Multi-Index Binary Response Analysis of Large Data Sets

Abstract

We propose a multi-index binary response model for analyzing large databases (i.e., with many regressors). We combine many regressors into factors (or indexes) and then estimate the link function via parametric or nonparametric methods. Neither the estimation of factors nor the determination of the number of factors requires ex ante knowledge of the link between the response and regressors. Furthermore, applying perturbation theory, we furnish a new asymptotic result to facilitate significance tests of factor loadings. We illustrate this approach with an empirical application in which we reduced dimensionality from 124 regressors to 4 factors.