We assess the role of common factors in the prediction of downside risk for a large panel of US macroeconomic time series obtained from the FRED-MD database. The panel exhibits pervasive cross-sectional tail dependence, with the joint occurrence of extreme events being power-law distributed. The pool of candidate factors we consider includes macroeconomic, financial, text-based and statistical factors. Results show that the downside risk is captured by one factor that can be associated with macroeconomic volatility and appears to be best captured by the macroeconomic uncertainty index of Jurado et al. (2015).