Computationally efficient coordination in game trees

The solution concept of “correlated equilibrium” allows for coordination in games. For game trees with imperfect information, it gives rise to NP-hard problems, even for two-player games without chance moves. We introduce the “extensive form correlated equilibrium” (EFCE), which extends Aumann’s correlated equilibrium, where coordination is achieved by signals that are received “locally” at information sets. An EFCE is polynomial-time computable for two-player games without chance moves.