This study reviews recent developments in the analysis of non-stationary panels, focusing on empirical applications of panel unit root and cointegration tests in the context of PPP. It highlights various drawbacks of existing methods. First, unit root tests suffer from severe size distortions in the presence of negative moving average errors. Second, the common demeaning procedure to correct for the bias resulting from homogeneous cross-sectional dependence is not effective; more worryingly, it introduces cross-correlation when it is not already present. Third, standard corrections for the case of heterogeneous cross-sectional dependence do not generally produce consistent estimators. Fourth, if there is between-group correlation in the innovations, the SURE estimator is affected by similar problems to FGLS methods, and does not necessarily outperform OLS. Finally, cointegration between different groups in the panel could also be a source of size distortions. Some empirical guidelines are offered to deal with these problems, but the study concludes that panel methods are unlikely to solve the PPP puzzle.
