Research

We study optimal simple rating systems that partition sellers into a finite number of tiers. We show that optimal ratings must be threshold partitions, and that for linear supply and Cournot competition with constant marginal cost, optimal thresholds solve a k-means clustering problem requiring only the quality distribution. For convex (concave) supply functions, optimal thresholds are higher (lower) than the k-means solution. For log-concave distributions, two-tier certification captures at least 50% of maximum welfare gains from full disclosure, with five tiers typically achieving over 90%. Applications to eBay and Medicare Advantage data illustrate our method.

This paper addresses two central questions in markets with adverse selection: How does information impact the welfare of market participants (sellers and buyers)? Also, what is the optimal information disclosure policy and how is it affected by the planner's relative welfare weight on sellers' surplus versus consumers' surplus? We find that as a result of improved information, prices become more strongly associated with the true quality of sellers and, thus, more dispersed. This will result in higher total surplus. Furthermore, we find that better information has opposing welfare effects on consumers and producers that could lead to limited disclosure depending on the social objective and market characteristics.

Certification of sellers by trusted third parties helps alleviate information asymmetries in markets, yet little is known about the impact of a certification's threshold on market outcomes. Exploiting a policy change on eBay, we study how a more selective certification threshold affects the distribution of quality and incumbent behavior. We develop a stylized model that shows how changes in selectivity change the distribution of quality and prices in markets. Using rich data from hundreds of online categories on eBay.com, we find support for the model's hypotheses. Our results help inform the design of certification selectivity in electronic and other markets.

How can a marketplace introduce mechanisms to overcome inefficiencies caused by adverse selection? In this article, I use a unique data set that follows eBay sellers to show that reputation is a major determinant of price variations. I develop a model of sellers' dynamic behavior where sellers have heterogeneous qualities unobservable by buyers. Using reputation as a signal of quality, I structurally estimate the model to uncover buyers' utility and sellers' costs and underlying qualities. I show that removing the reputation mechanism increases low-quality sellers' market share, lowers prices, and consequently reduces sellers' profit by 66% and consumer surplus by 35%.

Markets prone to asymmetric information employ reputation mechanisms to address adverse selection and moral hazard. In this paper, we use a change in such a reputation mechanism to examine its effect on improving adverse selection and moral hazard. In May 2008, eBay changed its reputation mechanism to prevent sellers from giving negative feedback to buyers. This change was intended to prevent sellers from retaliating against buyers who gave them negative feedback. We observe an improvement in the overall quality of the marketplace as a result of this change. We attribute 49%–77% of this improvement to reduced adverse selection as low-quality sellers exit the market or their market share drops, and the rest to a reduction in moral hazard as sellers improve the quality of their service.

To mitigate inefficiencies arising from asymmetric information, some markets rely on government interventions, whereas others rely on reputation systems, warranties, or guarantees. This paper explores the impact of two mechanisms, namely, reputation badges and buyer protection programs, and their interaction on eBay's marketplace. Adding buyer protection reduces the premium for the reputation badge and increases efficiency in the marketplace. These efficiency gains are achieved by reducing moral hazard through an increase in sellers' quality and by reducing adverse selection through a higher exit rate for low-quality sellers. Our estimates suggest buyer protection increases the total welfare by 2.9%.

We study mechanisms for aggregating information divided across a large population of biased senders. Each sender privately observes an unconditionally independent signal about an unknown state, so no sender's report can be verified against another’s. A receiver makes a binary accept/reject decision whose payoffs depend on the state. Even though cross-verification is impossible, we show the receiver can benefit from informational division. We introduce a novel incentive-compatibility-in-the-large approach that studies optimal design via the large-population limit. For fixed population size, optimal mechanisms are in general complex. However, we show that in the limit they converge to a simple mechanism that depends only on the payoff from acceptance, and punishes excessive consensus in the direction of the common bias. These surplus burning punishments yield payoffs bounded away from the first best; the resulting inefficiency demonstrates how our concept of informational division is distinct from standard models of information in large populations.

We introduce a tractable methodology for analyzing dynamic decision problems and games with irreversible decisions under uncertainty. Leveraging regularity properties, we propose an intuitive method for solving these problems using equivalent certain values, a concept that extends the idea of certainty equivalence to dynamic environments with irreversibility and delivers properties of optimal strategies and comparative statics. We show that irreversibility is analogous to information loss in the Blackwell order sense, making agents behave as if they had worse information than under reversible actions. As an application, we use our methodology to analyze design features of a class of dynamic auctions.

An earlier NBER version incorporating theory and empirics is available as "Bidding Dynamics in Auctions" (with Hugo Hopenhayn).

Digitization has radically transformed news consumption. While social media are now essential information channels, their lack of editorial oversight allows sophisticated disinformation campaigns that threaten democratic discourse. We leverage endogenous patterns of disinformation supply to significantly reduce disinformation consumption without impacting legitimate news spread. Our Network-based Disinformation Labeling (NDL) approach identifies false information through network structure analysis rather than content review. Using comprehensive data, we show that disinformation originates from accounts with distinct network characteristics that coordinate on news origination. Simulations show that NDL reduces disinformation posts by 70% and maximum engagement by 50%, correctly identifying 85% of disinformation events.

We examine the strategic interaction between an expert (principal) maximizing engagement and an agent seeking swift information. Our analysis reveals: When priors align, relative patience determines optimal disclosure — impatient agents induce gradual revelation, while impatient principals cause delayed, abrupt revelation. When priors disagree, catering to the bias often emerges, with the principal initially providing signals aligned with the agent's bias. With private agent beliefs, we observe two phases: one engaging both agents, followed by catering to one type. Comparing personalized and non-personalized strategies, we find faster information revelation in the non-personalized case, but higher quality information in the personalized case.

We study optimal rating design under moral hazard and strategic manipulation. An intermediary observes a noisy indicator of effort and commits to a rating policy that shapes market beliefs and pay. We characterize optimal ratings via concavication of a gain function, accommodating violations of monotone likelihood ratios and distributional concerns. When effort increases tail risk, optimal ratings use lower censorship, pooling poor outcomes to encourage risk-taking; when effort reduces tail risk, upper censorship discourages negligence. In multi-task settings with window dressing, reduced informativeness of ratings deters manipulation, and redistributive test design can feature mid-censorship.

We study how price transparency affects prices in an environment where prices are negotiated rather than posted. There is rising prevalence of policies that require health care providers or insurers to reveal previously hidden negotiated prices. We study the 2007 rollout of New Hampshire’s price transparency policies. We first estimate a Nash-in-Nash bargaining model, and find that hospitals’ bargaining weights rose after transparency, rationalizing the higher post-transparency prices. We then map these higher bargaining weights to a model with changes in insurers’ information.