Generalized Moments (gmm) Estimation Of The Impact Analysis

Detection of outliers and influential observations is very important in assessing the adequacy of model assumptions and identifying unusual features of the data that may seriously influence the conclusions to be drawn or may require special attention. The detection of influential observations is called influence analysis, which is an important part of statistical diagnosis. In econometrics, particularly in macroeconomics and fiance, the generalized method of moments (GMM) is a generic method for estimating parameters in statistical models. This thesis studies the problem of detecting outliers and influential observations in GMM estimators. Two topics, namely case deletion diagnostics and local influence analysis are studied.Under case deletion, the influence of a subset of observations on the GMM estima-tors is investigated. We derived the deletion formula for GMM estimator in a unified from. Some influence measures based on Cook’s distance are obtained and an approx-imate update formulae is developed. The results are applied to efficient instrumental variable estimation and dynamic panel data model. In addition, the generalized residuals and leverage measures in the framework of GMM estimators are defined and discussed. These measures can be used to study the influential observations for GMM estimators.In local influence analysis, we focus on simultaneously perturbing the moment con-ditions. Generalized influence function and generalized Cook distance for GMM estima-tion under this perturbations are derived and local diagnostic measures are obtained to assess joint influence of observations.Particularly, the instrumental variables (Ⅳ) estimation is an important application of GMM estimators. Influence function (IF) can be used to detect observations that are highly influential on the estimator. In the chapter5, we discuss the behavior of influence function for IV estimation and compare it with the case deletion.