Speaker: Bob Mungamuru, Stanford University and Stephen Weis, Google, Inc.
Title: Competition and Fraud in Online Advertising Markets
Advertising fraud, particularly click fraud, is a growing concern to the online advertising industry. Unlike many online security threats, ad fraud is primarily motivated by financial gain. Successfully committing ad fraud yields direct monetary gains for attackers at the expense of the victims. Thus, it is natural to consider online ad fraud in an economic context.
The online advertising market can be modeled as a dynamic game between three classes of players: publishers, advertisers, and ad networks. Publishers produce content that draws traffic and sell advertising space to advertisers. Ad networks act as intermediaries who connect publishers and advertisers. In practice, there are many publishers, many advertisers, and relatively few ad networks. Often a single organization may play two roles in the system, for instance, being both a publisher and advertiser or being both a publisher and ad network.
A complete specification of the player types, action spaces, and signals in this dynamic game can quickly become intractable. A publisher's type includes, among other things, the volume of traffic they receive, the quality of their content, and their user demographics and interests. Advertiser types are differentiated by their budget, the value of traffic they receive through ads, the quality of their campaign, and their relevance to particular demographics. Ad networks will differ in their ability to detect ad fraud, as well as the quality and relevance of their ad serving mechanisms.
The players' action spaces can also be of high dimensionality. Publishers may choose to commit fraud by engaging in click inflation, must decide how to allocate both legitimate and fraudulent traffic to competing ad networks, and may choose the amount of resources to invest in developing content or improving their site. Advertisers must decide how to allocate their budget across competing ad networks and may choose to defraud rivals through impression spam and competitive clicking. Ad networks must decide how much revenue to offer each publisher, how to filter fraudulent ad impressions and click-throughs, and which mechanism to use to sell ads, e.g. which type of auction. Modeling keyword-based advertising adds a further layer of complexity.
In our work, we first pose the ad fraud problem in detail. We then make a sequence of simplifications, leading to a tractable model that isolates the strategic interactions between competing ad networks. In particular, each ad network makes a single decision per period - how "aggressively" to filter for ad fraud. If an ad network filters more aggressively, advertisers realize a higher return on their investment, but the publishers and the ad network incur a higher false positive rate. Conversely, if an ad network filters less aggressively, the publishers and the ad network see a short term gain, but advertisers realize lower returns. The publishers and advertisers are assumed to observe the past history and play their best responses. We characterize the equilibria of this game, study convergence properties, and discuss the effect of reputation.