On Information Policy Design and Strategic Interactions
Zeng, Yishu
2022
Abstract
My dissertation investigates the design of information policy in three different types of strategic interactions: Chapter 1: "Performance Evaluation Design in Dynamic Incentive Contracts", examines how to motivate employees in an organization by strategically disclosing evaluations. Consider a continuous-time principal-agent setting where the agent exerts effort to generate output and the principal subjectively evaluates and pays for the agent's performance. If the principal can commit, what type of evaluation system should she implement? I show that the optimal evaluation system is not based directly on the output, but rather on an adjusted evaluation of output. In particular, it assigns inflated evaluations when the agent's continuation value is low and deflated evaluations when the continuation value is high. Adding such adjustment into evaluations allows the principal to recalibrate the agent's continuation value, which improves the contract by capturing gains from concavification that are not feasible in contracts directly based on output. As a result, adjusting evaluations also induces weakly higher volatility in the agent's continuation value even though the agent is risk-averse. Moreover, I show that additional contractual possibilities such as leaving the firm for an outside option, promotion, and reciprocity in output, could result in strengthening different evaluation biases at the optimum. My results help explain evaluation biases that have been empirically observed in appraisal systems. Chapter 2: "Persuasion of Interacting Receivers", investigates how the consideration of realistic features could complicate the structure of information policy in situations of strategic interaction. Its main contribution is to propose a general framework that allows a formal characterization for several relevant features and provides a tractable way to design optimal information policy in such settings. Specifically, I consider a setting where a sender could communicate privately with multiple receivers before they engage in a one-shot strategic interaction. To understand optimal information policy, standard approaches would focus on signals that recommend actions. However, if there are realistic features outside the scope of the standard model, such a focus could be suboptimal. I consider the following four features: (i) the equilibrium selection rule may be different from the sender-preferred one, (ii) receivers may have private information, (iii) receivers may have non von Neumann--Morgenstern utilities, and (iv) receivers may have heterogeneous priors. I establish a generalized obedience principle. In this version, the sender recommends actions and conjectures. I further provide a sufficient condition under which it is without loss of generality to reduce the messages involved in any signal from continuum to countably many. The construction provides a tractable way to compute optimal signals. I also apply my result to study information policy design in two applications in which the equilibrium selections differ from the sender-preferred selection and receivers may be privately informed. Chapter 3: "Derandomized Persuasion Mechanisms", investigates when focusing only on information policy that either fully reveals or (partially) pools the underlying states without adding extra noise is without loss of generality. I consider a setting where one sender can communicate with several privately informed receivers through a persuasion mechanism before the receivers play a game. I show that for any potentially randomized persuasion mechanism, under certain conditions, there is an effectively equivalent derandomized persuasion mechanism, and these two mechanisms have the same set of equilibria. Overall, this paper provides a rationale for the fact that persuasion mechanisms are often deterministic in practice.Deep Blue DOI
Subjects
Information transmission Persuasion Games of incomplete information Information design
Types
Thesis
Metadata
Show full item recordCollections
Remediation of Harmful Language
The University of Michigan Library aims to describe library materials in a way that respects the people and communities who create, use, and are represented in our collections. Report harmful or offensive language in catalog records, finding aids, or elsewhere in our collections anonymously through our metadata feedback form. More information at Remediation of Harmful Language.
Accessibility
If you are unable to use this file in its current format, please select the Contact Us link and we can modify it to make it more accessible to you.