Joint bidding in ad auctions

Abstract

In traditional advertising auctions, commodity suppliers as advertisers compete for adverting positions to display commodities. As e-commerce platforms become more prevalent, offline retailers are also opening online virtual shops, and retailers are starting to pay a fee for extra exposure of their shops. This has led to situations where a single commodity may be sponsored by both the retailer and the supplier, offering opportunities for more profit. In order to explore this novel advertising pattern, we propose a new model called the joint advertising system (JAS), where retailers and suppliers jointly bid for advertising positions. In the context of this realistic scenario, conventional mechanisms such as GFP, GSP and Myerson auction cannot be applied directly. Besides, the VCG mechanism results in negative revenue in JAS. To solve this issue, we modify the payment rule of VCG to create a revised VCG mechanism that guarantees incentive compatible, individually rational and weakly budget-balanced. Additionally, we leverage the structure of the affine maximizer auction (AMA) and the technique of automated mechanism design to train joint AMA. Finally, we conduct several experiments to demonstrate the performance of the joint AMA. It turns out that our mechanism maintains good economic properties and outperforms other mechanisms in various settings.