A Theory of Leadership in Online Advertising

September 15th , 2011

The investigation of the European Commission on Google is focusing a lot of attention on the market for search advertising, whose particular structure has attracted recent attention of the theoretical and empirical economic research on multisided markets. Understanding this special market, its borders, its structure and the role of its leader is crucial for antitrust analysis. As of now, we lack a theory of leadership in multisided markets (most theory in the field is about symmetric competition), which is essential to evaluate abuse of dominance in online advertsing in a rigorous way. Below we analyze what we believe to be the main aspects of the structure of the market for online advertising, which would deserve a formal analysis in leader-followers models.

The main aspect of search is the pervasive role of network externalities: the more Internet users reach a search platform, the more effective is an advertising campaign on the same platform. Since Internet users often join the platform with a commercial purpose in mind and use it to find information on products, advertising can be tailored on them in a much more effective way than with other means. Moreover, continuous technological innovations open new ways for advertisers to be always more effective on a platform online. This generates a very limited substitutability between online and traditional advertising.

In spite of this, some researchers (Ratliff and Rubinfield, 2010) have advanced the hypothesis that some substitutability between online and traditional advertising could exist because they both serve broad advertising goals. However, this neglects a key difference: traditional advertising is aimed at building a general brand-awareness usually without a specific target audience, while search advertising is largely aimed at generating market transactions online. From this perspective, traditional and online advertising are almost always complements rather than substitutes - Analogously, Amazon may compete with traditional bookstores, but does not compete with traditional advertisers of books (whose services actually promote the business of Amazon rather than being substitutes for it). Empirical evidence on the subject, however, is still too limited.

Having defined search advertising as a reasonably separate market, we can move to analyze its structure. Search advertising is aimed at direct demand fulfilment, as witnessed by the "text-only" composition and the payments on a cost per click basis; moreover, platforms compete to conquer the largest number of visitors, that is they compete in quantities, and charge advertisers for the clicks they receive on their ads. All the search engine platforms choose a quantity of advertisements to be made available in a specific order for any search query. Given the space allocated to these sponsored links, an auction pins down the market clearing price for these advertisements. The highest bid for each keyword association wins, with the price given by the second highest bid: this tends to force bidders to reveal their true evaluation of the click, which allows the auctioneer to choose the profit maximizing auction mechanism. Contrary what is usually claimed, these auctions do not necessarily add a competitive element to online advertising, but allow a dominant firm to adopt complex forms of price discrimination aimed at excluding competitors in search advertising or at extracting all the surplus from the bidders. Moreover, the auction process is made more complex by the different places where the ad can appear on the search page, and remains largely obscure to advertisers and competitors. Finally, most platforms have recently introduced forms of click-weighted auctions that weight bids according to secret algorithms to give priority to advertisements with a larger chance to be clicked on: this mechanism is aimed at increasing the effective willingness to pay.

The structure of the market for search advertising depends on a particular form of network externality which is quite different from the one emerging in other markets of the New Economy. This is due to two main reasons, one related to technological issues and one to the role of multihoming. Let us start from the first. Network effects in search are combined with a form of learning by doing: search engines find more relevant results for each query when there are more queries and the subsequent clicks of the users provide information on what are the most relevant websites associated with keywords. Through this feedback, users improve the algorithms that govern the search engine (with a larger impact on tail queries compared to common queries on which all search engines reach a relatively large feedback). Therefore, not only more search generates more demand for advertising (as in any market with network effects), but more search generates also the scale needed to improve the search technology and provide more relevant results and ads, which in turn generates more search. This is a key difference compared to other markets with network externalities, as those of other software platforms: on one side, in traditional platforms the number of consumers determines the demand of application developers but does not affect the quality of the software platform for a given number of applications, on the other side, in search platforms the number of visitors determines the demand of advertisers and also the (future) quality of the search engine. This combination of network effects and learning by doing induces initial increasing returns to scale in this market, that are reflected in the pattern of the "revenue per search", which has a typical S shape in the size of the web visitors.

As a consequence, for a platform to enter the search advertising market (or increase its market share) it is crucial to rapidly gain scale and close the gap on the search queries. At the same time, for a leading platform to maintain its dominance it is crucial to protect the information gained through search and limit the scale of the rivals. Any exclusivity agreement between the dominant firm and hardware or software distributors to install only its search-related products and services or between the dominant firm and advertisers to rely only on its platform may jeopardize the hopes of the competitors to gain scale and compete on the merit. This may be the case of the exclusivity agreements on the Google toolbar or on the search default settings between Google and software vendors as Adobe, hardware vendors as Apple or browser distributors as Firefox, Safari and Opera. Moroever, scale requires that any search engine must be able to access all websites and "crawl" them to find new information. Clearly, if the dominant platform obtains a privileged access to some relevant websites and limits the access of competitors, competition is penalized: such a strategy of raising rivals' costs may exclude some competitors (or accommodate market outcomes with high prices that are detrimental to advertisers). In this case, innovation by the followers is penalized as well. This is what may have happened since the acquisition by Google of YouTube, the main website for video contents, whose access for competing search engines does not appear to be as direct and immediate as for Google.

The other reason for which network effects in search are different from those of other markets is that multihoming on both sides (by web visitors and advertisers) can easily spread benefits between different providers. Of course, multihoming by consumers is key to drive the accumulation of information needed to build scale for a minor search engine. The fact that search engines are free and, most of all, are always "one click away" on the Internet plays a double role in this context. On one side, it allows consumers starting from the dominant search engine to easily try alternatives to test their capabilities, which enhances the chances for minor search engines to take off. On the other side, multihoming by consumers allows those who are experimenting a minor search engine to quickly return to the dominant one jeopardizing the chances of the former to develop and protecting the advantage of the latter.

Equally important is multihoming on the advertisers' side. Since advertisers are uniquely interested in the effectiveness of their spending, they have good reasons to diversify their investments between alternative platforms in such a way that the marginal returns are equalized. Multihoming guarantees that different Internet users can be reached with different search engines, typically with a higher budget destined to the leading channel (currently Google AdWord) and a smaller budget shared across the others (as Yahoo! Panama or Microsoft AdCenter). Moreover, data analysis can easily allow for a comparison of the return on investment in each channel to optimize spending. It is clear that multihoming by advertisers would contribute to the development of scale and efficiency for minor search platforms. Notice that this is quite different from what emerges in some of the other software platforms where multihoming on both sides is often not necessary (think of the market for videogames, where consumers choose a single platform, but multihoming by game developers guarantees network effects for all the platforms). As a consequence, any policy aimed at limiting multihoming creates obstacles to network effects. This may be the case for the contracts with which Google prohibits advertisers from using competing platforms, and for the exclusive use by Google of the data on its clients which prevents them from performing data analysis to compare the return on investment of different advertising channels.

In conclusion, the impact of these strategies should be evaluated theoretically in models of leadership in the competition between platforms, and empirically with new investigations on online advertising: this should be a priority of research in the field.

Send us your feedback and comments about this article