Learning Financial Investment Strategies using Reinforcement Learning and 'Chan theory'

Abstract

We explored the potential of the Double DQN (Deep Q-learning) framework to trade on the stock market based on the S&P500 and Chinese Future market. In our work, human crafted features were used to capture the trends from the complex unstable, and dynamic financial time series. The feature engineering is based on `Chan' theory, which is a very popular trading system in Chinese financial market. Combined with the Double DQN's capability for searching policies, our experiment's results show the Double DQN model has competitive performance, which also shows the 'Chan theory' is a successful technology analysis methodology. In this research, we solved two problems: 1. Firstly introduce 'Chan Theory', a financial time series analysis method based on its past configurations, to formal paper. 2. The first time to combine the 'Chan theory' with machine learning models Keywords: Double-DQN; Chan theory; financial time series; feature engineering;