| Large-scale VAR models exist widely in economic,financial,monetary,medical,dem ographic and other fields.In recent years,the modeling of large-scale VAR models has attr acted widespread attention from economists.Numerous studies have shown that it is an eff ective research method to model large VAR model data using Bayesian estimation.With th e rapid development of computer technology,the Bayesian VAR(BVAR)model has becom e more and more widely used.Its advantage is that it can establish a flexible model that is r elevant to experience and capture key data characteristics,but it is not so flexible that it is s o that it is seriously overparametric to help estimate.The impact of multiple variables in th e whole model improves the practicability and accuracy of the VAR model and reduces pre diction errors.In this context,it is very important to determine the prior information of the BVAR model.Based on the advanced Minnesota adaptive shrinkage a priori and modern MCMC sampling technology,this paper studies the improvement of two superparameters o f this transcendental.The main work of this article is described in detail below.In order to obtain a more accurate prediction effect,the first part of this article studies the optimisation of the shrinkage parameters of the BVAR model.The data-driven estimati on method is used for the shrinkage parameters corresponding to the lag order,the NormalGamma contraction a priori is introduced,and the data information is used to determine dif ferent lag orders.For the different contraction degrees of the parameter matrix,a more flex ible Bayesian estimation method for the high-dimensional VAR model is proposed,a more flexible Minnesota adaptive contraction a priori is proposed,And through a large real data set,it is verified that the improved a priori has better predictive performance compared wit h the BVAR package that has been developed.Considering that the a prior lag order is fixed to have a certain subjectivity,in the seco nd part of this article,the optimisation of the lag order of the BVAR model is further studie d,using the method of maximising the approximate marginal likelihood of the model,appl ying the grid search technology to find the optimal lag order,and finally applying the mode l to a large real data set.The results show that the lag order selected by the method of maxi mising marginal likelihood can improve the prediction accuracy of more than 90% of varia bles and greatly improve the prediction ability of the model. |
