The project will unify a variety of approaches to studying information transmission as special cases of a model with stochastic evidence, which captures to situations where agents choose among actions that generate random signals that depend on their types. Agents can then choose what to present to a principal who decides, based on what she observes, what decision to take. This incorporates and generalize aspects of models in the literature where outcomes or signals sent depend on agents' types, such as models with adverse selection combined with moral hazard, Bayesian persuasion, random costly signaling, hard evidence, DeMarzo, Kremer, and Skrzypacz (2017) model of testing and Deb, Pai, and Said (2017) model of forecasting. A first step in analyzing mechanisms in such environments is to provide a useful analog of the Revelation Principle. The general class of mechanisms for these problems is quite complex, involving numerous steps of communication between the agent and the principal. The PIs will analyze conditions under which this can be reduced to a much less complex restricted class of mechanisms. Using this simplification games (without commitment) will be compared to mechanisms (with commitment). This comparison continues the research in Glazer and Rubinstein (2004, 2006), Sher (2011), Hart, Kremer, and Perry (2017), and Ben Porath, Dekel, and Lipman (2017) who studied it in deterministic evidence environments; here it will be extended to stochastic evidence which will be shown to encompass, under some conditions, a wide class of economic environments not typically described as stochastic evidence. An essential question is what class of games (without commitment) to consider. To address this issue the second step is to consider various notions of a mediator-based solution concept. This exploration resembles, in a very different context, the study of the various notions of a mediator and of correlated equilibrium in Bayesian games as in Forges (1993) and Bergemann and Morris (2016). The third step will characterize when commitment does and does not have value. Based on initial examples, it is clear that there are interesting environments where commitment does have a benefit and where it does not. Consider the standard disclosure game in which a principal wants to guess an agent?s type and the agent wants to be thought of as having a high type. While in the deterministic evidence environment there is no benefit to commitment here, with stochastic evidence commitment may have value. By contrast if a principal wants to decide if an agent's type is high enough to merit receiving some good, it turns out that commitment has no value. The fourth step of this work will be to explore the outcomes in particular problems of interest, such as allocation problems and public-good problems. In particular the focus will be on the outcomes and distortions in games and mechanisms in such environments.
|Effective start/end date||9/1/19 → 8/31/22|
- National Science Foundation (SES-1919494)
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.