Single stock trading with deep reinforcement learning: A comparative study

In this paper, we apply Deep Reinforcement Learning (DRL) methods to automate the trading of single stock. The A2C, PPO, DDPG, TD3 and SAC deep reinforcement learning models are built and studied comparatively. Shanghai Composite Index (SH00001) is used as the trading stock, where the stock data before the Covid-19 is used as the training set, and the data after the Covid-19 is used as the testing (trading) set to back-test the performance of these models. Experimental results show that the DDPG, TD3, and SAC models outperform the benchmark, among which the DDPG model shows the most obvious advantages in returns and risk control, achieving a cumulative return rate of 25%, while the TD3 and SAC models achieve a cumulative return rate of 16-17%. The A2C and PPO models have inferior performance comparing to the benchmark.