Reinforcement Learning for Stock Exchange

Abstract

This project aims at using the Reinforcement Learning (RL) AI technique to perform stock exchange to maximize the investment. Data acquired from the Yahoo Finance, calculated as several common trading indicators, became the inputs of RL. Output actions are the Buy, Hold, and Sell accordingly for simplicity. At the machine learning phase, RL will adjust their actions continually according to their next state reward results. RL is the core framework in this project. Google’s TensorFlow, the former AlphaGo AI engine, is then be used. During the machine learning phase, I found that lots of areas, i.e., input parameters, reward strategies, discount factor, number of training loops, purchasing strategies etc., can significantly affect the reward. For the ease of analysis, I mainly consider the 388.HK stock in this project. Once successful, we can apply this AI technique to other stocks in a similar manner but with different parameters and strategies.