Testing cross-section correlation in panel data using spacings

This article provides tools for characterizing the extent of cross-section correlation in panel data when we do not know a priori how many and which series are correlated. Our tests are based on the probability integral transformation of the ordered correlations. We first split the transformed correlations by their size into two groups, then evaluate the variance ratio of the two subsamples. The problem of testing cross-section correlation thus becomes one of identifying mean shifts and testing nonstationarity. The tests can be applied to raw data and regression errors. We analyze data on industrial production among 12 OECD countries, as well as 21 real exchange rates. The evidence favors a common factor structure in European real exchange rates but not in industrial production.