A Novel Deep Reinforcement Learning-based Automatic Stock Trading Method and a Case Study

Abstract

As the most important capital market, how formulating a reasonable stock trading strategy to improve capital return and reduce trading risk has always been the focus of people’s attention. In recent years, with the development of data-driven AI technology, people have begun to try to apply it to stock market data analysis to minimize the trading risk caused by the uncertainty of price fluctuations in the stock market. Accordingly, in this paper, we solve the complex stock trading decision-making problem by exploiting the deep reinforcement learning techniques. Briefly, in this paper, we consider two trading models: 1) for trading a specific single blue chip stock, we design a new trading agent based on Double Deep Q-Network (DDQN) to maximize the return for such specific blue chip stock purchase and sale; 2) To further reduce the risk of stock trading, for the more common multi-stock trading scenarios, we utilize twin-delayed deep deterministic policy gradient (TD3) technique to design a multi-stock collaborative trading agent for achieving the goals of risk hedging and maximizing returns. We further evaluated the efficiency of the proposed trading agents in stock price prediction accuracy and returns based on the actual U.S. Stock Market data.