Effect of cross correlations in error terms on the model selection criteria for the stationary VAR process

Abstract

In this paper we investigate the finite sample properties of several model selection criteria in case of bivariate Vector AutoRegressions (VARs) of order one and two by a simulation study, particularly focusing on the effects of the degree of cross correlation in the error terms in combination with the values of more or less extreme values of the characteristic roots of the VAR-process. The Monte Carlo experiments show that the degree of cross correlation has an influence on the model selection criteria.