On the use of panel unit root tests on cross-sectionally dependent data: an application to PPP

A Monte Carlo exercise demonstrates the different size distortions that two of the most commonly used panel unit root tests have when the sections of the panel are affected by correlated errors, when they are cointegrated, or both. For a specific form of sectional correlation, the limiting distribution is derived and asymptotic normality of the test statistic is established. To determine the nature of contemporaneous cross-sectional correlation in real data, covariance matrix estimation techniques are discussed and an appropriate bootstrap method for the estimation of standard errors is suggested. In an application to a panel of real exchange rates it is found that both aforementioned dependencies are present, and therefore the results of panel unit root tests if applied at all should be interpreted accordingly.