Research on Stock Trading Algorithm Based on Multi-Objective Deep Reinforcement Learning with Multiple Rewards

In this paper, we propose a stock trading method based on deep reinforcement learning with multiple rewards for the multi-objective problem in quantitative stock trading. The method uses three reward functions, which are the rate of change of profits (Returns), Average Hindsight Returns (AHReturns) and Sharpe Ratio (SR). However, the weight assignment of these three rewards is a more difficult problem. In this paper, we design an analytic hierarchy process(AHP) based on experimental data to calculate the weight of each reward. Then, multiple reward functions are synthesized into a single reward function through weights for multi-objective deep reinforcement learning. Finally, we conducted several comparative experiments to compare the performance of the method proposed in this paper with other methods, such as single reward methods and random weight methods, and the experimental results show that the proposed method is effective.