Decentralized control of a multi-agent stochastic dynamic resource allocation problem

We study a stochastic system consisting of multiple decentralized control agents who allocate shared system resources in response to requests that arrive stochastically over time. Decentralized agents in our system maximize their own objectives subject to potentially mis-specified models of the way in which system resources are consumed by other agents in the system. We introduce the notion of a transfer price, which is a contract that describes the way in which decentralized agents compensate other agents in the system for using shared resources whenever they satisfy a request. We characterize transfer prices under which there is no efficiency loss relative to the optimal system-wide expected revenue when it is managed by a fully-knowledgeable centralized controller. We also show that optimality of these contracts is insensitive to mis-specification by each of the agents of the behavior of other agents in the system.