Probability and Severity of Recessions

This paper tackles the prediction of the probability and severity of US recessions. We employ parsimonious Probit models to estimate the probability of a recession h periods ahead, for h varying between 1 and 8 quarters. A novel goodness-of-fit measure derived from the Kullback-Leibler Information Criterion is developed and used to select the regressors to include in the Probit models. Next, an autoregression (AR) augmented with inverse Mills ratio (IMR) and diffusion indices (DI) is fitted to selected measures of real economic activity. The resulting “IMR-DI-AR” model is used to generate forecasts conditional on optimistic and pessimistic scenarios for the horizon of interest. The severity of recessions is defined as the gap between the pessimistic scenario and the recent trend of the series. For a time series of GDP growth, our measure of recession severity has the interpretation of the output loss. Our results support that U.S. recessions are predictable to a great extent, both in terms of occurrence and severity. All recessions are not alike: some are more predictable than others while some are more severe than expected.