Multiple comparisons and combinations of significance in nonstationary panel data

Viac o knihe

The use of panel data has become a key method for enhancing small sample sizes in empirical macroeconomics, allowing for the validation or rejection of economic concepts reliant on the stationarity of specific variables. Standard tools for assessing stationarity include unit root and cointegration tests, which can yield individual-specific p-values when applied to each panel member. These p-values facilitate testing for (non)stationarity through either panel or individual-specific hypotheses. This thesis explores the application of p-value combinations, providing a flexible means to test panel hypotheses and enabling meaningful inference for each individual in the panel. Through extensive Monte Carlo simulations involving unbalanced and cross-correlated panels, the author analyzes the finite sample behavior of three widely-used p-value combinations, both individually and in combination. An innovative approach, extending the work of Cheng and Sheng (2011), effectively integrates three methods in a cointegration context. The practical significance of these testing approaches is illustrated through their application to OECD interest rate data. Notably, the multiple testing approach allows for a more nuanced understanding of each individual hypothesis, enhancing insights into the (non)stationarity of the panel.