Granger Causality in the Presence of Structural Changes

We focus in these paper on Granger shifts or structural breaks. We show that when the assumption of parameter constancy is violated, due to occurrence of structural breaks, Granger causality tests can provide misleading inference about the underlying relationship of causality. We consider a Bayesian model for the detection of structural breaks which can make Granger causality tests ‘robust’ to the presence of structural instabilities in the sample. An application of the method to the Canadian series of GNP and M1 is presented.