Multiple comparisons and combinations of significance in nonstationary panel data

The use of panel data has established itself as a means to increase insufficiently large sample sizes in empirical macroeconomics. These studies thus involve multiple individuals for which it is sought to validate (or reject) economic concepts that depend on the stationarity of particular economic variables. Standard tools for checking this stationarity condition are unit root or cointegration tests. Applied to each of the individuals in the panel, many of these testing procedures are able to provide individual-specific p-values which can in turn be used to test for (non)stationarity by formulating either a panel hypothesis or individual-specific hypotheses. This thesis is concerned with the application of p-value combinations which offer a flexible approach to test a panel hypothesis as well as multiple testing procedures which enable meaningful inference on each individual in the panel. In extensive Monte Carlo simulations including unbalanced and cross-correlated panels, the author examines the finite sample behavior of three popular p-value combinations individually as well as for combinations of these tests. The latter, which is an innovative approach suggested by Cheng and Sheng (2011), is extended to a combination of three methods and is shown to work well in a cointegration setting. Furthermore, the author demonstrates the practical relevance of both testing approaches by applying them to OECD interest rate data. In particular, the benefit of the multiple testing approach to consider each individual-specific hypothesis separately permits a more informative statement regarding the (non)stationarity of the panel.