A novel approach to dynamic portfolio trading system using multitree genetic programming

Abstract

Dynamic portfolio trading system is used to allocate one’s capital to a number of securities through time in a way to maximize the portfolio return and to minimize the portfolio risk. Genetic programming (GP) as an artificial intelligence technique has been used successfully in the financial field, especially for the forecasting tasks in the financial markets. In this paper, GP is used to develop a dynamic portfolio trading system to capture dynamics of stock market prices through time. The proposed approach takes an integrated view on multiple stocks when the GP evolves and generates a rule base for dynamic portfolio trading based on the technical indices. In the present research, a multitree GP forest has been developed to extend the GP structure to extract multiple trading rules from historical data. Furthermore, the consequent part of each trading rule includes a function rather than a constant value. Besides, the transaction cost of trading which plays an important role in the profitability of a dynamic portfolio trading system is taken into account. This model was used to develop dynamic portfolio trading systems. The profitability of the model was examined for both the emerging and the mature markets. The numerical results show that the proposed model significantly outperforms other traditional models of dynamic and static portfolio selection in terms of the portfolio return and risk adjusted return.