Selected Works and Working Papers.
1. Opaque by Design: In many principal-agent environments—such as hiring consultants, executives, or technical experts—effort is unobservable and monitoring is infeasible. At the same time, agents often receive private signals about fundamental project characteristics. We show that the principal can strategically use information provision to substitute for monitoring and reduce both incentive costs and rents for truthful reporting. When effort costs are convex, the optimal contract provides only coarse information—even when signals are costless, accurate, and verifiable. Full transparency and complete opacity are both strictly suboptimal. This result offers a strategic, non-behavioral rationale for the use of vague evaluations and standardized review systems commonly observed in firms and bureaucracies.
​
​2.The Arbiter's Gambit:Pacifying interventions like ceasefires, strike bans, and mandated mediation aim to de-escalate conflict. But by lowering the immediate cost of disagreement, they can dull incentives to settle—paradoxically prolonging disputes. This paper formalizes that mechanism in a dynamic reputational bargaining model with indivisible stakes and evolving leverage. Unlike canonical models, where delay vanishes with perfect information and zero frictions, delay here persists due to pacifying policies themselves. Even in frictionless settings, interventions distort concession dynamics and induce inefficient delay. The framework applies broadly—from war and litigation to labor disputes—and yields a sharp empirical prediction: interventions that reduce conflict intensity increase dispute duration. We test this using a 200-year panel of 92 interstate wars and an instrumental variables strategy exploiting exogenous variation in military preparedness and domestic unrest. The results confirm the theory: pacifying interventions systematically prolong conflict, even as they reduce short-run costs.
​
3. Optimal Sequential Experimentation: I study a sequential experimentation problem where a decision-maker (DM) learns about competing alternatives by managing an experiment i.e., a noisy data-generation process. Extending the widely used Drift-Diffusion (DD) framework, I allow for a broad spectrum of evidence-generation strategies: from continuous updates to rare, decisive signals. Despite this added flexibility, I find that the unique, optimal strategy remains feasible in the DD framework, as experimental constraints compel gradual evidence accumulation. This finding challenges recent results suggesting a preference for infrequent, decisive evidence under flexible learning. The model's generality captures diverse applications, from R&D and due diligence to policymaking, where balancing precision, speed, and evidentiary flow is crucial. I further propose a novel experimental design to test cognitive cost structures and clarify when outsourcing experimentation to consultants is optimal.
​
​
4. The Game of Snake (Upcoming, latest research (early stage), slides available here): Technological progress is key to wealth generation and nations, consequently, invest heavily in research and development (R& D). Over time, however, the direction of research is increasingly narrow and unresponsive to new findings (Park et al 2023): thus, technological progress stalls. In this paper, I rationalize the trends above when a scientist's skills are valued outside of R&D and, in discovery, there are overall agglomeration gains. This is because new discoveries increases non-research productivity and thus its wages. Over time, rising, non-research wages prompts increasing concentration in scientific topics with decreasing social and private value.
5. Learning to Commit: I study the relation between limited commitment and learning in auctions. In each period, the seller sets the terms for an auction selling an indivisible good among multiple buyers; but if the item fails to sell, he cannot pre-commit to the terms of future offerings. I find that, in interdependent value settings, the seller's equilibrium revenues are greater than immediately running an efficient, Vickrey auction. In contrast with private value settings, this result persists regardless of how often the seller may interact with buyers. This is because learning among buyers both limits how many times a good can be gainfully re-offered and the information rents that each buyer can demand the seller. Intuitively, buyers lower their valuations in response to their peer's lack of interest. This progressively lowers the trade surplus and compresses the support of valuations. As the dispersion in valuations falls, the seller further extracts an increasing share of the remaining trade surplus.
​