Dynamic Online-Advertising Auctions as Stochastic Scheduling

Abstract

We study dynamic models of online-advertising auctions in the Internet: advertisers compete for space on a web page over multiple time periods, and the web page displays ads in differentiated slots based on their bids and other considerations. The complex interactions between the advertisers and the website (which owns the web page) is modeled as a dynamic game. Our goal is to derive ad-slot placement and pricing strategies which maximize the expected revenue of the website. We show that the problem can be transformed into a scheduling problem familiar to queueing theorists. When only one advertising slot is available on a webpage, we derive the optimal revenue-maximizing solution by making connections to the familiar cμ rule used in queueing theory. More generally, we show that a cμ-like rule can serve as a good suboptimal solution, while the optimal solution itself may be computed using dynamic programming techniques.