Testing for the Cointegration Rank when Some Cointegrating Directions are Shifting

In this paper we propose a multivariate analysis of a cointegrated vectorial autoregressive model with structural breaks affecting the cointegrating vectors. These changes are previously recognized using a single-equation methodology. Asymptotic properties of the breaks dates estimators allow us to implement a full information maximum likelihood analysis with the breaks identified considered as fixed. Thus, a VECM is estimated, providing a relevant framework to test for non-causality and neutrality tests, as well as impulse response analysis of the dynamics. The methodology is applied to the trivariate system analysed by Gregory and Hansen (1996), with a money variable, an interest rate and the output for the US over the 1960-1990 period.