结合趋势的深度强化学习股票交易策略

Deep Reinforcement Learning Stock Trading Strategies Combining Trends

摘要: 机器学习广泛应用于股票交易决策中。如何在交易过程中获得有效的市场信息，实现利益最大化和风险最小化，是一个值得长期研究的话题。基于深度强化学习的传统交易模型无法提前识别剧烈的股价波动，导致投资收益不稳定。本文提出了一种结合趋势的深度强化学习股票交易模型，选取根据趋势指标RSI指数调整后特定条件下的利润作为奖励函数，模型能有效识别股价波动风险，获得稳定收益增长。实验选取中国股市的3只股票进行模拟交易，与对照组相比，本文结合趋势的深度强化学习模型训练良好，在实验期间的平均年回报更高，年波动率更低，且夏普比率更好。通过实验数据验证了模型的稳定性和有效性。

Abstract: Machine learning is widely used in stock trading. How to obtain effective market information and maximize benefits and minimize risks in the process of stock trading is a topic worthy of long-term research. The traditional trading model based on deep reinforcement learning can’t identify the violent stock price fluctuations effectively, resulting in the instability of investment return. In this paper, we propose a deep reinforcement learning model incorporating trends in stock trading strategies. The reward function selects the profit under specific conditions adjusted according to the trend indicator RSI index. It can identify the risk of stock price fluctuation effectively and obtain stable growth of return. We evaluate our method with 3 stocks in the Chinese stock market. Compared with the control group, the proposed model is well trained and achieves higher average annual returns, lower annual volatility, and better Sharpe ratio during the experimental period. The experiments results demonstrate the stability and validity of the model.