Abstract:

[Objective] Stock trading represents a critical topic within the field of finance and has garnered significant attention for the integration of reinforcement learning models into trading algorithms. [Coverage] This paper collects literature related to Reinforcement Learning algorithms designed for stock trading in recent years. [Methods] This paper systematically outlines the core challenges of intelligent computing in stock market modeling, discusses the limitations of traditional statistical methods and machine learning models in the stock domain, and analyzes state-of-the-art stock reinforcement learning algorithms. By comparing the performance of three representative algorithms, this study conducts an empirical evaluation from three dimensions: return stability, risk control, and computational efficiency. Based on the experimental results, the paper thoroughly examines the strengths and weaknesses of each algorithm. Building on these findings, it proposes future research directions for this field, providing theoretical foundations and practical references for further studies. [Conclusions] Although the integration of stock reinforcement learning algorithms with models such as LSTM and NLP has shown promising performance in backtesting, the increased model complexity has significantly raised computational time costs, reducing their efficiency in real-time analysis scenarios. Future research needs to advance both efficient algorithm design and hardware adaptation optimization to address current limitations.

Key words: reinforcement learning, stock trading, high-performance computing