Carbon implied volatility and return volatility of EU emissions: Evidence from machine learning methods

Abstract

In contrast to recent research that has primarily focused on predicting carbon price return volatility, the present study investigates whether the information content of a newly introduced carbon implied volatility (CVIX) index improves the forecasting accuracy of realized volatility in EU carbon returns. To this end, it uses data from September 7, 2013 to December 31, 2022, and employs Machine learning (ML) based shrinkage techniques, including the least absolute shrinkage and selection operator (LASSO), elastic net (ENET), and ridge regression (RR). In the process, Mean Absolute Percentage Error and Diebold-Mariano (DM) statistics are used to evaluate the forecasting performance of LM techniques relative to traditional linear models, such as autoregressive (AR) processes. The main results indicate that the CVIX index significantly improves volatility forecasts for EU market returns. All shrinkage methods outperform both baseline and extended AR models, and these findings remain robust when forecasting both good and bad volatilities. Therefore, the information content of carbon market implied volatility should not be ignored when forecasting return volatility of EU emissions. The results have important implications for investors and risk managers aiming to construct low-carbon portfolios with EU carbon assets and improve their forecasting models.