Reserve price optimization in sponsored search auctions : a multi-stage stochastic programming approach

Details

Supervisors

Jungers, Raphaël M. ; Wibisono, Andre

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en mathématiques appliquées, à finalité spécialisée

Abstract

Sponsored search advertising, a technique that displays one or more ads in search engines such as Google or Facebook whenever someone searches for the products or services of an advertiser, has a remarkable and increasing economic importance. In sponsored search ads, auction mechanisms are usually used for selling the ad slots. When observing the problem from the viewpoint of the search engines (also known as publishers), setting a reserve price is the main mechanism through which they can influence their revenues in the auctions. While there is a relevant body of literature related to the best way to choose reserve prices in sponsored search advertising, there are practical constraints that make the problem more complex. This thesis proposes a model to optimize the reserve price in single-item and generalized second-price auctions, two of the main types of auctions used in sponsored search advertising, while taking the advertisers’ budgets into account. Assuming that publishers are able to learn bid distributions of the advertisers participating in the auctions, the model is formulated as a multi-stage stochastic mixed-integer linear program (MSSMILP), whose objective is to maximize the publisher’s revenue in a set of subsequent auctions. It is proposed that, for the case of a single-item second-price auction, the revenue resulting from this model is always greater or equal to the revenue when using four other approaches to choose the reserve price. Moreover, still for the single-item case, the performance of the model is analyzed and compared to the same four approaches by a numerical example and simulations, in which the solution of the proposed model led to significantly higher revenues.