A Hybrid Reinforcement Learning and NLP-Based Framework to Navigate Earnings Volatility

Stock trading during earnings periods is highly volatile, with increased retail participation and potential market manipulation amplifying risks. Existing automated trading models often fail to recognize manipulation and struggle with optimal decision-making in unpredictable conditions. While reinforcement learning has been applied to financial markets, many approaches overlook sentiment-driven volatility, leading to suboptimal trades. This study addresses these gaps by developing a reinforcement learning-based framework that identifies optimal entry and exit points, while integrating sentiment analysis to detect manipulation risks. The model is trained on 20 years of earnings data, utilizing technical indicators and news sentiment analysis through a BERT-based model. The framework shows strong predictive capabilities, effectively balancing profitability and risk mitigation. By considering sentiment and manipulation factors, it enhances trading strategies and improves decisionmaking during volatile periods, offering a more robust approach to navigating market fluctuations and boosting trading performance.