Collusive Behaviour, Efficiency and Cheap Talk Negotiation in Repeated Games

This paper addresses the relationship between cheap talk negotiation and collusion/efficiency in repeated games by explicitly modelling the negotiation. At each date, players bargain over how to play the continuation game and thereafter they play a stage game G. We consider equilibria that are measurable with respect to the latest agreement - as they are salient/focal features of the past. We show equilibrium payoffs are bounded below by that of a Nash equilibrium of G and are weakly renegotiation-proof. Our main results are: a non-babbling efficient equilibrium exists in many games if the discount factor δ is high, and every equilibrium payoff is either babbling or efficient in the limit as δ → 1. Finally, we check the robustness of the latter result to two perturbations: complexity costs and trustworthy/honourable players. Equilibria that survive the former perturbation are efficient or induce one-shot Nash equilibria and equilibria that survive the latter are efficient.