China's Macroeconomic Data Forecast Based On Mixed Frequency Bayesian Vector Auto-Regression Model

The cyclical fluctuation of the economy is the content of macroeconomics research.The prediction and analysis of economic indicators provide important support for the decision-making of national economic regulation.The VAR model is one of the most commonly used models in macroeconomic research.When the traditional VAR model is used for macroeconomic research,the monthly data is often directly added to the quarterly data or the annual data,and then the VAR model containing only the quarterly data or the annual data is used.This simple summation of monthly data leads to the loss of high frequency data information.In order to make full use of the high-frequency monthly data information,the prediction and analysis of the national macroeconomic indicators are more accurate.This paper regards the low-frequency data as the missing of the high-frequency data,and establishes a state space model,which can be actually observed in each period.The low-frequency quarterly data and the high-frequency monthly data are used as observation variables,and the potential monthly data of each variable is used as the state variable.The VAR model of the mixed frequency is used as the state equation of the state space model,and the correspondence between the state variable and the observed variable is used as the measurement equation.In order to overcome the problems caused by too many parameters of the VAR model,the paper uses the Minnesota prior Bayesian estimation method to obtain the posterior distribution of the parameters of the VAR model.The Kalman filter and Kalman smoothing method are used to obtain the potential.The posterior distribution of the potential state variables is obtained by Kalman filtering and Kalman smoothing.Then according to the posterior distribution of VAR model parameters and the posterior distribution of potential state variables,Gibbs sampling method is used to obtain the estimation of VAR model and state variables,so as to predict and analyze national macroeconomic data.In the empirical part,this paper uses the Mixed Frequency Bayesian Vector Auto Regressive model(MF-BVAR)to predict China's quarterly GDP growth rate.There are three main findings:(1)It is found that the MF-BVAR model has a good performance in the case of real GDP growth rate growth,decline and stability.(2)The increase in monthly data can significantly improve the accuracy of the MF-BVAR model for China's quarterly GDP growth forecast,which reflects the advantages of the mixed data model.(3)The prediction effect of the MF-BVAR model is significantly better than that of another commonly used mixing data model,U-MIDAS model.