In order to reduce the unpredictability of carbon prices caused by their increasingly prominent environmental and market attributes, and to minimize their negative impact on carbon trading, further research on forecasting models for carbon price is urgently needed. To improve the accuracy of prediction, this paper proposes a carbon price forecasting method based on SSA-NSTransformer. The method includes four main steps: Firstly, decomposition of carbon price signals, using Singular Spectrum Analysis to remove noise signals; Secondly, analysis of influencing factors, using Random Forest to identify and select key influencing factors of carbon price signal components from energy price, financial market, socio-economic, and environmental aspects; Furthermore, influencing factors prediction, considering the impact of different carbon reduction targets and predicting future trends of influencing factors; And finally, carbon price prediction, considering the impact of factors based on multi-stage carbon reduction targets, using Non-stationary Transformer to predict the signal components of carbon prices, reconstructing the carbon price time series, and testing the model accuracy. This paper uses historical data of ICE-EUA closing prices and influencing factors as samples, and the model error comparison results confirm that the proposed hybrid model for carbon price forecasting has significant advantages, which can provide scientific and accurate reference for carbon market participants' behaviors.
Read full abstract