The Importance of Correct Model Specification : A Regime Switching GARCH MIDAS Approach with Application to Oil Volatility

Events such as pandemic, changes in government policies and wars result in structural breaks in many areas including oil markets. RS GARCH MIDAS models, which consider both structural changes and macroeconomic factors affecting the oil prices have been studied by very few authors where they assumed innovations are normally distributed. In this article, we consider different error distributions to analyse how effective they are in capturing the characteristics of oil returns compared to Normal innovations. In a Monte Carlo simulation, we investigate how model misspecification affects the estimation results and find that misspecified models have greater bias, overestimation in the long-term component and problems with identification of two volatility regimes. The results obtained in simulation are also confirmed in an empirical application to WTI crude oil returns. Finally, the forecast performance of RS GARCH MIDAS-t model is compared with various competitor models. Considering models with long-term component we find that RS GARCH MIDAS-t with realised volatility achieves the lowest MSE and QLIKE, thus indicating that, in our case, production and demand do not provide useful information regarding oil volatility