Quasi Maximum-Likelihood Estimation of Dynamic Panel Data Models

摘要

This paper establishes the almost sure convergence and asymptotic normalityof levels and differenced quasi maximum-likelihood (QML) estimators of dynamicpanel data models. The QML estimators are robust with respect to initialconditions, conditional and time-series heteroskedasticity, andmisspecification of the log-likelihood. The paper also provides an ECMEalgorithm for calculating levels QML estimates. Finally, it uses Monte Carloexperiments to compare the finite sample performance of levels and differencedQML estimators, the differenced GMM estimator, and the system GMM estimator. Inthese experiments the QML estimators usually have smaller --- typicallysubstantially smaller --- bias and root mean squared errors than the panel dataGMM estimators.