Learning to Cooperate and Relational Contracts

The purpose of this project is to use Q-learning algorithms to model the interaction of individuals in long term relationships. There is a large literature that studies the set of subgame perfect equilibria in repeated relationships (see https://docs.iza.org/dp15881.pdf). The literature typically assumes that individuals make no errors. In practice, individuals make errors and learn over time. Q-learning algorithms from the machine learning literature provide a framework to study the performance of parties over time. The goal is to understands the types of behaviors that lead to more efficient and productive relationships.