Time-Varying Covariance in Major Energy Portfolios

Abstract

Utilizing the primary crude oil markets of Average, Brent, Dubai and West Texas Intermediate from 1982 to 2023, this research aim to estimate the Multivariate VECH between returns on Average, Brent, Dubai, and WTI crude oil price to see how the conditional covariance matrix of the crude oil market variables have a flexible dynamic structure, report time-varying covariance and impact of lagged shocks on conditional volatility and select the appropriate model to model the energy markets. After preliminary investigation have been conducted, like the time graph to determine trend in the evolution of the series, Augmented Dickey-Fuller test to ascertain unit root and logarithmic return for stability, the Multivariate Vector Error Conditional Heteroskedasticity (MVECH) and Diagonal Conditional Correlation (DCC) models were applied to the study variables. The findings demonstrate that the "positive semidefinite" property is satisfied by the diagonal multivariate VECH model as the estimates on the leading diagonal of the variance-covariance matrix are positive, it may be inferred that the variables are traveling in the same direction. Also, each asset in the portfolio exhibits time-varying volatility as captured by the significant ARCH and GARCH coefficients. Market returns and crude oil prices exhibit time-dependent oscillations, according to the Diagonal Conditional Correlation (DCC). By applying the DCC-GARCH's constant conditional correlation to the relationship between Dubai raw prices and Brent, Average and West Texas Intermediate, the impact of lagged shocks on the conditional variance was statistically significant. According to the model selection approach that use the Akaike information criterion (AIC), the Diagonal Conditional Correlation (DCC) model performs better than the Diagonal Multivariate VECH model. All sufficiency tests also show that the model is adequate. Since crude oil market factors such as time-varying covariance, and volatility imbalance are interdependent, it is necessary to use a multivariate GARCH Model to assess the advantages of this dependency. Some recommendations were proffered.