What can we learn from applying machine learning to bargaining?

We collect a unique dataset from an online experiment on bilateral distributive bargaining. Using machine learning techniques, we construct a probabilistic model that predicts the share demands made by human participants. This model forms the basis for two bargaining algorithms: one that imitates human behavior, and another that optimally responds to the former. We then simulate additional bargaining games between these agents and estimate a range of models to analyze how structural features and past interactions influence agents’ decisions and bargaining outcomes. This approach yields new insights into the strategies employed by both human negotiators and artificial agents.