A High-dimensional VAR Model Averaging Method And Application Based On Factor Analysis

In recent years,high-dimensional data modeling has been widely concerned by statisticians.In practice,high-dimensional time series data is very common,such as stock index,macroeconomic indicators,port throughput data,etc.It has great significance to reduce the dimension of the data and then build a model to achieve accurate predictions.Recently,the model averagging method is recommended by many scholars because of its excellent theoretical properties and application performance.This paper focuses on the averaging method of high-dimensional time series model.Firstly,this paper considers using factor model to reduce the dimension of high-dimensional data to a low dimensional dynamic part and a static part.Then we use vector autoregressive model(VAR)to discribe the reduced data.A stepwise forward model averaging criterion(SFMA)is developed in the framework of VAR model.It is proved that the criterion minus a constant is an unbiased estimate of the stepwise forward expected loss function.This paper compares the small sample performance of SFMA with some commonly used model averaging methods,such as SAIC,SBIC and MMA.It is shown that SFMA is the best method.Finally,this method is applied to predict container throughput of seven major ports in China,and the results also show the superiority of SFMA.