TESTING FACTORS IN CCE

One of the most popular estimators of interactive effects panel data models is the common correlated effects (CCE) approach, which uses the cross-sectional averages of the observables as proxies of the unobserved factors. The present paper proposes a simple test that is suitable for testing hypotheses about the factors in CCE and that is valid provided only that the number of cross-sectional units is large. The new test can be used to test if a subset of the averages is enough to proxy the factors, or if there are observable variables that capture the factors. The test can also be used sequentially to determine the smallest set of averages needed to proxy the factors.