Forecasting carbon price using empirical mode decomposition and evolutionary least squares support vector regression. (1st April 2017)

• Record Type: Journal Article
• Authors: Zhu, Bangzhu; Han, Dong; Wang, Ping; Wu, Zhanchi; Zhang, Tao; Wei, Yi-Ming
• Is Part Of: Applied energy. Volume 191(2017)
• Page Start: 521
• Page End: 530
• Publication Date: 2017-04-01
• Subjects: Carbon price forecasting -- Empirical mode decomposition -- Least squares support vector regression -- Particle swarm optimization; EMD empirical mode decomposition -- LSSVR least squares support vector regression -- IMF intrinsic mode function -- PSO particle swarm optimization -- EU ETS European Union Emissions Trading System -- GARCH generalized autoregressive conditional heteroskedasticity -- ANN artificial neural networks -- ARIMA autoregressive integrated moving average -- RBF radial basis function -- ECX European Climate Exchange -- RMSE root mean squared error -- Dstat directional prediction statistic -- DM test Diebold–Mariano test; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2017.01.076
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: A multiscale least squares support vector regression is built to predict carbon price. Carbon price is decomposed into several simple modes via empirical mode decomposition. Evolutionary least squares support vector regression is used to forecast each mode. The proposed approach can achieve high statistical and trading performances. Abstract: Conventional methods are less robust in terms of accurately forecasting non-stationary and nonlineary carbon prices. In this study, we propose an empirical mode decomposition-based evolutionary least squares support vector regression multiscale ensemble forecasting model for carbon price forecasting. Firstly, each carbon price is disassembled into several simple modes with high stability and high regularity via empirical mode decomposition. Secondly, particle swarm optimization-based evolutionary least squares support vector regression is used to forecast each mode. Thirdly, the forecasted values of all the modes are composed into the ones of the original carbon price. Finally, using four different-matured carbon futures prices under the European Union Emissions Trading Scheme as samples, the empirical results show that the proposed model is more robust than the other popular forecasting methods in terms of statistical measures and trading performances.