Learning structure in nested logit models

Abstract

This work is about developing an estimation procedure for nested logit models that optimizes over the nesting structure in addition to the model parameters. Current estimation practices require an a priori specification of a nesting structure. We formulate the problem of learning an optimal nesting structure as a mixed integer nonlinear programming (MINLP) optimization problem and solve it using a variant of the linear outer approximation algorithm. We demonstrate that it is indeed possible to recover the nesting structure directly from the data by applying our method to synthetic and real datasets.