In this paper we examine the out-of-sample forecast performance of high-yield credit spreads regarding real-time and revised data on employment and industrial production in the US. We evaluate models using both a point forecast and a probability forecast exercise. Our main findings suggest the use of few factors obtained by pooling information from a number of sector-specific high-yield credit spreads. This can be justified by observing that, especially for employment, there is a gain from using a principal components model fitted to high-yield credit spreads compared to the prediction produced by benchmarks, such as an AR, and ARDL models that use either the term spread or the aggregate high-yield spread as exogenous regressor. Moreover, forecasts based on real-time data are generally comparable to forecasts based on revised data. JEL Classification: C22; C53; E32 Keywords: Credit spreads; Principal components; Forecasting; Real-time data.