Forecasting Value-at-Risk and Expected Shortfall: A comparison of non- and parametric methods for crude oil amidst extreme volatility

Abstract: Practitioners primarily utilise nonparametric methods when estimating Value-at- Risk (VaR) and Expected Shortfall (ES) for computing capital requirements. However, various researchers assert that there are issues with those estimates, particularly amidst periods of market turmoil. Academia produces novel parametric methods to estimate extreme risk measures to address these deficiencies. Nevertheless, empirical findings of the discrepancies in the performance between non- and parametric estimation methods are inconclusive. Various authors discover that the nonparametric methodologies display superior backtesting results, while others demonstrate contrary results. Therefore, this thesis contrasts the backtesting performance of non- and parametric estimation methods for VaR and ES in the context of crude oil amidst a significant geopolitical event which had major implications for West Texas Intermediate (WTI) and Europe Brent: the COVID-19 pandemic. We contrast the performance of the nonparametric BHS, AWHS, and VWHS methods with the parametric Gaussian, Student’s t, and conditional EVT methods. Our backtesting results demonstrate that the conditional EVT is the superior method for estimating VaR, whilst the Student’s t-distribution displays the most rigorous performance in estimating ES. These results are robust across WTI and Brent crude oil and for the duration of the backtesting period. Hence, we recommend that practitioners utilise parametric methods for estimating measures of extreme risks for crude oil.