IZA

Logo
Fixed Effects Bias in Panel Data Estimators
by Hielke Buddelmeyer, Paul H. Jensen, Umut Oguzoglu, Elizabeth Webster
(May 2008)

Abstract:
Since little is known about the degree of bias in estimated fixed effects in panel data models, we run Monte Carlo simulations on a range of different estimators. We find that Anderson-Hsiao IV, Kivietís bias-corrected LSDV and GMM estimators all perform well in both short and long panels. However, OLS outperforms the other estimators when the following holds: the cross-section is small (N = 20), the time dimension is short (T = 5) and the coefficient on the lagged dependent variable is large (γ = 0.8).
Text: See Discussion Paper No. 3487