«Information, Privacy, and the Internet: An Economic Perspective Susan Athey Stanford Graduate School of Business1 Contents Introduction 5 1. ...»
In search advertising, advertisers place standing bids on keywords (for example, “tennis shoes”) that are stored in a database. Bids may be exact match, so that ads are displayed only if the user enters exactly the same term; or “broad match,” where the ads may be displayed on terms (for example, “tennis footwear”) that search engine algorithms determine are “related” (a nebulous concept whose definition may change over time). There are several possible “match types” in practice. Advertisers bid and pay “per click,” so that they pay only when a user clicks on the ads. The ads are “text” ads which appear in a similar format to the “algorithmic” or “natural” results that a search engine returns when a user enters a search. Thus, search advertising is very targeted to the user’s intent at the time: the user is expecting to get links they can click on that relate to the search term, and both the “natural” and “sponsored” results meet that criteria. Although in principle the prices for the advertisements are set in what is known as a “generalized second price auction” (see Varian (2006)), 2 in practice prices are determined by the “pure” auction in combination with various “reserve prices” (minimum prices that must be paid in order to be shown). Reserve prices are set at a very granular level, so that in principle the search engine can control prices at the level of the advertiser X search term.
From the perspective of competition, search advertising is a multisided market. A key feature is that there are economies of scale and indirect network effects across all sides of the market. Like most markets, competition generally enhances welfare of customers on all sides (though perhaps at the cost of duplicated fixed costs).
For other academic articles describing online advertising auctions see, among others, Jain et al.
(2005) and Mahdian, Nazerzadeh, and Saberi (2006) who propose an early system of allocating advertising space, Meek, Chickering, and Wilson (2005) who study incentive-compatible Vickrey auctions, and Aggarwal, Goel, and Motwani (2006), Edelman and Ostrovsky (2007) and Varian (2007) who analyze the generalized second price auction model and its relationship to Vickrey auctions.
The user side of the market is the one most people are most familiar with—most internet users also use search engines frequently to find things on the internet. The user side has economies of scale: the more users, the more data the search engine has to determine what links are most relevant (they record previous clicking behavior on the same or similar searches) and the more users can be harnessed for ongoing experimentation with new and improved algorithms. Search engines run thousands of experiments per year to improve the algorithms used for ranking.
Competition among search engines is important because it provides the incentive for ongoing research and development. It also provides the incentive to create the best possible ranking of results.
If having a scale advantage implies that a dominant firm has an advantage at providing accurate rankings, it also gives the firm wiggle room to manipulate results. If a firm advantages its own affiliated websites, or delay releasing innovations that might be good for users but distract users from clicking on the ads that make you money, it doesn’t suffer too much in terms of lost users if it already has a quality advantage. Edelman and Lai (2013) provide some empirical evidence about manipulation in search, which has been one of the main focuses of the European Commission’s ongoing investigation into Google. See also Edelman and Wright (2012). As discussed below, Athey (2013) provides some evidence from a real-world field experiment about the possibility of manipulation.
On the advertiser side of the market, advertisers are willing to pay the fixed costs of joining a platform and maintaining campaigns if there is enough traffic. Why would a small business bother joining a search engine that just brought a few clicks a month? The more traffic a search engine has, the more advertisers it attracts. In Europe, only a small fraction of mid-to-small sized advertisers advertise on Bing in addition to Google. A dominant search engine realizes that it is a “must buy” for online advertisers, and that it can exercise market power, raising prices without losing advertising dollars. Search engines can engage in very sophisticated pricing schemes, where each advertiser gets a personalized price. The algorithms are not transparent, and so it is difficult for an advertiser to know why their prices might suddenly one day increase. It could just be that they got a bad draw from the algorithm, or it could be that they have been deliberately penalized.
New businesses are particularly vulnerable to excessive prices charged by search engines since they rely more heavily on search advertising to be discovered initially.
Now consider a less obvious side of the market, web publishers— the innovators who create original content on the internet, with advertising revenue as their primary revenue source in many cases.
Many web pages have a little search box in the corner. Those boxes can be very efficient for publishers, since they don’t take up much room on the page, and they are directly useful for consumers.
Search engines compete to provide the results when people search there, because that generates more revenue and gives them more user data. Search engines share some of the revenue back with the publishers. How much? That is determined by competition. If the second place search engine is small and doesn’t attract many advertisers, it doesn’t have as much revenue to share. But then, the dominant firm doesn’t have to share much either, as the competition is not very strong. As a result, the dominant search engine keeps more of the revenue generated by searches. Since the publisher then gets less revenue for every user, there are lower incentives to create web sites and innovate. On the other hand, if there are two evenly matched search engines, most of the revenue goes to the publisher. In large deals, sometimes a smaller search engine will offer to share with publishers an amount in excess of 100% of the revenue they expect to generate, as a way to increase scale. Anticipating that, a dominant search engine might offer to share even more, as a way to prevent the smaller search engine from gaining scale and becoming a more effective competitor.
Keeping the competitor small allows the dominant firm to spend less on R&D to attract users, and to share less revenue with all other publishers in the future on the publisher side of the market.
Linn (2005) describes the bidding between Microsoft and Google for a deal to serve searches for AOL, in which both sides bid aggressively in order to capture the large chunk of consumer traffic. Google eventually paid more to AOL than what it reportedly earned in advertising revenue.
Thus, competition in search advertising affects welfare on all sides of the market: users, advertisers and publishers. Behavior on one side of the market can affect all other sides. Even though search is “free” to users, search engines make decisions that affect consumer welfare, such as investments in R&D and decisions about whether to demote links to competitor sites. Competition creates incentives for firms to make decisions that are more beneficial to the constituents on all sides of the market.
Stutz (2011) provides a detailed discussion of antitrust issues involved in one of the many examples of vertical acquisitions and mergers in search from the last few years, Google’s acquisition of ITA, the online travel search company.
Now consider display advertising. Most internet publishers, particularly online news, relies on advertising as a major or primary source of revenue. Athey, Calvano, and Gans (2013) discuss some impacts of the internet on markets for online advertising in news media, taking the perspective of the theory of multi-sided markets.
Evans (2009) provides an overview of the online advertising industry. Internet publishers may choose to either sell their own “inventory” (spaces on their web pages) directly to advertisers, or they may rely on a third party display advertising platform to provide the ads. In the latter case, the advertising platform selects the ads and charges the advertisers, and shares the revenue with the publishers, less the platform’s fee for providing the matchmaking service.
In some ways, the economics are similar to search advertising, with some important differences. First, in search, a lot of user traffic is “Owned and Operated,” meaning that, for example, users go to Google.com to conduct a search, and Google shows search ads on its own site. Above, we discussed the publisher side of the market, where search engines get traffic from third parties, and share some of the advertising revenue with them. In display advertising platforms, most of the “inventory” of advertising space comes from firms other than the owner of the advertising platform. In addition, the advertisements are usually either only loosely targeted at users (e.g. all users on a particular news website) or targeted by user demographics or past behavior (as inferred by past behavior and recorded in “cookies”). Unlike search, the user’s intent on a page is often only loosely related to the advertising, and the user is often not in the mindset to click on advertising links when, for example, reading online news. Advertisers typically view the display and search markets as distinct.
Levin and Milgrom (2010) provide a discussion of some market design issues in online advertising markets. There are a number of subtleties related to the fact that advertisers may be differentially informed about the value of individual users, due to differential access to cookies. See also Abraham, Athey, Babaioff, and Grubb (2014) for a discussion of the challenges created for auction design in this context.
Competition in display advertising markets has large effects on the internet’s content creators, such as the news media, but also innovative websites that provide consumers information about things like travel. If two display advertising marketplaces are similarly efficient, then they will compete strongly for “inventory” from publishers, charging low fees. On the other hand, if one display advertising platform has a large advantage (e.g. a larger advertiser pool, or unique data about users, their interests, or the set of advertisements the users have seen), they can attract publishers without sharing as high a fraction of revenue with the publishers. That is, they can charge higher fees.
Higher fees charged by display advertising platforms translate into lower revenues for online publishers, and reduced incentives to create content.
The prospect of collecting higher fees gives ad platforms the incentive to amass as much data as possible about users, as well as to ensure that competing ad platforms do not have access to this data. Clearly, this incentive will figure prominently when considering privacy policies.
Overall Importance of Competition in Online Advertising Summarizing, competition in internet search and in online advertising broadly is very important for welfare. In principle, data creates enormous value for advertisers and publishers, but those parties only benefit from the data if the gains are shared with them rather than paid to ad platforms as fees.
Competition policy must deal with the complexities of the multisided nature of the markets. It must also confront the fact that when a business is an information gatekeeper, increased competition may not lead to better behavior. Instead, it can lead to worse behavior from a welfare perspective: the information gatekeeper may choose to decrease rather than increase its quality as a gatekeeper in order to divert customers away from its competitors.
3 Information Gatekeepers, Data, and Competition on the Internet In the early days of the internet, many observers predicted that search costs for consumers would be dramatically lowered by the internet, and thus e-commerce would soon resemble a stylized model of perfect competition. Prices for products would quickly converge to the marginal cost of the products, and consumers would benefit greatly.
However, this view was resoundingly rejected by empirical evidence that showed that price dispersion remained alive and well on the internet. Although search cost did go down, one thing that early observers missed was that the internet would also bring an enormous amount of new content, and so search might still be challenging. (Of course, there are other economic forces behind price dispersion as well). Brynjolfsson, E., and M. Smith (2000), Baye and Morgan (2001), Baye, Morgan, and Scholten (2004), and Ellison and Ellison (2009) all document price dispersion on the internet; see Ghose and Yao (2010) for a survey of some of the findings.
Search Engine Rankings and Consumer Choice Perhaps even more unexpected was the fact that the internet didn’t just fail to make search across websites costless, but that consumers still behave as if search costs are important even when a set of results is presented in an ordered list on a single page, such as on search engines, which help roughly two billion people find information on the internet. An extremely robust finding is that the order of search results matters, not just a little bit, but a lot; and more broadly, the design and layout of web pages, the prominence of various elements of content, and subtle factors like color and font make a large difference in what links consumers click and thus what information they ultimately discover. A key component to how links (both ads and algorithmic results) are ranked in internet search engines is how well these links have performed in the past in getting clicks. Thus, the amount of data available to a website to rank content is crucial to its quality.
As search technology has evolved, search engines have evolved far beyond the original “10 blue links” from the early days of Google.
Today, search engines have become “commerce platforms,” pointing people to the products and services they seek by connecting potential customers to a growing number of online business.
A key question for efficiency in internet services concerns what happens if a search engine decides to place links to its own products and services at the top of the search results page, crowding out results identified by its algorithm as the best. This can have a huge effect, if search rankings are impactful.