The study examine how the evidence of predictability in asset returns affects optimal portfolio choice for investors with long horizon . According to Bayesian, we estimate uncertainty about the true values of VAR(Vector Autoregression) model and decide optimal portfolio allocation with the max expected utility of investor. The empirical results show that the weight of optimal portfolio can be very different from short-horizon and long-horizon. Considering estimation risk and predictor makes investors allocate substantially more to stock with the long horizons. Finally, we calculate the VaR (Value at Risk) to analyze the different portfolio.