This paper proposed a new evolutionary algorithm for generating trading rules on stock market, which is called Probabilistic Genetic Network Programming (P-GNP). P-GNP represents its solutions using graph structures based on probability. It has been clarified that P-GNP works well especially in dynamic environments. In the proposed hybrid stock trading model, P-GNP is applied to generating stock trading rules using variance of fitness values and probability. The unique point is that the generalization ability of P-GNP is improved by considering the robust fitness function and the Q value of the branch obtained by Sarsa Learning. Generally speaking, the hybrid intelligent system consists of three steps, the priority selection by portfolio β, the optimization by Genetic Relation Algorithm (GRA) and stock trading by P-GNP. In the simulations, the stock trading system is trained using the stock prices of 10 brands selected from the Nikkei 500 Index, then the generalization ability is tested. From the simulation results, it is clarified that the trading rules created by the proposed P-GNP model obtain much higher profits than the traditional methods and its effectiveness has been confirmed.
展开▼
