Research On Price Forecasting Of International Carbon Futures Based On Machine Learning

Since the entry into force of the Kyoto Protocol,the concept of resource conservation and environmental protection has been strengthened.Countries have paid more and more attention to carbon emission economy and their participation has been increasing.They are focusing on building carbon market and developing new carbon derivatives.In May 2020,the central government,together with relevant ministries and commissions,proposed to support the construction of Guangzhou futures exchange and explore and develop carbon emission trading.Taking the international mature carbon futures market as the standard,the construction of Guangzhou futures exchange will promote the participation of national investors in trading,which will help to compete with the global advanced level,create a market with world influence,and promote the early possession of product pricing power.The price of carbon futures is affected by market equilibrium and reflects the real carbon price.Grasping the price of carbon futures can stimulate investors’ enthusiasm for trading and improve the efficiency of market resource allocation.China is in the initial stage of the construction of carbon futures market.Taking the price as the connecting point,we can compare the development differences of domestic and foreign carbon markets,and explore the development path suitable for China’s carbon futures market.The domestic research is in the preparatory stage,and it does not have the necessary data sample conditions.Therefore,the largest and most mature EUA price in the international market is taken as the research object.In terms of model,machine learning has good generalization ability and can be used in price forecasting.In this paper,machine learning is used as a tool to predict the price of carbon futures.First of all,the part of model construction.On the one hand,support vector machine(SVM)modeling,in order to determine the most suitable SVM for carbon futures price prediction,this paper breaks the previous research on the specified kernel function mode,according to the data itself,uses grid search,genetic algorithm and particle swarm optimization algorithm to optimize the parameters,and creates the model optimization combination with kernel function.After cross validation,the combination results are input into SVM,The optimized and upgraded prediction model is constructed.On the other hand,random forest modeling is carried out,and the decision tree model is established and constructed by bootstrapping method.Secondly,in the data processing part,min max standardization and principal component analysis are used to eliminate dimensional differences and collinearity.Then the adjusted data is imported into the model to obtain the prediction results.Finally,according to the standard,the prediction results are evaluated and analyzed to realize the optimization of precision.The experimental results show that GA-SVM(rbf)is better than random forest in terms of prediction accuracy and fitting degree.This paper uses machine learning to predict the price of carbon futures,and puts forward some suggestions from the macro and micro perspectives.The research is expected to help regulators improve carbon trading management ability,maintain its good operation,help participants regulate market activities,improve investment returns,create a fully functional carbon futures market,and promote the development of green and low-carbon economy.