FALL 2025

SPRING 2026

Abstract. Auctioneers suspecting bidder collusion often lack the formal evidence needed for legal recourse. A practical alternative is to design auctions that hinder collusion. Since Abreu et al. (1986), economic theory has emphasized imperfect monitoring as a constraint on collusion, but evidence remains scarce on whether: (i) information frictions meaningfully limit real-world collusion; and (ii) auctioneers can effectively exploit these frictions. Indeed, transparency concerns prevent the introduction of explicit randomness in auction design. We make progress on this issue by studying the impact of subjective scoring in auctions run by Japan’s Ministry of Land, Infrastructure, and Transportation. The adoption of scoring auctions significantly reduced winning bids in ways inconsistent with competition. Model-based inference suggests that the cartel’s dynamic obedience constraints were binding and tightened by imperfect monitoring. Subjective scoring can successfully leverage imperfect monitoring frictions to reduce the scope of collusion.


Abstract. We develop a modular Cardell variance-components framework for panel discrete-choice demand. By choosing which latent components are shared across products and purchase occasions, the framework generates application-specific dependence patterns in unobserved utility while preserving the extreme-value structure that makes logit models tractable. The representation extends Cardell beyond its standard role in nested logit and provides a unified framework for single- and multi-level nesting, cross-classified within-period correlation, persistent match effects, and autoregressive shocks. It is also compatible with standard empirical IO tools, including random coefficients, instruments for endogenous prices, and the joint use of micro data and aggregate shares. We develop one specification in detail: an AR-persistent model that captures persistent preferences through both stable consumer-product match quality and transitory shocks that decay across purchase occasions. For long panels, we estimate this specification with a pairwise composite likelihood that keeps the single-period marginal logit likelihood exact and uses simulation only for the dependence correction that identifies persistence. Monte Carlo experiments show good finite-sample recovery of structural parameters and elasticities. We conclude with an empirical application using linked household-panel and scanner data on over-the-counter cough remedies and consider a counterfactual that removes a popular product. Compared to mixed logit benchmarks with and without lagged-choice state dependence, the AR-persistent model better matches repeat-purchase patterns, return-after-stockout patterns, and has better long-horizon out-of-sample prediction. We find that, in our setting, models ignoring AR and persistent errors overstate price sensitivity and counterfactual welfare losses.


Abstract. Industrial decarbonization requires significant investments in abatement technology. Yet if firms have market power, their incentive to invest may deviate from the socially optimal level -- even after carbon externalities have been appropriately priced -- because they do not fully internalize the consumer gains from reduced prices or they face preemption motives in deterring rival firms. We explore the implications of market power for technology adoption and efficient policy in the context of the U.S. cement industry. We first estimate a dynamic model drawing on data over 1974-2019, where we endogenize firm decisions to adopt new technology or retire old technology. We then use counterfactual simulations to examine how firms respond alternative policy design.