A note on a Bayesian order determination procedure for vectorautoregressive processes (Q1383247)

The most widely used model in time series analysis is the vectorautoregressive linear model (VAR model). In the empirical application of this process one of the problems to solve is the question of order selection, i.e. the determination of the number of lags. This paper presents an analytical order determination procedure from a Bayesian viewpoint where straightforward incorporation of prior knowledge about all model parameters (including the order) is possible. We will describe our Bayesian order identification criterion, give some remarks on the computation and compare the method with the standard criteria AIC and BIC in some simulations. In a final section some advice on the implementation of prior information is given.