Value of Information in Incentive Design: A Case Study in Simple Congestion Networks

It is well-known that system performance can experience significant degradation from the self-interested choices of human users. Accordingly, in this article, we study the question of how a system operator can exploit system-level knowledge to derive incentives to influence societal behavior and improve system performance. Throughout, we focus on a simple class of routing games where the system operator has uncertainty regarding the network characteristics (i.e., latency functions) and population characteristics (i.e., sensitivity to monetary taxes). Specifically, we address the question of what information can be most effectively exploited in the design of taxation mechanisms to improve system performance. Our main results characterize an optimal marginal-cost taxation mechanism and associated performance guarantee for varying levels of network and population information. The value of a piece of information cannot be known a priori, so we adopt a worst-case interpretation of the value a piece of information is guaranteed to provide. Several interesting observations emerge about the relative value of information, including the fact that the value of population information saturates unless we also acquire more network knowledge.