Uncertainty and Forecasts of U.S. Recessions

We use a Boosted Regression Trees (BRT) approach to study the potentially nonlinear link between various standard predictors (stock-market returns, term spread, a short-term interest rate, among others), components of a news-based uncertainty index, and U.S recessions. The BRT approach shows that, according to a relative-importance measure, war-related uncertainty is among the top five predictors of recessions at three different forecast horizons. In the second half of the 20th century, uncertainty regarding the state of securities markets has gained in relative importance. Partial-dependence curves show that the probability of a recession is nonlinearly linked to war-related and securities-markets uncertainty. An analysis based on receiver-operating-characteristic (ROC) curves shows that including war-related uncertainty in the list of predictors improves out-of-sample forecasting performance at a longer-term forecasting horizon, where the predictive value of this component relative to other components of uncertainty has fallen in the second half of the 20th century. Estimation results for a dynamic version of the BRT approach recover the relative importance of various lags of government-related uncertainty for recession forecasting at a longer forecast horizon.