Great Paper on NN Out of University of Florida

I’m back from a vacation in Israel to discover an amazing economic analysis of network neutrality posted by my good buddies at Consumers Union on Written by University of Florida Economists Hsing Cheng, Subhajoyti Bhandyopadhya and Hong Guo, Net Neutrality: A Policy Perspective applies game theory to the network neutrality debate. They conclude that abandoning network neutrality would create a disincentive for broadband network providers to build fatter pipes.

If this analysis seems familiar, it’s because I wrote something similar (but without the fancy math) about a year ago. As always, I get warm fuzzies whenever economists confirm my Econ 101 “gut check.”

Of course, these guys being real economists (as opposed to undergrad posseurs like yours truly) have a bit more to say on the subject and use lots of fancy math that I will not try to reproduce. But I offer some brief plain language explanation (including what I think are the brilliant points in the analysis) below….

This paper applies a game theory analysis to the problem of predicting ISP behavior in a network neutrality world and in a non-network neutrality world. Here are the key points:

1) Assume a limited number of players. This assumption keeps the math manageable. You then generalize from the basic model.

I figure its worth stating this up front because I expect the anti-NN crowd to jump on the mathematical tractibility assumptions as a flaw in the analysis. But it’s pretty standard and applies with equal force to the theoretical papers produced by the Phoenix Center and the rest of the anti-NN crowd. As always, the question is whether the assumptions are justified and whether they generalize out to reality.

Here, the paper assumes a monopoly ISP provider and two providers of “internet content.” One of the providers has more customers than the other. The analysis focuses on the “indifferent customer” to measure impact on consumers of the various scenarios discussed.

The paper offers a reasonable justification for a monopoly ISP provider analsysis, observing that many customers in the U.S. face a practical monopoly choice for a variety of reasons (too far from the central office for DSL, inside wiring access issues). Below, I provide a theoretic justification for expanding the analysis to duopoly providers, although I will not be providing any math. As policy-wonk Barbie says: “Math is hard! Lets do policy!”

2) A key insight of the paper (in my opinion) is its recognition that broadband ISPs are not, in fact, producers. As a result, to speak of this as a problem of “producer incentives” misses the point. Instead, the question becomes how does an ISP behave as a transporter of information between parties. Or, as the authors explain:

Thus, the debate is not about how Tier-1 or Tier-2 ISPs charge content providers, but how local Tier-3 ISPs serving the end consumers propose to charge content providers . . . the role of the broadband service provider that we need to model is not as a producer of the service of providing hosting services to the content providers (and, in most cases, the hosting service provider is different from the local broadband service provider at the consumer’s end) but as that of a gatekeeper who determines how the content producers reach the consumers, after it reaches the broadband provider’s local switching office.

(Emphasis in original)

So lets get to the game. Assume a single broadband ISP serving a number of customers. Assume two service providers, offering comparable services. (The paper, in a moment of unbearable but forgivable preciousness, calls these “Provider G” and “Provider Y.”) Assume Provider G controls most of the market share. Further assume that some number of customers are indifferent to Provider G or Provider Y, while others care which service they use. Finally, assume that the ISP must announce the price to both players and offer equal treatment to the two players in accordance with the most favorable “tiering” argument (i.e., no discrimination in the offering of tiered services).

First run the network neutrality universe to determine what happens as a baseline.

Now run four scenairios where the ISP can charge for premium delivery: (1) neither G nor Y pay; (2) Y pays and G does not; (3) G pays but Y does not; (4) Both G & Y pay.

Cheng et al. find that you will never have a situation in which Y pays but G does not, since G has too much to lose by not paying. While it is possible that G pays and Y doesn’t, this case is extremely unlikely (assuming both can afford to pay). So the most likely scenrios are either no one pays (in which case the result is the same as with NN) or both pay.

The spoiler is the ISP, which will set its price for premium access to drive G & Y to pay. Where G is substantially bigger than Y, the ISP will set prices to extract the maximum payment from G that shafts Y. Where G and Y are more comparable, the game more likely devolves into the “both pay” scenario.

What does this do to the universe. Well, no surprise, G & Y are worse off in scenario “D” (both pay). G is better off in Scenario “C” (G only pays) because it uses the premium treatment to capture more customers. Y, of course, is much worse off, as are customers that value Y over G. But customers that value G over Y or are indifferent to G over Y experience a “consumer surplus,” because G is available faster at no additional cost to the customer.

This doesn’t help “competition” mind, and in reality means that the non-premium services are marginalized (as predicted). But, under the terms of the theoretic game, this is a “consumer welfare surplus” because more people (those that prefer G or are indifferent) are better off under the terms of the game. I expect this to be another point that anti-NN folks seek to use to foster confusion by applying the term “consumer welfare” outside the very narrow context of the specified game.

But Cheng, et al. have not completed their analysis. They are interested not merely in the pricing behavior of the ISP, but in modelling whether permitting such behavior will provide incentive to the ISP to build faster pipes. After all, this is a key argument of those opposing network neutrality — that permting tiered pricing will encourage providers to offer larger broadband pipes.

Turns out, that allowing tiering creates a disincentive to build bigger pipes. As the paper explains (I’ve omitted the math):

The intuition behind the fact that the ISP has more incentive to expand capacity under NN is as follows: In scenario 1 (i.e. under NN), the ISP’s profit comes from only the consumers. In scenario 3, the ISP’s profit consists of two parts: fees from consumers and fees from content provider G. In scenario 4, the ISP’s profit consists of three parts…When capacity increases in Scenario 3 [only G pays], consumers face less congestion as a whole, and therefore value the preferential treatment of content provider G’s packets relatively less as compared to the case when capacities were lower (and congestion was higher).

The same logic holds true in Scenario 4. Because an increase in capacity makes preferential treatment less noticeable, the third-party content provider is less likely to pay if the ISP increases capacity.

Or, as I argued last year:

First, Whitacre tiering removes the incentive to build a bigger pipe for the end-user customer. Worse, it creates an incentive not to build a bigger pipe. Why? Because the value of the “premium” service offered to third parties depends upon it being both scarce and necessary. No party will pay more to deliver its content than it has to. So the ISP has to make Whitacre tiering worthwhile. As a result, as services become congested, the ISP has every incentive to simply sit there and charge a higher price for “fast lane” service to the customer. In doing so, the ISP also saves money, because it does not need to invest in a general upgrade of its systems and wait for the gradual migration of customers to pay for its system upgrade. So not only does the ISP make more money by substituting Whitacre tiering in place of an option to customer tier, it avoides expenses.

Result: instead of an incentive to expand bandwidth availability to the customer, the ISP now has an incentive to supress bandwidth availability as an absolute matter (since it can charge both the customer and a third party for the equivalent of a bandwidth upgrade by using Whitacre tiering).

However you say it, it comes out the same way: a network neutrality rule encourages ISPs to build bigger pipes, while permitting tiering discourages ISPs from investing in the pipes.

Amazing Sausage Factory Bonus Analysis
As promised, here is the solution to why the monopoly provider generalizes out to a duopoly provider in the same game.

As we have observed, a monopoly provider experiences a rise in profit if permitted to charge third parties for premium access. But does this work in a duopoly situation?

Lets assume we now have ISP A and ISP B available to the same population in the problem. ISP A begins charging for premium access. Under what circumstances will ISP B decline to charge for premium access?

Because ISP B must forgo the additional profits available to ISP A if it refuses to charge for premium access, there must be some other incentive to ISP B that outwieghs these profits. Since ISP B will only derive profit from customer subscribers, it must assume that it’s “neutral” treatment will attract subscribers from ISP A in sufficient number to outwiegh the lost profits.

Indeed, if this turns out to be true, then ISP A will not seek to extract rents from third parties, because doing so will cause it to lose an equal or greater amount in customer revenue.

But such a situation is so unlikely as to constitute zero probability. Consider that in Scenario 1(both content providers refuse to pay) and Scenario 4 (both content providers pay), consumers notice no difference. They are therefore nor more likely to switch than under a NN rule. ISP B therefore has no incentive to forgo the profit under Scenario 4 by refusuing to charge.

This leaves Scenario 3 (G pays, Y does not) as the only possible situation in which ISP B might gain an advantage by not charging for premium delivery, as those that significantly favor Y over G are suffering a consumer welfare loss in Scenario 3. But even assuming there are enough people who really favor Y to make it profitable to forgo the profit of providing G premium access, ISP B has not offered sufficient incentive to those who strongly favor Y. Customers of ISP B that favor Y are no worse off relative to those who favor G, but they are no better off either. Indeed, ISP B risks losing customers that value speedier access to G to ISP A by its failure to sell premium access to G. Only if ISP B offers premium access to Y does ISP B hope to attract customers that favor Y. Given that more customers favor G, however, ISP B is profoundly unlikely to favor Y over G.

So, even if we assume a duopoly market rather than a monopoly market, the model still holds.

I would expect that there is some level of competition that is sufficiently high that the marginal loss of customers by favoring one content provider over another outweighs the potential profit. Further, the more the customer market atomizes, the less any specific ISP can extract rents from third parties because the number of customers it can cut off from the the third party diminishes. That’s what we used to have in the old dial-up days, when the FCC forced the network operators to allow rivals to access their physical networks. It’s why countries that have wholesale and retail separation, like Japan and France, have very competitive markets that offer high-speed connections at a fraction of the cost you can get here. But I don’t see us (re)adopting such a pro-competitive policy anytime soon.

Stay tuned . . . .


  1. Welcome back, Harold, we missed you!

    In particular, see this shout-out from me:

    I’m going to give you a week to get back into the swing of things and then we need to talk about wetmachine/book projects. Spring is here; time to shake off winter sluggishness. (For me, anyway. I’ve just about gone into hibernation. . .)

  2. Thanks, Harold! That explains things well even to the fuzzy minded, though I don’t suppose I’ll get my dear pal Senator Specter to see eye to eye with this anytime in the next eon.

    And yes, we missed you!

Comments are closed