Interpreting cointegrating vectors and common stochastic trends

It often proves difficult to give a satisfactory economic interpretation to estimated cointegrating vectors derived from the maximum likelihood estimation of unrestricted vector correction models. This paper shows that this is because, without introducing a priori information, they are not identified. This is even true where there is only a single cointegrating vector. It is also shown that the common stochastic trends derived using VAR analysis in the presence of cointegration are not identified, nor can they be obtained uniquely from the estimated cointegrating vectors. The implication is that cointegration analysis needs to take account of structural restrictions after all. Consequently it is likely to be of less use in econometric model building than was first thought.