AEMPS

Forecasting Crude Oil Price Volatility: A Comparative Study of ARIMA, GARCH, and Hybrid Machine Learning Models

Author

Jikai Tang * 1

1 Shanghai Jiao Tong University

Corresponding Author

Jikai Tang

Keywords

Volatility Process of Crude Oil Price, Process of Volatility Forecasting, Hybrid Forecasting Models of Such Kind, GARCH Family Models, Artificial Neural Networks (ANN)

Abstract

Predicting the volatility of crude oil prices is of crucial significance, yet it poses challenges because of the intricate interaction process between geopolitical factors and economic factors. In this study, a systematic and comparative analysis is carried out on traditional econometric models and modern machine-learning models for the volatility of WTI crude oil. This study starts by making use of ARIMA and GARCH-family models (including GARCH and APGARCH) to capture the linear patterns and the asymmetric volatility clustering phenomenon. In order to deal with the nonlinear limitations of these parametric models, this study then puts forward a brand-new hybrid framework that combines the APGARCH model with an Artificial Neural Network (ANN). By using the WTI spot price data, the empirical results of this paper show the better performance of the hybrid approach. Specifically speaking, the APGARCH-ANN model reduces the forecast errors by more than 80% when it comes to RMSE in comparison with the standalone APGARCH model. This notable improvement emphasizes the synergy that exists between the econometric feature extraction process and the machine learning’s ability to perform nonlinear approximation. The findings offer a robust and advanced forecasting tool for energy firms as well as financial institutions, which can enhance the risk management process and the strategic decision - making process when facing the volatile oil markets.

Citation

Jikai Tang. Forecasting Crude Oil Price Volatility: A Comparative Study of ARIMA, GARCH, and Hybrid Machine Learning Models. AEMPS (2026) Vol. 259: 90-103. DOI: 10.54254/2754-1169/2026.LD31510.

References

[1]: Patidar, A.K., Jain, P., Dhasmana, P. and other scholars (2024). The impact that global events have on the crude oil economy: a comprehensive review that is carried out on the geopolitics of energy and economic polarization. GeoJournal, 89: 50.

[2]: Ho, S. L., and Xie, M. (1998). The utilization of ARIMA models in order to carry out reliability forecasting and then analysis. Computers and Industrial Engineering, 35(1 then 2), 213 then 216. https: //doi.org/10.1016/s0360-8352(98)00066-7

[3]: Herranz, E. (2017) carried out a study on the topic of unit root tests. The study was published in the WIREs Comput Stat, and the publication details show that it was in volume 9, specifically corresponding to e1396.

[4]: Engle, R. F. (1982). Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation. Econometrica, 50(4), 987-1007.

[5]: Dhaher Alwan, H. and Najm Abdulla, S. (2022). An analysis is carried out on models (that is, the NAGARCH and the APGARCH) through the process of using simulations. Journal of Economics and Administrative Sciences, 28(132), pages from 62 then 73.

[6]: Bollerslev, T. (1986). Generalized autoregressive and then conditional heteroskedasticity. Journal of Econometrics, 31(3), 307 and then 327.

[7]: Kristjanpoller, W.D., along with Hernández, E. (2017). In the situation where the volatility of main metals is being considered, it is forecasted by a hybrid ANN - GARCH model that has regressors. This research is published in Expert Syst. Appl., and specifically, in the 84th volume, the content ranges from page 290 to page 300.

[8]: Jadhav, V., Chinnappa Reddy, B. V., and Gaddi, G. M. (2017). Application of the ARIMA Model for Forecasting Agricultural Prices. Journal of Agricultural Science and Technology, 19(5), 981-992.

[9]: Zhao, P., Zhu, H., NG, W. S. H., and Lee, D. L. (2024). In the process of transitioning from the GARCH model to the neural network for the purpose of conducting volatility forecasting. Proceedings of the AAAI Conference on Artificial Intelligence, 38, number 15, ranging from 16998 to 17006.

[10]: Han, H., Liu, Z., Barrios Barrios, M. and other scholars (2024). In the process of carrying out research, they developed a time series forecasting model for the extraction of non - stationary series patterns. This development was achieved through the utilization of deep learning and the implementation of GARCH modeling. The relevant research results were published in the journal Cloud Computer, specifically in volume 13, issue 2.

[11]: Amirshahi, B., & Lahmiri, S. (2023). Models of the hybrid deep learning type and GARCH - family models which are of the kind for the purpose of making forecasts regarding the volatility of cryptocurrencies. Machine Learning with Applications.

Copyright © 2021 Eliwise Academy. Unless Otherwise Stated