Common correlated effects and international risk sharing

International risk sharing has been among the most actively researched areas of macroeconomics for the last two decades. Empirical contributions in this field make extensive use of so called "consumption insurance" tests evaluating the extent to which idiosyncratic shocks in income get transferred to consumption. A prerequisite of such a test is the isolation of country specific variation in the data. We show that the cross-sectional demeaning technique frequently used in the literature is in general inadequate to eliminate global factors from a panel data set, and can lead to misleading inference. We argue that international risk sharing tests should instead be based on a method that more reliably deals with global factors. We claim and illustrate in our empirical application that the fairly simple common correlated e ects estimator for cross-sectionally dependent panels introduced by Pesaran (2006), and Kapetanios et al. (2010) is a tool that satisfies this requirement.