Predict China Macroeconomic Indexes Based On TSPCA-FTNet

Macroeconomic indexes include gross domestic product(GDP),consumer price index(CPI),retail price index(RPI),etc.Through these indicators,we can know the economic condition of a country or region from a macro perspective.Therefore,these indicators are of vital significance for knowing macroeconomic conditions and controlling macroeconomic trends.If a more accurate forecast of the macroeconomic indexes can be made in advance,the government and relevant macro-control departments can propose more forward-looking and targeted macro-control policies and deploy necessary control measures.In addition,if individuals can know the macroeconomic indexes in advance,they can also make more reasonable decisions about personal savings and investment.From this aspect,forecasting macroeconomic indexes has important significance of research.There are many models or methods for predicting macroeconomic indexes,among which the vector autoregressive model(VAR)is a typical representative.With the deepening of research,scholars have found that the vector autoregressive model(VAR)is over-parameterized,that is,if you want to achieve better estimation,the establishment of a vector autoregressive model often requires a larger sample size,which brings great difficulties to the estimation of parameters of model,and the macroeconomic index is updated on a monthly,quarterly or annual frequency.Limited by statistical costs,the sample size of macroeconomic data is often small.On the other hand,the vector autoregressive model(VAR)is a linear model,and the macroeconomic indexes may have some nonlinear relationships.In order to comprehensively consider these aspects,this article combines the TS-PCA method of Yao Qiwei and the Flexible Transmitter Network model of Zhou Zhihua's team to obtain the TSPCA-FTNet model.The model not only reduces the number of parameters,but also takes into account the nonlinear relationships of the data,thus it can achieve better prediction results.In the empirical part,this article refers to relevant literatures and obtains 12 macroeconomic indexes of China from January 1997 to October 2020 from the statistical database of China Economic Network.This paper uses the data from 1997 to 2018 and 1997 to 2019 as the training set,and the data from 2019 and 2020 as the test set respectively.We make a comprehensive comparison of vector auto regression(VAR),Bayesian vector auto regression(BVAR),recurrent neural network(RNN),long and short-term memory neural network(LSTM),TS-PCA,TSPCA-RNN,TSPCA-LSTM and TSPCA-FTNet from four perspectives: model prediction accuracy,prediction stability,number of model parameters and running time of these 8 models.The results show that the TSPCA-FTNet method can achieve the best prediction accuracy,better prediction stability and shorter running time with fewer parameters.Finally,this article uses the TSPCA-FTNet model and the U-MIDAS model to predict the macroeconomic index in the fourth quarter of 2020.The results of forecast show that my country's macroeconomic indexes are relatively stable,and the current year-on-year GDP growth rate in the fourth quarter of 2020 is expected to be 8%.Finally,we summarize the research content and results of this article,and propose the suggestions and directions for improvement.