Multi-step-ahead and interval carbon price forecasting using transformer-based hybrid model.

Area of Science:

• Environmental Economics
• Financial Forecasting
• Data Science

Background:

• Accurate carbon price forecasting is crucial for market stability and investment decisions.
• Existing methods struggle with the non-linear and non-stationary nature of carbon price data, especially for multi-step and interval predictions.
• Addressing these challenges is vital for reliable carbon market assessment.

Purpose of the Study:

• To propose a novel hybrid model for accurate multi-step-ahead and interval carbon price forecasting.
• To enhance the reliability of carbon price predictions for policymakers and investors.
• To address the limitations of existing models in handling complex carbon price dynamics.

Main Methods:

• A hybrid model combining Hampel identifier (HI), time-varying filtering-based empirical mode decomposition (TVFEMD), and a transformer model.
• HI for outlier identification and correction; TVFEMD for decomposing and reconstructing carbon price series.
• Orthogonal array tuning for hyperparameter optimization and quantile loss function for forecasting.

Main Results:

• The proposed HI-TVFEMD-transformer model significantly outperformed benchmark models in multi-step-ahead forecasting (e.g., MAE of 0.6546 for one-step ahead).
• Interval forecasts consistently achieved a PICI above 0.95 at a 0.1 confidence interval, demonstrating effective uncertainty quantification.
• The model proved reliable across five pilot Chinese carbon trading markets.

Conclusions:

• The novel hybrid model offers a significant advancement in carbon price forecasting accuracy and reliability.
• It effectively handles the non-linear and non-stationary characteristics of carbon price data.
• The model provides a dependable tool for informed investment and operational decisions in carbon markets.