Distilling Models of Bounded-Rational Choice: A Constraint Programming Approach

We provide an analytical framework that allows for distilling the full explanatory and welfare-relevant content of influential yet computationally hard models of bounded-rational general choice. We do so by introducing constraint programming methods and tools from the optimization literature. We focus on the prominent "shortlisting" and "limited-attention" models. Applying our framework on imperfectly rational human choice data, we find that these models jointly account for nearly all behaviors, with limited-attention ones explaining better while being more permissive. Selection criteria that we introduce narrow down the models' welfare-relevant predictions, considerably alleviating their indeterminacy and contributing toward their practical applicability.