Identification Restrictions and Posterior Densities in Cointegrated Gaussian VAR Systems

We derive the postenor density of the cointegrating coetficients in a Gaussian VAR system. The density does not belong in general to a family of densities with known properties. If there is one cointegrating vector, the density belongs to the class of poly-t densities. It is integrable if the coefficients are identified and it has finite moments to the order of overidentification. The identifying restrictions we consider are linear restrictions on the cointegrating vectors. The structure or the posterior density is exploited to implement Monte Carlo integTi\tion nwthods that are needed when there is more than one cointegrating veetor. The paper contains two empirical illustrations.