China's GDP Forecast Based On ARIMAX-BP Combination Model

As the core economic index to evaluate the economic development of a country or region,gross domestic product plays an important role in judging whether a country's economy is developing healthily.Accurate GDP forecast will also provide statistical theoretical support for relevant departments to make economic development plans and policies.Therefore,it is of great practical significance to find an effective prediction method for GDP predictionThis paper takes the GDP series of China from 1952 to 2019 as the research object,uses MATLAB software,establishes a number of models to predict China's GDP,compares and evaluates the results of each model,and determines the relatively optimal model and prediction method.First,according to the theory of time series,a traditional ARIMA model is established for the GDP series.At the same time,in order to improve the prediction effect of the time series model,we select 10 economic variable indexes which are highly related to GDP,and use lasso method to select variables.Finally,we select fiscal revenue series and household consumption level series as input series to establish ARIMAX model.Second,according to the theory of neural network,aiming at the shortcomings of the traditional BP neural network,the improved LM-BP neural network is established to forecast GDP.Third,considering that the combination model can usually combine the advantages of each single model and make up for the shortcomings of each single model,we try to establish a combination model for our GDP series.At the same time of building traditional series and parallel combination model,ARIMAX-BP combination model is put forward and established.The specific method is to select ARIMAX model which has better prediction effect than ARIMA model to depict the linear characteristics of China's GDP,and LM-BP neural network is used to excavate the nonlinear characteristics of the ARIMAX residual sequence.The final prediction value of ARIMAX-BP combination model is obtained by adding the results of the two models.To sum up,there are three single models in this paper:ARIMA model,ARIMAX model,LM-BP neural network model.Based on the single model,the paper discusses the method of establishing the combined model.,some conclusions are drawn:Firstly,the ARIMAX model with input variables is better than ARIMA model.This shows the feasibility of lasso method to select model variables,and it also shows that the introduction of fiscal revenue series and household consumption level series is helpful to depict the development trend of GDP series.Secondly,all the models established in this paper predict China's GDP accurately,and ARIMAX-BP combination model has the best relative prediction effect.This shows that both time series and neural network methods can effectively predict China's GDP,and the combination of the two methods is more advantageous.