A memory-trait-driven decomposition–reconstruction–ensemble learning paradigm for oil price forecasting

Abstract In order to improve the prediction performance in oil price forecasting, a novel memory-trait-driven decomposition–reconstruction–ensemble learning paradigm is proposed for oil price forecasting. The proposed methodology consists of four steps, i.e., data decomposition for original complex time series, component reconstruction for decomposed components, individual prediction for the reconstructed components, and ensemble output based on the individual component prediction results, which are all driven by memory traits. For verification purpose, the West Texas Intermediate (WTI) crude oil spot prices are used as the sample data. The experimental results demonstrated that the proposed methodology can produce the better and more robust results relative to the benchmarking models listed in this study. This indicates that the proposed memory-trait-driven decomposition–reconstruction–ensemble​ methodology can be used as a promising solution to oil price prediction with the traits of memory.