Revenue Management beyond “estimate, Then Optimize” a Dissertation Submitted to the Department of Electrical Engineering and the Committee on Graduate Studies of Stanford University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

Modern revenue management systems enable firms to make sophisticated pricing decisions over the course of a sales season. Many of these systems operate within what one might refer to as the “estimate, then optimize” paradigm where estimation and optimization are two distinct, interleaved activities. This thesis will discuss two research efforts that explore moving away from such a paradigm. We begin with the study of a simple model of one-resource revenue management that incorporates uncertainty in demand statistics. The opportunity to learn more about demand over the course of the sales season introduces a tension between “exploratory” pricing that attempts to learn quickly and “exploitative” pricing that attempts to exploit existing demand knowledge so as to maximize revenues. We present a simple heuristic that addresses this trade-off in a transparent, operationally intuitive, manner. We establish that pricing decisions that account for this trade-off offer significant increases in revenue over repeated cycles of “estimate, then optimize.” We next turn our attention to the dynamic capacity allocation problem that airlines face. We will present new approximate dynamic programming based algorithms for this large-scale problem that allow for a seamless integration of complex demand forecast models within the optimization framework. The algorithms we develop are scalable to large problems and we present computational results that suggest a significant improvement over methods that rely on frequent re-optimization using a popular linear programming based heuristic.