Point and interval prediction of carbon price: A novel hybrid model considering the severe volatility and the complexity

Abstract

The development of global economic suffers from the serious problem of carbon emission. Accurate carbon price prediction is of great significance for carbon emission reduction. However, it is difficult for the existing carbon price prediction model to simultaneously solve the severe volatility and the complexity of carbon price. Therefore, this paper proposes a novel hybrid model composed of econometric model, machine learning model and optimization algorithm to realize point and interval prediction of carbon price. In the proposed model, an adaptive variational mode decomposition algorithm is proposed to explore the characteristics of carbon price sub-series. In point prediction, different from previous studies, this paper uses unsupervised clustering to distinguish the different complexity of the intrinsic modal functions. The high complexity components are predicted by BP neural network based on war strategy optimization algorithm, and the low complexity components are predicted by econometric model, which improves the prediction accuracy and the interpretability of the model. In interval prediction, the paper uses kernel density estimation and nonparametric bootstrap to obtain the probability distribution of the predicted value, and makes interval prediction according to different significance levels, which can provide more reliable information for decision-making. According to the empirical results of China’s Shenzhen carbon trading market and Beijing carbon trading market, our proposed model is superior to the other 23 benchmark models in point prediction and can perform effective interval prediction.