Ghysels, Horan, and Moench (2017) show that extracting principal component (PC) factors from real time as opposed to revised macro variables substantially reduces their power in forecasting bond excess returns. In this paper, we propose a predictive principal component (PPC) approach to extract factors from information pertaining to expected bond excess returns contained in real time macro variables. In so doing, the new PPC factors remove common noises in real time data and exhibit significant bond return predictability. The inand out-of-sample R2s improve by more than 50% relative to the PC factors. Moreover, the forecasted bond excess returns are countercyclical, consistent with standard asset pricing models.