Todd Sarver (Duke University)

Abstract. We introduce and characterize complexity constrained expected utility (CCEU) theory. Our representation can be interpreted as if the decision maker refines their understanding of an act (or lottery) by forming a partition of the set of outcomes and then taking a pessimistic evaluation of the act given this partition. The partition is generated sequentially using a greedy algorithm that examines each cell in the partition and only subdivides it further if the split would lead to a sufficiently large refinement of the understanding of the act. Our model can address well-known anomalies related to choice under uncertainty, including Allais-type behavior and violations of non-obvious dominance.