Research On The Forecasting Methods Of GDP In China

GDP not only plays an irreplaceable role in reflecting a country's national income and consumption capacity,economic development and other aspects,but also helps people to know and understand a country's economic operation from a macro perspective.It is the key basis for the formulation of national economic policies,as well as the key means to test whether economic policies are effective and scientific.Therefore,if we use appropriate statistical methods to reveal the law of the change of GDP quantity,and to predict short-term GDP with high accuracy,then it is of great practical significance for macroeconomic planning.China's GDP historical data is a non-stationary time series with both linear and non-linear characteristics.Based on four models,this paper models GDP data,uses R software to simulate and calculate,and compares the prediction accuracy and generalization ability of the four models to obtain the optimal model.Firstly,the traditional ARIMA model is established.After the sequence is smoothed,the model is identified and ranked according to BIC criterion combined with autocorrelation function graph and partial autocorrelation function graph.After white noise test,the optimal model is determined to be ARIMA(0,1,2).Secondly,the BP neural network with good non-linear fitting ability is studied,and GDP data are preprocessed according to the characteristics of China's GDP data.Transform into input matrix,set up reasonable network structure and activation function,excavate the non?linear characteristics of GDP data;Thirdly,considering the defects and advantages of single model,combine ARIMA model with BP neural network to establish a combination model,that is,use ARIMA model to predict the linear main body of GDP data,and then use BP neural network to estimate the non-linear residual,and add it up.The final predicted value of the combined model is obtained.The empirical analysis shows that the combined model partially realizes the complementary advantages among the models.Fourthly,considering that compared with the single model,the combined model has certain advantages,but there are still some shortcomings,which need to be further optimized.This paper speculates that GDP data is not a complete exponential trend.Therefore,two ways are adopted to achieve stabilization,and corresponding ARIMA models are established respectively.The predicted values of the two models are weighted as the linear part of the improved combination model,and the Bagging ensemble algorithm is incorporated into the prediction of the non-linear residual of GDP data.The results show that the improved combination model has great advantages in all aspects.Compared with the first three models,the improved combination model has higher overall prediction accuracy and more stable prediction effect.To sum up,the four methods of GDP prediction studied in this paper are:ARIMA model,BP neural network model,combination forecasting model and improved combination model.Through the analysis and comparison of the forecasting results of the four models,it is proved that the improved combination model has the best forecasting effect and the most effective forecasting of China's GDP.Finally,according to this model,the whole GDP data are modeled to predict China's GDP in the next three years.