View article

The empirical analysis of economic behavior often entails specifying dynamic econometric models; that is, models with lagged dependent variable (s) among the regressors. As in time-series context, when the model is dynamic, standard estimation methods based on least squares generally do not lead to estimators having good properties. Indeed, for dynamic panel data models, methods such as OLS or the Within estimators are not consistent. This results from the fact that, due to the unobserved effects, the lagged dependent variable and the disturbance terms are correlated.

Therefore one has to resort to alternative methods. The most commonly used approach is that of GMM, relying on a properly defined set of instrumental variables, or equivalently, a set of orthogonality conditions (see Chap. 4). A large part of this chapter is devoted to the presentation of a number of such consistent estimators. We consider …