"Disclosure and Endogenous Attention" (latest draft- updated Nov' 24)
Abstract: We study voluntary disclosure incentives when attention is costly for receivers. A sender seeks to persuade a receiver with privately known outside option, who chooses a costly probability of observing disclosures (for example, an inspection duration). We fully characterise unique equilibria for generic costs and priors, and show equilibria feature partial disclosure, because inattention rationally mutes scepticism. We find strategic withholding distorts attention relative to full disclosure, with optimal attention lower for low receiver types but higher for intermediate types, creating a role for mandatory disclosure to be welfare improving for all sender and all receiver types. Our results offer an explanation for empirical failures of unravelling when disclosure is costless and the sender is known to be informed; for example, strategic withholding of restaurant health grades, or student test scores.
Older drafts: Feb' 24
"Optimal Frictions in Disclosure Games" (latest draft - updated Sep' 24) [R&R at Games and Economic Behavior]
Abstract: Partial disclosure can emerge as the optimal policy for a sender with commitment power; we ask when simple frictions can serve as a proxy for commitment in disclosure games. In a voluntary disclosure setting in which a sender can publicly control the probability of obtaining evidence, but cannot influence its distribution, we study sender preferences over evidence acquisition. When sender utility depends only on receiver posterior mean, we derive prior-free conditions under which the sender benefits from an intermediate probability of obtaining evidence, and show sender utility switching from concave to convex is a necessary and sufficient condition. We then turn to comparative statics, and identify shifts in payoffs and priors that induce more information acquisition from the sender, showing that improvements in full-disclosure payoff may reduce optimal information acquisition.
"Costly Disclosure" (latest draft - updated Dec' 24) [submitted]
Abstract: We study costly communication between an upwardly biased sender and receiver when the sender has access to verifiable messages, and fully characterise the sender's equilibrium disclosure strategy as a function of messaging costs. Unless bias is large, disclosure rules can vary non-monotonically with costs, with the sender withholding evidence of low states for low costs, intermediate states for intermediate costs, and either low or high states for high costs. Our results imply when lobbying is costly, lobbyists may withhold evidence that would move policy in the direction of their bias, and policy makers may do worst meeting lobbyists with intermediate bias.
"Costly Advertising and Disclosure" (latest draft- updated Nov' 24) [preliminary]
Abstract: How does the availability of costly advertising affect what firms disclose? We study a sender who can voluntarily disclose private information to a receiver, and increase the probability messages are observed at a cost. We apply our model to informative advertising by a firm, and fully characterise optimal advertising expenditure in equilibrium. We then turn to welfare, and show that decreasing the cost or increasing the reach of a given advertising policy can lead to a Blackwell garbling of equilibrium disclosures, by crowding out expenditure on policies with greater reach. Similarly we find that mandating disclosures at some base visibility level need not improve equilibrium information provision.
"Receiver-Optimal Communication Protocols" (latest draft - updated Nov' 23)
Abstract: A sender with access to both costly verifiable messages and cheap unverifiable messages provides advice to a receiver. In a uniform-quadratic set-up, we show that if disclosure costs are small relative to bias and the sender-optimal equilibrium is selected, the receiver benefits ex-ante from restricting the sender's message space to include only one cheap-talk message.
"Noisy Disclosure" [new draft coming]
Abstract: Even when experts are unable to lie, they may be misunderstood. We model a setting of verifiable disclosure in which a privately informed sender chooses between disclosing the state and sending no message to a receiver who observes noisy realisations of messages, due to a language barrier. We show that full disclosure will only occur if the sender is sufficiently biased, and that the receiver-optimal disclosure rule does not feature full disclosure. When sender bias is small and noisy messages do not narrow posterior supports, we show both upper and lower censorship are possible equilibrium disclosure strategies. We show that, ex-ante, in general a receiver prefers to face an unbiased rather than biased sender, and that noise in message creates a commitment problem for biased senders.
"Reputational Incentives with Networked Customers" (latest draft - updated Aug' 22)
Abstract: We propose a model of reputational incentives for a firm with privately known type whose customers share reviews via a social network. The firm can choose a non-contractible effort level for each customer, which stochastically improves that customer's review. When customers base their purchase decisions on their friends' reviews, firm incentives for effort are stronger if a customer's review will be read by many friends, and if those friends are not too connected (since their beliefs are easier to influence). From the perspective of ex-ante social welfare, this creates a trade-off between providing incentives and generating learning; more connected networks may allow customers to learn more, but remove firm incentives to exert effort when serving them. When effort is sufficiently productive, ex-ante expected total surplus can be higher when the social network has disjoint components. Our results imply that interventions to increase customer review visibility (such as auto-translate features) may be harmful for social and consumer welfare, if firm effort is sufficiently productive.
"Strategic Information Release on a Communication Network" [draft available on request]
Abstract: Information quality can deteriorate as information spreads by word of mouth, due to accumulated misunderstandings. We analyse a game in which a sender seeds a communication network with information, which travels through the network to receivers, acquiring noise. If seeding is observed, when receiver preferences are homogeneous the sender wants to provide information to the most central receiver if the sender's preferences are sufficiently aligned with the receivers, else they want to provide information to the least central receiver. With heterogeneous receiver preferences or unobserved seeding, it may remain optimal for the sender to seed the most central receiver even when all receivers on the network are very misaligned with the sender. In a setting with two competing senders, we show that a sender whose preferences are misaligned with receivers may find it optimal to copy the seeding decision of the other sender, even if this implies seeding the most central receiver.
"Optimal Signalling Incentives" [presented at EWMES 23]
Summary: When agents are motivated to exert effort on a task solely to signal their ability to an observer, which signal distributions maximise expected effort? We characterise the maximum expected effort a signal can induce in equilibrium, and construct a signal which induces this effort. The optimal signal is generically stochastic, suggesting a school maximising student effort may benefit from building luck into testing, and that a monopolist may benefit from selling status goods via lotteries. We also find distortion at the top, with grading less informative when a school can set students of different abilities different tests.
"Firm Experimentation and Social Network Spread" (with Stephen Nei)
"Limits of Persuasion" (with Dimitri Migrow)
"Evidence, Effort, and Interim Evaluations" (with Dimitri Migrow)