Research Papers
Coordination and Learning in Dynamic Global Games: Experimental Evidence (Job Market Paper)
This paper uses a laboratory experiment to test the predictions of a dynamic global game designed to capture the role of information and coordination in speculative attacks. The game has two stages and a large number of heterogeneously informed agents deciding whether to attack a status quo. The status quo collapses if enough agents choose to attack. In the first stage, the equilibrium size of the attack is decreasing in both the underlying strength of the status quo and the agents' cost of attacking. In the second stage, the knowledge that the status quo has survived the first-period attack decreases the incentive to attack, implying that a new attack is possible only if agents receive new information. Our experimental evidence supports these theoretical predictions in both stages. We identify the agents’ beliefs about the actions of others to be the main channel through which the relative strength of the status quo, the cost of attacking, and learning impact observed behavior. However, we also find that the subject's actions are overly aggressive relative to the theory’s predictions. Once again, we find that the excess aggressiveness in actions stems from the aggressiveness of their beliefs about others’ actions.