Learning Strategies for Outsourcing Problems With asymmetric Information and Uncertain Execution

Abstract

In this contribution, we consider an outsourcing problem based on a specific principal–agent relationship with hidden characteristics. Under the assumption that the principal knows the probability distribution on the agent’s discrete type space, a standard solution technique for the resulting contracting problem is stochastic optimization on the set of incentive compatible menus of contracts from which the agent can choose a single contract according to the take-it-or-leave-it principle, respectively. Admittedly, this approach neglects any sort of uncertainties in the post-contract phase which is not realistic in many practical environments like production and logistics. To address this issue, we present a novel and holistic problem formulation that links the contracting phase to an uncertain execution phase in a logistical context containing the possibility of renegotiating contracts as a reaction to environmental changes. Since the resulting problem has the character of a multi-round game, we apply well-known concepts from the trendy AI-area of Deep Reinforcement Learning to exploit clever contracting strategies for the principal. Finally, we evaluate our approach inside a computational study.