A new hybrid model for forecasting Brent crude oil price

Received a plethora of attention by both practitioners and researchers, oil price forecasting remains a challenging issue due to the particular characteristics of oil price and its prodigious impact on various economic sectors. Motivated by this issue, the authors aim to introduce a robust hybrid model for reliable forecasting of Brent oil price. For this purpose, the Adaptive Neuro Fuzzy Inference System (ANFIS), Autoregressive Fractionally Integrated Moving Average (ARFIMA), and Markov-switching models are employed in the proposed hybrid model. The cardinal merit of this hybridization lies in the fact that the constituent models are capable of capturing particular features like nonlinearity, lag, and market interrelationships existing in oil price time series. Then, specific weights are assigned to each model to achieve an accurate prediction of the empirical time series. Three weighting scenarios, namely equal weights, error-value-based weights, and genetic algorithm weighting function, are applied. The authors use root mean square error, mean absolute error, and mean absolute percentage error to measure errors. Robustness of results and prediction quality of the hybrid model compared with counterparts are also investigated by Diebold-Mariano test. Finally, numerical results reveal that the hybrid model weighted by genetic algorithm generally outperforms the constituent models, hybrid model with equal weights, and hybrid model weighted based on the error values. Reliable forecasting of crude oil prices is especially beneficial to producer and importer nations to optimize their production and order rates and mitigate the adverse effect of possible shocks. (C) 2020 Elsevier Ltd. All rights reserved.