I’m not sure whether what follows is a tentative set of hypotheses, a deliberate provocation, or a game of spot-the-fallacy at my expense. Perhaps it’s all three. In any event, a new way of thinking about nonexclusivity (new to me, at least) struck me over the weekend, and I would particularly value your help in working through the implications.
I’ve been arguing since day one that the exclusivity of the Google Books settlement is problematic. My preferred remedy has been nonexclusivity. As I said in How to Fix the Google Book Search Settlement:
Any other entity willing to assume the same payment and security obligations that Google assumes in the settlement should be allowed to offer the same services that Google will, or any subset of them.
While I continue to think that the settlement’s exclusivity is problematic, I’ve recently come to question my assumption that making it nonexclusive would be straightforward. I now think the pricing issues in a world that features not just Google Books but also Twitter Books and other competitors would be substantially more complicated than I’d previously thought.
I’ll focus on Consumer Purchase, although interestingly related issues also arise under the Institutional Subscription. Our ideal here is “competitive” pricing, in which each copyright owner faces competition from other copyright owners and must take their prices as given when choosing her own. She is, however, insulated from competition in the sale of her own book; that’s what it means for her to have exclusive rights over it. To measure how far the settlement could diverge from competitive pricing if everything goes wrong, imagine a world in which Google faces absolutely no competition (including from itself).
Some copyright owners who claim their books will set their own prices in the Consumer Purchase program. They, of course, will want to price to maximize their individual revenue with their choices, which is exactly what “competitive” pricing expects. So far, so good.
Other copyright owners will choose algorithmic pricing. When they do, Google’s algorithm is supposed to “simulate how an individual Book would be priced by a Rightsholder of that Book acting in a manner to optimize revenues.” (Amended Settlement § 4.2(c)(ii)(2)). That is, if Google does its job well, it will choose the same price a fully-informed copyright owner would choose in a competitive market. For unclaimed books, algorithmic pricing is the only option. Here again, Google must maximize revenues “without regard to changes to the price of any other Book.” (Amended Settlement § 4.2(b)(2)). Although there is no copyright owner in the picture to make decisions, the goal is that Google would act as though there were one and maximize revenues for her.
That’s what Google is instructed to do, at least. But we can ask whether it has the right incentives. What would happen if the algorithm turned out to be Larry and Sergey picking numbers to benefit Google’s bottom line? In one important way, the settlement gets the incentives right. Since Google is a 37% participant in Consumer Purchase revenues, whatever maximizes revenue for copyright owners also maximizes revenue for Google. (We’ll leave Google’s costs asides in this simplified model.) Google is an agent for copyright owners whose interests align with theirs.
Perhaps the alignment is too good. The fear for readers is that copyright owners—like sellers in any competitive market—might be better off conspiring to raise prices. It’s a classic cartel story: higher prices, lower output, higher profits for sellers, lower social welfare. Since Google proportionately shares in copyright owners’ revenue, it also shares in their incentive to cartelize. In a world with no outside competitive constraints and no checks on how the algorithm was actually implemented, Google might well design the algorithm to overprice books.
The takeaway here is not that the settlement authorizes Google to cartelize Consumer Purchase. (As amended, it expressly does not.) Nor is it that Google will cheat on the algorithm. (Einer Elhauge has given reasons to think that it would not.) My point is just that Google faces a temptation to overprice, and will, unless somehow kept in check. The specification of the algorithm in the settlement is designed to keep prices down.
Against that backdrop, nonexclusivity struck me as an effective counter to fears of excessive prices. Introducing competition from Twitter Books et al. would give consumers additional options. If Google’s algorithmic prices were supracompetitive, Twitter Books would step in and price competitively, taking market share—and revenues—away from Google. This is a surer, safer check on overpricing than internal controls in the settlement or external monitoring. It directly hits Google (and Twitter) where it matters: in the incentives.
On further reflection, however, I’m now concerned that nonexclusivity might also work too well. Again, let’s try to derive the worst-case story. For books whose copyright owners specify a price, nothing changes. Copyright owners can choose a price on Twitter Books just as they choose one on Google Books. They have an incentive to maximize their private revenue, of course, but as long as they act individually, that just means they’ll price competitively. (Remember, this is is “competition” between books, not competition to sell the same book.)
But when Google and Twitter are setting the prices, everything changes. Before we introduce the algorithm let’s ask what Google’s and Twitter’s incentives look like. By way of example, let’s suppose the book in question is the latest Dan Brown potboiler, The Watched Pot.
If Google offers The Watched Pot at $10 but Twitter prices it at $9, then Google will pocket 37 cents more per copy. Twitter, however, will capture more market share, since readers would love to save the dollar. And if readers are sufficiently discerning—if they all realize that Twitter has lower prices—Twitter will take the whole market. Google’s additional 37 cents per copy isn’t worth much unless it’s moving copies. Google, of course, won’t take Twitter’s price attack sitting down. It will cut its own prices—The Watched Pot for $8. Twitter will come back with a $7 copy, and whad’ya know, we’re in an all-out price war!
In the offline world, there’s a floor. Dan Brown (and his publisher) are selling The Watched Pot to bookstores at $5 wholesale. Prices will drop to $5, then stop. Under, a nonexclusive version of the settlement, however, both Google and Twitter would be selling on commission. The only hard floor is at $0. The endgame involves both Google and Twitter selling remainder-priced copies of The Watched Pot. As much as this would be a nice deal for readers, Dan Brown will hate it. In their price war with each other, Google and Twitter will have thrown away not just their 37% but also his 63%. He’ll pull The Watched Pot out of algorithmic pricing on both Google and Twitter and move it back up to $10.
We can expect a similar unwinding process for any other book exposed to algorithmic pricing, so copyright owners will withdraw their books from it en masse. There is, however, one class of books that will stay put, no matter how underpriced they are: unclaimed books. Not only is there no one to act to remove them, there’s also nowhere for them to go. The result is rock-bottom prices for unclaimed books.
Of course, the (hypothetical nonexclusive) settlement wouldn’t officially permit Google and Twitter to do this. They’d have to use algorithms that look for competitive prices, rather than pricing in response to each other. But we might ask how effective a constraint this is. There are a lot of subtle choices to be made in designing a book-pricing system. Google and Twitter will draw from different sources of data; there will be a lot of choices to be made. $10 vs. $9 situations will emerge all the time through normal variations; each time they do, the market will slosh over to the lower-priced option. And that’s assuming no one tries to cheat. (Software engineers aren’t stupid; they can often tell that it’s in their employer’s interests for the algorithms to return lowball results.)
The point is that the incentives are now all right from readers’ perspective and all wrong from copyright owners’. Moreover, note what the algorithm now is doing. It’s not holding prices down; it’s holding them up. It protects copyright owners, not readers.
Perhaps some tweaks can get the incentives right for nonexclusive Consumer Purchase? Certainly, if the Registry or others were able to inspect the algorithms more closely, they could detect certain forms of cheating. Still, Twitter Books would have all the wrong incentives, making for a constant struggle. Part of the beauty of the present settlement is that it generally aligns Google’s and copyright owners’ incentives for the revenue-generating uses.
Perhaps we could try to emulate the offline world and replace the percentage formula with set wholesale prices that Google and Twitter would need to pay. But that simply begs the question. Someone still needs to choose the price—it’s just a wholesale price, rather than a retail price. We’re no closer to having an incentive-compatible system for picking it.
Or maybe what the nonexclusive settlement needs is some kind of entity to act as a fiduciary for the owners of unclaimed works. We could call it the Unclaimed Works Fiduciary, and give it the power to set the prices for unclaimed works. That sounds great, until you realize that this change would make the UWF into a cartel: a single entity with the power to set binding prices for an entire group of competitors’ products, and no good market substitutes. Pushing the decision-making upstream gets the incentives right from copyright owners’ perspectives: too right.
A less dramatic alternative might allow the UWF merely to remove books from Consumer Purchase entirely. That “solves” the problem by making some unclaimed works go away. Run the film forward a few frames to see why. Google and Twitter have their price war, and the price drops. The UWF steps in and pulls the books out. Now the unclaimed books are unavailable at any price—so we’re back where we started, before the settlement.
This conclusion doesn’t look good for my original suggestion that the settlement could easily be made nonexclusive. As much as readers might like Twitter’s incentives to lowball its prices, it’s palpably dangerous to copyright owners. Given that the nonexclusive settlement would still need to be justified on the basis of its fairness to copyright owners, these misaligned incentives would pose a Rule 23 obstacle worth worrying about.
The underlying problem here is that it’s hard to come up with “competitive” prices for goods in which there is no existing market. Commercially unavailable books are, by definition, not being sold in an ordinary market flowing from copyright owner to reader. Nor, for unclaimed books, is there a copyright owner whom we can ask to set a price.
The settlement’s answer to this puzzle is to have Google create a pricing algorithm. Under most circumstances, saying “The computers will figure out ideal prices” would strike us as a dicey proposition. It’s a skyhook. Now, if anyone can make an algorithmic skyhook work, it’s Google. But even Google’s algorithms won’t get everything right. The algorithm is supposed to observe demand curves for each book, but there may not be very many data points to work with. It’s also suppose to base pricing on sales of comparable books, but every book is a unique good. Twitter will be even more likely to err.
That’s why we care so much about the incentives of the algorithm-maker. If it were a self-published author writing the algorithm for her own use, for example, her incentives would be right. She bears the costs of bad design choices and reaps the benefits of good ones. Google and Twitter don’t have the right incentives in a nonexclusive world, which means they may not invest properly in getting the algorithm right, compounding whatever other mistakes their algorithms make.
The regular old exclusive settlement doesn’t have the same problem. Remember that Google is a proportional participant in Consumer Purchase revenue. Its own revenue is maximized, for a given book, when it finds the revenue-maximizing price for the copyright owner, which fortunately happens to be the “competitive” price. What the exclusive settlement does have is a collusion problem, which is structurally different and potentially easier to prevent, as supracompetitive cartel pricing requires changing the prices of many books together, not just one at a time independently. Mispricing will be harder to detect and fix under a nonexclusive settlement than under an exclusive one.
But there is a sting in the tail. If algorithmic pricing as specified in the settlement works better when the settlement is exclusive, that’s because it depends on Google having no close competitors for selling individual copies of unclaimed works. Settlement Controlled Pricing for unclaimed books is monopoly pricing. That’s why it works.
If you prefer, the reason that the settlement is able to induce Google to price with copyright owners’ interests at heart is that Google will become a de facto copyright owner for unclaimed works. The settlement gives Google a default 37% ownership share of all commercially unavailable books. Copyright owners can take it back by claiming their books. But until they do, Google is a co-owner with them.
What’s more, the conclusion that nonexclusivity is dangerous for copyright owners is also problematic for the claim that Google competitors could and should obtain similar class-action settlements. Given that the result could be a destructive price war unless the book-sellers are watched closely—and that orphan owners are unable to watch closely—copyright owners ought to be rationally reluctant to enter into a second class-action settlement on identical terms.
You can accept all of the above and still favor the settlement. Its antitrust defenders are generally open about their belief that the settlement is good for readers even if Google never has any competition for unclaimed books. But these points give me pause, and I hope you’ll ponder them, too.