This paper investigates the effects of the omission of relevant variables from the statistical model on cointegration analysis, proposed by Johansen (1988, 1991). We show that underspecification of the statistical model leads to either failure in detecting cointegration or underestimation of the cointegrating rank. Although in the underspecified statistical model the estimator of the detected cointegrating vectors is shown to be consistent, this is not the case for the estimators of the adjustment coefficient matrix and the variance of the error term. The asymptotic analysis is supplemented by a Monte Carlo experiment and an empirical example. <br><br> Keywords; cointegration, omitted variables, asymptotics, Monte Carlo
