Natural gas volatility predictability in a data-rich world

Abstract

This study employs macroeconomic variables and economic indices to forecast natural gas volatility. The out-of-sample results show that the forecasting performance of the macroeconomic variables outperforms the economic indices. Additionally, the forecasting performance of the mixed data sampling model, which combines the least absolute contraction and the selection operator (MIDAS-LASSO), is better than that of other competing models, and it still has a good predictive ability under certain conditions (e.g., business cycles). Our study confirms the superiority of the MIDAS-LASSO model for natural gas volatility forecasting.