Forecasting Carbon Price in China: A Multimodel Comparison.

Area of Science:

• Environmental Economics
• Machine Learning
• Time Series Analysis

Background:

• The carbon emission trading market is increasingly important globally due to carbon dioxide concerns.
• Accurate carbon price forecasting is vital for understanding market dynamics and achieving emission reduction goals.
• Carbon price forecasting is complex due to numerous influencing factors and the nonlinear nature of time series data.

Purpose of the Study:

• To forecast carbon prices in China using a deep learning approach.
• To evaluate the effectiveness of Multivariate Long Short-Term Memory (LSTM) for carbon price prediction.
• To compare Multivariate LSTM with other models like MLP, SVR, and RNN.

Main Methods:

• Utilized Multivariate Long Short-Term Memory (LSTM), a deep learning model.
• Collected historical time series data for carbon prices in Hubei (HBEA) and Guangdong (GDEA) from May 2014 to July 2021.
• Incorporated traditional energy prices as factors influencing carbon prices.
• Compared Multivariate LSTM performance against Multilayer Perceptron (MLP), Support Vector Regression (SVR), and Recurrent Neural Network (RNN).

Main Results:

• The Multivariate LSTM model demonstrated superior performance compared to MLP, SVR, and RNN.
• Lower Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) were achieved with Multivariate LSTM.
• The model's accuracy was validated against recent deep learning-based forecasts for HBEA and GDEA.

Conclusions:

• Multivariate LSTM is highly suitable for carbon price forecasting.
• This deep learning approach offers a novel method for predicting carbon prices.
• The findings provide valuable insights for carbon market policy implications.