This paper takes the accuracy of predictability of excess return into consideration to optimize the portfolio decision model. By using Bayesian Model Averaging, the accuracy of predictability of excess return is improved. The model is verified by the data from the stock market in China. The results indicate that: (i) the active portfolio structured by this decision model outperforms the benchmark portfolio; (ii) it is necessary to include the accuracy of predictability of the excess return in the decision model which enhances the returnrisk profile; (iii) the model works in the market with short selling with the same excess return level and more active risk. The excess return still exists when transaction cost is under consideration.