A hybrid adaptive trading strategy integrating investor sentiment for precious metal ETFs

Abstract This study develops a novel hybrid trading framework that integrates support vector regression (SVR) for price forecasting with proximal policy optimization (PPO) reinforcement learning for adaptive decision-making. Focusing on precious metal exchange traded funds (ETFs)–GLD, SLV, PPLT and PALL–the model incorporates both market-based volatility (VIX) and news-based sentiment (SSW) indices alongside key financial variables. By dynamically switching between rule-based trading actions (informed by SVR) and a learning-based PPO agent, our hybrid strategy consistently outperforms traditional benchmarks, including buy-and-hold and 5-day moving average. Empirical results show higher cumulative returns, improved Sharpe ratios and a higher proportion of profitable trades across all four ETFs. In addition, the inclusion of investor sentiment data significantly improves forecasting accuracy, particularly for gold, silver and platinum, and strengthens the stability of both short- and long-term predictions. These findings underscore the critical role of investor sentiment in predictive modeling and highlight the effectiveness of reinforcement learning for adaptive trading in volatile commodity markets. The proposed SVR-PPO approach provides a transparent, data-driven solution for investors and portfolio managers seeking to optimize returns, manage risk and respond quickly to changing market conditions.