Research On The Method And Application Of Carbon Market Price Forecasting

As an efficient way for controlling the cost of tackling climate change, carbon market has been paid more and more attention in theoretical and practical fields. In recent years, as the representative of European Union Emission Trading Scheme(EU ETS), global carbon market has influences on reduction performance significantly with its dramatic price volatility. Aiming at the carbon price of EU ETS, this dissertation explores prediction analysis, it’s contributions are as follows:(1) A novel method of carbon prices forecasting based on simultaneous optimization for phase space reconstruction(PSR) and least square support vector regression(LSSVR) using particle swarm optimization(PSO) is proposed. Firstly, the parameters optimization of PSR and LSSVR for carbon price forecasting is transformed into a combinatorial optimization of the parameters. Secondly, The optimal parameters are obtained simultaneously by using the PSO algorithm on the basis of data characteristics, which can overcome the drawbacks of separate optimization and alternative optimization. Finally, taking two carbon future prices under the EEU ETS as samples, the empirical results show that, compared with the traditional forecasting approaches, the proposed method has a better forecasting accuracy.(2) A novel parameters simultaneous optimization for PSR and LSSVR with uniform design(UD) is put forward for carbon price forecasting. Firstly, the parameters simultaneous optimization of PSR and LSSVR for carbon price forecasting is transformed into a combinatorial optimization of multiple factors and multiple levels. Secondly, the large sample combinatorial optimization of multiple factors and multiple levels is transformed into a small sample combinatorial optimization through UD, so as to enhance the optimization efficiency. Thirdly, all the parameters are simultaneously optimized using the self-invoking LSSVR to obtain the optimal parameters. Finally, the proposed method is verified by forecasting two carbon futures with different maturities under the EU ETS. The empirical results show that comparing with particle swarm optimization, the proposed parameters simultaneous optimized LSSVM predictor with UD can significantly improve the modeling efficiency at the same time ensuring a high prediction accuracy.(3) We develop a new adaptive multiscale ensemble forecasting model incorporating EEMD, PSO and LSSVR with kernel function prototype integrating to improve the accuracy of energy price prediction. Firstly, the extrema symmetry expansion EEMD is utilized to decompose the energy price into simple modes, which can effectively restrain the mode mixing and end effects. Secondly, by using the fine-to-coarse reconstruction algorithm, the high frequency, low frequency and trend components are identified. Meanwhile, ARIMA is applicable to predicting the high frequency components due to its strong ability of short-term memory. As to LSSVR, characterized by a favorable capture ability on nonlinear system, is therefore suitable for forecasting the low frequency and trend components. At the same time, in order to take full use of the advantages of various kernel functions types and make up the drawbacks of single kernel function, a universal kernel function prototype is introduced, which can adaptively select the optimal kernel function type and model parameters according to the specific data using PSO. Finally, the prediction results of all the components acquired by different models are aggregated into the forecasting values of energy price. It is proved that the proposed model can effectively improve the accuracy of carbon price prediction.