The Economist recently announced that the world’s most valuable resource is now data, displacing oil for the top spot. The “titans” of the digital era—Alphabet, Amazon, Apple, Facebook and Microsoft—look “unstoppable” as they are now the five “most valuable” firms in the world. Amazon captures “half of all dollars” spent online, and Google and Facebook accounted for “almost all the revenue growth in digital advertising in America last year.” The Economist claims that the “abundance” of data “changes the nature of competition,” and in particular that “with data there are extra network effects” (emphasis added). Ultimately, they declare, “antitrust authorities need to move from the industrial era into the 21st Century.” These statements are interesting, worth thinking about but basically wrong. Big data is about your personal preferences, perhaps even your thoughts and future actions, being for sale. It is about your privacy, not about competition.
“Big data” is the accumulation and digitization of sufficient information about the actions of particular individuals to be able to guess, with better accuracy, how that particular person will behave in a similar situation in the future. In the old days, your local shoe salesman to whom you’ve been going for years, would know that you wear a size 11 and prefer English leather shoes to Italian. He might even call you when a new pair you might like was coming in. You speak to him every 6 to 9 months when you think about getting a new pair of shoes, perhaps you know that he’s married and has one son who’s now at college. Now it’s Amazon, and that data is for sale. It’s no longer John Smith from the shoe store who knows your shoe size and affinity for British shoes, but Nike, Adidas and New & Lingwood, the computers of which send you a friendly good-morning email about their latest innovations every single day. This process is about perfecting knowledge about an individual’s particular demand—what they like to consume and how much they will pay for it. While some people pay more, it may very well be because they value the good or service more than the seller had first thought. Raising price to those individuals isn’t necessarily a reflection of market power; it’s a reflection of product differentiation. Big data allows sellers to identify new and ever-smaller groups to which it can price discriminate, groups, before big data, that were too small to locate. It is important to discuss these issues, but they are privacy issues, they are not antitrust issues. The risk with big data is imposing a regulatory or enforcement regime that limits the accumulation and analysis of data in the name of competition.
What is Big Data?
The purpose, and value, of data is that it allows others to predict behavior. If you observe a person in a variety of different contexts over a period of time, you can develop a pretty good sense of how that person will act in a given situation. That understanding can be highly valuable, particularly when selling something to that person, predicting an illness or, for the more nefarious of us, blackmailing them for personal gain. Data, without more, does not confer market power. Just because a seller knows you really well does not mean that he’s the only seller to whom you can turn for sources of supply. Just because a company has a really good advertising campaign that draws in a lot of customers does not mean that company has market power or that a purchaser couldn’t switch to an economic substitute.
Take a Google search for “pissed.” An American probably means “angry” when she says “pissed.” A Brit probably means “drunk.” So, a search at 830a on a Monday from the Lincoln Tunnel probably means “angry.” Monetizing that knowledge, Google would serve up ads to that person regarding anger management, alternative routes or vacations. On the other hand, a search at 1130p on a Friday from the streets of Cardiff probably means “pub” or “let’s drink ourselves sick…” Google would serve up ads regarding pubs to that person. A repetition of these interactions, where an individual types in a string of characters and clicks through to particular sites in response, gives Google a pretty good understanding of what different words mean to different people in different places. In effect, Google is a dictionary. Serving up anger management ads to the Welshie won’t produce any click-throughs. Serving up a list of “pubs” that are open and serving will. Capturing that process creates valuable data.
Now let’s compare Facebook and Google. Say you Google “flowers.” It’s spring time, and Google knows, through other people’s searches, that people in your area are looking to purchase cherry trees, so it serves up ads for nurseries. Now, let’s say you type that same word into Facebook’s search line. But, you’ve told Facebook that your great-grandmother, “Nana,” lives in Dallas. It knows further that she has just died. Rather than show you cherry tree sources in Washington, it shows you florists in Dallas. Moreover, because it’s your beloved Nana (because 83.7% of your posts on her page include the word “love” in them) and perhaps because you buy £400 shoes from England, it knows you are looking for a nicer bouquet appropriate for a wake. Analyses of these relationships (looking at terms, correlating clicks and purchases) give Facebook a more thorough understanding of who you are and what you are likely to buy and when.
Now take Amazon. They actually know what you search for and what you actually buy, when, and at what price.
Free delivery is designed to make Amazon your first choice when you are looking to buy something, increasing the number of sales but enhancing their data set as well. Google’s ancillary services, like email, phone and the like, are also designed to get you to their platforms so they can get even more data, in new and different contexts. Facebook’s web of relationships with small to medium sized ecommerce sites is designed to do the same. Its links with Amazon increase the value of the data. You search for and buy Saucony running shoes, but only at a discount, on Amazon. This fact tells Facebook not to serve up an ad for the most recent Nike release at $250 a pair, but a Nordstrom Rack Saucony at $75, 50 percent off. It also tells Facebook that you might like that new discount upstart sportswear provider…
Extra Network Effects?
Network effects are present if the value of participation in a network increases with the number of participants. There is a mild, lagging network effect with Google. The more people search, the better it understands what one means when one types something in. But that effect has limits. Knowing that someone from New York thinks “pissed” means angry is not helpful when trying to sell something to someone in Wales. That effect can also be fleeting. Some words don’t change meaning much over time; others come and go. “New Kids” would have had a very different meaning 20 years ago than 10 years ago than 10 years from now. Contrast this effect with a classic network effect—ATM networks. I am a small bank with 10,000 local customers. Most of those customers travel. I can offer my own ATM network, consisting of the machine in front of my retail locations. I could join “Dave’s ATM Network” which has three or four local chains in the tri-state area. Or I could join Cirrus, with over 2 million ATMs around the world. The more banks that participate on a network, the more desirable it is to be a bank on the network, and the more desirable it is to be a customer of one of those banks. With search, the network effect is lagging. A user uses the service; it becomes more valuable for the next user. With ATMs, the network effect is immediate. A system with 2 million banks is more valuable than one with 1 million or 500.
These extra network effects are even more limited when it comes to purchasing. Just because I will only buy discount running shoes doesn’t mean my neighbor will. Google’s purchasing tool does not become more valuable to me because it knows my neighbor is marginally less price sensitive than I am. Perhaps with many purchases in its database, Google would be able to identify others with very similar profiles to mine and then be able to sell me things I didn’t know I wanted. That value, attributable to more participation, however, seems to be marginal.
The more data a platform has, however the more desirable it is to advertisers. The analysis is similar to the one used to analyze television advertising markets. A particular television program may generate a lot of viewers at any given moment, but those viewers aren’t locked in. Their tastes can change; the shows can change. Search may have a little bit more longevity than a television show as most words don’t change meaning that quickly. Switching costs are very low, however. It’s just as easy to type “bing.com” as it is “google.com” or to select a button on your search bar. Moreover, since the value of search is in its ability to understand what people want, as soon as there is meaningful movement on an engine for a particular term, that engine begins to gain value and desirability.
That’s the error the Economist makes with Facebook. Facebook is desirable because a good many of your friends and family are on it. It’s an easy way to communicate with friends. The more that are on it, the more desirable a social network platform it is. But just because your Nana and 175 of your closest friends from high school you’ve not spoken to in years are on it doesn’t mean it suddenly becomes the exclusive means for you to acquire shoes or flowers. Moreover, being on one social network does not diminish the value of being on others. LinkedIn is a business network. The now-moribund Vine was popular with video-oriented kids, as is Instagram. Indeed, I could see some networks losing value depending on who was on it. I might not want my mother seeing an Insta of me doing shots in Cabo with my frat brothers.
It simply is insufficient to say “data is oil” and be done with it. You need to look at what the data is, what it tells you, and, most importantly, what product market the data is relevant to. Fundamentally, all data is at this point is a way of predicting sales and maximizing profits for underlying products. These entities are perfecting their ways of finding data, digitizing it and correlating it to behavior. It truly is an innovation to figure out how to scan a series of text entries made by a particular user and figure out that he has a Nana, who lives in Dallas, and who is now dead. It’s an even greater feat to understand that that person will spend $150 on a floral arrangement right then and there if you show him the right image. That ability does not confer market power to the Dallas florist or to Facebook. There is nothing preventing that individual from going to Google to look up florists, or to call a relative or the funeral home and ask. That data does not confer market power. A network effect in the platform does not correlate to a network effect in the market that includes the subject good.
So what is market power for big data?
The classic articulation of market power is the ability to sustain a supracompetitive price or exclude competition. Search engines and social networks may have large followings, but that doesn’t mean individuals cannot switch easily. I can be a member of Facebook, LinkedIn and Instagram. I can “be” different people on all of them. I can have multiple accounts. And their existence and availability do nothing to dissuade me from joining Musical.ly, an entirely new and different platform, so that I can spy on my children. Moreover, one doesn’t naturally think of Facebook or Amazon as a search engine, but they are quickly becoming them. Indeed, their growing usefulness as search engines is the very disruption the Economist believes doesn’t exist in data anymore.
If big data is the product, then the customer is the entity that buys ads in the hopes of selling their wares to the platforms’ users. They are the ones buying your data. They are the ones that want to know to whom they can sell their product for the most money and at what time. You are not the customer.
A few scenarios:
Scenario One. Facebook contracts with Amazon to purchase Amazon’s customer data. Cross-licensing data between big data firms shouldn’t be anticompetitive if the firms are free to use the data as they wish. If the licensing includes restrictions on the entities to which big data can sell or use the resulting, combined data, that may be a restraint. It prohibits the parties from competing on the merits for their product, the data analyses.
Scenario Two. A merger between two big data companies. One would want to look at the common customers. The reason is to get an idea about how similar the data sets are. If they are very closely related, then customers could be pitching one entity off the other. If they are complementary—say Big Data One knows boomers well; Big Data Two knows millennials well—then the merger should be less of a problem. The key here would be to look at what exactly the customers are buying rather than what the platform does. A purchaser of an ad may in fact just be looking to move product. It may very well look at the number of sales a particular platform can close rather than how it draws customers to that platform. Nike doesn’t care if it is selling a shoe to a text poster or a photograph sharer. In that vein, an ad on Google could very well be a viable substitute for an ad on Facebook even though the users of those platforms are using them for very different purposes.
Scenario Three. A text-based social network acquires a picture-based social network. The text-based network is dominant, with millions of student users across the country. The picture-based network is a startup and offers great picture editing features. From a data perspective, the merger adds to the text-based networks data set and adds to its efficiency. Ad purchasers likely do not view the two as close substitutes, and so wouldn’t necessarily be losing a major constraint on their other primary source’s ability to raise price. The ability to offer cutting edge photo editing features to its customers may make the text-based network more useful to existing users and draw in new ones. The picture-based network users may not find being added to the larger network desirable, however.
The fact that users don’t like the merger is interesting. They are getting access to the platform and its services for free. Is it appropriate to ignore their concerns since they aren’t paying? And what’s the proper metric to measure potential harm? A price increase of any amount is of infinite proportion (division by zero error). And if you start to charge something that was for free, even if for a single penny, you will lose customers. So, the standard metrics don’t really lend themselves to this type of product. If you raise price by 1¢, you’ll be able to keep customers, and some will leave—some out of principle, some because they don’t want to be bothered giving over credit card information. Their loss to the platform provider appears to be costless. So the price increase, of infinite size, is profitable. That would mean, however, that every merger involving a free platform service would be anticompetitive. But, since the price increase could be effected absent the merger (the platform provider could raise price at any time) and achieve the same “profitability,” the ability to raise price profitably would not necessarily be merger-specific, so, at least under this view, no merger involving a free platform service would ever involve a merger-specific increase in market power.
One could argue that the loss in eyeballs could reduce revenues associated with advertising. If a platform lost 100,000 eyeballs, for example, it would lose a number of advertisers. If that price increase did not result in a significant enough loss in advertising revenue, one could argue that the price increase did reflect market power. The issue of whether that’s merger-specific would still be there, though. One might be able to look at eyeball losses pre-merger and post-merger and see whether the curve itself shifts at all. In this sense the advertising revenue could very well be a rough approximate of the value customers’ presence imparts to the site.
Using advertising switching as a proxy for platform user welfare also avoids the need to address whether users can accurately assess the value of their participation on those sites to the site owners. How much is my personal information really worth? Is that additional penny of cost, and its associated administrative burdens and opportunity costs, small or big in comparison to the total value of the personal information I’m parting with? Using the advertising market as a proxy monetizes that value in at least a fairly consistent way across the user base. A deep analysis into assessing user value may in fact not be necessary. This conceptualization also means that a platform user’s welfare is in fact not ignored. To the extent a user’s experience between platforms differs, the “advertising profile” of those sites should also differ. The less similar, the less likely they are substitutes to an advertiser, and so the less harm to competition their merger might pose.
Using advertising dollars as a metric should also take into account “innovation.” The more innovative and desirable a platform, the more likely users will flock to that platform, and the more desirable it would be to advertise with it. If there is significant switching by advertisers from a dominant platform to an upstart platform, one could view that switching, particularly if supported by extrinsic evidence, as evidence that the upstart is a market maverick.
Kill the geezers
Big data isn’t oil. It brings more products to more people at optimally efficient prices. I don’t wake up in the morning and enjoy a life-enriching draught of data. I don’t hop into my big data and warp to Paris for the day. The antitrust process is designed to find unfair accumulation of market power and disperse it. The lag comes when deciding whether a new business activity is “innovative” or “exploitative.” Standard Oil was formed in 1870. The Supreme Court sustained its breakup in 1911. In 1983, Steve Salop and Dave Scheffman described Rockefeller’s rail freight conspiracy in the Pennsylvania oil fields as “raising rivals’ costs.” It took economists almost a century to articulate why Rockefeller’s rail freight conspiracy harmed consumers. But it didn’t take that long to break it up.
The fundamental question antitrust addresses is whether an activity is good or bad for the economy and society. Over the years, the metric for how one makes that assessment has changed. At present, the Chicago School still dominates, asking whether the activity is efficiency enhancing. Under that metric, big data reflects a new and innovative product that deserves protection under Grinnell. But if big data is “bad,” according to The Economist, antitrust itself needs a realignment away from Grinnell and the Chicago School; it needs to reassess the usefulness of the Chicago School as a metric. But eliminating efficiency as a standard takes us back to the Von’s Grocery Store style “Big is Bad” type of antitrust. The Economist also dismisses this approach—“size is not a crime”—leaving us with some passing and unsatisfying suggestion to look at deal value.
Antitrust can also be viewed as a tool to disperse socio-economic power to relieve pressure on the proletariat. The ability to identify ever-smaller price discrimination markets means, to some extent, that certain classes of people were receiving a benefit from seller ignorance. They would have paid more had the seller charged more. These customers could very well be described as free riders, exploiting a lack of seller knowledge, an informational asymmetry, that benefited them. To the extent big data gives large corporations the ability to identify and exploit informational asymmetries to its own benefit, which it can do because consumers misunderstand the value of their own data, perhaps big data is the appropriate target of enforcers. But this is just “big is bad” using fancier terms. And, there’s also no real way to break up big data. You cannot order a company to divest some of its understanding of a particular person or product.
Fundamentally, this is the problem with big data—people can’t value the information they are giving away. That is a consumer protection problem, not antitrust. As such, we really don’t have to kill all those elderly industrialist antitrust enforcers and replace them with the young and hip. The Oldes are doing just fine.