On the power of panel cointegration tests: a Monte Carlo comparison

This paper enlarges on Karlsson and Lothgren's results [Economics Letters, 66 (2000) 249] on panel unit root tests to panel cointegration tests. We show that for a homogeneous panel, Kao's tests [Advances of Econometrics, 15 (1999) 7] have higher (lower) power than Pedroni's tests [Oxford Bulletin of Economics and Statistics, Special Issue (1999) 653] when a small-T (high-T) number of observations are included in the panel and both tests show better performance than Larsson et al.'s test [Econometrics Journal, 4 (2001) 109]. In addition, depending on the T-dimension of the panel, cointegration tests can have high power when a small or high fraction of the relationships are cointegrated. This result suggests that when rejecting the null hypothesis of no cointegration for the whole panel not all the relationships can be really cointegrated. (C) 2003 Elsevier Science B.V. All rights reserved.