Forecasting global crude oil price fluctuation by novel hybrid E-STERNN model and EMCCS assessment

Abstract

Energy futures are a very significant part of commodity futures, no less than the influence of the spot market. A novel hybrid neural network (denote by E-STERNN) is proposed through combining Elman recurrent neural network model with stochastic time strength (ST-ERNN), and ensemble empirical mode decomposition (EEMD) is also introduced to improve the performance of forecasting neural network system for energy markets. ST-ERNN model is established for taking into account the weight of energy historical data with time variations. EEMD is an algorithm that decomposes any non-stationary and nonlinear time series into simple and independent time sequence. From the empirical research for four global energy market prices, the proposed hybrid E-STERNN model is verified to have higher prediction accuracy compared with the original ERNN and the ST-ERNN models. Moreover, a new error evaluation approach, called the exponent of multi-scale composite complexity synchronization (EMCCS), is utilized to analyze and estimate the prediction performance, and the demonstration analyses confirm that the hybrid E-STERNN model has higher prediction accuracy for global energy futures indexes.