Emily Eisner, UC Berkeley, Economics Department
Abstract: We show that the standard implementation of X13-ARIMA yields predictable revisions to initial releases of employment growth. At the onset of severe recessions, initial growth rates tend to be revised downward. This causes the severity of recessions to be underestimated in real-time-- a major concern for policy-makers. The problem arises from the combination of concurrent seasonal adjustment, and the standard forecasting procedure used as part of the seasonal adjustment algorithm, when applied to abrupt movements such as those that occur at the onset of a recession. This problem was avoided in the COVID crisis only due to the triggering of outlier detection clauses, which counteracted this effect.