Abstract
The development of computational intelligence based trading strategies for financial markets has been the focus of research over the last few years. To develop efficient and effective automated trading strategies, we need to understand the workings of the market and the patterns emerging as a result of the traders interactions. In this paper, we develop an adaptive Genetic Programming (GP) agent-based trading system under Intraday Seasonality Model (ISM), which is abbreviated as GP-ISM trading system. ISM is used for creating maps and visualizing the dynamic price evolution of the asset during the day. This new model permits the recognition of periodic patterns and seasonalities in the price time series and hence eliminates any unnecessary data input. We use a high-frequency dataset of historical price data from Saudi Stock Market, which enables us to run multiple market simulation runs and draw comparisons and conclusions for the developed trading strategies. The goal of our work is to develop automated computational intelligence-based strategies for real markets, and this study facilitates a more thorough understanding of a specific market's workings and constitutes the basis for further exploration into such strategies designed for the stock market. We evaluate the intelligence of the GP-ISM trading system through agent-based simulation market index trading. For comparison, we also include four other types of trading agents in this contest, namely, zero-intelligence agents, Buy-and-Hold agents, fundamental agents and technical analysis agents. As a result, GP-ISM performs the best, which provides a general framework for the further development of automated trading strategies and decision support systems.