How important are climate change risks for predicting clean energy stock prices? Evidence from machine learning predictive modeling and interpretation

Abstract

The clean energy equity sector plays an important role in the transition to a low-carbon economy. This paper explores the role of climate change risks in predicting the direction of clean energy stock prices (solar, wind, nuclear). We employ machine learning models, including random forests, boosting, extremely randomized trees, and support vector machines, to make our predictions. Variable importance is determined using Shapley/SHAP values. Notably, tree-based ensemble and boosting models show an accuracy exceeding 85 % for the 10 day to 20 day forecast period. For the stock prices of solar, wind, and nuclear energy, inflation expectations and technical indicators (which account for behavioral factors) such as on-balance volume and Williams’ accumulation/distribution are important features within this forecast range. For wind and solar energy stocks moving averages are also important additional features while for nuclear energy stocks economic policy uncertainty and stock market volatility are additional important features. In the five day to twenty day forecast horizon, climate change risks are not important features. These results align with a body of literature that raises concerns about equity prices not fully reflecting climate change risks. An equally weighted portfolio of wind, solar, and nuclear energy stock prices that used trading signals from an Extra Trees prediction model outperformed a buy and hold portfolio in terms of risk adjusted returns. These results are robust to trading costs and weekly or monthly portfolio rebalancing.