Using a Monte Carlo approach, we find that the bias of LSDV for dynamic pan
el data models can be sizeable, even when T = 20. A corrected LSDV estimato
r is the best choice overall, but practical considerations may limit its ap
plicability. GMM is a second best solution and, for long panels, the comput
ationally simpler Anderson-Hsiao estimator performs well. (C) 1999 Elsevier
Science S.A. All rights reserved.