Pseudo‐Bayesian updating

I propose an axiomatic framework for belief revision when new information is qualitative, of the form “event A at least as likely event B .” My decision maker need not have beliefs about joint distribution signal she will receive and payoff‐relevant states. three axioms, Exchangeability , Stationarity Reduction to characterize class pseudo‐Bayesian updating rules. The key axiom, requires that order in which arrives does matter if different pieces neither reinforce nor contradict each other. show adding one more Conservatism adjust her just enough embrace information, yields Kullback–Leibler minimization: selects posterior closest prior terms divergence from probability measures consistent with newly received information. agents are susceptible recency bias, may be mitigated by repetitive learning.