Incidental trends and the power of panel unit root tests

摘要

The asymptotic local power of various panel unit root tests is investigated. The (Gaussian) power envelope is obtained under homogeneous and heterogeneous alternatives. The envelope is compared with the asymptotic power functions for the pooled t-test, the Ploberger and Phillips [2002. Optimal testing for unit roots in panel data. Mimeo] test, and a point optimal test in neighborhoods of unity that are of order n~(-1/4)T~(-1)and n~(-1/2)T~(-1), depending on whether or not incidental trends are extracted from the panel data. In the latter case, when the alternative hypothesis is homogeneous across individuals, it is shown that the point optimal test and the Ploberger-Phillips test both achieve the power envelope and are uniformly most powerful, in contrast to point optimal unit root tests for time series. Some simulations examining the finite sample performance of the tests are reported.