Consistent estimation of dynamic panel data models with time-varying individual effects

This paper proposes a new specification for dynamic panel data models, where unobserved heterogeneity is modeled by the sum of the usual additive individual effect, and a multiplicative, time varying individual effect. We show that usual GMM estimators based on first-difference or quasi-difference transformations are generally inconsistent with our model specification, and we propose a consistent GMM estimation procedure based on a double-difference transformation. Small sample properties of alternative GMM estimators are investigated through Monte Carlo experiments.