| The problem of global warming is becoming more and more serious.More and more countries accept the establishment of carbon market for carbon emission trading to realize market-oriented control of total carbon emission.As the world’s second largest economy,China has innovatively proposed a strategy of "carbon peaking and carbon neutrality" between economic development and environmental protection.To achieve this long-term goal,China is also gradually piloting carbon markets in Beijing,Shanghai,Guangdong and other regions,aiming to use the invisible hand of the market to conduct macro-control of carbon emissions.Therefore,the prediction of carbon emission trading price in carbon market has farreaching theoretical value and practical significance.In order to predict the carbon price of eight major carbon markets in China,the GARCH-MIDAS/VMD-BiLSTM-GPR model is proposed.Since many literatures have proved that the original carbon price sequence is nonlinear and non-stationary,VMD method is firstly used to decompose the original carbon price sequence of each carbon market,and the carbon price sequence of each market is decomposed into several sub-bands.GARCH-MIDAS modeling is used to predict the high frequency band with ARCH effect and stable sequence after decomposition.The remaining bands were modeled and predicted by BiLSTM.Finally,the predicted values of carbon price obtained by the two methods are integrated and compared with the observed values of the original carbon price series.In order to improve the prediction accuracy and the economic value of the research content,this paper selects four influencing factors including similar products,energy structure,macro economy and climate environment,and quantifies the text information about"carbon" through Baidu index.In addition,in order to eliminate redundant variables,random forest(RF)was used to screen variables according to the importance of variables(VIM)method.The influence factors of different time frequencies were incorporated into the model through mixed frequency data sampling(MIDAS).Then maximum likelihood estimation(LLM)was used to obtain the model parameters of this part.In order to improve the stability of prediction,Gaussian random regression(GPR)is also used to make interval prediction based on point estimation,so as to improve the robustness of the whole model.Finally,MAE,MSE,RMSE,PICP and other evaluation criteria are used,and only GARCH model,BiLSTM model and this model are used for comparative analysis.The empirical results show that:compared with mere BiLSTM and mere GARCH-MIDAS models,the hybrid model with multiple factors engaged in this paper shows stronger robustness in predicting the accuracy of carbon price in China,and the model has the best prediction effect for Hubei carbon market.At the same time,this article also found the strong index,China’s economic policy uncertainty index,SO2 and so on to our carbon price also has a strong interpretation power.Admittedly,this paper also has a number of shortcomings,such as the failure to consider internal and external impacts such as carbon network and emergencies,single data conversion mode,and the setting of model parameters mainly following empiricism. |
