Oil volatility forecasting and risk allocation: evidence from an extended mixed-frequency volatility model

This paper proposes a novel GARCH-MIDAS-SK (G-M-SK) model that improves the basic GARCH-MIDAS (G-M) model by incorporating time-varying skewness and kurtosis. We employ our model with data concerning macroeconomic fundamentals to investigate in-sample fit and out-of-sample prediction of volatility in crude oil prices. Empirical results suggest that G-M-SK models produce the better in-sample fit than basic G-M models do. This result is also robust for the subsample of the oil market. It is equally noteworthy that modelling low-frequency macroeconomic variables better reveals the long-term volatility from time-varying skewness and kurtosis. More importantly, G-M-SK model significantly and robustly improves accuracy in predicting oil volatility. In particular, we find that data regarding macroeconomic fundamentals contribute more to forecasting volatility in oil prices than their variance does. Finally, our G-M-SK model more precisely calculates the utility incident to minimizing risk and allocating portfolios. Its results are consistent with out-of-sample forecasting results. These results are beneficial to the decision-making of the crude oil investors and policymakers alike.