Utilizing fuzzy reinforcement learning for prediction in volatile stock markets

Nowadays, the stock market plays an important role in the global economy, underscoring the need for accurate stock price prediction. However, forecasting stock prices remains challenging due to the complex, volatile, and uncertain nature of financial data. Despite recent advances in artificial intelligence, machine learning, or reinforcement learning, stock price prediction accuracy has still not improved substantially. Moreover, reinforcement learning approaches are less frequently applied because of the complexity associated with large state and action spaces. In this study, we show that fuzzy least-squares policy iteration (FLSPI), a fuzzy reinforcement learning technique, performs effectively for stock price prediction in high-dimensional and uncertain environments. By combining least-squares policy iteration with fuzzy systems, FLSPI is able to handle stock price volatility and uncertainty more robustly. We evaluate FLSPI against classical and modern baselines, including autoregressive integrated moving average (ARIMA), Long Short-Term Memory (LSTM), or the Chronos-Bolt-Tiny foundation model in both zero-shot and few-shot settings. Experimental results across multiple stocks and prediction horizons demonstrate that FLSPI provides robust and competitive predictive performance, particularly in the multivariate setting. Beyond prediction accuracy, directional and transaction-cost-aware financial metrics indicate that FLSPI achieves a balanced trade-off between accuracy, stability, and risk. Overall, these findings suggest that FLSPI represents a reliable alternative for stock price forecasting and extends the set of tools available for financial prediction.