Carbon price forecasting: a novel deep learning approach.

Area of Science:

• Environmental Economics
• Machine Learning
• Time Series Analysis

Background:

• Carbon emission trading markets are essential for effective carbon reduction.
• Accurate carbon price forecasting is critical for policy and investment decisions.
• Existing prediction models struggle with the non-linearity and complexity of carbon prices.

Purpose of the Study:

• To propose an advanced deep neural network, TCN-Seq2Seq, for accurate carbon price forecasting.
• To leverage a sequence-to-sequence architecture with fully convolutional layers for temporal dependency learning.
• To demonstrate the model's suitability for small datasets using few-shot learning principles.

Main Methods:

• Development of the TCN-Seq2Seq deep neural network model.
• Utilizing a sequence-to-sequence framework with fully convolutional layers.
• Applying parallel training for efficiency with fewer parameters, suitable for few-shot learning.

Main Results:

• The TCN-Seq2Seq model significantly outperforms traditional statistical and current deep learning models.
• Achieved highest Directional Accuracy (DA) of 0.9697.
• Recorded lowest Mean Absolute Percentage Error (MAPE) of 0.0027 and Root Mean Square Error (RMSE) of 0.0149.

Conclusions:

• The TCN-Seq2Seq model offers superior predictive ability and robustness for carbon price forecasting.
• Accurate forecasting provides valuable insights for policy makers and carbon market investors.
• The proposed model enhances decision-making in carbon trading markets.