Abstract
We compare the predictive performance of a standard economic model to a variety of machine learning models by presenting nearly 1,000 subjects with a consumer decision problem -- the selection of a bundle of contingent commodities from a budget set. Our experimental dataset allows us to compare predictions at the individual level and relate them to the consistency of individual decisions with revealed preference axioms. We show that there is a strong relationship between relative model performance and consistency with an underlying preference order: regardless of estimated risk preferences, as choices align more with an underlying preference order, economic model performance increases relative to all machine learning models’ performance.