The problem of modeling economic exchange on networks is one that has traditionally been examined through solving for an equilibrium set of prices. However, this method falls short in practice as human behavior often deviates from the assumptions that are generally made in order to predict an equilibrium. I expand on a current line of research by Professor Kearns, which augments equilibrium analysis by examining human interaction on networks through behavioral experiments. In particular, I focus on a recent experiment based on the "Milks and Wheats" network model, and an extension of the model that I have created.
This project consisted of two stages. In the first stage, I analyzed data from Professor Kearns' behavioral experiment in order to understand how the strategies and behaviors of the participants differ from "rational" behavior as predicted by equilibrium analysis. In the second stage I analyzed the theoretical properties of a new model that I have developed after taking into consideration the results of the behavioral experiments. The analysis was facilitated through the construction of three artificial intelligence agents - one employing adversarial search, another reinforcement learning, and a third heuristics. Taken together, the goal of the project has been to develop a theory of trade on economic networks that advances our current understanding of both rational and real-life decision-making in such situations.