Stock trading system using fuzzy candlesticks and reinforcement learning
Lee, Wen Chong
Date of Issue2018-12-03
School of Computer Science and Engineering
One commonly used technical analysis is the candlestick charts. By studying historical stock data in candlestick charts, experts hypothesize and propose patterns that can predict price trends ahead. Inspired by this methodology, fuzzy logic is generally used to model raw stock data into fuzzy candlesticks, providing autonomous predictions. Most literature that used this approach tries to model existing patterns established by experts. The objective of this research is to discover candlestick patterns and propose a trading system that takes advantage of these patterns. Firstly, the necessity of expert knowledge is circumvented by discovering candlestick patterns using genetic algorithm. A trading system that incorporates the top performing patterns is then developed and used to evaluate their competence. Additionally, an experiment is conducted to determine the potential of using fuzzy candlesticks and the discovered patterns in a reinforcement learning technique (Double Deep Q-Network). The results of the experiments show promise for this novel approach. The discovered patterns have an accuracy rate of approximately 70 – 80%. Furthermore, the trading system is found to do remarkably better when trading with multiple stocks. With the proposed trading systems, the performance of trading with 28 stocks from the S&P 500 index outdoes the average return rate.
DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
Final Year Project (FYP)
Nanyang Technological University