Abstract
The note comments on the application of Granger non-causality tests of short-run vector error correction models as attempted by Ghali. It is noted that when the dimension of VAR is greater than two, block Granger causality needs to be performed, and that checking variable to variable causality has little meaning in multivariate context as the covariance matrix is positive definite. It is also noted that identification of cointegrating relationships does not provide information on short-run structural relationships, which requires modelling of contemporaneous innovations.