A Robust Stablizing Transformer with Deep Reinforcement Learning for Risk-Adjusted Equity Trading Strategies

With the development of financial markets, an increasing number of financial practitioners are engaging in stock trading activities. This trend not only enhances market vitality but also increases the uncertainty associated with portfolio risks faced by financial practitioners. This paper proposes a novel framework for risk-adjusted automated stock trading, based on the Stable Transformer model and the Proximal Policy Optimization (PPO) algorithm, to address challenges such as portfolio risks. First, a reinforcement learning environment for stock trading is constructed using historical stock trading data. Subsequently, an intelligent stock trading agent is designed using the Stable Transformer model with a shared feature extractor. Modifications are made to the original PPO algorithm to improve the training efficiency and stability of the agent, leveraging the network characteristics of the Stable Transformer. Experimental results demonstrate that this trading strategy surpasses other baseline models in its ability to mitigate portfolio risks in stock markets. Additionally, this trading strategy exhibits significant profitability, expanding the research frontiers of financial risk mitigation strategies.