Disparate Outcomes in Buyer–Seller Matching: A Formal Analysis of Discrimination in Two-Sided E-Commerce Platforms

Abstract

Two-sided e-commerce platforms increasingly intermediate trade by selecting which sellers are shown to which buyers, often through search, recommendation, and sponsored ranking systems. These systems can generate systematically different outcomes across seller groups even when listings are similar in measured quality or price, raising questions about the mechanisms that produce disparate matching and about the scope of discrimination in algorithmically mediated markets. This paper develops a formal analysis of disparate outcomes in buyer--seller matching on a platform that controls exposure and information while buyers and sellers respond strategically. The framework accommodates taste-based discrimination in buyer preferences, statistical discrimination arising from heterogeneous beliefs and noisy signals about seller quality, and algorithmic discrimination that emerges from optimization under partial observability, feedback, and constraints. The model yields equilibrium conditions linking exposure, clicks, conversions, seller pricing, and platform objective functions. It also provides a decomposition of disparity into components attributable to preference heterogeneity, information and inference, and platform policy. The analysis highlights how subtle differences in priors, measurement error, and exploration rules can produce persistent gaps in exposure and sales, including regimes where outcomes diverge despite symmetric underlying quality distributions. The paper also characterizes design interventions based on constrained optimization and counterfactual parity concepts, clarifying when they can reduce disparities without inducing large efficiency losses, and when they primarily shift rents between market sides.