Mobile web advertising: maximum entropy banner allocation
Description
The worldwide mobile advertising market, currently $3 billion in size, is expected to grow to $20 billion by 2011. Online and mobile advertising employs two main pricing models: pay-per-click (CPC) and pay-per-impression (CPM). To date, most of the academic literature on online advertising has concentrated on auctions designed for the CPC model such as Google's Generalized Second Price auction. This paper will instead focus on the CPM model, also known as banner advertising.
We take the point of view of an ad-serving platform that has made a number of CPM commitments with various advertisers. This ad platform is able to forecast advertising inventory based on historical user behavior. We describe a maximum entropy method for allocating future inventory to advertisers that fulfills commitments in expectation. We also prove that our method is robust against poor predictions by the forecasting oracle.