A K-line pattern combinations stock return prediction method using deep deterministic policy gradient

This paper studies the stock return prediction under specific K-line pattern combinations in the domestic Chinese A-share market. Firstly, derivative factors such as the upper (lower) shadow rate are designed according to the basic stock information and trader psychology. Secondly, the Deep Deterministic Policy Gradient algorithm, reconstructed to adapt to the stock trading market, is applied to build the reinforcement learning framework. Numerical comparison experiments show that the factor combination based on the K-line pattern can obtain higher profit and lower risk than other technical factor combinations. Our newly designed technical factors enable the agent to achieve a substantial amount of abnormal return. Notably, the linearly correlated derived factors do not significantly influence the agent’s decision-making process. Furthermore, a more diverse set of factors with varying significance in the state space led to increased abnormal return for the agent’s decisions.