Estimation of Spatial Autoregressive Econometric Models with Continous and Limited Dependent Variables

Ohio State University Research Foundation -Do Not Use, Columbus OH

Investigators

Abstract

The field of spatial econometrics is concerned with the use of statistical and econometric techniques to handle spatial effects in multiregional economic models and economic interaction of agents located in space. Proper spatial econometric models need to be developed to empirically validate modern spatial economic theories. Further empirical developments are motivated by the growing importance of Geographic Information Systems (GIS) in regional and urban policy analyses. This project develops econometric methodologies for the estimation and testing of complex spatial econometric models including spatial models with limited dependent variables. This project proceeds in several directions. Statistical properties of some popular estimation methods in the literature are often assumed without detailed investigation into possibly distinctive features of a spatial econometric model. While some claims are correct under certain spatial scenarios, they might not be so in others. The existing literature on spatial econometrics has mainly focused on models with small group interaction but models with large group interaction have many interesting potential applications. This project continues the investigation of statistical properties of popular estimators for spatial autoregressive models with large group interaction. Estimators for spatial models with large group interaction have quite different statistical properties from those of models with small group interaction. This project explores alternative models to capture social interaction effects and addresses issues of random sample and possible incomplete spatial interactions in spatial models. In addition, statistical procedures are developed for distinguishing spatial models from social effect models. Spatial models with agents making interactive discrete choices are useful for research on innovation diffusion processes. The estimation of a discrete choice model with spatial interaction can be quite challenging as its likelihood function involves high dimensional integral. The dimension of integration can be as large as the number of sample observations. In order to develop these models to their full capacities, computationally tractable methods must be developed. This project develops estimation methods based on simulation estimation methodologies. The effectiveness of various simulation estimation methods, which include the method of simulated maximum likelihood, the method of simulated EM algorithm, the method of simulated scores, and the Gibbs sampler, are investigated. Statistical properties of those estimators are studied. In addition to estimation, test statistics for spatial correlations and diagnostics are also developed. This project develops computationally tractable generalized method of moments for the estimation of spatial autoregressive models of any finite order with or without the presence of exogenous variables. Asymptotic properties of such estimators are investigated. Dynamic discrete choice models capture various notions of dynamic effects, state dependence, heterogeneity, and spurious correlation in a panel data setting. This project generalizes existing models to incorporate possible contemporaneous and intertemporal spatial interaction effects. Spatial dynamics in discrete choices are of special interest. Special attention is paid to the specification and estimation of such panel data models. Even though the development of econometric methods is the main focus of this project, empirical studies with panel data from developing countries illustrate the useful of the new models and the feasibility of the methodologies.