Reinforcement learning in online principal­-agent problems

• Main Author: Yue, Ming Long
• Other Authors: PUN Chi Seng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Social sciences::Economic theory::Microeconomics; Science::Mathematics::Statistics
• Online Access: https://hdl.handle.net/10356/148487
• Institution: Nanyang Technological University

The principal-agent problem arises when an entity (the agent) acts or makes decisions on behalf of another (the principal) that goes against the best interests of the principal, typically the result of asymmetric information. To address the problem, the principal can align incentives through the use of appropriate contracts. In this paper, we focus on online principal-agent problems. We propose the utilisation of reinforcement learning methods to allow the principal to learn to generate optimal contracts in an end-to-end fashion, solving the principal-agent problem in a model-free manner without the need for prior knowledge of the environment or explicit modelling of the problem.