Research On The Forecasting Methods Of China Commodity Price Index

Commodities is a popular concept nowadays.It often refers to commodities that can enter the circulation field but not the retail link.Commodities has the characteristics of homogenization,tradable and large demand.Generally,commodities can be divided into three categories:agricultural and sideline products,metal products,energy and chemical products.Commodities involve many fields and industries,which have a profound impact on the stability and development of the national economy.Therefore,the study of the development trend of China commodity market has important theoretical and practical significance.This paper selects the China Commodity Price Index(CCPI)as the research sample,because of its high reliability and wide application.This paper models and analyzes the weekly CCPI historical data from 2017 to 2021,forecasts the CCPI data in the first five weeks of 2022,and finds out the most optimal prediction model according to the prediction effect.Firstly,this paper establishes ARIMA model,BP neural network model and support vector regression model to analyze and predict CCPI historical data,and evaluates those single models according to the comparison between the predicted value and the real value of each model.Secondly,in order to comprehensively analyze the information of CCPI historical data and integrate the advantages of each model,this paper establishes three combination models by the following methods:Dominance matrix method,Reciprocal variance method,Entropy method.Then,this paper determines the weights of those three combination methods according to the genetic algorithm,in order to further improve and integrate the combination model and obtain the final improved combination prediction model.Finally,this paper uses some common indicators in mathematical statistics of MAE,RMSE and MAPE to prove that the final improved combination prediction model has the best prediction effect.The innovation of this paper is reflected in the following two aspects.On the one hand,in addition to the common time series analysis model like ARIMA model,this paper also establishes BP neural network model and support vector regression model,which extends the prediction of CCPI to the field of machine learning.On the other hand,this paper creatively uses genetic algorithm to determine the weights of three combination methods about Dominance matrix method,Reciprocal variance method and Entropy method to further improves the combination model,and obtain the best prediction effect.In summary,this paper through combining several single models and combination methods to improve the prediction effect and obtain the best prediction model.Therefore,the research results of this paper have practical significance in the application of predicting CCPI.Finally,based on the research results,this paper provides relevant references and suggestions for the government,enterprises,financial institutions and individual investors.