Title: Relative Maximum Likelihood Updating of Ambiguous Beliefs

Abstract: When ambiguous beliefs are represented by multiple priors, a decision maker (DM)'s updating rule may also include a step of refining the initial belief. Maximum Likelihood (ML) updating indicates one example of such a refinement, whereas Full Bayesian (FB) updating does not allow for any refinement. In fact, ML and FB are the two extremes of refining beliefs with likelihood in updating. To capture behaviors between these two extremes, the present paper proposes and axiomatizes a new updating rule, Relative Maximum Likelihood (RML), to represent conditional preferences that are not as extreme as either case. RML updates an intermediate set of priors, which can be expressed as a linear contraction of the initial set with respect to the set of maximum likelihood priors. The linear contraction parameter captures the DM's relative inclination towards ML with respect to FB, it is also the threshold of a relative likelihood ratio test. Moreover, RML includes both FB and ML as the two extreme special cases.
Additionally, the present paper provides an axiomatization of ML for preferences admit Maxmin Expected Utility (MEU) representation as a preliminary result for the exposition of RML.