Abstract: We study a manufacturer's demand-investment decisions in distribution channels subject to double marginalization. Casting this as a mechanism design problem, we show that demand-enhancing investments strengthen retailers' incentives to exploit market power, forcing manufacturers to concede greater rents. Manufacturers therefore optimally restrict product quality or market coverage. We fully characterize which demand parameters create these perverse incentives: increases benefit manufacturers in segments where they control pricing but harm them in segments with binding incentive constraints. This reveals fundamental limits to demand-side investment in vertical relationships.
working paper (November 2025)
Abstract: We study competition and entry in information markets where buyers combine sellers' signals. Price competition yields efficient trade but need not leave the buyer any surplus. Entry may be anti-competitive, reducing buyer surplus, but is never socially excessive. Competition works through the coalitions of sellers the buyer can credibly exclude. We introduce two notions of complementarity that characterize buyer surplus extraction. Weak competitive complementarity is equivalent to full extraction in some equilibrium. Strong competitive complementarity is equivalent to full extraction in every equilibrium. A global test fully characterizes the weak notion; adding a local test characterizes the strong notion.
working paper (July 2026)
Abstract: When multiple parties contribute to a shared reputation, free-riding arises. We study whether internal monitoring by a global player helps. We identify two instruments. Based on his observations, the global player can (i) condition his own effort and (ii) pay discretionary bonuses. These instruments are substitutes; their optimality depends on the global player's effort technology---whether his effort is common or market-specific (separate). Under common effort, conditional effort is generally optimal, dominating relational contracting. Under separate effort, relational contracting dominates when there are two local players or players are patient. In general, the best equilibrium involves conditional contracting, conditional effort, or no use of internal monitoring.
slides (April, 2024)
Abstract: We revisit the tension between the legal doctrine of renegotiation and economic efficiency. We introduce \textit{self-revealing mechanisms} that combine bidirectional communication (the agent sends and receives information) with conditional disclosure (communication remains private during renegotiation but becomes verifiable at contract execution). Within the canonical \citet{FudenbergTirole1990} framework, we design a self-revealing mechanism that fully mitigates the renegotiation threat by uniquely implementing the second-best allocation. Thus, we achieve the full-commitment outcome while satisfying renegotiation-proofness. Our optimal mechanism is structurally simple, and exploits signal disclosures to the agent to construct incentive-compatible punishments, which she activates upon observing a renegotiation offer. It satisfies standard commitment assumptions by only conditioning decisions on public information, without requiring any third-party enforcement. In practical terms, it can be implemented using existing smart-contract techniques. Our results extend to general settings of renegotiation.
working paper (February, 2025)
slides (Paris, 2025)
video presentation (June, 2024)
Abstract: We develop a novel approach for mechanism design with evidence using \citet{Mye82}, allowing a characterization of implementability for general evidence structures. This is so because the revelation principle of \citet{Mye82} applies, yielding standard notions of incentive compatible direct mechanisms. The controllability of evidence however determines their specific structure. We show a general value of controllability. For deterministic implementation, we offer two independent conditions under which this value vanishes, one on evidence (WET) and another on preferences (TIWO). Allowing for fully stochastic mechanisms, we show a general value of randomization and clarify to what extent this value vanishes under the common assumption of evidentiary normality (NOR). Neither control nor randomization have value when NOR holds with WET or TIWO, implying that, without loss, the mechanism design problem can be analyzed assuming controllable evidence and deterministic implementation.
working paper version (March, 2025)
slides (Toronto, 2022.03.22)
video presentation (90min; newer version)
video presentation (20min; older version)
presentations: Cambridge (UK), University of Toronto, Yale University, Paris School of Economics, Michigan University, Penn State
This work is based on and supersedes my working paper "Mechanism design with partially verifiable information" from 2016.