Oil value-at-risk forecasts: a filtered semiparametric approach

Wei Kuang

Abstract

The Covid-19 pandemic has set the stage for greater volatility in oil prices. Given this unprecedentedly volatile environment, protection against market risk has never been more important. Value-at-risk (VaR) is a popular metric to measure and control risk. However, the widely used historical simulation approach is unresponsive to upticks in stress. Therefore, the need has arisen for an alternative method that is easy to implement while still achieving forecast accuracy. We propose the generalized autoregressive conditional heteroscedasticity (GARCH) model combined with the Cornish–Fisher expansion (a semiparametric approach to address skewness and excess kurtosis as well as volatility dynamics) for the oil VaR forecast. We compare the performance of the proposed approach with that of historical simulation and GARCH-type models with alternative residual distributions: historical simulation, normal, skewed Student t and generalized Pareto. The study is based on the daily spot data from the Energy Information Administration for the period from December 19, 2012 to October 30, 2020 for Brent and from November 13, 2012 to October 30, 2020 for West Texas Intermediate, each with a total of 2001 observations. We find that the historical simulation approach significantly underestimates the risks for both long and short positions during the recent market turmoil, which confirms the importance of the filtering process in VaR forecasts. Moreover, the proposed approach provides the most accurate VaR forecasts, especially at high confidence levels for the long position. The analysis serves as a useful guide to energy market risk quantification for practitioners and policy makers.