We analyze quarterly occupation-level data from the US Current Population Survey for 1976-2013. Based on common cyclical employment dynamics, we identify two clusters of occupations that roughly correspond to the widely discussed notion of "routine" and "non-routine" jobs. After decomposing the cyclical dynamics into a cluster-specific ("structural") and an occupationspecific ("idiosyncratic") component, we detect significant structural breaks in the systematic dynamics of both clusters around 1990. We show that, absent these breaks, employment in the three "jobless recoveries" since 1990 would have recovered significantly more strongly than observed in the data, even after controlling for observed idiosyncratic shocks.
employment polarization jobless recoveries factor model