Information Design for Platform-Enabled Operations
Information design studies how an informed player (planner) can optimally provision information over an uncertain payoff relevant random variable to influence the actions of less-informed players (agents). Codified through a ``signaling mechanism", the informed player can design distributions o...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Massachusetts Institute of Technology
2024
|
| Online Access: | https://hdl.handle.net/1721.1/155500 |
| _version_ | 1811072896390397952 |
|---|---|
| author | Shah, Sohil |
| author2 | Amin, Saurabh |
| author_facet | Amin, Saurabh Shah, Sohil |
| author_sort | Shah, Sohil |
| collection | MIT |
| description | Information design studies how an informed player (planner) can optimally provision information over an uncertain payoff relevant random variable to influence the actions of less-informed players (agents). Codified through a ``signaling mechanism", the informed player can design distributions over informative signals to reveal depending on the value of the uncertain random variable. Through the design of the signaling mechanism, a planner can affect the agents' posterior beliefs of the uncertainty and the agents' consequent actions. While an abstract representation, solving for the optimal signaling mechanism provides valuable real-world intuition into how platforms or public entities should provision information.
However, the practical design of these optimal signaling mechanisms more generally is associated with five key technical challenges. First, the uncertainty set can be continuously-valued which leads to an uncountably large set of decision variables over which to optimize. Second, the objective of the planner and the response of agents to the induced beliefs can be arbitrarily complex which can lead to intractable optimization formulations. Third, in dynamic settings, planners with multi-period objectives need to provision information across time, but provisioning information in the present affects what information the planner can provision in the future. This necessitates the use of computationally intensive multiperiod dynamic programming. Fourth, if agents in these dynamic settings are long-run and subject to subgame perfection, agents anticipate what information the planners will adaptively provision in the future necessitating one to solve a coupled dynamic program. Fifth, planners may find themselves in competition over the provision of information, aiming to gain favor in strategic interactions where both the quality and the content of the information revealed matter.
This thesis presents a study of information design as a means to improve platform operations. We formulate models that address each of the five technical challenges described in the context of a particular practical application. Examples of the practical applications we consider include pandemic management, ride-hailing, and incentivizing research and development. In the first chapter, continuous-valued uncertainty and optimization methods robust to planner preference are addressed. We consider a planner using information design to manage a population of hybrid workers amidst the spread of a disease with uncertain infectious risk. We identify closed-form solutions for the optimal signaling mechanism over the risk for a general class of set-based objectives and we identify computationally efficient algorithms to approximate the optimal signaling mechanism for general Lipschitz-continuous objectives. In the second chapter, dynamicity is addressed. We consider a dynamic model where the planner iteratively provisions information to agents with time-varying preferences. The third chapter addresses long-run agents. We focus on a setting where agents serve a transportation platform affected by surge pricing and we identify the optimal signaling mechanisms that provision information over the timing of the uncertain surge. In the final chapter, we address a competitive information design setting. We consider a Bayesian Stackelberg game with a malicious attacker performing a sequential search over a set of firms under costly inspection. Firms can pay to mitigate the probability that the attacker succeeds should they be chosen as a target. Firms can then also choose how much information to reveal on inspection about the attacker's probability of success. We characterize the equilibrium mitigation and signaling strategies of the firms. |
| first_indexed | 2024-09-23T09:20:11Z |
| format | Thesis |
| id | mit-1721.1/155500 |
| institution | Massachusetts Institute of Technology |
| last_indexed | 2024-09-23T09:20:11Z |
| publishDate | 2024 |
| publisher | Massachusetts Institute of Technology |
| record_format | dspace |
| spelling | mit-1721.1/1555002024-07-09T03:54:59Z Information Design for Platform-Enabled Operations Shah, Sohil Amin, Saurabh Jaillet, Patrick Massachusetts Institute of Technology. Operations Research Center Sloan School of Management Information design studies how an informed player (planner) can optimally provision information over an uncertain payoff relevant random variable to influence the actions of less-informed players (agents). Codified through a ``signaling mechanism", the informed player can design distributions over informative signals to reveal depending on the value of the uncertain random variable. Through the design of the signaling mechanism, a planner can affect the agents' posterior beliefs of the uncertainty and the agents' consequent actions. While an abstract representation, solving for the optimal signaling mechanism provides valuable real-world intuition into how platforms or public entities should provision information. However, the practical design of these optimal signaling mechanisms more generally is associated with five key technical challenges. First, the uncertainty set can be continuously-valued which leads to an uncountably large set of decision variables over which to optimize. Second, the objective of the planner and the response of agents to the induced beliefs can be arbitrarily complex which can lead to intractable optimization formulations. Third, in dynamic settings, planners with multi-period objectives need to provision information across time, but provisioning information in the present affects what information the planner can provision in the future. This necessitates the use of computationally intensive multiperiod dynamic programming. Fourth, if agents in these dynamic settings are long-run and subject to subgame perfection, agents anticipate what information the planners will adaptively provision in the future necessitating one to solve a coupled dynamic program. Fifth, planners may find themselves in competition over the provision of information, aiming to gain favor in strategic interactions where both the quality and the content of the information revealed matter. This thesis presents a study of information design as a means to improve platform operations. We formulate models that address each of the five technical challenges described in the context of a particular practical application. Examples of the practical applications we consider include pandemic management, ride-hailing, and incentivizing research and development. In the first chapter, continuous-valued uncertainty and optimization methods robust to planner preference are addressed. We consider a planner using information design to manage a population of hybrid workers amidst the spread of a disease with uncertain infectious risk. We identify closed-form solutions for the optimal signaling mechanism over the risk for a general class of set-based objectives and we identify computationally efficient algorithms to approximate the optimal signaling mechanism for general Lipschitz-continuous objectives. In the second chapter, dynamicity is addressed. We consider a dynamic model where the planner iteratively provisions information to agents with time-varying preferences. The third chapter addresses long-run agents. We focus on a setting where agents serve a transportation platform affected by surge pricing and we identify the optimal signaling mechanisms that provision information over the timing of the uncertain surge. In the final chapter, we address a competitive information design setting. We consider a Bayesian Stackelberg game with a malicious attacker performing a sequential search over a set of firms under costly inspection. Firms can pay to mitigate the probability that the attacker succeeds should they be chosen as a target. Firms can then also choose how much information to reveal on inspection about the attacker's probability of success. We characterize the equilibrium mitigation and signaling strategies of the firms. Ph.D. 2024-07-08T18:55:25Z 2024-07-08T18:55:25Z 2024-05 2024-05-30T21:19:09.428Z Thesis https://hdl.handle.net/1721.1/155500 Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) Copyright retained by author(s) https://creativecommons.org/licenses/by-nc-nd/4.0/ application/pdf Massachusetts Institute of Technology |
| spellingShingle | Shah, Sohil Information Design for Platform-Enabled Operations |
| title | Information Design for Platform-Enabled Operations |
| title_full | Information Design for Platform-Enabled Operations |
| title_fullStr | Information Design for Platform-Enabled Operations |
| title_full_unstemmed | Information Design for Platform-Enabled Operations |
| title_short | Information Design for Platform-Enabled Operations |
| title_sort | information design for platform enabled operations |
| url | https://hdl.handle.net/1721.1/155500 |
| work_keys_str_mv | AT shahsohil informationdesignforplatformenabledoperations |