📝 SummarXiv

Abstract

We develop a non-parametric methodology to quantify preference heterogeneity in consumer choices. By repeatedly sampling individual observations and partitioning agents into groups consistent with the Generalized Axiom of Revealed Preferences (GARP), we construct a similarity matrix capturing latent preference structures. Under mild assumptions, this matrix consistently and asymptotically normally estimates the probability that any pair of agents share a common utility function. Leveraging this, we develop hypothesis tests to assess whether demographic characteristics systematically explain unobserved heterogeneity. Simulations confirm the test's validity. Applying the method to grocery expenditure data, we find robust evidence that larger families, higher income, and younger age are associated with greater similarity in revealed preferences, whereas low income and older age predict lower similarity.