Stock trading rule discovery with double deep Q-network

Abstract

Stock market serves as an important indicator of today’s economy. Predicting the price fluctuation of stocks and acquiring the maximum gains has been the main concern of investors. In recent years, deep learning models are widely applied to stock market prediction and have achieved good performances. However, the majority of these deep learning based models belong to supervised learning methods and are not capable of dealing with long-term targets. Therefore, in this paper we proposed a deep reinforcement learning based stock market trading model, which is suitable for predicting stock price fluctuation and stock transactions. We carefully devise the reward function and deep learning based policy network, which enables the model to capture the hidden dependencies and latent dynamics in the stock data. In order to evaluate the superiority of the proposed model, stock price trend forecasting and transaction is conducted on randomly selected stocks and stock market indices. Experiment results demonstrate that our model outperforms baseline methods on several indicators.