Carbon price prediction based on multiple decomposition and XGBoost algorithm.

Area of Science:

• Environmental Economics
• Computational Finance
• Data Science

Background:

• Carbon trading is a key mechanism for controlling global carbon dioxide emissions.
• Carbon pricing significantly influences market participant and policymaker decisions.
• Accurate carbon price prediction is crucial for effective market management.

Purpose of the Study:

• To propose and evaluate a novel carbon price prediction model named multi-decomposition-XGBOOST.
• To enhance the accuracy of carbon price forecasting in diverse market conditions.
• To provide a robust tool for market participants and policymakers.

Main Methods:

• Utilized Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) for initial price series decomposition.
• Applied Variational Mode Decomposition (VMD) to the intrinsic mode function (IMF) with the highest sample entropy.
• Recombined IMFs based on sample entropy, followed by further CEEMDAN decomposition and XGBoost prediction.

Main Results:

• The multi-decomposition-XGBOOST model demonstrated superior predictive performance in both the Beijing and Hubei carbon markets.
• In the Beijing market, the model improved RMSE, MAE, and MAPE by 30.4%, 44.5%, and 42.9% respectively, compared to single XGBoost.
• In the Hubei market, RMSE, MAE, and MAPE decreased by 28.5%, 39.4%, and 39.4% respectively.

Conclusions:

• The proposed multi-decomposition-XGBOOST model offers significantly improved prediction accuracy for carbon prices.
• The model's effectiveness spans both highly volatile and comparatively stable carbon markets.
• This approach provides a reliable tool for understanding and predicting carbon market dynamics.