Rational Inattention with Ambiguity Aversion

We develop a tractable framework incorporating ambiguity aversion into rational inattention. We model uncertainty using smooth ambiguity (Klibanoff, Marinacci, and Mukerji, 2005) and define entropy-based information costs on the predictive prior distribution, which allows us to rewrite the problem in terms of stochastic choice rules. Our solution generalizes the multinomial logit formula of Matějka and McKay (2015) by adding a multiplicative state fixed-effect, that depends on ambiguity, to every payoff. We provide an axiomatic characterization of this formulation. Under stronger conditions, we show that our solution follows Luce’s (1959) multinomial logit model, providing a new foundation robust to ambiguity aversion