Informative Imitation

We study steady-state equilibria of a sequential learning game in which each agent chooses between a safe action and a risky action of unknown profitability. Before choosing, each agent observes how many of the last $m$ agents adopted the risky action, but not whether they were informed or what their adoption costs were. We show that uninformed agents’ imitation decisions can be strategic complements: the more likely uninformed agents are to join risky-action herds, the more informative those herds become about the action’s profitability. When this complementarity gives rise to multiple equilibria, the one that maximizes the probability of imitation by uninformed agents also maximizes agents’ welfare.