The Relationship between Power and Level for Generic Unit Root Tests in Finite Samples.

The author shows that the maximum power of a generic unit root test against any stationary alternative is equal to the true level of the test. He then uses Monte Carlo methods to investigate the implications for several such tests. He shows patterns of rejection probabilities over a variety of unit root and stationary processes. He discusses the implications of these results for some of the uses of unit root tests in applied work. Copyright 1992 by John Wiley & Sons, Ltd.