Finite Sample Performance of Semiparametric Binary Choice Estimators

Strong assumptions needed to correctly specify parametric binary choice probability models make them particularly vulnerable to misspecication. Semiparametric models provide a less restrictive approach with estimators that exhibit desirable asymptotic properties. This paper discusses the standard parametric binary choice models, Probit and Logit, as well as the semiparametric binary choice estimators proposed in Ichimura (1993) and Klein and Spady (1993). A Monte Carlo study suggests that the semiparametric estimators have desirable nite sample properties and outperform their parametric counterparts when the parametric model is misspecied. The semiparametric estimators show only moderate eciency loss compared to correctly specied parametric.