Natural gas volatility prediction via a novel combination of GARCH-MIDAS and one-class SVM

Abstract

Research has focused on whether information spillovers from external influences play a role in clean energy–natural gas volatility forecasts. However, the climate and energy crises caused by the intensification of extreme events, such as recent extreme weather and geopolitical risks, have led the public to turn their attention to research in the field of clean energy. Therefore, this paper uses one-class SVM (support vector machine) techniques to identify extreme volatility in natural gas prices induced by significant occurrences (e.g., wars, financial crises, and COVID-19) and then investigates whether considering extreme volatility in natural gas over different volatile periods (short- and long-term periods) improves volatility forecasting accuracy within the context of a GARCH-MIDAS framework. The in-sample analyses demonstrate that extreme shocks increase natural gas price volatility and that the asymmetric effects are more influential than the short- and long-term extreme volatility effects. The out-of-sample results indicate that the GJR-GARCH-MIDAS-one-class-SVM-SLES model outperforms the other models and achieves the best forecasting performance of the remaining extended models. In addition, robustness tests confirm these findings.