Spatial logit for large samples with local spatial lag and regional spatial random effects using linearized GMM: an application to land use models
Résumé
Following Klier and McMillen (2008), we propose a linearized spatial logit GMM (LGMM) estimator for particular SAR models combining 2 spatial parameters and two spatial weights matrices. This model, a double spatial filter, is particularly suitable for modeling spatial binary choices where land use choices are influenced both by close neighbors and by neighboring municipalities’ urban policies, and/or by urban public policy at a higher level (county, region). We show that under realistic empirical conditions, the estimators of full GMM are consistent and asymptotically normal. Moreover, using MC experiments with various forms of spatial weights matrix including spatially aggregated locations, we show that the proposed LGMM estimators accurately identify the presence of spatial effects for large sample sizes and provide accurate estimates of the unknown parameters when the average number of neighbors of the two spatial weights matrices is reasonable. Finally, we use the estimation techniques developed for the proposed double SAR model to analyze urban sprawl and urban policy spillover, investigating a huge sample for land use change at parcel level in France.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...