We reassess the predictability of U.S. recessions at horizons from three months to two years ahead for a large number of previously proposed leading-indicator variables. We employ an efficient probit estimator for partially missing data and assess relative model performance based on the receiver...