The realized volatilities of China’s financial futures is forecasted by constructing a Bayesian factor augmented heterogeneous autoregressive model ( DMA( DMS) -FAHAＲ) with time-varying parameters and sto-chastic volatility． The Bayesian inference is employed to obtain the latent factors of the daily，weekly，and monthly predictor sets including the lagged volatility variables，jump variables，and signed jump variables． Speculation variables are used to investigate the impact of speculation activities on the volatility forecast． The results suggest that the Bayesian factor augmented HAＲ model performs best for short-term，mid-term，and long-term forecasts among all candidate forecast models． Meanwhile，the time-varying Bayesian HAＲ models have superior forecast performances compared with the fixed parameter HAＲ models． In addition，better fore-cast performances are achieved after incorporating the speculation variables into the forecast models for both the stock index futures and the Treasury futures．