We extend the density combination approach of Billio et al. (2013) to feature combination weights that depend on the past forecasting performance of the individual models entering the combination through a utility-based ...