An axiomatic foundation for additive utility representations on simplex domains, applied to preferences over probability distributions.
Research
Information, competition, and cooperation in financial markets and AI systems.
Publications
Publications
A weakening of separability that retains tractable structure while admitting richer interaction across choice dimensions.
Working Papers
Working Papers
Should credit registries disclose borrowers’ application histories? We study this question in embedded finance, where lenders differ in how accurately they assess each borrower. Some borrowers can apply only to their compatible lender; others choose their application order. A prior rejection may reflect borrower–lender mismatch rather than low quality, and disclosure can prevent lending despite a favorable second assessment. We measure welfare by expected project surplus net of funding costs. The welfare effect of disclosure depends on the high-quality share of the loans it blocks. Disclosure lowers welfare when these loans have positive expected net project surplus but negative expected lender profit. Holding average type-conditional screening rates fixed, heterogeneous screening and endogenous routing can reverse the optimal disclosure policy relative to common screening in either direction. We give sufficient conditions on lender-specific access and the value of delayed funding under which all perfect Bayesian equilibria yield the same funding probabilities and welfare for each disclosure policy.
Work in Progress
Work in Progress
Distribution drift interacts with selective labeling in credit scoring to break myopic-optimal screening, producing cyclical model degradation and an exploration trade-off.