Abstract
Forecasting the value at risk (VaR) of crude oil futures can be a challenging task for investors due to the high volatility of these prices. It is crucial to describe the return in the tail distribution, as extreme values can trigger larger price fluctuations and market risks. In this study, we proposed a hybrid model, ARIMA-SVR-POT, which uses a combination of the autoregressive integrated moving average (ARIMA), support vector regression (SVR), and peak over threshold (POT) method from the extreme value theory. We compared the performance of our hybrid model with three other models, namely ARIMA-EGARCH, ARIMA-SVR, and ARIMA-EGARCH-POT. We demonstrated the effectiveness of our model using crude oil WTI Futures as a sample from June 23, 2016, to September 30, 2022. Our findings show that the ARIMA-SVR-POT hybrid model provides accurate predictions of the returns and volatility. Furthermore, the model performs exceptionally well in capturing the extreme tail of returns and outperforms the other models. We also conducted back-testing in the proposed model and the results show that the ARIMA-SVR-POT model passed the Kupiec test at confidence levels of 95 %, 99 %, 99.5 %, and 99.9 %. Our proposed model provides a more precise reflection of potential losses when estimating VaR. The predicted loss probability is closer to the actual loss occurrence probability, indicating superior performance compared to traditional statistical models, which enhanced crude oil risk management tools and suggested effective measures to manage market risks.