Tail risk assessment is crucial in financial markets, especially for commodities such as Brent crude oil, where extreme price fluctuations pose a risk for investors and policymakers. Risk models such as generalised autoregressive conditional heteroscedasticity (GARCH) often struggle to capture these extreme movements accurately, leading to potential underestimation of risk exposure. To solve this problem, we combine an alpha-recurrent neural network with a generalised Pareto distribution to better predict extreme price changes and improve tail risk estimation. Our findings demonstrate that this approach effectively captures downside risk, with backtesting results yielding high p-values, confirming its statistical reliability. The results from the shape parameter reveal that losses in crude oil markets are significantly riskier than gains, highlighting the asymmetric nature of price movements. Risk estimates indicate that the model provides robust assessments for both long- and short-term trading positions, making it a valuable tool for risk management. Nevertheless, these results have broader implications for financial risk modelling, particularly in commodity markets, where macroeconomic and geopolitical factors influence price volatility. Future work should focus on expanding the data set, enhancing computational efficiency and adding external risk factors such as liquidity constraints and regulatory shifts. Better calibration methods and the ability to adjust in real-time can make predictions more accurate, helping risk assessment models stay useful in changing market conditions.
Bantsi et al. (2026) studied this question.