GBS: Who’s Afraid of the Algorithm?


I’m never going to get around to writing about the antitrust issues if I save the whole thing for one big post. Let me try hacking smaller chunks off and see if that approach works.

The question for today is whether we should be afraid that the individual e-books available through the Consumer Purchase program will be priced unfairly high. Note that I’ll be looking only at individual books, not at the all-you-can-eat Institutional Subscription, which raises very different economic issues. I’m also going to look only at books whose rightsholders have claimed them with the Registry, since the antitrust issues with unclaimed and orphan works are also subtler. In this post, I’ll be looking only at whether booksellers will take advantage of Google’s marketplace for e-books to collude in setting their prices for individual e-books above the competitive level.

My current conclusion, following the arguments made by Einer Elhauge, is that there is not an economic danger here. Any potential cartel is unstable; Google’s role in algorithmically setting prices is not likely to lead to trouble. The only potential issue is transparency; if Google successfully pulls one over on both the public and on copyright owners with its algorithm, it could put the squeeze on readers.

An Antitrust Economics Review

You can skip this section if you feel comfortable with basic antitrust economics.

The standard economic model of antitrust is that in a competitive market, sellers are forced to lower their prices to compete with each other. This drives the prices down to the point at which the sellers just barely cover their costs. The result is that the price charged will be below what many buyers are willing to pay; the difference is consumer surplus and is good for society.

In a cartel, the sellers collectively agree to raise their prices to some higher level. (I’m going to assume no price discrimination; that’s a topic for another day and another blog post.) These higher prices have two effects. First, some buyers are driven out of the market because the price is higher than they’re willing to pay. Second, the buyers who stay in the market pay more. Fewer units are sold, but at a higher price. The difference in price, times the number of buyers who stay in the market, is now money going into the pockets of the sellers, rather than benefits accruing to buyers who paid less than the value they derive from the product. That means some consumer surplus has been transformed into producer profits: good for sellers and bad for buyers.

But notice that some buyers have dropped out entirely. The consumer surplus they used to enjoy is now gone. It’s a social benefit that the cartel’s pricing has destroyed. Consumers overall have kept some surplus, given some to the producers, and watched the rest vanish into thin air. Cartel pricing causes more harm to consumers than benefit to producers, and thus, while it’s in sellers’ self-interest to cartelize if they can, it’s bad for society.

In a market with only one seller, it’s relatively easy to hold prices at a supracompetitive level. It just sets a take-it-or-leave it price. But in a market with many sellers, it takes coordination. They have to agree to charge supracompetitive prices, and they have to enforce the deal against each other. That’s because an agreement to monopolize a market doesn’t magically take away the competitive pressure to lower prices; each cartel member would love for the other suckers to set their prices high while it undercuts them and steals the whole market. That’s why cartels are more often formed when there are good ways for members to monitor each other and punish each other for cheating.

Copyrighted works are a funny kind of market for antitrust purposes. On one level, each individual work is its own, legitimate, micro-monopoly. No one but the author (or her licensees) can legitimately sell a book. Since the marginal cost of making more copies can approach zero (especially for e-books), supracompetitive pricing is actually the incentive the copyright system gives to create new books. It’s a conceptual mistake to use antitrust law to try to force that price down.

On another level, though, books clearly compete with each other. I’d say that Dan Brown is a poor substitute for Umberto Eco, but will do in a pinch—and there are others who’d say the exact opposite. That means that the availability of cheap titles exerts downward pressure on the price of other titles. Publishers can’t ignore what others are charging, as much as they’d like to, and there are many out there distressed by what they see as a downward price spiral in the digital age. (By way of comparison, note how music is effectively becoming cheaper each year.)

Specified Pricing

The settlement provides two options for pricing books. Each copyright owner is free to choose either, and to switch between them at any time. Let’s start with the first, Specified Price, which lets each copyright owner choose a price. That choice belongs solely to the copyright owner; Google has no influence over it.

I find it hard to see how pure Specified Pricing could result in cartel-like behavior. I’ll put it this way: has the publishing industry successfully cartelized physical book sales or e-book sales? The situation there is very similar to the situation under Specified Pricing: publishers set the prices at which they sell copies. There are lots of terms having to do with promotions, discounts, bulk orders, returns, and so on—but the basic reality is that books are priced to sell.

Specified Pricing through Google Books doesn’t change that reality. It adds another distribution channel (one in competition with existing ones, I might note), but provides no institution capable of coordinating prices or enforcing discipline against low-pricers. True, copyright owners can monitor each others’ prices easily, but they can do that fairly easily already. Google’s Specified Pricing adds no enforcement teeth. There’s no reason to expect that pricing under Specified Pricing will be any less competitive than the current book retail market, and thus no reason to veto the settlement on that ground.

Settlement Controlled Pricing

That leaves Settlement Controlled Pricing, in which a Google algorithm will sort books into one of twelve pricing bins ranging from $1.99 to $29.99. The Pricing Algorithm must “find the optimal such price for each Book and, accordingly, to maximize revenue for each Rightsholder.”

There are two ways of reading this provision. One is that the Pricing Algorithm must mimic the behavior of sellers in a competitive book market (or, rather, sellers in a competitive book market if they had algorithms as smart as Google’s). Each seller’s goal is to maximize revenue, so it cuts price as long as that increases its sales and market share enough to make up for the price cut. Google’s algorithm will have access to lots and lots of data, so it will presumably be very good at finding the revenue-maximizing point.

If so, then Settlement Controlled Pricing is a good deal for everyone. It will keep copyright owners from stupidly leaving money on the table by setting silly prices. Some prices will fall, others will rise. Overall, copyright incentives will more clearly match the value of books to readers. Google’s algorithm will remove inefficiencies in the market, without introducing any anticompetitive elements.

The other, scarier reading is that Google’s algorithm must try to collectively maximize the revenue of copyright owners. If so, Google might discover, from its mountains of data, that if it simultaneously raised the price of all books from $10 to $15 (to take a simplified scenario) that the increase in revenue would look a lot like the output of a cartel to raise prices. This is a point made by Randy Picker and Eric Fraser. Einer Elhauge thinks that the settlement, to avoid antitrust dangers, should simply be interpreted to require individual maximization rather than collective; I agree, but I would rather have the settlement explicitly say so than rely on a canon of construction.

But Is the Cartel Stable?

Even assuming the settlement has the dangerous, collective-maximization reading (or that Google secretly twists its algorithm to follow that rule), what will happen? The key question here is whether this new equilibrium is stable. Fraser’s simultaneity argument, in essence, is that individual copyright owners might not be able to raise prices unilaterally, but if Google does it for all of them at once, they reach a point where they’re all better off.

Elhauge, however, works through the game theory of the situation and persuasively concludes that no, each copyright owner has full incentives to defect from the collectively raised prices. That is, if Google raises the prices on all books at once, each copyright owner now has a chance to seize market share by dropping its price back down to the lower level. Everyone else looks like a sucker. Copyright owners still have the right incentives form society’s point of view; Settlement Controlled Pricing alone can’t boost them into a realm where those incentives fail to apply.

Put another way, the copyright owner’s option to switch back to Specified Pricing means that any cartel dependent on Settlement Controlled Pricing is as unstable as a cartel crated through Specified Pricing would be. It’s possible, but the instrument of conspiracy would be some kind of separate agreement, not mere participation in Google Books. And that means that the possibility of excessive pricing here is not a reason to reject the Consumer Purchase terms of the settlement.

Transparency and Algorithmic Review

There is one potential scenario in which I can see trouble: if Google implements its algorithm in a price-fixing manner and copyright owners are asleep at the switch. Perhaps Google raises prices to a supracompetitive level and most copyright owners simply don’t notice, or, lacking good information, don’t realize they could do better by cutting their prices. Google, after all, holds all the valuable information here, in its pricing algorithm.

The settlement does include an auditing provision, which allows the Registry, “to validate, through the use of a reasonable number of third-party experts, the reasonableness of the Pricing Algorithm and to verify that the conclusions it produces are statistically valid.” The audits are confidential and Google only needs to disclose details of the algorithm that are “necessary” to verify its conclusions.

This might not be sufficient oversight. Much depends on the Registry and its experts. I would note that the Registry has the right incentive to keep Google from systematically underpricing books, but not the right incentive to keep Google from systematically overpricing them in a cartel-mimicking fashion. Were I the Registry, I would simply not design my audit in a way likely to catch coordinated overpricing, I would display a studied incuriosity about the details of the algorithm, and I would not put pressure on Google to provide copyright owners with much other information for them to make pricing decisions.

The information really needs to flow either to individual copyright owners, so they can check whether their prices are right (and cut them if they aren’t), or to antitrust enforcers (so they can check the algorithm itself), or to both. Of course, one might believe for other reasons that Google’s pricing choices are too constrained to be dangerous. You could argue that these e-books must compete with all of the other usual channels of trade and that copyright owners can experiment with putting their books into Specified Pricing and gather their own data. My point is just that if you start from the hypothesis that the prices here should be set competitively, it’s not fully clear that Settlement Controlled Pricing gets copyright owners all the information they need to have the right incentives to defect from a Google-created cartel-by-default.

Conclusion

I am not very worried about pricing in the Consumer Purchase program, at least for books whose copyright owners are around to exercise control. An explicit cartel for this part of the market is not an immediate worry. The choice to name their own price, which choice is always available, is a powerful constraint on Google’s ability to impose supracompetitive prices on readers. The only risk of collusion that I see in Consumer Purchase for actively-managed consists in Google keeping secret too many of the details of its pricing algorithm for copyright owners to know when it would be in their interest to set a different price. That kind of potential dishonesty, though, can be fixed with the right kind of transparency and oversight. The Registry audit procedure provides a good starting point, even if it is not yet a sufficient guarantee.


I find it puzzling that the AntiTrust analysts (Picker, Elhauge etc) have by and large accepted the way in which the Settlement has structured the services envisaged by the parties. Should one not step back a bit and consider the surprising asymmetry in the way in which a library service is predicated on a ‘big deal’ (but with annual subscriptions) and the individual consumer service is structured round title by title sales (but for perpetual access)? I have seen no justification for this surprising asymmetry, which on the face of it appears to be geared towards a monopsony/monopoly marketplace where the trend will be for gradually raising prices rather than having a competitive situation. I would be interested in an analysis which considered whether library purchases of individual volumes for perpetual access, and individual purchases of large collections for short term access are not inherently more competitive than the reverse.