Abstract:
Using monthly data from 1871 to 2024 and logistic models with shrinkage estimators, we compare the contribution of stock and oil-market moments (returns, volatility, skewness, and kurtosis) to the accuracy of out-of-sample forecasts of U.S. recessions at various forecast horizons, while controlling for standard macroeconomic predictors and the total connectedness indexes of the moments. Adding stock-market moments to the potential predictors improves significantly the accuracy of out-of-sample forecasts at an intermediate forecast horizon, where the lagged recession dummy, term spread, and stock returns are top predictors. Oil-market moments and connectedness indexes do not contribute much to forecast accuracy.
