The Role of Coincident Information in Real-Time Business Cycle Forecasting

Abstract

Official NBER recession dates are announced with substantial delay. Therefore, real-time forecasters cannot condition on the most recent business cycle states even though recessions and expansions are highly persistent. I study whether real-time coincident releases can substitute for this missing information. At each monthly forecast origin, I construct a recession nowcast, using four coincident indicators and several supervised classifiers, and add this nowcast probability to standard probit forecasting models. In an out-of-sample evaluation for US monthly data from 1986 to 2021, nowcast augmentation improves forecast accuracy at short horizons and at the 1-year horizon relative to a term spread benchmark, while including the raw coincident indicators directly is less effective. The gains are incremental once strong leading indicators are included, and model rankings are sensitive to resampling variation.