Monte Carlo tests of cointegration in a bivariate normal common factor system

In this paper the power and size distortions of three representative cointegration tests, the Augmented Dickey-Fuller (ADF), Z-super-ˆα and Stock-Watson (SW) statistics are evaluated in some large-scale Monte Carlo simulations. Consistent with previous evidence, the ADF-statistic is seen to be sensitive to the validity of the common factor restriction in the data generating process of our study. The SW-statistic is stable but not high in terms of power. The Z-super-ˆα -test shows the best balance between power and size. In the light of previous evidence, the test statistics seem to be sensitive to the DGP used more than to parametric variation of the active DGP.