Topics in generalized method of moments estimation with application to panel data with measurement error

This dissertation investigates three topics in generalized method of moments (GMM) estimation.;Firstly, I address the application of GMM to the measurement error problem in panel data. Griliches and Hausman (1986) and Wansbeek (2001) are two major existing works in this area. I study some issues that are not addressed or not fully addressed by these two approaches, but are crucial to their applications, these include: how to choose instruments; what is the optimal weighted instrumental variable estimator; how to explicitly construct GMM estimators; how to remove the redundancy of the moment conditions constructed by Wansbeek (2001); and whether optimal GMM estimators exist. I unify Griliches and Hausman's and Wansbeek's approaches by establishing their equivalence. I then consider models with exogenous regressors and models with nonclassical assumptions. I also extend the GMM methods to models where multiple covariates are measured with errors, and to unbalanced panel data models. Three applications illustrate the implementation of the GMM methods.;Secondly, I extend the GMM method to models where some of the moment conditions are redundant. I show that in such cases, any reflexive generalized inverse of the relevant covariance matrix is an optimal weighting matrix. The convergence and small sample bias of the estimator of the optimal weighting matrix are also discussed.;Finally, I propose the weighted method of moment (WMM) for estimation in moment condition models. I show that the WMM estimator using a particular set of weights is as efficient as the efficient GMM estimator. The WMM estimator is also shown to be the same as the GMM estimator in linear models. Limited simulation studies reveal that the WMM estimator has better small sample properties than the GMM estimator and the EL estimator for certain models.