关键词: 碳市场价格预测, 多尺度预测, 局部回归, 集成经验模态分解

Abstract: The carbon price exhibits non-linear and non-stationary characteristics. It is a challenging task to accurately predict the carbon price. Based on the framework of “divide and conquer”, a multi-scale carbon price forecasting model based on local regression is proposed. The ensemble empirical mode decomposition (EEMD) is used to decompose the original carbon price time series into several simple components. Motivated by the fully local characteristics of a time series decomposed by EEMD, the local regression methods are adopted to forecast each component. The forecasting results of components are aggregated to obtain the final results. The localregression methods include local linear regression (LLP), local polynomial regression, local ridge regression, local principal component regression, local partial least squares regression, and local lasso regression. The two-carbon market futures price time series of the European Climate Exchange are selected as samples. The experimental results show that the proposed multi-scale forecasting model based on local regression performs effectively. Note that the EEMD-LLP model has a simple structure and shows better performance. Further analyses are implemented for the adaptability of parameters of the EEMD-LLP model. We compare EEMD-LLP with the newly-proposed methods, demonstrating the effectiveness of the EEMD-LLP model for carbon price forecasting.

Key words: carbon price forecasting, multi-scale forecasting, local regression, ensemble empirical mode decomposition