Forecasting volatility of EUA futures: New evidence

Abstract

In this paper, we judge the predictability of EUA's own short- and long-term asymmetry, extreme observations, as well as jump components on its volatility by comparing the GARCH mixed frequency data sampling model and its asymmetry, extreme observation and jump extensions. The in-sample estimation results show that both long- and short-term asymmetries, extreme observations, and jump information have substantial effect on the EUA volatility. The out-of-sample forecast assessment results further illustrate the predictability of these volatility components on EUA volatility. Specially, among all asymmetric extension models, the model extended by both short-term asymmetry, long-term asymmetry and long-term leverage has better predictive performance. Among all extreme observation extension models, the model extended by only short-term extreme observations has better predictive performance. Among all jump extension models, the model extended by only the short-term jump information and the model extended by both the short-term and long-term jump information perform better, and their forecasting performance outperform all the other extension models in most cases. These findings are robust even if the assessment method, the rolling window length and the lag order are changed. It is worth mentioning that the advantage of these extension models in predicting EUA volatility is mainly seen in periods of low volatility. However, even during periods of COVID-19 pandemic, the predictive performance of the two well-performing jump extension models cannot be underestimated.