Reinforcement learning (RL) based stock trading system via support vector machine

• Main Author: Ong, Zhi Yuan.
• Other Authors: Quek Hiok Chai
• Format: Final Year Project
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Computer science and engineering
• Online Access: http://hdl.handle.net/10356/20775
• Institution: Nanyang Technological University

The stocks market is one of the widely traded financial instruments. During the recent economic crisis, lots of investors suffer lost in their investment as stocks prices fell to new low. This study uses algorithms to trade and had produced promising results despite the current market condition. Support vector machines (SVMs) have produces promising results in various applications such as text categorization, hand-written character recognition, image classification and time series prediction. This study applies SVM to predict Singapore Stocks pricing. In addition, this study compares the performance of SVM with other financial forecasting tool such as back-propagation neural networks. Reinforcement Learning (RL) is a computational approach to automate goal-directed learning and decision making in agent-based systems and has been successfully applied to problems solving. Moody et al [3] have used Direct RL model for stock trading and it is concluded that model-based RL approaches find better policies efficiently. In this study, we propose a RL based SVM approach in stock prediction. Firstly, the system use Q-Learning to estimate the optimum input dimension for the SVM stock price predictor based on different states in the time series. This provides a more flexible approach in adjusting the input dimension catered for the changing market condition to produce better results. Finally, we make use of the above RL based SVM price predictor with trading rules that are tuned via Reinforcement Learning to trade over a 5 years period. It is found to generate significant profits as compared to other benchmark systems despite the recent economic crisis the world face.