What predicts US recessions?

We reassess the in- and out-of-sample predictability of US recessions at horizons of three months to two years ahead for a large number of previously proposed leading indicator variables, using the Treasury term spread as a benchmark. We estimate both univariate and multivariate probit models, and evaluate the relative model performance based on the receiver operating characteristic (ROC) curve. At the three- and six-month-ahead horizons, various alternative predictor variables increase the accuracy of recession forecasts significantly relative to the term spread, with the annual return on the S&P500 index providing the strongest improvement. While the Treasury term spread is more difficult to outperform systematically at longer horizons, manufacturers' new orders of capital goods and balances in Broker-Dealer margin accounts increase the precision of recession predictions significantly at horizons of more than one year.