June 2007 SGE Monthly Luncheon
"The Evolution of Household Income Inequality"
by Karen Dynan, Elmendorf, and Dan Sichel
Rapporteur: Ben Page, Congressional Budget Office*
According to Elmendorf, this research began with the widespread observation that the aggregate U.S. economy has grown more stable since the mid-80s (the “Great Moderation”). In previous work, the authors examined the possibility that the increased stability stemmed in part from financial market innovation (e.g., more widespread access to credit). However, testing that hypothesis requires examining what has happened to stability on the micro level. Previous research on that level found mixed results: some researchers have found greater micro-level volatility recently, while others have not. However, there has been relatively little research on the trends in volatility at the household level. This research sought to fill that gap.
The study uses data from the Panel Study on Income Dynamics (PSID), and takes a very simple metric as its measure of volatility: the cross-sectional distribution of two-year percentage changes in various measures of income. That measure is transparent and model-independent, but does not identify the persistence or source of income shocks, particularly whether they stem from voluntary or involuntary events.
The study begins by examining the volatility of labor earnings of heads of household. In those earnings, transitions to (and from) zero dollars from (and to) earnings over $10,000 rose sharply beginning in 1992. That corresponds to the timing of significant methodological changes in the PSID, including a shift from personal to telephone, computer-assisted interviews. Because that shift could not be tied to specific economic events, and there was no corresponding shift in transitions to and from low (but non-zero) earnings of heads, or any transitions for spouses, the authors ascribe that shift to the measurement change. They correct for the shift by leaving out observations that include zero earnings. The standard deviation of percent changes in that adjusted sample of earnings changes for household heads shows a slow and fairly steady rise over time since the early 70s. (Leaving in observations with zero earnings results in a shift to a higher trend in the 90s).
In addition, a cyclical pattern is clearly evident in the data; negative income changes are relatively more frequent during recessions, while income rises are more frequent during recoveries. Dividing the sample by educational attainment, all categories show some increase in volatility since the 70s, but there was relatively less increase among the college-educated until recently, when volatility has increased more rapidly.
When the earnings of spouses are pooled with those of heads of household, there is little trend in volatility. Both this and the finding of increased volatility for household heads run somewhat counter to earlier findings by the Congressional Budget Office (CBO). Such differences could stem from differences in the treatment of self-employment income, the age range covered, the treatment of large percentage income changes, and the data source (the CBO uses the Continuous Work History Sample). The study finds less increase in volatility than studies by Gottschalk and Moffitt (GM). Possible explanations for those differences include different age cutoffs, lack of adjustment for measurement changes by GM, and GM’s model-based decomposition into permanent and transitory variance in earnings.
The study proceeds to examine broader measures of earnings. It finds an upward trend in combined head and spouse earnings. This is reconciled with the flat trend in pooled head and spouse data by the fact the earnings of heads are higher, so the upward trend in volatility of earnings of heads of households has greater weight in head plus spouse earnings. The study finds that earnings of spouses do tend to offset changes in earnings by the household head, but the degree of offset is small. The study also finds that the volatility of household capital income has remained fairly steady over the sample period, but the volatility of transfer income has increased somewhat. Transfer income tends to offset some of the changes in market income, but the magnitude of the offset is small and trend-free.
Total household income, including earnings, transfer, and capital income, shows an upward trend in volatility over the sample period, rising by about 25%. Most of the increase appears to come in the form of increased frequency of large changes (both positive and negative). Increases observed in all three decades of the sample period.
Elmendorf concluded with a discussion of the drawbacks of the research: the study examined changes in volatility, but that does not translate into corresponding changes in risk, because an unknown amount of the volatility may be voluntary; there was no examination of changes to desired spending, such as health shocks requiring out of pocket expenditures; the study did not account for changes in saving and borrowing that could help buffer against volatility; the study covered only prime-age households; and the findings are based on a particular methodology applied to a single dataset.
* The analysis and conclusions expressed in this summary are those of the author and should not be interpreted as those of the Congressional Budget Office.