A novel carbon price prediction model based on optimized least square support vector machine combining characteristic-scale decomposition and phase space reconstruction

Carbon trading is an important market mechanism to promote carbon emission reduction and clean development. Accurate carbon price prediction is significant for environmental policymaking and improvement of carbon market efficiency. However, the existence of end effect and chaotic characteristics of carbon price sequence have limited the improvement of carbon price prediction accuracy. In this paper, a novel carbon price prediction model is proposed, which is based on local characteristic-scale decomposition (LCD), phase space reconstruction (PSR) and least square support vector machine (LSSVM) optimized by artificial fish swarm algorithm (AFSA). Firstly, carbon price is decomposed into several intrinsic scale components (ISC) by LCD to capture carbon price characteristics. Secondly, the maximum Lyapunov exponent is used to detect the chaos of the intrinsic scale components, and the chaotic ISC is further reconstructed by phase space reconstruction (PSR). In the meantime, the influence variables of non-chaotic ISCs are selected through partial autocorrelation analysis. Finally, the LSSVM optimized by AFSA is established to predict the ISC components of carbon price series and the ISC components are combined into carbon price prediction results. The empirical analysis shows that LCD-PSR-AFSA-LSSVM model has better prediction accuracy than Comparison models, and the MAPE values of the three carbon markets are 1.23%, 1.49% and 3.27%, respectively. The results suggest that the LCD-PSR-AFSA-LSSVM model is validity, generalization and stability. The application of the model will improve the operation efficiency of carbon market trading and advance clean development of various industries. •The local characteristic-scale decomposition is suitable for carbon price prediction.•Combination of local characteristic-scale decomposition and phase space reconstruction is effective.•The optimized least square support vector machine improves the prediction accuracy.•This method could be applied to mature and emerging carbon markets.