Non-Causality in VAR-ECM Models with Purely Exogeneous Long-Run Paths.

We propose in this paper a framework based on a canonical representation of the long run matrix, which can constitute a basis for Granger non-causality testing in a VAR-ECM model using asymptotically Chi-square distributed Wald test statistics, and that unlike Giannini and Mosconi (1992), permits to clearly distinguish the nullity of some parameter blocks we can always achieve without any loss of generality, of the nullity of the parameter blocks resulting from the non-causality property. This canonical representation requires to determine the specific rank of a particular sub-matrix, which can be done using sequential test procedure, whose properties are analysed in small and large samples with Monte Carlo experiments.