Cointegration tests in the presence of structural breaks

Structural breaks in stationary time series can induce apparent unit roots in those series. Thus, using recently developed recursive Monte Carlo techniques, this paper investigates the properties of several cointegration tests when the marginal process of one of the variables in the cointegrating relationship is stationary with a structural break. The break has little effect on the tests' size. However, tests based on estimated error correction models generally are more powerful than Engle and Granger's two-step procedure employing the Dickey-Fuller unit root test. Discrepancies in power arise when the data generation process does not have a common factor.