Agreeing to Disagree with Multiple Priors

We present an extension of Aumann's Agreement Theorem to the case of multiple priors. If agents update all their priors, then, for the Agreement Theorem to hold, it is sufficient to assume that they have closed, connected and intersecting sets of priors. On the other hand, if agents select the priors to be updated according to the maximum likelihood criterion, then, under these same assumptions, agents may still agree to disagree. For the Agreement Theorem to hold, it is also necessary to assume that the maximum likelihood priors are commonly known and not disjoint. To show that these hypotheses are necessary, we give several examples in which agents agree to disagree.