Enhancing support vector machine model for stock trading using optimization techniques

Abstract

This study aims to optimize the stock market forecast model using the Support Vector Machine. Two types of SVM optimization techniques were tested, Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), in order to find which optimization technique provides the most appropriate parameters for the SVM Dot Kernel and results in improved stock market forecast efficiency. The experiment uses data from five stocks from the Stock Exchange of Thailand, which are tested and compared with results from before and after application of the optimization techniques. In addition, the SVM model, which was enhanced by the optimization technique, was backtested for trading in order to analyze the return on investment by comparing trading results with technical analysis techniques that are popular with general investors, such as Relative Strength Index (RSI), Stochastic Oscillator (STO), and Moving Average Convergence/Divergence (MACD). The results showed that genetic algorithm can improve the SVM Model for increased forecasting accuracy, and the average return results were higher than the compared results from the technical analysis indicators. In addition, this study also analyzes the SVM model trading performance based on price trends, which include uptrend, downtrend and sideways trend. This concludes that SVM with Dot Kernel optimized by GA is able to provide effective forecasting for all trends, with the best performance during the uptrend.