Kailin Chen

Research

1.  Communication with Multiple Senders. arXiv

Previously circulated under the title ``Learning from Strategic Sources".

Extended abstract in EC’24. 

This paper analyzes a cheap talk model with one receiver and multiple senders. Each sender observes a noisy signal regarding an unknown state of the world. Existing literature (e.g., Levit and Malenko, 2011; Battaglini, 2017) focuses on scenarios where the receiver and senders have aligned preferences in each state. We further explore situations with disagreement states in which the receiver and the senders have misaligned preferences. We first show that, when the number of senders grows large, each sender's message must convey almost no information to the receiver. Furthermore, we identify a discontinuity in information transmission: with moderate conflict between the receiver and the senders, introducing an arbitrarily small probability of disagreement states causes complete unraveling, contrary to full learning predicted by the literature. Finally, we demonstrate that the receiver cannot fully learn the state even when receiving messages from arbitrarily many senders.

2. Ranking Statistical Experiments via the Linear Convex Order and Lorenz Zonoid: Economic Applications. arXiv

Previously circulated under the title ``Experiments in the Linear Convex Order".

Extended abstract in EC’25. 

This paper introduces a novel ranking of statistical experiments based on the linear convex order. This ranking encompasses a broader range of scenarios where intuition suggests that one experiment is more informative than another, and offers more tractable characterizations than the Blackwell order, which is based on the convex order. Moreover, this ranking admits an elegant geometric characterization through the Lorenz zonoid, which generalizes the Lorenz curve to multidimensional settings. We apply the ranking to compare experiments in binary-action decision problems and in problems with quasiconcave payoffs, as analyzed by Kolotilin, Corrao, and Wolitzky (2025). Furthermore, the ranking enables comparison of experiments in moral hazard problems, complementing the findings in Holmström (1979) and Kim (1995). Finally, the ranking applies to comparing experiments generating ex post signals in screening problems.

3.  Collective Experimentation with Correlated Payoffs

Revise & Resubmit, Theoretical Economics. 

This paper studies an exponential bandit model in which a group of agents collectively decide whether to undertake a risky action. This action is implemented if the fraction of agents voting for it exceeds a predetermined threshold. Building on Strulovici (2010), which assumes the agents' payoffs are independent, we explore the case in which the agents' payoffs are correlated. During experimentation, each agent learns individually whether she benefits from the risky action; in this way, she also gains information about its overall desirability. Furthermore, each agent is able to learn indirectly from the others, because in making her decisions, she conditions on being pivotal (i.e., she assumes her vote will determine the collective outcome). We show that, when the number of agents is large, increasing the threshold for implementing the risky action leads to increased experimentation. However, information regarding the overall desirability of the risky action is effectively aggregated only if the threshold is sufficiently low. 

4.  Fishing for Approval Supplementary Appendix (with Stephan Lauermann and  Mehmet Ekmekci)

There is a concern that a biased agent might fish for approval if one assent undoes all past rejections. In this paper, we study a sequential voting model in which a biased organizer engages in a costly search to solicit one vote for his preferred policy. Voters have common, state-dependent preferences. The organizer is informed about the realized state while the voters obtain noisy information via private signals. We show that, somewhat paradoxically, rather than hurting, the organizer's ability to fish for approval helps the voters, often leading to the voters' first-best, full-information equivalent outcome.