Latent utility and permutation invariance: A revealed preference approach

This paper provides partial identification results for latent utility models that satisfy an invariance property on unobservables such as exchangeability. We employ a simple revealed preference argument to "difference out'' unobservables, obtaining identifying inequalities for utility indices. We show the differencing argument is also useful for counterfactual analysis. The framework generalizes existing work in discrete choice by allowing latent feasibility sets and by allowing individuals to purchase multiple (possibly continuous) goods. We present a new framework leveraging nesting structures that generalizes nested logit. In a panel setting, we innovate by allowing preferences for variety.