Asymptotically Normal Unit Root Testing: Extensions to Panel Data and Structural Breaks 1 Uwe Hassler, Matei Demetrescu and Adina Tarcolea Goethe-University Frankfurt 2 November 12, 2004 Abstract We study the asymptotically normal LM integration test. It is ob- served that the test is robust to shifts in the mean under the null hypothesis, while it retains nice power properties under the alterna- tive. If the occurrence of a mean shift is taken into account, power will increase. Assuming a finite number of independent cross-sections the sum over individual test statistics will again follow a standard distri- bution as time dimension tends to infinity. This is true for unbalanced panels and for panels where in some cross-sections structural breaks may have been accounted for. We allow for cross-correlated panels and propose a simple correction. The validity of our asymptotic results is assessed in extensive Monte Carlo experimentation. Keywords: LM type tests; mean shifts; panel data; cross-correlation; finite N large T asymptotics; Monte Carlo evidence 1 Introduction Since the seminal work by Dickey and Fuller [DF] (1979), formal tests for unit roots (or integration of order one, I(1)) have become standard in applied time series analysis and econometrics. Moreover, many theoretical extensions of 1 Earlier versions of the paper were presented at the 14th EC 2 conference, London, December 2003, at the 11th International Conference on Panel Data, Texas A&M Univer- sity, June 2004, and at the 59th European Meeting of the Econometric Society, Madrid, August 2004. 2 Statistics and Econometric Methods, Goethe-University Frankfurt, Gr¨afstr. 78, D-60054 Frankfurt, Germany, Tel: +49.69.798.23660, Fax: +49.69.798.23662, email: hassler@wiwi.uni-frankfurt.de. 1