Modelling and forecasting oil market volatility: A regime switching GARCH MIDAS approach

Tirkishova, Menli

Abstract

In this thesis, we consider a regime switching GARCH MIDAS model with Student-t innovations. By allowing the error term to be non-Gaussian, we want to see how effective it is in describing the volatility displayed in financial time series. For the long-term volatility component, as natural explanatory variables we also consider realised volatility (RV) calculated as absolute returns. This extension is particularly important since regime switching GARCH MIDAS models have been partially implemented by very few authors where they assumed innovations are normally distributed and calculated RV as squared returns.
In addition, by Monte Carlo simulation and a real data application we provide evidence to support our proposed model setting RS GARCH MIDAS-t with RV. We consider the misspecification in terms of; (a) not considering regime switching, (b) misspecifying the error term, (c) omitting the long-term volatility component, (d) all three combined. The simulation results confirms the importance of correctly specifying volatility models. We show that when models are misspecified, the bias of parameter estimates increase while the ability of regime switching process to correctly identify regimes deteriorate quite considerably. Moreover, the long-term volatility in misspecified models leads to an overestimation.
The validity of our proposed model is then explored through a real data application. Empirical analysis of West Texas Intermediate crude oil returns show that regime switching models outperform single-regime models. Specifications with Student-t innovations are superior to their Gaussian counterparts in terms of higher log-likelihood value, lower model selection criteria, and ability to better identify the volatility regimes. This provides strong within-sample estimation evidence in favour of our non-Gaussian assumption. We also find that production has a significant positive effect on crude oil volatility while demand has insignificant negative effect. For out-of-sample forecasting evaluation, while considering models with long-term volatility component, we show that under loss functions models with t innovations are favoured over those with a normal innovation, while RS GARCH MIDAS-t with RV outperforms other models.

Citation

Tirkishova, M. (2023). Modelling and forecasting oil market volatility: A regime switching GARCH MIDAS approach. (Thesis). Keele University

• Thesis Type: Thesis
• Deposit Date: Oct 12, 2023
• Publicly Available Date: Oct 12, 2023
• Award Date: 2023-10