Identifiability of cointegrated systems

This paper stipulates conditions for identifiability of the parameters of a cointegrated VAR model under general linear restrictions, possibly including cross-equation restrictions. An algorithm is given to obtain the maximum likelihood estimators under such restrictions on both the cointegrating vectors and the adjustment parameters. Then the asymptotic distribution of the estimator and of the likelihood ratio statistic for the over-identifying restrictions is given. The importance of the observed information matrix for identification issues is emphasized