Interval forecasting of carbon price: A novel multiscale ensemble forecasting approach

Abstract

Aiming at the limitations of carbon price point forecasting, we propose a novel integrated approach of binary empirical mode decomposition (BEMD), differential evolution (DE) algorithm, and extreme gradient boosting (XGB) for carbon price interval forecasting. Firstly, BEMD, which is suitable for interval time series, is introduced into decomposing complex carbon data into simple components. Secondly, XGB is used to forecast the obtained components, and DE is used to synchronously optimize all parameters of XGB. Thirdly, the individual component forecasting values are aggregated into carbon price forecasting values. Taking Guangdong and Hubei carbon markets as samples, in comparison with other popular prediction models, the proposed approach has a higher coverage rate and lower prediction error. The sensitivity analysis verifies that the proposed approach is robust.