skip to primary navigationskip to content

Theory Workshop - Ali Jadbabaie (University of Pennsylvania)

When Oct 09, 2014
from 01:00 PM to 02:00 PM
Where Marshall Room
Add event to calendar vCal

Speaker: Ali Jadbabaie

Title: Learning to Coordinate in Social Networks

With Pooya Molavi (MIT), Ceyhun Eksin and Alejandro Ribeiro (Penn)


We study a  game in which a group of players attempt to coordinate on a desired, but only partially known, outcome. The desired outcome is represented by an unknown state of the world. Agents’ stage payoffs are represented by a super modular utility function that captures the kind of trade-off exemplified by the Keynesian beauty contest: each agent’s stage payoff is decreasing in the distance between her action and the unknown state; it is also decreasing in the distance between her action and the average action taken by other agents. The agents thus have the incentive to correctly estimate the state while trying to coordinate with and learn from others. We show that short-lived Bayesian agents who  play this game once and observe the actions of their neighboring roles over a network (that satisfies some weak connectivity condition) eventually succeed in coordinating on a single action.  The agents also asymptotically receive similar payoffs in spite of differences in the quality of their information. When the tilities are quadratic, we show that the resulting equilibrium is unique, and agents also reach consensus in their  estimates. Finally, we show that if the agents’ private observations are not functions of the history of the game, then the private observations are optimally aggregated in the limit. Therefore, agents asymptotically coordinate on choosing the best estimate of the state given the aggregate information available throughout the network.