GNP-Sarsa with subroutines for trading rules on stock markets

The purpose of this paper is to study how to improve the evolution of GNP-Sarsa with subroutines and its application to trading rules on stock markets. Recently, a successful study, namely GNP-Sarsa, shows us its effectiveness and powerfulness, which combines sophisticated diversified search ability for structures using evolution and intensified search ability of RL for many technical indices and candlestick charts. But, another advantage of GNP, the compact structure becomes weak by GNP-Sarsa, and then the choice of trade-off between efficiency and compactness is difficult. So, we intend to solve this problem by extending the basic structure of GNP. A new method named GNP with subroutines has been proposed, which adds new kind of nodes named subroutine nodes to main GNP to call a corresponding subprogram (subroutine). This reusable subroutine works like small scale GNP and has also judgment nodes and processing nodes like GNP. This paper introduces the concurrent evolution of GNP and subroutine. In the simulations, the stock prices of different brands from 2001 to 2004 are used to test the effectiveness. The results show that the proposed approach can provide reasonable opportunities for complex solutions to evolve.