Author(s)

Guanyu Ji

Corresponding Author

Guanyu Ji

Abstract

Accurate carbon price forecasting is critical for carbon trading markets to function properly and to meet carbon reduction objectives. However, carbon pricing data are highly nonlinear, non-stationary, and complicated, offering considerable problems to typical econometric models. To solve these challenges, this work introduces a new hybrid forecasting model, the SSA-VMD-ARIMA-CNN-LSTM model. The model uses Variational Mode Decomposition (VMD) to break down the original carbon price series into linear and nonlinear components, with an adaptive Sparrow Search Algorithm (SSA) optimizing VMD settings. An empirical analysis using carbon trading data from Fujian Province shows that the proposed model significantly outperforms single machine learning models and traditional forecasting approaches in terms of accuracy and robustness. The Autoregressive Integrated Moving Average (ARIMA) model is used to model the linear component, while Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and hybrid CNN-LSTM models are applied to various nonlinear components based on their characteristics. The findings offer important insights for investors, businesses, and policymakers in carbon markets.

Keywords

Carbon Price Forecasting, SSA-VMD Decomposition, Hybrid Model