Distribution free estimation of spatial autoregressive binary choice panel data models

This paper proposes a semiparametric estimator for spatial autoregressive (SAR) binary choice models in the context of panel data with fixed effects. The estimation procedure is based on the observational equivalence between distribution free models with a conditional median restriction and parametric models (such as Logit/Probit) exhibiting (multiplicative) heteroskedasticity and autocorrelation. Without imposing any parametric structure on the error terms, we consider the semiparametric nonlinear least squares (NLLS) estimator for this model and analyze its asymptotic properties under spatial near-epoch dependence. The main advantage of our method over the existing estimators is that it consistently estimates choice probabilities. The finite-dimensional estimator is shown to be consistent and root-n asymptotically normal under some reasonable conditions. Finally, a Monte Carlo study indicates that the estimator performs quite well in finite samples.