We investigate analytically and via Monte Carlo simulations the effects of the inclusion of irrelevant variables in the statistical model, on the cointegration analysis of Johansen (1988, 1991). We show that overspecifying the statistical model does not affect inference about the cointegrating rank, as Johansen (1996, p. 42) suggests. Estimators of the cointegrating vectors and adjustment coefficients remain consistent, but efficiency is affected for the cointegrating vectors. Simulations show that overspecifying the statistical model can considerably reduce the power of tests of cointegrating rank. <br><br> Keywords; cointegration, misspecification, irrelevant variables, asymptotics, monte carlo
