pairwise likelihood inference for spatial regressions estimated on very large datasets

This paper proposes a pairwise likelihood specification of a\r\nspatial regression model that simplifies the derivation of the\r\nlog-likelihood and leads to a closed form expression for the\r\nestimation of the parameters. With respect to the more traditional\r\nspecifications of spatial autoregressive models, our method avoids\r\nthe arbitrariness of the specification of a weight matrix, presents\r\nanalytical and computational advantages and provides interesting\r\ninterpretative insights. We establish small sample and asymptotic\r\nproperties of the estimators and we derive the associated Fisher\r\ninformation matrix needed in confidence interval estimation and\r\nhypothesis testing. We also present an illustrative example of\r\napplication based on simulated data.