Estimating the Number of Common Factors in Serially Dependent Approximate Factor Models∗

The Bai-Ng criteria are shown to overestimate the true number of common factors when panel data exhibit considerable serial dependence. We consider filtering the data before applying the Bai-Ng method as a practical solution to the problem. Despite possible bias and model misspecification, the AR(1) least squares dummy variable (LSDV) filtering method is shown to be consistent for panels with a wide range of serial correlation, and a combination of the LSDV and the first-difference filtering methods is shown to be consistent in both weakly and strongly serially correlated panels. According to simulations these filtering methods considerably improve the finite sample performance of the Bai-Ng selection methods. We illustrate the practical advantages of LSDV filtering by considering three economic panels that exhibit moderate serial dependence. In each case, LSDV filtering yields a reasonable factor number estimate, whereas conventional methods do not.