Two-Sided Markets and Restricted Boltzmann Machines

Abstract

We extend the standard model of the two-sided market (e.g., Uber or Tinder) by allowing benefits from pairwise interactions to differ across pairs of agents. We show that these benefits can be estimated consistently and tractably by observing agents’ repeated decisions whether to participate in the market. The proposed method mitigates the inherent combinatorial explosion by exploiting a formal connection to a prominent class of neural networks called restricted Boltzmann machines. We show that combinatorial explosion does not hinder the discovery of a simple pricing scheme that maximizes both welfare and revenue in a special case of the model.

Keywords: two-sided markets, platforms, restricted Boltzmann machines, contrastive divergence

JEL Classification: L10, C45, D85, C51