A group of experts with different prior beliefs must make a collective

decision over the choice of a treatment or policy. We propose a model

where such disagreements are resolved through bargaining. We show

that, when the outcome is determined according to the Nash bargaining

solution, the collective decision is made as if a planner maximized expected

utility with respect to a “compromise belief” that places greater

weight on the more pessimistic experts. In interesting classes of environments,

bargaining leads to an inefficient use of information in a

strong sense: experts receive a lower payoff in every state, and thus for

any prior belief. This inefficiency takes the form of under-reaction to

information. We connect these findings to speculative betting and to

the admissibility of statistical decision procedures.