The Forecast Of China’s GDP Based On Time Series Model

Gross domestic product(abbreviated as GDP)is the result of the production activities of all resident units in a country within a certain period of time calculated according to the national market price.GDP can not only reflect the scale of economic development of a country or region,and judge the speed of its economic development and overall economic strength,but also an important basis for macroeconomic decision-making.Therefore,it is very meaningful for us to use statistical methods to forecast changes in GDP.In this paper,the object of our study is the real data of our country’s GDP from 1978 to 2021(data source:China Statistical Yearbook).This set of data can be viewed as a time series.We select a total of 39 real data of our country’s GDP from 1978 to 2016 as a training set for modeling and prediction,and select a total of 5 real data of our country’s GDP from 2017 to 2021 as a test set for comparative analysis with the predicted data.First,we use Rstudio to establish two single-item forecasting models for the training set GDP time series,and make predictions and analysis.Secondly,on the basis of the prediction results of the two single-item forecasting models,we establish two combined forecasting models and made predictions and analysis.Finally,we comprehensively analyze the prediction results of the four forecasting models to obtain the optimal forecasting model.In this paper,we build an autoregressive integrated moving average model(abbreviated as ARIMA model).First,we perform differential processing on the GDP time series of the training set,and observe the difference graph to judge that the differenced sequence is stable.Secondly,we perform the augmented Dickey-Fuller test(abbreviated as ADF test)on the differenced sequence to verify whether it reaches a stationary state.If the differenced sequence does not reach a stationary state,it is necessary to continue to differentiate it until it passes the ADF test to ensure that the differenced sequence reaches a stationary state.Finally,after the differentially stationary sequence passes the white noise test,we observe the autocorrelation graph and partial autocorrelation graph of the differential sequence,combined with the minimum information criterion(abbreviated as AIC criterion)and Bayesian information criterion(abbreviated as BIC criterion)to identify and order the model.In this paper,we build an exponential smoothing forecasting model.We use Rstudio to calculate the smoothing coefficient based on the fitting principle,and use the Holt two-parameter exponential smoothing method to iterate continuously to obtain the parameter estimates of the last period.Then we establish a prediction model.Because the single-item forecasting models have some advantages and disadvantages,in order to test whether the single-item forecasting models have a good forecasting effect,we combine the ARIMA model and the exponential smoothing forecasting model to construct combined forecasting models.According to the different methods of determining the weight coefficients,we obtain two combined forecasting models,namely,the combined forecasting model by the variance reciprocal method and the combined forecasting model by the residual reciprocal method.We use four forecasting models to forecast our country’s GDP from 2017 to 2021,and make a comparative analysis with the test set.After comparative analysis of these four forecasting models,from the short-term forecast results,we found that the prediction effect of the combined forecasting model by the residual reciprocal method is better than the ARIMA model,better than the combined forecasting model by the variance reciprocal method,and also better than the prediction effect of the exponential smoothing forecasting model.In summary,in this paper,we study four GDP forecasting models,namely ARIMA model,exponential smoothing forecasting model,combined forecasting model by the variance reciprocal method and the combined forecasting model by the residual reciprocal method.After comparative analysis,we find that the prediction effect of the combined forecasting model by the residual reciprocal method is the best,and the prediction effect of the ARIMA model is better than that of the combined forecasting model by the variance reciprocal method.Therefore,the combined forecasting models can make up for the deficiencies of the single-item forecasting models to a certain extent.