Does Mixed Frequency Information Help To Forecast the Value at Risk of the Crude Oil Market?

We test the value at risk (VaR) forecasting accuracy of seven generalised autoregressive condition heteroskedasticity (GARCH)-mixed data sampling (MIDAS) models, which potentially provide superior forecast accuracy than traditional GARCH models by capturing different forms of mixed frequency information from the market. The main empirical results are as follows. First, most traditional GARCH models have difficulties forecasting the VaR of the crude oil market. Second, although GARCH-MIDAS models generally produce more accurate forecasts than the traditional GARCH models, some specific GARCH-MIDAS models have poor forecasting accuracies. Third, we find that the mixed frequency information on the demand side of the crude oil market is most helpful for forecasting the VaR. The model that integrates the world industrial production index (GARCH-MIDAS-IP) robustly demonstrates good forecasting performance