4 | 2.8 | 1.6 | 1.5 |
| Spend same as income | 16.3 | 29.2 | 18.0 | 11.5 | 6.5 | 5.7 | 7.0 | 4.4 |
| Save “left over” income | 30.2 | 30.8 | 33.0 | 30.3 | 27.0 | 25.2 | 28.5 | 25.0 |
| Some saving plan | 46.5 | 26.5 | 41.4 | 54.1 | 63.0 | 66.3 | 62.8 | 69.1 |
| Saving last year: | ||||||||
| Spent more than income | 15.4 | 24.8 | 16.4 | 11.2 | 10.4 | 6.8 | 9.3 | 5.9 |
| Spent same as income | 28.5 | 40.4 | 30.0 | 26.7 | 17.2 | 21.5 | 12.8 | 4.4 |
| Spent less than income | 56.1 | 34.8 | 53.6 | 62.2 | 72.4 | 71.6 | 78.0 | 89.8 |
Capital gains and wealth. Many families accumulate wealth simply by continuing to own certain assets. The Federal Reserve Board’s flow of funds accounts provide data on aggregate capital gains for several assets owned by the household sector (Z.1 Release, table R.100).19 For most of the years considered in this paper, holding gains on assets explain a very large fraction of the change in net worth in the flow of funds accounts; for example, in the fourth quarter of 2004, holding gains accounted for about 92 percent of the change in the net worth of the household sector.
19The household sector includes
actual households as well as non-profit organizations.
For assets held directly by households, gains are relevant mainly for real estate investments, private businesses, mutual funds, and publicly traded stocks. The SCF contains information on the unrealized gains embedded in all of these assets.20 For principal residences, the original purchase price and subsequent improvements are known; for other real estate, only the original price is known; for private businesses, the tax basis is known; and for mutual funds and directly-held stocks, the amount of gain or loss in the current value is known.
20An important omission
in the survey for this purpose is the lack of information about capital
gains
on assets held through retirement accounts and trusts.
For families in the lowest 25 percent of the wealth distribution (table 7), capital gains explain very little of a typical family’s wealth, and even in aggregate the fraction is quite small for the group (or negative in periods where the total wealth of the group is negative); this finding reflects the fact that ownership of assets that experience capital gains is relatively less common in this group. For the second 25 percent of the wealth distribution, the median ratio of gains to assets ranged only between 2.9 and 8.4 percent over the 1989–2004 period and the aggregate ratio of gains to assets ranged between 11.2 and 16.0 percent. But for the remaining half of the wealth distribution, both the median and aggregate ratios are generally above 20 percent, and for the highest 1 percent of the wealth distribution, the ratio ranges even higher — over 40 percent; the difference reflects the much higher rate of ownership among this group of the relatively risky assets that experience capital gains.
| Table 7: Median ratio of capital gains to assets and ratio of aggregate capital gains to aggregate assets, by percentiles of the wealth distribution, 1989–2004, percent. | ||||||||||||
| |
1989 |
1992 | 1995 | 1998 | 2001 | 2004 | ||||||
| Median ratio | Agg. ratio | Median ratio | Agg. ratio | Median ratio | Agg. ratio | Median ratio | Agg. ratio | Median ratio | Agg. ratio | Median ratio | Agg. ratio | |
| All | 11.1 | 34.0 | 7.9 | 32.1 | 5.6 | 27.3 | 7.6 | 28.3 | 8.8 | 28.8 | 11.2 | 30.7 |
| 0-25 | 0.0 | -6.2 | 0.0 | -7.5 | 0.0 | 1.1 | 0.0 | 2.2 | 0.0 | 0.0 | 0.0 | 5.7 |
| 25-50 | 5.4 | 14.5 | 4.6 | 13.7 | 2.9 | 11.2 | 5.5 | 13.3 | 6.8 | 13.6 | 8.4 | 16.0 |
| 50-75 | 30.2 | 28.1 | 22.0 | 24.5 | 19.5 | 21.4 | 18.8 | 21.1 | 20.1 | 23.5 | 24.8 | 27.0 |
| 75-90 | 32.0 | 32.0 | 28.9 | 29.1 | 24.2 | 25.6 | 21.0 | 23.9 | 20.7 | 25.7 | 24.3 | 27.5 |
| 90-95 | 32.2 | 32.7 | 27.5 | 29.1 | 21.4 | 20.9 | 21.0 | 24.9 | 21.3 | 24.3 | 26.7 | 28.0 |
| 95-99 | 31.9 | 33.3 | 31.1 | 33.5 | 21.1 | 26.0 | 26.1 | 29.0 | 24.5 | 28.0 | 30.0 | 32.7 |
| 99-100 | 42.1 | 44.2 | 41.1 | 43.8 | 33.6 | 38.6 | 39.6 | 42.0 | 33.7 | 38.9 | 31.2 | 37.0 |
Some families may turn over their assets frequently and receive capital gains as a regular part of their financial management, but have little in the way of unrealized gains; others may have realized more sporadic gains in the past. Although the survey does not provide historical data on gains and losses, it does provides information on realized gains and losses in the calendar year preceding the survey.21 The 2004 survey’s information for 2003 income indicates that 10.7 percent of families had gains or losses, the median amount for those having losses or gains was $20,600, and the ratio of the total amount of realized gains and losses to total assets was 0.4 percent.
21SOI data for the same period
shows about twice as much in realized gains as the SCF. Although the
SCF income data should, in principle, follow the classification of a
personal income tax return, it may differ in some instances. For
example, families that experience gains and losses through a sole
proprietorship or other business may report gains as business income
that should appear separately on a tax return.
Inheritances and wealth. Depending on how they are divided, inheritances and gifts may affect the shape of the overall wealth distribution. In principle, the SCF provides information on all inheritances and substantial gifts received by the family, though nothing is known about the wealth of the person making the transfer; substantial gifts given by the family are less well covered in the survey.22
22Transfers such as payments
for
education and intangible transfers such as “connections” are not
captured in the survey.
From 1989 to 2004, the share of families that reported having ever received an inheritance or substantial gift fell from 23.4 percent to 20.4 percent (table 8), with nearly all of that decline occurring from 1989 to 1992. Although the set of families in existence changes over time, it seems likely that the drop in 1992 and the similarly large dip in 2001 are at least partially reflective of variability due to sampling error. Receipt of such transfers varied somewhat during the 1989–2004 period for the lower half of the wealth distribution, but was almost unchanged from 1989 at 12.6 percent of families in 2004. Receipt is much more common in the higher wealth groups, but it appears that the proportion of families who had received these transfers tended to decline. For example, in 1989 50.8 percent of the wealthiest 1 percent of families had received such transfers at some time, whereas in 2004 the figure was 33.3 percent.
| Table 8: Percent of families having ever received an inheritance or substantial gift, by wealth percentile group, 1989–2004. | ||||||
| Year | Wealth percentile group | |||||
| All | 0-50 | 50-90 | 90-95 | 95-99 | 99-100 | |
| 1989 | 23.4 | 12.5 | 31.5 | 47.7 | 45.7 | |