Performance of multiple testing in nonstationary panels and empirical applications

This paper investigates the small-sample performance of the multiple testing based on Romano and Wolf’s (2005) testing framework in nonstationary panel settings and shows some empirical applications in macroeconomics. In empirical analysis, while researchers apply panel-based unit root/cointegration tests, when the tests reject the joint null hypothesis (which assumes the presence of unit roots or the absence of cointegration for all cross-sectional units), they often face the difficulty of identifying which individual null hypothesis is false. Under such circumstances, the application of the multiple testing method is a useful alternative way to individually identify false null hypothesis. Moreover, this paper considers the method that can deal with a time series with a structural break at unknown dates or with nonlinearity. The multiple testing method is also designed to overcome the multiplicity problem (Savin 1984; Westfall and Young 1993), which means that when several individual tests are implemented simultaneously, even if all the null hypotheses are true, at least one hypothesis tends to be rejected at a much higher significance rate than the prespecified rate. The simple aggregation of some individual test results may lead to this problem. Therefore, the use of the multiple testing is expected to allow us to control the familywise error rate, which is defined as the probability of erroneously rejecting at least one null hypothesis in the entire test, under a prespecified desired rate (e.g., 5%). In this paper, the Monte Carlo simulation study shows the better performance of the multiple testing than that of an individual test.