Forecasting U.S. Recessions with Macro Factors

Dynamic latent factors estimated from panels of macroeconomic indicators are used to predict future NBER recession dates. Three monthly macro factors are considered: (1) a bond and exchange rates factor extracted from 22 financial time series; (2) a stock market factor extracted from 4 stock market indicators; (3) a real factor extracted from 4 macroeconomic time series. Three main results emerge from this work. First, probit models that use the macro factors as predictors of NBER recessions yield better in-sample fit than models that use macroeconomic and financial variables directly. Second, while probit models based on the financial and macroeconomic variables exhibit an important deterioration in fit after 2006, probit models that use the macro factors maintain their fit throughout the sample. Finally, pseudo out-of-sample forecasting exercises designed to mimic real-time conditions show that forecasts from the probit models that use macro factors dominate forecasts from models previously considered in the literature. As a result, probit models based on macro factors appear to provide the best and most robust predictive accuracy for NBER recessions at all horizons considered.