The finite sample behaviour is analysed of particular least squares (LS) andmethod of moments (MM) estimators in panel data models with individual effectsand both a lagged dependent variabIe regressor and another explanatory variabIewhich may be affected by lagged feedbacks from the dependent variabIe. Asymp-totic expansions indicate that the order of magnitude of bias of (generalized) MMestimators tends to increase with the number of moment conditions exploited. Forvarious estimation procedures we examine the analytical effects of feedbacks andother model characteristics such as prominence of individual effects. Simulation re-sults corroborate our theoretical findings and show that in small samples of modelswith dynamic feedbacks none of the techniques examined dominates. However, asimple bias corrected LS estimator which presupposes strict exogeneity is found tobe rather robliSt, showing often smaller root mean squared errors than GMM.