Alex Boche
PhD Candidate, Managerial Economics and Strategy
I am an applied theorist with a focus on incentive design. In my job market paper, I study the design of efficient health insurance in the presence of evolving risk, adverse selection, and ex-post moral hazard. In a companion paper, I study health-insurance design when patients possess multidimensional private information about their health status and face multiple potential treatments. In other work, I explore how dynamically sophisticated AI agents might interact with their human overseers when they anticipate being retrained on the basis of their actions.
I am on the 2025–26 academic job market.
Subfields: Microeconomic Theory, Industrial Organization, Health Economics
Job Market Paper
Efficient Health Insurance with Moral Hazard, Adverse Selection, and Evolving Risk
I study efficient health insurance under evolving risk with adverse selection and ex-post moral hazard. I show that, under a simplification of incentive constraints known as the first-order approach (FOA), optimal linear coverage will be set higher than would be efficient without adverse selection. This distortion relaxes upward incentive constraints for lower-risk consumers, thereby allowing for greater cross-subsidization from lower-risk to higher-risk consumers, which is important for insuring dynamic risk. I then characterize the optimal menu of nonlinear coinsurance schedules under the FOA. Results inform the design of both private-sector dynamic health insurance contracts and social health insurance programs administered by governments or employers.
Work in Progress
Efficiency of Bundled Copays in Multidimensional Health Insurance
I consider a model of consumer copayments where the consumer privately realizes a two-component health state corresponding to two independently distributed health ailments with no interaction between them. Each ailment can be treated by a corresponding binary treatment. I investigate conditions under which optimal copays should be subadditive in the sense that the copay for buying both treatments simultaneously is less than the sum of the copays for buying each on its own. I show that when the patient’s utility function satisfies constant marginal rate of substitution between health status and monetary consumption as well as prudence (i.e., positive third derivative), subadditive copays can strictly improve upon any additive copay scheme. I then show that if, further, the problem of choosing an optimal vector of copays is concave, subadditive copays are globally optimal.
Deceptive AI Alignment with Retraining
I provide a stylized model of a dynamically sophisticated AI agent whose actions are observed by an overseer, but whose preferences are not. Its preferences are misaligned with those of the overseer to an unknown degree and in an unknown direction (up or down). After observing the AI’s initial behavior, the overseer can “retrain” the AI by translating its preferences for future actions by a chosen vector. I study equilibrium behavior in this environment in a two-period model. I consider two cases depending on whether the overseer can commit to a policy specifying how retraining will occur as a function of the AI’s first-period action. I show that without commitment, if the AI is sufficiently patient, all equilibria involve pooling of types into one or more first-period action pools. With commitment, I provide a first-order condition for the overseer’s optimal nonlinear retraining policy. At the optimum, increasing the aggressiveness of marginal retraining trades off a beneficial direct effect on the second-period action against an elasticity effect that is overall harmful but not unequivocally so. Specifically, the elasticity effect leads to a more deceptively aligned first-period action (which is myopically beneficial to the overseer) but a more misaligned second-period action (which is harmful to the overseer).
[Draft Coming Soon]