Research On Economic Growth Forecast Based On Model Average Method

Economic growth analysis is an important topic in the subject of econometrics,especially in the critical period of China's economic transformation.It is necessary to analyze the key factors affecting economic growth,and accurately predict economic growth in order to achieve healthy and sustained economy.From the macro perspective,economic growth forecast is an important basis for policy making.From the micro perspective,economic growth forecast affects individual investment and savings strategies.In this case,the study of economic growth is of great significance.Starting with the analysis of factors affecting economic growth,this paper chooses seven representative contributing variables according to the analysis and previous literatures,and takes GDP as an index to measure economic growth,128 linear models are fitted,empirical analysis and prediction of economic growth are carried out with different model averaging methods.The empirical analysis consists two parts: the economic growth forecast based on Frequency Model Averaging method and the economic growth forecast based on Bayesian Model Averaging method.In the first part,OPT,MMA,JMA,S-AIC and S-BIC methods are used to calculate the model weights,and then all the model predictions are weighted and combined into the final predictions.According to the absolute error,the optimal rate and the standard deviation of absolute error,the accuracy and stability of the predicted values of the different model averaging methods are analyzed,and their advantages and disadvantages are evaluated.A modified OPT method is proposed,and using the same way to check whether the modified OPT method effectively improves the OPT method.In the second part,Bayesian Model Averaging are used to forecast the economic growth of the same data,the basic principle of Bayesian Model Averaging and the similarities and differences of two packages,BMA and BMS,are analyzed.The forecasting results are also analyzed according to absolute error,optimal rate and standard deviation of error.Finally,the robustness of the model is tested,and the advantages and disadvantages of the overall methods are analyzed.The results show that.Firstly,the JMA method performs best in Frequency Model Averaging methods,with the minimum average prediction error,the highest prediction accuracy and the highest optimal rate.Secondly,the modified OPT method can effectively improve the prediction effect of the original OPT method,significantly reduce the prediction error and error standard deviation,and improve the prediction accuracy and optimal rate.Thirdly,BMA algorithm of Bayesian Model Averaging is obviously superior to BMS algorithm.The prediction error is the lowest,the optimal rate is highest and the standard deviation of error is the smallest,and the prediction result is the most stable.Fourthly,there is a gap between the variables selection of different algorithms in Bayesian model,so the average prediction results of Bayesian model are generally not as stable as that of frequency model.Fifthly,comparing all the methods,JMA has the least prediction error,the highest accuracy,the highest optimal rate and the most stable prediction results.Sixthly,the model averaging method has a good robustness in predicting economic growth,and the conclusions can be consistent when the parameters are changed.