Short-run parameter changes in a cointegrated vector autoregressive model

This paper addresses the question of whether a conventional approach to cointegration is applicaple to the case where changes are allowed in the parameters for the short term dynamics. We reparametrise a vector autoregressive model such that the short-run parameters exhibiting changes at known points are explicitly given. We then show that the likelihood ratio test statistic for cointegration rank is based on reduced rank regression and has the usual asymptotic distribution. An empirical illustration using US gasoline prices is presented.