Income Risk in Recessions

This paper studies how recessions affect individual income risk. We employ a unique and confidential administrative data set with tens of millions of observations on individual earnings histories from the Social Security Administration records. We use a dataset that is a 10% random sample of the US population and covers the period from 1978 until 2010, encompassing the Great Recession of 2007--2009. We use these data to measure the effects of the recession on workers with different backgrounds. Our first set of findings concerns the cyclical nature of idiosyncratic shocks. We find that income shock variances are not countercyclical. Instead, it is the left skewness of shocks that is countercyclical. That is, during recessions, the upper end of the shock distribution collapses---i.e., large upward wage movements become less likely---whereas the bottom end expands---i.e., large drops in incomes become more likely. Moreover, the center of the shock distribution is very stable and moves very little compared to either tail. Second, we examine the systematic component of business cycle risk. We study three separate dimensions: whether the income loss during recession depends on (i) one's ranking in the lifetime income distribution, (ii) one's inherent sensitivity to business cycles as measured by the correlation of one's earnings with the cycle (excluding the recession under study), and (iii) one's income growth right before the recession. We find all three to matter for the fortunes of workers during recessions. In fact, for deep recessions (e.g., 1980--83 and 2008--2010), most of the rise in income inequality is due to this “factor structure†---i.e., the systematic divergence of incomes between groups of workers that are observationally different before the recession. To give one example, during the Great Recession, workers in the 20th percentile of the lifetime income distribution, on average, experienced a decline in their income that was nearly 10% more than the decline for workers who were in the 90th percentile of the distribution.