The Great Reversal

by Thomas Philippon

  Synergies exist in the new economy, but they also exist in the old economy. Leaders of the new economy, just like leaders of the old economy before them, tend to overestimate the positive externalities from their activities. Rana Foroohar, writing in the Financial Times about the gig economy (August 2018), mentions that several years ago, Travis Kalanick, the founder and former chief executive of the ride-sharing company Uber, told a group of business executives that we were heading toward a world in which “traffic wouldn’t exist” within five years. Well, if recent experience in New York City is any guide, that is not happening. The average travel speed of cars in Midtown was 6.4 miles per hour in 2010. In 2017, it was 5 miles per hour, according to the Department of Transportation’s Mobility Report.

  Matias Covarrubias, Germán Gutiérrez, and I (2019) have studied the economies-of-scale hypothesis, and we have not found much support for it. Using detailed data from the US, we estimate the degree of returns to scale across various industries. We perform this analysis separately for the 1988–2000 period and then for the 2001–2016 period. We did not find evidence that returns to scale have increased significantly over the past thirty years. One model suggests that returns to scale are about the same today as they were in the past, and another model suggests they might be about 5 percent higher. Few industries show signs of returns to scale significantly above one.

  There is also no reason to think that intangible assets are more likely to create positive externalities than tangible ones. Patents are a prime example of intangible assets. Many patents today, and most of the litigation surrounding them, come from patent trolls. They abuse the system and create negative externalities. Another example is market research and advertising, or marketing more generally. These have become more important in the digital economy, yet they contain a higher fraction of zero-sum activities than other types of research and development because the gains of one firm are directly related to the losses of another.

  This is not to say that there are no examples of positive externalities in the digital economy. There are indeed increased possibilities of positive synergies in information sharing. Wikipedia is an amazing example. GitHub and Stack Overflow offer developers valuable online help and access to millions of pieces of code.b But the idea that the digital economy is full of positive synergies and returns to scale that can be achieved only through high concentration is misleading, to say the least.

  What’s the Trouble with Big Data?

  Over the past decade, the GAFAMs have developed a large network advantage over potential competitors. They possess troves of data about users that make it difficult or even impossible for rival startups to get a foothold in the marketplace. The collection of data also gives the GAFAMs opportunities to exploit their customers or suppliers.

  As Cornell University professor Saule Omarova observed, “If Amazon can see your bank data and assets, [what is to stop them from] selling you a loan at the maximum price they know you are able to pay?”c

  This highlights the growing issue of price discrimination, defined formally in Box 14.2. Is price discrimination in a market good or bad? The answer depends mostly on whether there is free entry in that market, or at least if the market is contestable. Price discrimination is efficient in the sense that it maximizes the total surplus of all transactions. When the firm has all the information, it can propose a price or a contract that is specific to each client and acceptable to each client. With efficient discrimination, as long as the transaction is economically viable, it will take place. Without discrimination, there are cross-subsidies, and some people can be priced out. The concern, however, is that a monopoly with full information can extract all the surplus. This is the fear expressed by Omarova. The key point here is that free entry becomes more important when firms increase price discrimination.

  Platforms use a variety of tools to limit competition, and sometimes that involves preventing price discrimination. Nobel Prize–winning economist Jean Tirole (2017) emphasizes the role of price coherence, also called the “most favored nation” clause. That name is a metaphor that comes from international trade agreements. The idea is that a platform will prevent its merchants from offering lower prices outside the platform. Online booking services require that restaurants or hotels not offer cheaper prices on their own websites. Amazon imposes similar restrictions on its suppliers in many countries. American Express requires that a merchant not charge a higher price to consumers using their Amex cards even though Amex fees are often higher than those of other credit card companies. An important point here is that the extra costs are paid by customers who do not use the platform because the merchants are forced to charge everyone the same price. To understand why this happens, imagine an economy in which half of the people are rich and enjoy using an expensive card with high merchant fees of 4 percent. The other half use a card with a 2 percent fee. If the merchants could pass on the fees, for a good worth $100 they would charge regular consumers $102 and rich ones $104. In both cases, they would net only $100 from each customer. If they are forced to offer the same price to all, they will charge everyone $103. On average they will still receive $100 net of fees, so they can stay in business. But now the poor and the rich both pay $103. And that is not the end of the story. The rich probably enjoy fancy rewards programs; otherwise, they would be less likely to buy the expensive card in the first place. What happens in this example, then, is that price coherence forces the poor to subsidize the rich.

  Box 14.2. Price Discrimination

  Consider a market with two types of consumers, A and B, and one firm with production cost c. Consumers of type A value the good at va > c. Consumers of type B value the good at vb > va.

  First, let’s look at the case of no price discrimination. If the firm cannot tell A from B, it must offer the same price to both. It has two choices. It could offer the price p = va, and all consumers would accept the offer and buy the good. Its profits would be (va − c) × (na + nb) where na is the number of consumers of type A. But it could also offer the price p = vb and give up on the A consumers. Its profits would be (vb − c) × nb. Giving up on A is a better strategy when (vb − va) nb > (va − c) na. The tradeoff is clear. On the left you have the extra profit from charging B a higher price. On the right you have the loss from not trading with A. The firm is more likely to give up on type A consumers when there are few of them (na is small) or when there is a lot of inequality (vb − va is high). When that happens, type A consumers are priced out, and that can be terribly inefficient and unfair.

  Now we will consider the case of price discrimination. Imagine that the firm can tell A from B. It can then offer two prices, pa = va and pb = vb. There is no risk of type A being priced out. In that sense, price discrimination is efficient. On the other hand, the firm makes a killing: it extracts all the surplus from the consumers. This is why free entry is so important when firms can discriminate. With discrimination and free entry, the market is efficient, and the consumers end up with the surplus.

  What is the solution? You guessed it: free entry! With efficient price discrimination and free entry, we can have the best of both worlds. The combination of good information and free entry should lead to efficient markets with maximum consumer surplus.

  This is an important lesson. Big data makes free entry more important than ever. Big data without free entry might be worse than no big data at all. If we cannot ensure free entry (or credible contestability), then we are better off restricting the ability of firms to gather data. And if we are going to let these firms use our data on a large scale, then we must ensure free entry. What is unacceptable in the current environment is the combination of big data with no meaningful contestability of markets.

  One of the keys to market contestability in the digital age is giving people the property rights to their data. This is where competition and privacy become deeply intertwined.

  Big Data and Privacy

  Privacy and data protection issues have made the headlines. In 2015 Faceb
ook claimed that it had put in place strict policies to restrict outsiders’ access to personal information. Three years later, it disclosed that it had given dozens of companies special access to users’ data. In over 700 pages of responses sent to Congress, Facebook acknowledged that it had shared its users’ data with fifty-two hardware and software companies, many of them previously undisclosed. The new list includes Apple, Microsoft, and Amazon, as well as several Chinese companies, including Huawei. Facebook had been sharing data with device makers, mobile carrier AT&T, and chip designers such as Qualcomm.

  A scoop from Douglas MacMillan and Robert McMillan in the Wall Street Journal (October 8, 2018) showed how Google kept secret knowledge of a glitch involving the private data of hundreds of thousands of Google+ users. A memo reviewed by the newspaper prepared by Google’s legal and policy staff and shared with senior executives warned that disclosing the incident would likely trigger “immediate regulatory interest.”

  Perhaps they needn’t have worried. US regulators have been painfully slow to address these problems, and the debate has migrated to the other side of the Atlantic. The European Union took a bold step in regulating online privacy when it passed the General Data Protection Regulation (GDPR). The GDPR, which took effect in May 2018, gives new rights to individuals and imposes new responsibilities on companies.

  The GDPR restricts firms’ ability to gather personal data without the consent of EU residents. People can ask to see the information gathered about them and require that the information be deleted. Firms have to limit their data collection and delete data that are no longer needed. The GDPR also contains the requirement that data custodians notify users promptly in the event of a breach.d

  The GDPR is bold and ambitious. It is also a big mess. How could it be otherwise? It is the first effort to tackle an immensely important and complicated issue. One of the challenges for executives is that the legislation doesn’t specify how regulators will assess compliance, making it difficult for companies to decide if they have made sufficient changes to their data policies or invested enough in upgrading their systems.

  I was struck by the defensive reactions to this data privacy law that I heard in the US. Instead of acknowledging the issues and proposing improvements, many US commentators poked fun at the new regulation. I have heard countless lawyers complain that the GDPR’s vague and imperfect definitions make compliance more difficult. In other words, they are complaining that by trying to protect the privacy of half a billion people the Europeans are making the jobs of a few hundred compliance officers more complicated. Seriously? This is not just silly, it is also a sign of weakness.

  There are two ways to react when confronted by a daunting challenge. One is to sit down, procrastinate, and find excuses. The other is to try to do something, even if others complain that what you have done is not perfect. Everyone knows we need new data privacy regulations in the digital age. But the GDPR debate has US regulators and lobbyists in the role of the unhelpful complainers and EU members of Parliament in the role of the regulatory entrepreneurs. I would not have predicted such a reversal twenty years ago. It’s unusual for Americans to stand on the sidelines and criticize instead of being midfield and playing.

  Dealing with the GAFAMs

  The issue of concentration in tech raises this question: Are the GAFAMs using their enormous scale to unfairly crush competition? Is it time to break them up? Or would that do more harm than good?

  Let us start by saying that Google, Amazon, Facebook, and Apple in some sense owe their present success to the DoJ, which prevented Microsoft from monopolizing the internet in the late 1990s. It is therefore disingenuous to hear Google claim that antitrust enforcement is not needed. The beneficiaries of an open, competitive system often work to close the system and stifle competition once they are established (Rajan and Zingales, 2003). As successful firms grow large, they seek to alter the political system to their advantage and increase the cost of entry.

  There are three ways to deal with the GAFAMs. These are not mutually exclusive and are ranked in order of controversy: limit their acquisitions, limit their exercise of market power, and break them up.

  Limiting their acquisitions of small companies should be an obvious step at this point. With hindsight, it was arguably a bad idea to let Google buy Waze and DoubleClick, or to let Facebook buy Instagram and WhatsApp. These startups could have become real challengers. The key advantages of incumbents are their customer base and their financial resources. The key advantage of startups is that they are not held back by existing systems and are willing to make risky choices. In the case of the GAFAMs, another advantage of the incumbents is their understanding of the market. They are best positioned to understand before anyone else the potential of a startup. They can buy it early, before it becomes large enough to be noticed. This allows them to escape merger reviews. Generally, premerger notification is required if:

  either party to the proposed transaction has total annual net sales or total assets of at least $100 million and the other party has annual net sales or total assets of at least $10 million;

  and

  as a result of the impending merger, the acquiring party will hold more than $15 million of the acquired party’s stock and assets. An acquisition of another party’s voting securities of less than $15 million also requires reporting if, as a result of the impending acquisition, the acquiring person will hold 50 percent or more of the voting securities of an issuer that has $25 million or more in annual net sales or total assets.

  By buying their would-be competitors early, the GAFAMs exercise too much market power. These acquisitions need to be investigated and limited, but this is easier said than done. There are basically three ways to improve premerger notifications and merger reviews in the internet age.

  One is to lower the threshold for notification and investigation. Germany has tried this approach recently, and the results are not encouraging. To be effective, the threshold needs to be quite low, and it captures many mundane acquisitions between medium-sized companies that do not cause any antitrust concern.

  Another idea would be to use the price of the acquisition as an indicator, instead of the revenues of the target. Facebook paid around $20 billion for the messaging service WhatsApp. This was a shocking price considering that WhatsApp had very low revenues and fewer than fifty employees at the time. WhatsApp, however, had more than 450 million monthly active users, and Facebook knew better than anyone the threat this could pose to its own dominance. The price revealed the true economic importance of the acquisition and could have been used to justify an investigation. The problem with this approach is that acquisition prices, unlike revenues, can be manipulated. Firms could find ways to artificially lower the transaction price in order to avoid scrutiny.

  A third potential improvement in merger review would be to allow the ex post control of some mergers. Ex post control has one particular advantage: it can rely on information about the competitive effect of a merger that was not available at the time of filing. On the other hand, ex post remedies are typically more costly because assets are already commingled. Ex post control also creates legal uncertainty. It is too early to tell which of these ideas, if any, will prove most practical, but it seems to me that the significance of an acquisition should depend on its price, and that ex post controls become more useful in a rapidly changing business environment.

  The next option to discipline the GAFAMs is a bit more controversial. Limiting the exercise of the GAFAMs’ market power directly targets their dominance in some markets. Although the US has been unwilling to follow this route, the EU has challenged market dominance by European companies as well as by American ones operating in Europe. Research by Yale economist Fiona Morton shows that European regulators have been more active in new areas of enforcement than American regulators. The European Commission’s DG Comp has taken on a host of issues related to large tech firms, such as loyalty rebates, dominant positions in IT platforms, and others. Regulators
in Germany and Brussels have begun investigating Facebook, Amazon, and Google for their market dominance and data-gathering activities.

  To limit the excessive dominance of one network, such as Facebook, the authorities could require two features: interoperability (the ability to interconnect with other networks) and data portability (the ability to move one’s data from one network to another). These features are quite similar to the ones that were imposed on Telecom firms in the past.

  Regarding data gathering, regulators could impose a clause to let users opt out of horizontal tracking. Today Google and Facebook are the only companies able to track billions of users on millions of websites, whether the users like it or not. This is how they maintain their dominance in online advertising. An opt-out clause would allow users to decide whether or not they want to let Google and Facebook track them on other websites. Market forces are unlikely to lead to effective opt-out clauses, however. As Dina Srinivasan explains, “First, Facebook itself did not and does not allow consumers to opt-out of the new off-site tracking. Second, Facebook chose to ignore consumers’ explicit requests, enacted via the browsers’ Do No Track option, to not be tracked. Third, when consumers installed ad blockers to circumvent tracking and targeted advertising, Facebook responded by circumventing the users’ installed ad blockers” (Srinivasan, 2019). An effective opt-out option is likely to require regulatory control. This would increase competition and consumer welfare.

  The last option is to break up the GAFAMs. This is the most controversial option, and it would indeed be complicated. One issue is that the activities of the GAFAMs are more integrated than those of AT&T. In the case of AT&T, there was a clear distinction between long-distance service and the local infrastructure companies. It is not clear how a breakup of Amazon or Google would proceed. A breakup might seem like putting the cart before the horse. The priority should be to define privacy regulation and property rights over digital data and give customers effective opt-out clauses.