Arbitration and Mediation L17 Goltsman et al. (JET 2006) Institutional approach
Institutions CS assumes a particular extensive form of communication game Some institutions improve upon CS Two approaches Design of institutions given incentive constraints Particular extensive forms Mechanisms though which they improve communication How they compare to best CS equilibrium?
Sender-Receiver model Two agents: Sender (S) and Receiver (R) Type space: Preferences S sends message to intermediary, who makes recommendations to R Recommendation rule Intermediary: Arbitration (binding recommendations) Mediation (not binding) We look for rules that are ex ante optimal for R
Observations Commitment: Myerson’s revelation principle Direct truth telling mechanism Welfare rankings for R: - Arbitration - Mediation - SPN equilibrium in ``any’’ communication game Upper bounds For mediation and SPN equilibrium ex ante welfare of R and S aligned
Arbitration Problem: Incentive compatibility: Truth telling required to be ex post optimal for S (IC-S) Rule need not be (ex post) optimal for R.
Important trick Fix recommendation rule Sufficient statistics for S expected payoff Expectation is an ``action’’ and variance is a ``transfer’’
Incentive compatibility (IC-S) L: satisfies IC-S condition iff a) Is nondecreasing b) c) Proof
Incentive compatibility (IC-S) Is IC-S for some Is IC-S for some
Optimal arbitration rule Assume T: Optimal mechanism is deterministic with Melumad and Shibano (RAND 91) Remarks: Delegation with cap is optimal For bubbling is optimal for R (no information transmission) S -optimal rule
Heuristic argument :
Mediation Problem: Incentive compatibility: Truth telling required to be ex post optimal for S (IC-S) Rule has to be (ex post) optimal for R. Can optimal arbitration rule be implemented in a mediation setting?
Optimal Mediation Mechanism L: Upper bound for receivers ex ante utility is Partition T: Optimal mediation mechanism Mechanism is random Arbitrage mediation most informative equilibrium in CS Mediator’s only role is to introduce noise
Interpretation ``Broken’’ phone paradox Blume, Bord and Kawamura (TE 2007) With probability instead of signal receiver observes Informative equilibrium implements optimal mediation outcome Two effects of noise Makes the signal less informative Relaxes IC constraint for S fostering more truthful revelation The latter effect dominates
Negotiations Multistage cheap talk Aumann and Hart 2003, Krishna and Morgan 2004 Better than most informative CS Upper bound for welfare of R –mediation outcome-sometimes achieved Intuition: multistage noise allows for randomisation
Conclusions Commitment of external party improves welfa Delegation is optimal if feasible Noisy cheap talk can improve information transition Can be implemented as multistage cheap talk