A carbon price hybrid forecasting model based on data multi-scale decomposition and machine learning

Abstract

Accurate carbon price forecasting is of great significance to the operation of carbon financial markets. However, limited by the non-linearity and non-stationarity of the carbon price, the accurate and reliable predictions are difficult. To address the issue of applicability and accuracy, a novel carbon price hybrid model based on decomposition, entropy, and machine learning methods is proposed, named as CEEMDAN-PE-LSTM-RVM. Adopting the advanced structure (i.e., the prediction under classification), the proposed model owns reliable performance in face of the cases with different complexity. Furthermore, the relationship between the data feature and prediction accuracy is discussed to provide a benchmark for judging the reliability of the prediction, in which the chaos degree is introduced as a feature to characterize carbon price quantitatively. The performance of the proposed model is evaluated through historical data of four representative carbon prices. The results show that the average MAPE and RMSE of the proposed model achieve 1.7027 and 0.7993, respectively, which is significantly greater than others; the proposed model owns great robustness, which is less affected by the complexity of predicted objects. Thus, the proposed model provides a reliable tool for carbon financial markets.

Keywords: Carbon price forecasting; Chaos theory; Complete ensemble empirical mode decomposition with adaptive noise; Permutation entropy; Time series analysis.