On Selection of Cross-Section Averages in Non-stationary Environments

Information criteria (IC) have been widely used in factor models to estimate an unknown number of latent factors. It has recently been shown that IC perform well in Common Correlated Effects (CCE) and related setups in selecting a set of cross-section averages (CAs) sufficient for the factor space under stationary factors. As CAs can proxy non-stationary factors, it is tempting to claim such generality of IC, too. We show formally and in simulations that IC have a severe underselection issue even under very mild forms of factor non-stationarity, which goes against the sentiment in the literature.