Nonparametric Estimation of Large-Dimensional Binary Choice Models

Speaker: Dr Guo Yan

Affiliation: University of Melbourne

Online via Zoom: https://uqz.zoom.us/j/82603079317 

Abstract

We propose a kernelized non-parametric (KNP) estimator for nonparametric binary choice models, which do not impose parametric structure either on the systematic function of covariates or on the distribution of error term. Motivated by the kernel trick used in machine learning, the proposed method combines (i) approximating the systematic function of covariates by functions in a reproducing kernel Hilbert space, with (ii) approximating the probability density of the error term by squared Hermite polynomials. We establish consistency of the KNP estimator for both the systematic function of covariates and the density of error term. Furthermore, we provide a non-asymptotic high probability bound for the plug-in estimator of conditional choice probability function, and asymptotic normality for the estimator of weighted average partial derivatives. Simulation studies show that, compared to parametric estimation methods, the proposed method effectively improves the finite sample performance in case of misspecification, and has a rather mild efficiency loss if the model is correctly specified. Using administrative data on the grant decisions of US asylum applications to immigration courts and the case-day variables on the weather and pollution, we estimate a model using KNP procedure to examine the effect of outdoor environments on court judges’ “mood”, and thus, their grant decisions. Our method allows for a general complex association among all environment variables and captures important patterns.

About the presenter's meeting

If you would like to meet with Dr Yan, please contact Dr Fu Ouyang