Panel Stationarity Tests with Cross-sectional Dependence

We present a test of the null hypothesis of stationarity against unit root alternatives for panel data that allows for arbitrary cross-sectional dependence. We treat the short run time series dynamics non-parametrically and thus avoid the need to fit separate models for the individual series. The statistic is simple to compute and is asymptotically normally distributed, even in the presence of a wide range of deterministic components. Taken together, these features provide a generally applicable solution to the problem of testing for stationarity versus unit roots in macro-panel based data. The test is applied to assess the validity of the purchasing power parity hypothesis and finds significant evidence against the hypothesis being true. Department of Economics, University of Melbourne, 3010, Australia. Fax: +61 3 8344 6899. Email: [email protected].