On the finite-sample size distortion of smooth transition unit root tests

The finite-sample size properties of smooth transition unit root tests are examined when applied to unit root processes subject to breaks in either level or drift. In contrast to the weighted symmetric and recursively mean-adjusted unit root tests which have been shown to be robust in these circumstances, it is found that the empirical sizes of smooth transition tests are dependent upon the form, location and magnitude of the break imposed. It is concluded that while smooth transition unit root tests are capable of capturing breaks under an alternative hypothesis of stationarity, spurious rejection can occur when breaks occur under the null.