A New Multivariate Decomposition–Ensemble Approach With Multisource Heterogeneous Data for Crude Oil Price Forecasting

Abstract

Forecasting crude oil prices is pivotal for every individual and even the entire country. Previous studies have encountered difficulties in forecasting highly volatile crude oil prices, especially when conflicts, wars, and other irregular events occur. In light of this, this study introduces an innovative hybrid multifactor decomposition–ensemble approach with heterogeneous data from diverse sources. First, the multivariate forecasters including unstructured news text based on keywords and structured financial variables are processed. Second, multivariate empirical mode decomposition (MEMD) is used to decompose the crude oil price and its predictors, and sample entropy (SE) is employed to reconstruct the subcomponents obtained from the decomposition. Thereafter, some effective forecasters are screened from the reconstructed subcomponents of forecasters through statistically testing approaches. Finally, the crude oil prices are forecasted using a hybrid forecasting technique, and the performance of the proposed model is assessed from various viewpoints. The empirical conclusions indicate that the proposed model performs well in forecasting the weekly spot price of West Texas Intermediate crude oil.