Level-based estimation of dynamic panel models

Abstract

This paper develops an alternative estimator for linear dynamic panel data models based on parameterizing the covariances between covariates and unobserved time-invariant effects. A GMM framework is used to derive an optimal estimator based on moment conditions in levels, with no efficiency loss compared to the classic alternatives like Arellano and Bond (1991) and Ahn and Schmidt (1995, 1997). Still, we show analytically and by Monte Carlo simulations that the new procedure leads to efficiency improvements for certain data generating processes. The framework also leads to a very simple test for unobserved effects.