This paper examines the welfare implications of third-party informational intermediation. A seller sets the price of a product that is sold through an informational intermediary. The intermediary can disclose information about the product to consumers and earns a fixed percentage of sales revenue in each period. The intermediary's market base grows at a rate that increases with past consumer surplus. We characterize the stationary equilibria and the set of subgame perfect equilibrium payoffs. When market feedback (i.e., the extent to which past consumer surplus affects future market bases) increases, welfare may decrease in the Pareto sense.
Selling with Recommender Systems and Price Dynamics. (with Shuoguang Yang)
A long-lived seller sells a new product of unknown value through a recommender system that offers prices and recommends products to short-lived consumers in continuous time. The seller receives feedback about the product at a rate that increases with the instantaneous sales volume. The profit-maximizing selling mechanism features episodes of price discounts, during which the seller discontinuously lowers the price and selectively sends out unwarranted recommendations to consumers. The optimal price path may involve delayed discounts, below-cost pricing, and dynamic price back-tracking.
This paper studies sequential social learning, in which agents learn about an underlying state from others' actions. In contrast to the classic models that have a network observational structure, agents arrive in cohorts and observe action-signals regarding previous cohorts' actions. I identify a simple, necessary, and sufficient condition for asymptotic learning, called separability, which is a joint property of action-signals and agents' private information about the state. A necessary condition for separability is "unbounded beliefs" which require agents' private information to generate strong evidence of the true state, even if only with a small probability. With unbounded beliefs, separability is satisfied if action-signals have double thresholds so that at a minimum they reveal whether agents above a threshold number in each cohort choose actions below a choice-threshold. Without double thresholds, learning can be confounded so that agents' actions are forever nontrivially split among the available choices.
Platform Design for Costly Learning
This paper studies the optimal design of a platform to incentivize its users to collectively acquire costly information about the quality of a product (or a service, an investment opportunity, etc.). Users arrive in discrete time. Each user observes information disclosed by the platform and may acquire a costly private signal about the product quality before making his purchase decision. The platform observes users' past purchase history and can disclose any information about past histories to subsequent users. It is shown that if the information environment features negative feedback, it is optimal for the platform to release no information early on to induce user exploration and publish a list of potentially good products at a later point in time, once and for all. On the other hand, if the information environment features positive feedback, it is optimal for the platform to continuously flag projects that are potentially good for an extended period of time right after the product is released. Welfare comparisons with different benchmarks are discussed.
Social Learning under Information Control
The efficiency of market economies and democratic political system depends on the accuracy of individuals’ beliefs. However, centralized information control has, from time to time, hindered efficient societal information aggregation. In this paper, we study to what extent information aggregation in social learning environments is affected by centralized information control of a principal with a state-independent preference. We consider a population of agents who arrive sequentially and obtain information about the state of the world both from their private signals and by observing information about other agents’ actions either exogenously or through the principal. Contrary to the naïve intuition, information aggregation can be very resilient, rather than fragile, provided that the agents have access to some minimal amount of “expanding” observations of others’ actions that are outside of the principal’s control. In general, the learning outcome depends on whether and how the beliefs generated by the private signals are bounded, as well as the type of exogenous observations that agents have.