HMM-Based Market Regime Detection with RL for Portfolio Management

This paper studies the impact of combining a Hidden Markov Model (HMM) with a Reinforcement Learning (RL) framework, implemented within FinRL, to enhance algorithmic trading. Our objective is to improve trading performance and realism by dynamically adapting to market regimes. Leveraging a portfolio of 30 Dow Jones stocks, we analyze the impact of regime awareness by comparing hybrid RL-HMM agents to standard FinRL algorithms. We observe that RL-HMM outperforms baseline models in Sharpe Ratio, annual returns, and other risk-adjusted performance metrics.