Uncertainty, Volatility, and Renewable Energy Stock Returns: Evidence From Regime‐Dependent and Machine‐Learning Forecasting Models

This study examines the predictability of renewable energy stock returns under economic, financial, and policy uncertainty within a real‐time forecasting framework. Using monthly data from May 2007 to May 2025, we compare conventional econometric models, shrinkage estimators, machine‐learning algorithms, and regime‐dependent forecasting models using a recursive expanding‐window design. Forecast performance is evaluated using statistical accuracy measures, robustness analyses, and an economic portfolio evaluation. The results show that nonlinear forecasting methods consistently outperform conventional linear benchmarks. Gradient boosting achieves the lowest forecast errors, while principal component regression provides stable performance across alternative evaluation designs. Regime‐dependent forecasting models improve directional prediction during periods of heightened financial uncertainty but generate more modest gains in forecast accuracy. Predictor‐importance analysis identifies financial‐market volatility as the dominant source of forecasting information, followed by oil‐market volatility and broader policy uncertainty. Overall, the findings demonstrate that combining multiple sources of uncertainty with flexible forecasting methods improves both statistical forecasting performance and the economic value of investment decisions.