ABSTRACT
This paper reports on a project to construct and evaluate a wealth and saving dataset for those households who responded to the supplementary wealth and active saving modules of the Panel Study of Income Dynamics (PSID) over the period of 1984 to 2005. The PSID is unique in providing household-level estimates of active saving and net wealth accumulation. The dataset includes measures of wealth, net wealth accumulation, and active saving covering seven wealth surveys and six periods for saving and wealth accumulation. For individual years, the sample sizes vary between 7 and 8 thousand families. The number of families participating in multiple periods is less, but 1,963 have participated in all seven of the wealth surveys. In comparisons with the SCF, we find that the wealth data of the PSID yield very similar results for the bottom 95 percent of the wealth distribution, but the very wealthy are underrepresented in the PSID. The major advantage is that the PSID has multiple observations on a family’s wealth over the two decades of coverage.
The micro-level data are also very consistent in the differences in saving and wealth accumulations that they imply across various socio-economic groups. However, the data do not capture the phenomenon of a secular decline in household saving rates over the past two decades that is so evident in the macroeconomic analysis. The measures of wealth accumulation, in contrast, do accord with other evidence of a secular rise in the ratio of household wealth to income. The growth of total wealth over the period of 1994 to 2005 is very similar to that reported in the Flow of Funds Accounts, and it compares very favorable with the estimated wealth gains shown in the SCF. A plausible explanation for the differing results for saving and wealth accumulation is that the survey respondents do not adequately distinguish between active saving decisions and capital gains in accounting for their increases in wealth. Given the strong gains in asset values over the past decade, it is possible that active saving is overstated.1 At present, our analysis is insufficient to provide an informed answer.
Finally, we used the panel dimension of the data set to explore the question of the effect of capital gains on saving. However, the results were ambiguous and the exercise suggests that measurement errors are a particularly serious problem in the time series dimension of the dataset, and that more powerful methods of controlling for their influence need to be applied in future work.
