The Concentration of Job Destruction
A time series is concentrated if the expectation of its current value is a negative function of a moving average of past values up to all but the most recent past. Job destruction has the property of concentration in a model of heterogeneous jobs because an adverse shock destroys jobs in plants close to the margin of shutdown. Until other plants drift close to that margin, there are fewer plants that are vulnerable to another adverse shock. Concentration is easy to spot in the autocorrelations of a time series, which will be negative except for the first few lags. A simple model generates data displaying concentration. Data on job destruction and employment change for U.S. manufacturing show unambiguous evidence of concentration. According to the simple model, job creation is more persistent and thus less concentrated than is destruction, a property reflected in the data as well.