GMM estimation with cross sectional dependence (Q113633): Difference between revisions

This paper presents practical estimation methods for econometric models of cross-sectional dependence. These models use information on agents interdependence -- their economic distance -- to characterize dependence structures without parametric assumptions. The restriction on the dependence structure is that it can be characterized as a configuration of points in the Euclidean space, presumably of a low dimension relative to the sample size. Two important restrictions are imposed on the process determining which agents and locations are observed. Spatial models of dependent data show complications when using a GMM estimator (GMM= Generalized Method of Moments). GMM estimators remain consistent with such dependent data but the covariance matrix estimation procedure and the GMM estimators are different with the spatial model of dependence than with a time series model of dependence. Here, covariance matrices are estimated nonparametrically. The paper is organized as follows: Section 1 discusses the notion of economic distance, the complications with GMM estimators, and applications in rural developing economies. Section 2 provides the basic econmetric model, presents large sample results for GMM estimators and covariance matrix estimators. Section 3 shows a simple model where agents live on an integer lattice, and economic distances are known. A more general model allows agents to live at real-valued locations with measurements of distances to be distorted by bounded errors (section 4). Section 5 provides an empirical illustration of the potential impact of using this basic econometric model. Section 6 concludes limitations and possible extensions.