Predicting renewable energy stock volatility: A GARCH-CNN approach with indicator analysis

The renewable energy (RE) stock market is experiencing rapid growth, driven by environmentally conscious investors seeking to support a greener future while pursuing profitable opportunities. This study aims to forecast RE stock volatility, which is critical for managing risks in the RE stock market. We employ a Convolutional Neural Network (CNN) model combined with Generalized Autoregressive Conditional Heteroscedasticity (GARCH) forecasts to predict RE stock volatility. Three groups of indicators – internal stock, financial market, and policy uncertainty – are incorporated as additional inputs. The results demonstrate that integrating internal stock and financial market indicators significantly reduces prediction errors compared to the traditional GARCH model. Conversely, models incorporating the policy uncertainty indicator produce higher errors, suggesting that these indicators may introduce noise. SHapely Additive exPlanations (SHAP) analysis identifies the internal stock indicator, particularly the squared log returns of RE stocks, as a dominant factor, with the financial market indicator serving as a complementary factor. By integrating deep learning with econometric models, this study enhances the prediction of RE stock volatility and underscores the importance of selecting appropriate indicators. The findings provide valuable insights for investors and policymakers seeking to better understand and manage RE investment risks, highlighting the key drivers of RE stock volatility.