Electronic copy available at: http://ssrn.com/abstract=959944

The Debate on Net Neutrality: A Policy Perspective

Hsing Kenneth Cheng*, Subhajyoti Bandyopadhyay Department of Information Systems and Operations Management

Warrington College of Business Administration, University of Florida Gainesville, FL 32611-7169, U.S.A.

{hkcheng, shubho}@ufl.edu

Hong Guo Mendoza College of Business

University of Notre Dame Notre Dame, Indiana 46556-5646, U.S.A.

hguo@nd.edu

* Corresponding author. Phone: (352)392-7068; Fax: (352) 392-5438

Electronic copy available at: http://ssrn.com/abstract=959944

The Debate on Net Neutrality: A Policy Perspective

Abstract

The status quo of prohibiting broadband service providers from charging websites for preferential access to their customers the bedrock principle of net neutrality is under fierce debate. We develop a game-theoretic model to address two critical issues of net neutrality: (1) Who are gainers and losers of abandoning net neutrality; and (2) Will broadband service providers have greater incentive to expand their capacity without net neutrality? We find that if the principle of net neutrality is abolished, the broadband service provider stands to gain from the arrangement, as a result of extracting the preferential access fees from content providers. Content providers are thus left worse off, mirroring the stances of the two sides in the debate. Depending on parameter values in our framework, consumer surplus either does not change or is higher in the short run. When compared to the baseline case under net neutrality, social welfare in the short run increases if one content provider pays for preferential treatment, but remains unchanged if both content providers pay. Finally, we find that the incentive to expand infrastructure capacity for the broadband service provider and its optimal capacity choice under net neutrality are higher than those under the no net neutrality regime except in some specific cases. Under net neutrality, the broadband service provider always invests in broadband infrastructure at the socially optimal level, but either under- or over-invests in infrastructure capacity in the absence of net neutrality.

Keywords: Net Neutrality, Economics of Net Neutrality, Broadband Service Providers, Content Providers, Consumer Surplus, Social Welfare

1

The Debate on Net Neutrality: A Policy Perspective

Introduction

The net neutrality debate has reached a fever pitch as Congress mulls legislation that would

allow Internet service providers to charge Web sites for preferred delivery of digital content.

Should the Net Be Neutral, The Wall Street Journal Online, May 24, 2006 (WSJ 2006)

The recent proposal by such broadband providers as Verizon, Comcast and AT&T to charge

popular websites for preferential access to their residential and commercial customers, has generated

widespread attention in the media (Helm 2006, Waldmeir 2006). The proposal goes straight to the heart of

the debate on the idea of Net neutrality the phrase that was first coined by Columbia Law School

professor Tim Wu, and is used to signify the concept that the Internet is merely a carrier of online content

that does not distinguish one website from another. The central idea inherent in this concept is that a

maximally useful public information network aspires to treat all content, sites, and platforms equally

(Wu 2003, p. 142), and while a formal definition of the operationalization of the principle does not exist,

(Hahn and Wallsten 2006) point out that net neutrality usually means that broadband service providers

charge consumers only once for Internet access, do not favor one content provider over another, and do

not charge content providers for sending information over broadband lines to end users. (p.1)

Popular online content providers such as Google, Yahoo!, Microsoft would like to maintain the

status quo, which they claim would preserve the egalitarian philosophy on which the Internet was

founded. Other supporters of the concept include online start-ups, that claim it would be almost

impossible for them to pay these proposed fees when their revenue streams are almost non-existent, since

they have to give away most of their content in order to build a loyal customer base (Sydell 2006). Some

venture capitalists have even argued that abandoning net neutrality would result in would-be

entrepreneurs becoming more hesitant to start a business, which might hurt the competitiveness of

American online firms in the long run (Sydell 2006, Wu 2006a). Vint Cerf, the renowned computer

scientist who is commonly referred to as one of the founding fathers of the Internet, contends that such

jaminb720Highlight

2

a payment structure would result in the Internet increasingly resembling todays mass media, where a few

Internet service providers (ISPs) control what the customers effectively may access (Waldmeir 2006).

Tim Berners-Lee, the founder of the World Wide Web, also favors keeping net neutrality in place, since

[the Internet] is the basis of a fair competitive market economy (Berners-Lee 2006, blog entry). Finally,

some people have voiced their fears of the Internet service providers starting to offer such services as

Internet telephony to their consumers at rates that undercut rival providers who would struggle to remain

competitive if they have to pay these fees. This again might result in stagnation in what has so far

remained one of the most open marketplaces.

The Internet service providers have argued that they have put their resources to maintain and

upgrade the physical infrastructure to provide the services to consumers, while the popular web sites have

thus far gotten a free ride on their resources1 (Waldmeir 2006), and that the Internet service providers

should be allowed to strike deals to give certain Web sites or services priority in reaching computer users

(Krim 2005, p. D05). With online content increasing exponentially over the years, and consumers

increasingly becoming used to broadband access, it will be necessary to meet the rising costs of

increasing capacity and serving an expanded consumer base. Not having these sources of revenue might

act as a disincentive to upgrade the service providers infrastructure and affect their plans of increasing

existing capacities. That, in turn, would affect many emerging online services such as real-time

broadband video, which by design require preferential treatment of their packets. In some ways, the ISPs

contend, the new payment mechanisms might herald the beginning of new business models that demand

preferential treatment of their packets, and that the vertical integration of new features and services by

broadband network operators is an essential part of the innovation strategy companies will need to use to

compete and offer customers the services they demand (Thierer 2004, p. 1).

The proper usage and context of the term net neutrality itself has been subject to confusion (Wu

2006b); an extensive discussion of the issues can be found in (Economides and Tag 2007, Wu 2003). In

1 Economides (2008) notes that The claim that consumers, content, or applications providers use the Internet for

free is certainly incorrect. (p. 211) For a detailed discussion, see Economides (2008).

jaminb720Highlight

jaminb720Highlight

3

brief, network neutrality aims to address concerns raised by some specific behavior of the broadband

service providers: (a) blocking of some content providers; (b) preferential treatment of one content

provider over another; and (c) transparency failures, whereby a broadband provider fails to notify its

customers and content providers what service they offer in terms of estimated bandwidth, latency, etc.

(Wu 2006b). The current proposals by the broadband service providers (i.e., the ISPs) have raised

concerns around the second issue i.e., the possibility that one content or application provider pays the

broadband service provider for preferential treatment of its packets, as the ISP acts effectively as a

gatekeeper between the content providers and the customers it serves. 2

The entire debate has raised a number of unanswered questions that are of interest to researchers

and practitioners alike, not to mention the regulatory agencies. The intensity of the public debate, and the

stakes involved in the issue, were brought into focus during a House Committee hearing in April 2006

(Wu 2006a), where it was pointed out that [the Internet] has become part of Americas basic

infrastructure. It has become as essential to people and to the economy as the roads, the electric grid, or

the telephoneGiven this infrastructure, Americans are accustomed to basic rights to use the network as

they see fit. (p. 43) After a relatively quiet year in 2007, the net neutrality debate is back on top of the

technology agenda in Washington, with the lawmakers introducing a new bill, the Internet Freedom

Preservation Act of 2008 (H. R. 5353) (2008), on February 12, 2008.

From a policy perspective, two issues are of particular interest. First, the regulatory agencies

would like to know who are the gainers and losers if the principle of net neutrality is abolished.

Specifically, if social welfare increases as a result of abandoning net neutrality and more specifically,

the end consumers are better off the idea for the proposed payment mechanisms would gain traction

among policymakers. Conversely, if abandoning the principle of net neutrality results in helping a few

private agencies to extract more rent, the idea would find a much less sympathetic audience. In the first

part of our paper, we analyze this issue in the net neutrality debate within a game-theoretic model, where

2 We use the terms content and applications interchangeably throughout the paper; similarly, the terms broadband

service provider and Internet service provider (or ISP) are used interchangeably throughout.

jaminb720Highlight

jaminb720Highlight

4

we determine the equilibrium in the strategies employed by the content providers, which in turn prompts

the appropriate profit maximization strategy of the Internet service provider.

The second issue of interest to policymakers is to check the veracity of a key claim of the ISPs:

notably, that under net neutrality, the incentive to expand the capacity of the existing infrastructure for the

next generation of broadband services is much less as compared to when they are allowed to charge the

online content providers for preferential treatment. For policymakers, this is indeed a key issue. Higher

capacity broadband services will enable many services that are deemed important for the society as a

whole. Some examples of such services include disaster recovery, remote medical supervision, and the

like. For content providers, the next generation broadband services will enable instant delivery of high-

definition movies, consumer interactivity, a richer online shopping experience, and so forth, and in the

process open many new channels of revenue generation. In fact, many consumers who currently do not

feel the need for broadband services for their typical Internet activities of email or online shopping might

be ready to pay for such broadband services (Bandyopadhyay and Cheng 2006). In the second half of the

paper, we treat infrastructure capacity as a strategic variable for the Internet service provider in the long

run (as opposed to it being a constant in the short-run problem) and explore two related issues: (a) do

Internet service providers have a greater incentive to expand their capacities if the principle of net

neutrality is abandoned; and (b) how do the optimal capacity choices under net neutrality differ from

those when net neutrality is not longer enforced.

From the analytical perspective, what drives the problem is the increased latency of the

applications and content of those providers that are given less favorable treatment by the ISP. In other

words, the packets from these providers face increased congestion, which translates into a delay

disutility for the end users. Thus, while a consumer might have some intrinsic preferences of favoring

one content provider over another, these preferences can be modified by ones ISP by decreasing the

delay disutility of one providers packets over anothers. A proper analysis of the problem therefore

demands the modeling of the objectives of all the three players involved the content providers, the

consumers, and the ISP (Hahn and Wallsten 2006). This exercise is distinct and different from the two-

5

player models that analyze the broadband service provider that provides different classes of service to the

consumers (Bandyopadhyay and Cheng 2006, Bhargava and Sun 2005). It is imperative that this

distinction be highlighted, since the net neutrality principle has sometimes been misinterpreted as a

barrier to the ability of the broadband service provider to charge consumers different prices for different

classes of service. As existing broadband service offerings indicate, ISPs today already charge different

prices for different classes of broadband connection,3 and such price-discrimination strategies enhance

social welfare (Bandyopadhyay and Cheng 2006, Edell and Varaiya 1999, Hermalin and Katz 2007).

Finally, in analyzing the net neutrality debate, we are not concerned about how hosting service providers

charge content providers for hosting or transmitting their content over the World Wide Web (see Figure

1), or the various ways (like web caching, for example) by which content providers have their content

delivered faster. In the net neutrality debate, the issue is whether the broadband service provider should be

allowed to charge content providers for transmitting their content from its local switching office to the

consumers.4 In other words, the issue of interest is only at the local loop, the part encircled in the dashed

line in Figure 1. 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 users propose to charge the content providers. For a very good

discussion on the three tiers of ISPs, one is referred to Kurose and Ross (2003), Section 1.5. For a

discussion on some of the early works on Internet pricing itself, one is referred to Gupta et al. (1996) and

Gupta et al. (1997), two of the seminal papers in this area that rigorously analyzed the problem. In

contrast to that literature, which looks at the problem of congestion at the Internet backbone, the net

neutrality issue looks at the issue of the congestion over the last mile.

--- Insert Figure 1 about here ---

3 In the authors neighborhood (in a medium-sized University town), for example, the cable broadband service

provider has broadband download speed offerings of an introductory 256kbps, for $25 per month (excluding taxes and regular fees), and regular speeds of 7Mbps (for $40) and 12Mbps (for $55). 4 Technically, the service terminates in a cable modem termination system (CMTS) in the cable modem broadband

system or the Digital Subscriber Line Access Multiplexer (DSLAM) in the DSL broadband system, both of which have to be within a few miles of the end consumers. This is also called by the name Internet Exchange Point.

jaminb720Highlight

jaminb720Highlight

6

Thus, 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 consumers end) but as that of a

gatekeeper who determines how the content from the content producers reaches the consumer, after it

reaches the broadband providers local switching office.

As Economides and Tag (2007) point out, in sharp contrast to the large amount of literature that

discusses the legal issues surrounding net neutrality, there is a surprising lack of rigorous economic

analysis of the net neutrality debate. Economides and Tag (2007) identify six economic consequences of

abolishing net neutrality. First, two-sided pricing will be introduced by the last mile ISP to charge end

consumers on one side and content providers on the other side of the network. Second, the packets from

the content providers paying the ISP will receive priority over those from non-paying firms. Third, the

ISP can engage in identity-based discrimination. For example, the search engine firm with the highest bid

receives priority delivery of its search results to consumers, resulting in great distortion of the search

engine market. A more detailed discussion of this issue is covered in Economides (2007). Fourth, new

start-up firms will not likely win the prioritization auction, leading to less innovation. Fifth, ISPs can

impose preferential treatment for their own content and applications over those of other providers. To

consumers, online content and broadband service can be viewed as complementary products, where one

market is competitive while the other is not (see, e.g., Economides and Salop (1992), for the implications

on pricing of the joint product). Sixth, a significant reduction of trade on the Internet will occur, since

multiple fees are likely to be charged for a single transmission over the interconnected networks that

comprise the Internet.

Economides and Tag (2007) address the first issue in the above framework and find that net

neutrality regulation increases total industry surplus in the presence of a monopoly ISP. The same finding

also applies to a duopoly setting. In relation to the Economides and Tag (2007) framework, we examine

the second economic issue of net neutrality. The most crucial aspect of the network neutrality debate is

the fact that there is a constrained resource (the pipe between the ISP and the consumer) that is shared

jaminb720Highlight

7

by packets of all the content providers, and the very existence of this constrained resource has negative

externalities associated with it. If this pipe were not a constrained resource, the ISPs would not have

any credible mechanism to charge for preferential access. Analyzing this negative externality introduces a

significant amount of challenge in the modeling, but cannot be wished away if one has to do justice to the

quality of the debate.

The rest of the paper is arranged as follows. The following section outlines our model. The next

two sections analyze the model under net neutrality and without net neutrality, respectively. The section

on the gainers and losers addresses our first research question: the gainers and losers under net neutrality

versus a payment mechanism model without net neutrality. The section on the capacity expansion

decision considers the twin issues of the incentive to expand capacity for the ISP, and that of the optimal

capacity choice under net neutrality and in the absence of net neutrality. The final section concludes, by

summarizing the policy implications of our analysis and providing some possible directions for future

research. The objective of this paper, however, is not to make any policy prescriptions, but rather to

provide an objective economic analysis that clarifies some of the germane issues in this ongoing debate.

The Model

To analyze the problem at hand, we consider a stylized model with three types of players: (a) a

monopolist Internet service provider that not only serves consumers in a specific geographic market by

providing them with Internet access but also serves content providers by delivering their content to the

consumers in this market; (b) two competing content providers that provide their service for free to the

end users as they generate revenues from advertisers and associated click-throughs of the consumers;

and (c) consumers who consume content from their preferred content provider through the Internet access

provided by the local ISP. We develop the model both under net neutrality (hereafter shortened to NN),

and when net neutrality is abolished in favor of the regime where the ISP can charge the content providers

(i.e. no net neutrality, or NNN for short).

jaminb720Highlight

8

In the concluding section, we discuss in greater detail how germane this particular revenue model

is, but in short we believe that we faithfully capture the revenue model that is overwhelmingly prevalent

among online content providers, which is very different from traditional online retailers like

Amazon.com. The specifics of the revenue model change from one provider to another, but in general

terms, it involves no upfront fees from the customer, but rather a customers value is encapsulated in the

entire lifecycle of ones relationship with the firm. The idea was first proposed by the noted journalist and

commentator on digital technologies, Esther Dyson (1994), and is now considered mainstream by

industry observers like Chris Anderson (2008) and Nicholas Carr (2008), economics researchers like Paul

Krugman (2008) and Hal Varian (McKinsey Quarterly 2009), and IS researchers like Eric Clemons

(Knowledge@Wharton 2008). In this framework, the firm gets its revenues not directly from a particular

customer, but rather in a stochastic sense when these customers indirectly generate revenues through a

variety of means such as banner advertisements, affiliate revenues, rental of subscription lists, sale of

aggregate information, to name a few.

Without loss of generality, we normalize the total number of end consumers (i.e., the total

number of consumers served by the monopolist ISP) to 1. This unit mass of customers is uniformly

distributed on line segment [ ]0,1 in terms of their ideal content. There are two competing online content providers, Y and G, where content provider Y is located at zero, while content provider G is located at the

opposite end of the interval (see Figure 2). Let x be the marginal consumer that is indifferent to the

content between Y and G. Then, the market shares for Y and G are x and 1 x respectively. This amounts

to the market being fully covered, a standard assumption made in existing literature on two-sided markets

to achieve analytical closure (Armstrong 2006).

Both content providers offer their basic services at no cost to the end users.5 In our model, we

consider the revenue generation of the content provider as the average revenue generated (from all

5 In order to draw a meaningful comparison between the current and the proposed environment, we need to choose

the same revenue model for the content providers under both environments, and we have therefore chosen the revenue model that mirrors current reality.

9

sources) per packet requested by the end consumer. Let Yr and Gr denote the revenue rates of content

provider Y and G, respectively, per packet for content. In other words, these two parameters denote the

average rates at which the requests for content from the consumers provide revenues to the content

providers from myriad types of advertisers that want to reach them. Without loss of generality, we assume

that G Yr r> , which means that one content provider (G) is better than the other (Y) in getting the right

consumers for its advertisers (and its other revenue sources) and therefore can charge higher advertising

fees. This assumption does not affect our analysis results and is actually consistent with our empirical

findings.6 Without net neutrality, the ISP provides preferential delivery service for content at a price p

which is the unit price for priority data packet transmission per packet. The technology to discriminate

packets and streamline Internet traffic has been available at minimal fixed cost, and therefore the cost of

implementing this mechanism of priority delivery of some content is assumed to be negligible. In

response, the two content providers decide whether to pay for this preferential treatment. Then, the

service decisions (whether to choose the preferential delivery service) for the two content providers can

be represented by the indicator functions

1, if Y pays 1, if G pays0, if Y does not pay 0, if G does not pay

and .Y GI I= =

Let be the Poisson arrival rate of content requested by each consumer, and it is expressed in packets

per unit of time. Content provider Ys decision problem is ( ) ( ){ }max , ,Y

Y Y G Y Y GIr x I I I p x I I and

content provider Gs decision problem is ( ) ( ){ }max 1 , 1 ,G

G Y G G Y GIr px I I I x I I . The

6 We found, for example (from Nielsen Online and relevant 10-Q reports), that even though the search engines

Google and Yahoo! have very similar (in terms of numbers) unique audiences (the latter is about 90% of the former), the former garners nearly 6.5 times the advertising revenue per unique visitor than the latter. In terms of the incentives therefore, Google tends to gain a lot more from an additional consumer that it can lure from Yahoo! than the other way around, and this introduces a level of competitive tension in the game that is very important from the net neutrality standpoint as observers have mentioned, preferential access to the consumers can be very disadvantageous to new entrants, since their resources to compete against entrenched competitors are very limited. Our model gives us a way to model this aspect of the net neutrality debate that has so far not been done in the extant literature.

10

demand for content providers ( ),Y Gx I I and ( )1 ,Y Gx I I depend on their service choices YI and GI . We provide further analyses of the demand realization later.

--- Insert Figure 2 about here ---

A consumers net utility from using the services of either Y or G depends on ones individual

preferences, the distance of ones preferred provider (i.e. either Y or G) from ones ideal, and the cost of

delay that is a result of the general congestion in the access network between the ISP and the consumer.

Parameter t measures the fit cost of the deviation from a consumers ideal content in the Hotelling

framework (Hotelling 1929). Following Mendelson (1985), we denote ( )V to be the gross value

function of this content for each consumer, assumed to be twice differentiable and strictly concave.

Furthermore, consumers get a congestion disutility due to waiting for packets: the delay cost parameter d

multiplied by the expected time in such a queuing system, w . The ISP charges a fixed Internet access fee

F per unit time to the consumers for Internet access. Both the fixed fee F and the priority charge p are

expressed in the same unit of time. Therefore the utility function for an arbitrary consumer [0,1]x is

( )( )YU x V tx F d w = if the content provider is Y and is ( ) ( )( ) 1GU x V t x F d w = if the content provider is G. In order to determine the delay w , we consider to be the capacity that the ISP

provides to the consumers, expressed in packets per unit of time. This capacity constraint affects the

service that the ISP renders to the consumers in a unique fashion. Specifically, we can think of the packets

requested by the consumers as being serviced in an M/M/1 queuing system. We assume that customers

are homogeneous in terms of having the same rate of requests for content, valuation of content, and

sensitivity to delay. Currently under NN, the congestion delay is the same NN1

w = for all consumers

and does not figure into the consumers decisions. However, under NNN, the congestion delay plays an

important role. To understand the impact of abolishing net neutrality, consider a situation where the ISP

starts charging content providers Y and G for preferential treatment of their packets and suppose without

loss of generality that only Y decides to pay for the service. As a result, any packet from Y that is received

11

by the Internet service provider as a request from one of its customers now gets preferred treatment to the

top of the queue (these packets still face the congestion from other similarly preferred packets from Y).

We model the congestion in the network after Bandyopadhyay and Cheng (2006) and Mendelson (1985).

Packets from G do not receive any preferential treatment and at any point in time are in fact queued after

any packet from Y that might be requested at that point of time. Depending on the number of Ys packets

in the channel, some consumers that previously preferred G might now find the congestion of Gs packets

causing enough disutility that they might now prefer Ys service. The delays of the content providers

packets are thus dependent on their service choices, denoted by ( ),Y Gw I I . Table 1 gives the delays of the four different outcomes under NNN for a two-class priority M/M/1 queue with service preemption. The

xs in the delay expressions are the corresponding marginal consumer who is indifferent between Y and G.

Individual consumers then choose the content provider that yields the higher utility.

We assume that the consumer located at the two ends of the market is loyal to their corresponding

content providers as the consumer at the end point receives content of perfect fit. That is,

( )2( ) ( )d dV F V t F

>

, an expression that can be simplified to ( )2

2

td

< . This

assumption ensures the existence of a meaningful competition between the two content providers G and

Y, one that is similar to a standard assumption in two-sided markets literature, e.g., Eq. (8) of Armstrong

(2006) and assumption A3 of Armstrong and Wright (2007). Armstrong (2006) notes that this assumption

is the necessary and sufficient condition for a market-sharing equilibrium to exist. (p. 674). Since we

are modeling only broadband consumers, we assume that the Internet service provider captures all end

consumers in [0, 1] under net neutrality. Further, the ISPs have stated that their intention is not to degrade

the online experience for any current broadband subscriber even if net neutrality is abolished, and

therefore we assume that the Internet service provider continues to serve all the current consumers when

they start charging content providers for preferential delivery of their packets. In other words, we assume

12

that the consumers value function ( )V is sufficiently high that the utility for the indifferent consumer in

any of the outcomes that follow is nonnegative.

We consider a monopolist ISP that delivers digital content from its local switching office to the

end users. While the monopoly assumption is a simplification in some locales, unlike many other

countries, the extent of competition in the local broadband services market is very limited in the United

States, so much so that in many places, a single broadband service provider is often a de facto monopolist

(Economides 2008, Hausman, et al. 2001). The situation is aggravated by the high switching cost of long-

term service contracts and incompatible broadband technologies between cable and phone companies.

Further, many customers are not qualified for a digital subscribers line (DSL) broadband service from

phone companies, because they exceed the three miles distance limit from the phone companys nearest

switching office, making the cable operators the de facto monopolistic broadband service provider in

several local markets (Turner 2007). Thus, in addition to providing the benefit of making the analysis

tractable, the assumption closely reflects the reality of local broadband services in the U.S.. The ISP

charges consumers a fixed Internet access fee F per unit time. If it is allowed to charge the online content

providers, the ISP would charge the content provider a price p, a per packet charge for the priority

transmission of its packets. To keep the model tractable, we do not consider differential pricing that the

ISP might employ (Shapiro and Varian 1998). Then the payoff function for the ISP is F under net

neutrality and ( )1Y GF I p x I p x + + without net neutrality. A list of our notations is provided in Appendix A.

--- Insert Figure 3 about here ---

As shown in Figure 3, the timing of the game is as follows. The ISP announces the Internet access

fee F to consumers and the preferential delivery charge p to content providers under NNN. Based on the

announced fees, the two content providers decide simultaneously whether to pay the premium price for

priority delivery of their content. After the ISP and the content providers make their respective decisions,

consumers choose either content provider Y or content provider G. In the following sections, we use

13

backward induction to deduce the subgame perfect Nash equilibria of the game with and without net

neutrality regulation.

Net Neutrality

In this section we analyze the model under net neutrality where the ISP decides on the optimal

Internet access fee F, and consumers choose between content providers Y and G. The content providers do

not have any decision to make here.

Content decisions for consumers

Although individual consumers choose contents between Y and G independently, the consumers

decisions as a whole can be represented by the marginal consumer x with all the consumers located in

[ ]0,x choosing Y and all the consumers located in [ ],1x choosing G. The subscript NN denotes the case of net neutrality. Then NNx denotes the marginal consumer that is indifferent between content provider Y

and content provider G under NN and can be specified by

( ) ( ) ( )NN NN NN NN1d dV tx F V t x F = (1)

This leads to NN12

x = implying equal market share for the two content providers. The payoff to content

provider Y is NN_Y NN1

,

2Y Yx r r = = and the payoff to content provider G is

( )NN_G NN 11 2G Gx r r = = .

Pricing decisions for the ISP

Under NN, Internet access fees collected from consumers are the only revenues for the ISP.

Assume the ISP has negligible running costs. Anticipating consumers choices, the ISP solves the profit

maximization problem as follows:

14

( )( )

NNNN NN

NN_Y NN

NN_G NN

max

s.t. 0, 00, 1

FF

U x x xU x x x

=

(2)

This leads to ( )NN NN 2t dF V

= = .

No Net Neutrality

Next, we analyze the situation when the ISP is allowed to charge the content providers for

preferential treatment of the latters packets. Without net neutrality, the ISP charges the content provider

p per packet for priority over its competitors packets, should its competitor choose to not pay. When

both content providers pay the price p , both their packets receive equal treatment.

Content decisions for consumers

Given the ISPs choices of F , p and content providers choices YI and GI , consumers decide on

their preferred content provider. Based on content providers choices, there are essentially four possible

outcomes: neither content provider pay ( 0Y GI I= = ); one content provider pays and the other does not

(which results in two different outcomes 1YI = , 0GI = and 0YI = , 1GI = ); and both content providers

pay ( 1Y GI I= = ).

Outcome 1: Both content providers opt for not paying the priority price 1p ( 0Y GI I= = ). The

indifferent consumer 1x is signaled by ( )1 1 1 1( ) ( ) 1d dV tx F V t x F = which leads to

1

12

x = .7 Notice that this outcome amounts to the same result as in net neutrality (NN).

Outcome 2: Content provider Y pays 2p while content provider G chooses not to pay ( 1YI = ,

0GI = ). In Outcome 2, content provider Ys packets are prioritized and therefore face congestion only to

7 To facilitate understanding, we have made the numerical subscripts (1 through 4) of the various variables

correspond to the different Outcomes 1 through 4.

15

the extent of the traffic from Y, but content provider Gs packets are not so that Gs congestion is a

function of the entire traffic. The marginal consumer 2x who is indifferent between content provider Y

and content provider G under NNN in Outcome 2 is specified by:

( ) ( )( )2 2 2 22 2( ) ( ) 1d dV tx F V t x F

x x

=

(3)

This leads to 212

x > (see

Appendix B for the proof) meaning that content provider Y enjoys a larger market share 2x at the

price of 2p . Notice that Gs traffic faces delay costs of an M/M/1 priority queue with preemption.

Outcome 3: This case is the opposite of Outcome 2. Content provider G decides to pay the

preferential packet treatment price of 3p per packet, while content provider Y chooses not to pay ( 0YI = ,

1GI = ). Carrying out a similar analysis, we denote the marginal consumer 3x who is indifferent between

content provider Y and content provider G under NNN in Outcome 3 and is specified by:

( ) ( ) ( ) ( )3 3 3 333( ) ( ) 1 11d dV tx F V t x F

xx

=

(4)

It follows that 312

,x < meaning that content provider G enjoys a larger market share 31x at the

price of 3p .

Outcome 4: Both content providers pay the priority price 4p to have their content delivered (

1Y GI I= = ). Since both content providers packets are treated the same, this leads to 4 1 NN12

x x x= = = .

Delivery service decisions for content providers

Given certain values of F and p , content providers Y and G decide whether to pay for the

preferential delivery of data packets. Content provider Ys decision problem is represented by

16

( ) ( ){ }max , ,Y

Y Y G Y Y GIr x I I I p x I I , while content provider Gs decision problem is represented by

( ) ( ){ }max 1 , 1 ,G

G GY G Y GIr I px I I x I I . Table 2 presents the payoff matrix for the content

providers under the four outcomes.

Outcome 1: The revenues for content provider Y and G are 112 YY

r = and 112 GG

r = ,

respectively. The incentive compatibility constraint for content provider Y is 1 2 0Y Y and the

incentive compatibility constraint for content provider G is 1 3 0G G .

Outcome 2: The revenue for Y is 2 22 YY x r x p = and the revenue for G is ( )22 1 GG x r = . The incentive compatibility constraint for content provider Y is 2 1 0Y Y and the incentive

compatibility constraint for content provider G is 2 4 0G G . These two incentive compatibility

constraints can be reduced to

2

2

1 2Y

xp r

x

(5)

and

2 1 21 2 G

xp r

(6)

Compare the right hand side (RHS) of constraints (5) and (6) leads to (5)(6) 2

12

Y

G

RHS rRHS x r

= < since

2

1 and

2 G Yx r r> > . Thus, there is no feasible p such that content provider Y pays for the preferential

delivery while content provider G does not. Therefore Outcome 2 cannot be an equilibrium. Note that this

result is driven by the fact (or more correctly, the assumption) that G Yr r> . In other words, if G Yr r> , we

can never have an outcome where content provider Y decides to pay and G does not. The assumption

G Yr r> does not affect the key results of our analyses. If, however, this assumption is reversed, then

Outcome 3 instead of Outcome 2 cannot be an equilibrium.

17

Outcome 3: The revenues for Y and G are 33 YY x r = and ( ) ( )3 33 1 1GG x r x p = respectively. The incentive compatibility constraint for content provider Y is 3 4 0Y Y and the

incentive compatibility constraint for content provider G is 3 1 0G G . These two incentive

compatibility constraints can be reduced to

31 21 2 Y

xp r

(7)

and

3

3

1 21 G

xp r

x

(8)

Comparing the RHS of (7) and (8) gives ( )(7) 3(8)

2 1 YG

RHS x rRHS r

= . To analyze the magnitude of this ratio, we

consider two possibilities in the relative values of Gr and Yr : Case I: ( )32 1G Yr x r< ; and Case II: ( )32 1G Yr x r . If Case I holds, there is no feasible p such that content provider G pays for the

preferential delivery and content provider Y does not, and therefore Outcome 3 cannot be an equilibrium.

If Case II holds, Outcome 3 may be an equilibrium. We will further analyze the equilibrium results in

next subsection (i.e., stage 1 of the game).

Outcome 4: In this case, content providers Y and G get the same revenues (i.e. 12 Y

r and 12 G

r for

Y and G respectively) as those in Outcome 1. Both content providers, however, incur an extra expense of

12

p , which paid to the ISP. The incentive compatibility constraint for content provider Y is

4 3 0Y Y and the incentive compatibility constraint for content provider G is 4 2 0G G . We note

that Outcome 4 is a classical prisoners dilemma that has been described in other contexts (Brander and

Spencer 1983, Roller and Tombak 1990) both content providers know that they would be better off by

not paying, but given the relative proximity of their per-consumer revenue streams, they end up paying

the ISP.

18

Pricing decisions for the ISP

Expecting the best responses of both content providers and consumers, the ISP analyzes the

maximum profit that he can make under the various permutations of the choices of the content providers

to pay him (recall that the content providers can decide either to pay or not pay the priority delivery

charges) and the corresponding consumers choices represented by the marginal consumer x.

Outcome 1: Both content providers opt to not pay the priority price 1p ( 0Y GI I= = ). Similar to

the NN case, the Internet access fees collected from consumers is the only revenue for the ISP. The

broadband provider solves the following profit maximization problem:

( )( )

1 11 1

,

NNN_Y 1

NNN_G 1

1 2

1 3

max

s.t. , , 0, 0, , 0, 1

00

F p

Y G

Y G

Y Y

G G

F

U x I I x xU x I I x x

=

(9)

The first two constraints are participation constraints for consumers of content provider Y and G

respectively. The last two constraints are incentive compatibility constraints for the content providers.

The consumers participation constraints can be reduced to 212

1 2Y

xp r

x

and 313

1 21 G

xp r

x

. Since

2 3

1 12 2

x x = (this is shown to be true in Outcome 3, which is discussed later) and G Yr r> ,

3 2

3 2

1 2 1 21 G Y

x xr r

x x

, which implies that the ISPs optimal preferential delivery charge can be

specified as 313

1 21 G

xp r

x

. Therefore the results of the profit maximization problem of the ISP in

Outcome 1 are: the ISP makes a profit of 1 1 ( ) 2t d

F V

= =

and charges the content providers a

fee 1p such that 33

11 21 G

xp r

x

. The best response for both content provider Y and content provider G

is to Not Pay the fee.

19

Outcome 2: Content provider Y pays 2p while content provider G chooses not to pay (

1, 0Y GI I= = ). In addition to the revenue from end users, the ISP also gains revenue from content provider

Y for preferential delivery of Ys content. The broadband providers profit maximization problem is thus:

( )( )

2 22 2 2 2

,

NNN_Y 2

NNN_G 2

2 1

2 4

max

s.t. , , 0, 0, , 0, 1

00

F p

Y G

Y G

Y Y

G G

F x p

U x I I x xU x I I x x

= +

(10)

The first two constraints are the participation constraints of the consumers that prefer G and the

consumers that prefer Y respectively. The last two constraints ensure that while content provider Y will

pay, content provider G will not. As discussed in the second stage of the game, there is no feasible 2p to

induce Outcome 2.

Outcome 3: Content provider G pays 3p while content provider Y chooses not to pay (

0, 1Y GI I= = ). The ISPs profit maximization problem is given by:

( )( )( )

3 33 3 3 3

,

NNN_Y 3

NNN_G 3

3 4

3 1

max 1

s.t. , , 0, 0, , 0, 1

00

F p

Y G

Y G

Y Y

G G

F x p

U x I I x xU x I I x x

= +

(11)

The first two constraints are the consumers participation constraints, while the last two ensure

that this outcome actually holds i.e. Y does not pay, but G does. Recall the two cases discussed in

content providers decisions. If Case I holds, i.e., ( )32 1G Yr x r< , there is no feasible 3p and Outcome 3 is not possible. If Case II holds, i.e. ( )32 1G Yr x r , the ISPs optimal choice of pricing strategy and its

profits are given by the expressions ( ) ( ) ( )3 33 1 1d

F V t xx

=

, 3

33

1 21 G

xp r

x

=

, and

( ) ( ) ( )3 3 333 3 31

1 ( ) 11 2 Gd

F x p V t x x rx

= + = +

.

20

Outcome 4: Both content providers pay the priority price 4p ( 1Y GI I= = ), so that neither content

gets any relative advantage for delivery. The ISP now gains revenue from consumers as well as both

content providers and therefore solves the following optimization problem:

( )( )

4 44 4 4

,

NNN_Y 4

NNN_G 4

4 3

4 2

max

s.t. , , 0, 0, , 0, 1

00

F p

Y G

Y G

Y Y

G G

F p

U x I I x xU x I I x x

= +

(12)

It follows that the ISPs optimal pricing strategy is given by 4 ( ) 2t d

F V

=

, and

( )34 1 2 ,Yp x r= while his profit is ( )34 4 4 ( ) 1 22 Yt d

F p V x r

= + = +

.

We note that ( )32 1G Yr x r (i.e. Case II) is a necessary condition for Outcome 3 to be an equilibrium. When this condition is not satisfied, i.e., ( )32 1G Yr x r< (Case I), we have only two potential equilibria (Outcomes 1 and 4). Recall further that Outcome 2 is never an equilibrium as long as G Yr r> .

In order to determine his optimal pricing strategy * *( , )F p , the ISP compares the profits under the

various outcomes and for a given set of parameter values, chooses its pricing strategy (which drives the

equilibrium to one of the above-mentioned outcomes) to arrive at the highest profit. As the monopolist

gatekeeper between the content providers and the customers, the ISP can essentially drive the direction

of the equilibrium in such a way that it ensures the highest possible profits.

We can readily observe that 4 1 > in both Case I and Case II. As a result, in Case I, the

broadband provider will set the final ( )* *,F p to * 4 ( ) 2t d

F F V

= =

, ( )* 34 1 2 Yp p x r= = , and

realize the profit ( )* 34 ( ) 1 22 Yt d

V x r

= = +

since Outcomes 1 and 4 are the only two

potential equilibria.

21

In Case II, the ISP needs to compare 3 with 4 in order to determine the outcome that gives

the maximum profit, which leads to the following comparison:

( ) ( ) ( ) ( )3 3 333 41

1 ( ) 1 21 2 2G Yd t d

V t x x r V x rx

= + +

( )[ ]3 31 2 1 22 G Yx r r t x =

after applying Eq. (4) and some algebra.

We observe that if ( )32 1 2G Y tr r x + , then 3 4 . The broadband provider will then set the

menu of prices ( )* *,F p to ( ) ( )* 3 33 ( ) 1 1d

F F V t xx

= =

, * 3

33

1 2,

1 Gx

p p rx

= =

and will realize

the profit ( ) ( )*

3 33

31( ) 1

1 2 Gd

V t x x rx

= = +

.

Conversely, if ( )32 1 2G Y tr r x< + , then 3 4 < . The broadband provider will then set the

menu of prices * *( , )F p to * 4 ( ) 2t d

F F V

= =

, and ( )* 34 1 2 ,Yp p x r= = to attain a profit of

( )* 34 ( ) 1 22 Yt d

V x r

= = +

.

Note that the condition that ensures the ISP to realize more profit in Outcome 3 than Outcome 4

also satisfies the condition for Outcome 3 to be feasible since ( ) ( )3 32 1 2 2 1 .Y Ytr x x r+ > We thus

simplify the combination of the profit comparison condition (between Outcomes 3 and 4) and the

feasibility condition of Outcome 3 into the following two cases.

Case A: Outcome 3 is not only feasible but also more profitable to the ISP than Outcome 4. Therefore,

Outcome 3 is the equilibrium. The required condition for Case A is

( )32 1 2G Y tr r x + (13)

22

Case B: Either Outcome 3 is not feasible, or when Outcome 3 is feasible, Outcome 4 is more profitable.

Therefore, Outcome 4 is the equilibrium. The required condition for Case B is

( )32 1 2G Y tr r x< + (14)

Figure 4 summarizes these results graphically, by plotting Gr against Yr . The ISPs choice of

the equilibrium in the game between the two content providers is dictated by the relative magnitudes of

their revenue-generation capabilities. In Region A (corresponding to Case A), where Gr is significantly

greater than Yr , the ISP chooses its pricing strategy in such a way as to drive the game to Outcome 3,

where the content provider with higher profitability has the incentive to pay for priority delivery, while

the content provider with lower profitability does not have the incentive to pay that fee. Conversely, in

Region B (corresponding to Case B) where the content providers have more comparable revenue rates,

the game will end up with Outcome 4, when both content providers pay the ISP the priority delivery fee.

--- Insert Figure 4 about here ---

The choice of making G Yr r> is simply a matter of convenience of exposition, that the

generalized results are symmetric on either side of the line G Yr r= as indicated in Figure 5. We can

interpret Regions C and D be interpreted analogously as we did with Regions A and B.

--- Insert Figure 5 about here ---

Gainers and Losers Comparison between NN and NNN

In this section, we consider the resulting surpluses for the various players. These results can then

be used by the policymaker who has to decide whether to allow the ISP to charge for preferential service

(i.e., opt for NNN), or continue to maintain the NN status quo. The policymaker can proceed to compare

the equilibrium under NN and NNN, by evaluating the payoff for the ISP and content providers,

consumer surplus, and social welfare under each regime.

23

From the ISPs point of view, NNN is preferred to NN since the profits in either Case A or Case

B, *3 or *

4 is higher than *

NN . What is of interest to the policymaker is whether the other participants

gain from this arrangement too and whether social welfare as a whole increases. The results of this

analysis are summarized in the following proposition.

Proposition 1 (Gainers and losers in the short run):

The economic outcomes in the short-run under NN and NNN vary. Specifically, using NN as the

benchmark,

(a) Social welfare would either increase or remain unchanged depending on parameter values as stated in

conditions (13)-(14). Likewise, consumer surplus would increase or remain unchanged.

(b) Content providers are usually worse off under NNN except under condition (13) when the content

provider paying the priority delivery fee has the same surplus as under NN.

(c) The ISP is unambiguously better off.

Proof: See Appendix C.

These results are organized in Table 4. Appendix C contains the details of their derivation.

--- Insert Table 4 about here ---

Clearly, the gains of abolishing net neutrality are not experienced equally. While the monopolist

Internet service provider gains if no net neutrality were in place (in both Cases A and B of Table 4), the

content providers are definitely worse off under this arrangement. Only content provider Gs surplus is

unchanged under Case A. It is interesting to note that G does not get to enjoy the increase in the number

of consumers, since the extra rent is fully extracted by the Internet service provider. It is therefore no

wonder why the content providers and the Internet service providers have been on the opposite sides of

the net neutrality debate.

The fate of the end consumers is more nuanced. If the two content providers do not differ

significantly (regions B or D in Figure 5) in terms of their revenue generation rates, the consumer surplus

is unchanged. Consumers as a whole do stand to gain if one content provider is significantly better than

24

the other in revenue generation (regions A and C in Figure 5). This increase in overall consumer surplus,

however, is derived at the expense of the group of consumers whose content provider does not pay the

priority charge, a result contrary to the assertion of the ISPs that no consumer would be left worse off

under the new arrangement (WSJ 2006).

Social welfare as a whole, in contrast, is at least as high under NNN as it is under NN, and is

sometimes higher. Under NNN Case B, social welfare (like consumer surplus) does not change, but there

is a transfer of wealth from the content providers to the Internet service provider. This transfer is made

possible by the priority delivery charge that the Internet service provider extracts from both content

providers, since the subscription fee to end users does not change ( * *4 1F F= ). Under NNN Case A, the

consumer surplus increases due for the most part to the lower subscription fees for all consumers (* *

3 1F F< ). The winners under this arrangement are the consumers of content provider G (who are a

majority) and the ISP, while the losers are the consumers of content provider Y and content provider Y

itself. Content provider Gs surplus from the additional consumers that have migrated from Y is fully

siphoned away by the ISP.

Capacity Expansion Decision Does NN Hinder the Broadband Service Providers Incentive to Expand Infrastructure Capacity?

The other key question for the policymaker is the Internet service providers motivation to expand

capacity under NN. To discuss this question, we consider the long-run problem where the ISP can choose

its capacity .

Let ( )C be the cost associated with capacity . The long-run problem can be modeled as a three-stage game where the ISP chooses capacity and announces the Internet access fee F to consumers

and preferential delivery fee p to content providers under NNN in the first stage. Based on the announced

fees, in the second stage content providers choose whether to pay or not pay for preferential delivery, and

in the third stage consumers choose between content provider Y and content provider G. In the long-run

25

problem, we also need to consider the cost of capacity expansion (this was not an issue in the preceding

analysis, since in the short run the existing network capacity is fixed).

Our objective is to determine the ISPs incentive to expand capacity and its optimal capacity

decision under NN and NNN, and compare the two meaningfully in order to find the regime under which

the incentive for the ISP to expand capacity is higher. Finally, the choice of regime (NN or NNN) is not

under the control of the ISP, and is a choice that lies with the policymakers (who can calculate the

aforementioned incentives).

We observe that evaluating the incentive to expand for the ISP is different from calculating just

the magnitude of its profit.8 Further, there is a crucial distinction between whether the ISP has incentive to

expand the capacity and whether the ISP should expand the capacity as suggested in Jamison and Hauge

(2007).9

We go through a process similar to that employed in analyzing the short-run problem to

investigate ISPs incentive to expand capacity and his optimal capacity choice.

Decisions for both content providers and consumers in the capacity expansion problem

Given certain F , p , and , the analysis of the best responses for content providers and

consumers do not differ from those in the short-term problem and is therefore not repeated here. The

content providers profits and consumers utilities in the long-run problem will be multiplied by 11

where is the discount factor. In other words, the content providers long-run profits are 11Yi Yi

pi =

for Y and 11Gi Gi

pi = for G, where ( 1,2,3,4)i = represents the four different outcomes. The long-run

8 A simple example illustrates this. The incentive to expand is greater if the ISPs net profit goes up from $1 to $5

(an increase of $4) by expanding, than when the net profit goes up from $5 to $8 (an increase of $3). 9 Jamison and Hauge (2007) suggest three main propositions, all of which do not hold up to scrutiny. Propositions 1

and 3 are based on what they call a non-degradation condition which states that the ISPs should increase capacity, whether or not they have the incentive to do so; while Proposition 2 is based on assuming a peculiar consumer utility function wherein a consumer consumes all the content available to her on the Internet, regardless of quality, and this consumption increases her utility an assumption that counters reality.

26

utility of an arbitrary consumer [0,1]x is ( ) ( )11Y Y

u x U x=

if content provider Y is chosen, and is

( ) ( )11G G

u x U x=

if content provider G is chosen.

Pricing and capacity expansion decisions for the ISP in the capacity expansion problem

Expecting the best responses of content providers and consumers, we evaluate the optimal

decision ( )* * *, ,F p for the ISP.

Under Net Neutrality or Outcome 1: 112

x =

The ISP solves the following problem:

( )( )( )

1 1 11 1

, ,

NNN_Y 1

NNN_G 1

1 2

1 3

11

max1

s.t. , , 0, 0, , 0, 1

00

F p

Y G

Y G

Y Y

G G

F C

u x I I x x

u x I I x x

pi

pi pi

pi pi

=

where 2x and 3x are as defined in Outcomes 2 and 3.

The long-run objective function represents the net cash flow for the ISP. We know that the first

constraint is binding: i.e., *1 *

1

( )2t d

F V

=

. The optimal capacity *1 can be derived by maximizing

the long-term net cash flow ( )* *1 1*1

1 ( )1 2

t dV C

pi

=

. Eq. (15) gives the first order condition

of this optimization problem.

( )( )1

21 0

1Cd pi

=

=

(15)

The term 1pi

is the marginal increase in profit of the ISP with respect to , and as such we

can interpret this term as the ISPs incentive to expand capacity.

27

Outcome 2: It can be proved analogously as discussed earlier in the short-run problem that given

the nature of the parameter values, this outcome is not possible.

Outcome 3: 31

,

2x < determined by ( )[ ]( ) ( ) ( )3 33 3

3

3 3 3

11 1

d dtx t x

x x

+ = +

. As

before, ( ) ( )*

3 3 *3 3

( ) 11d

F V t xx

=

, and * 333

1 2.

1 Gx

p rx

=

The ISP will choose the optimal capacity *3 to maximize the long-term net cash flows once

again:

( ) ( ) ( ) ( )( ) ( )

* *

3 3 3 3*3 3

*

3 3

11

11

111 2

1 1 22 2

G

G

dV t x x r Cx

V t x r t t C

pi

= +

= + +

Then, *3 can be characterized by the first order condition in Eq. (16):

( )3 33

2 21 1 11 01 1 2G

t Cxt r x

pi

= =

(16)

Outcome 4: 412

x = , ( )*4 *42

t dF V

=

and ( )*4 31 2 Yp x r= .

Similarly, the corresponding net cash flows are given by

( ) ( ) ( )* *4 3 4*4

11

1 22 Yt dV x r C

pi

= +

and *4 is the solution of Eq. (17):

( )( )34

21 2 0

1 1 YCxd

rpi

= =

(17)

--- Insert Figure 6 about here ---

Comparing the ISPs incentive to expand capacity under NNN (either 3pi

or 4pi

) to that under

NN, the ISP has more incentive to expand capacity under NN when

28

( ) ( )3 32 1 2 2 or 1 2G Y Gtr r x tr x < + (18)

and the ISP has more incentive to expand capacity under NNN when

( ) ( )3 32 1 2 2 1 2Y Gtr x r t x + < (19)

These conditions are graphed in Figure 7 with Condition (18) corresponding to the shaded area and

Condition (19) corresponding to the unshaded area. The following proposition summarizes the result

concerning ISPs incentive to expand infrastructure capacity.

--- Insert Figure 7 about here ---

Proposition 2 (ISPs incentive to expand infrastructure capacity):

Except for the region defined in Condition (19), the ISP has more incentive to expand capacity

under net neutrality.

Proof: See

Appendix F.

To understand the nature of the result in Figure 7, note that the capacity costs under NN and

under NNN are identical given any same capacity level. Hence, we only need to focus on the revenue of

the ISP. Further, under NN, the ISPs revenue ( )11 2

t dV

is derived solely from the

consumers while the ISPs revenue under NNN has two components: the contribution from the consumers

and the contribution from the content providers. The revenue contribution from consumers increases in

(as consumers enjoy reduced congestion), while the revenue contribution from the content providers

decrease in (as the content providers have a reduced willingness to pay for priority delivery when

congestion is reduced). Specifically, when both content providers pay (Regions B and D in Figure 7), the

29

ISPs long-run revenue from consumers is ( )11 2

t dV

, which increases in . The ISPs

long-run revenue from the content providers is ( )31 1 21 Yx r , which decreases in . Thus, by

increasing the capacity , the ISP platform gains from the consumers side and loses from the content

providers side. Since the gains under NN and under NNN (Case B) are the same and the ISP incurs a loss

from the content providers side under NNN from expanding the infrastructure capacity, the ISP will

almost always have a higher incentive to increase capacity under net neutrality.

When only one content provider pays (Region A or C in Figure 7), and, the contribution from the

consumers is ( ) ( ) ( )3 31

11 1

dV t x

x

which increases in and the contribution from the

content providers is 31 1

1 2 Gx r

which decreases in . The ISP gains from the consumers side and

loses from the content providers side when increasing capacity , a result similar to the case where both

content providers pay. It is only in the small unshaded area in Figure 7 that under Condition (19) the gain

outweighs the loss to give the ISP more incentive to expand the capacity under NNN.

From the policymakers perspective, the questions of utmost interest concern the ISPs optimal

capacity choices under NN and NNN, and whether the ISPs optimal capacity choices under NN and

NNN are socially optimal. The following two propositions provide useful guidance to the policymaker in

addressing these questions.

Proposition 3 (ISPs optimal capacity choice):

Except for the region defined in Condition (19), the optimal capacity choice under net neutrality

(NN) is higher than under no net neutrality (NNN).

Proof: See Appendix G.

Proposition 4 (Whether ISPs optimal capacity choice is socially optimal?):

30

The ISP always invests at the socially optimal level under net neutrality. Abolishing net neutrality

results in underinvestment in infrastructure capacity by the ISP when both content providers pay the

priority delivery charge, and either underinvestment or overinvestment when only one content provider

pays the priority charge.

Proof: See Appendix H.

A corollary of Propositions 3 and 4 is that while the ISPs optimal capacity choice might be

higher under NNN than under NN in some specific instances, this higher capacity choice reduces social

welfare. As an aside, as the ISPs capacity increases, the parameter space for Outcome 3 becomes bigger

in the expense of Outcome 4 (i.e., the line separating the two Outcomes squeezes out Region B as shown

in Figure 6). The intuition is straightforward: notably, as capacity increases, congestion goes down and

some customers of G prefer Y, which therefore has a lesser incentive to pay, thus making Outcome 3 more

probable and therefore making it possible for a switch in equilibrium from Outcome 4 to Outcome 3).

The comparative statics for the various pricing variables and the surpluses of the various parties

involved with respect to the capacity are summarized in Table 5.

-- Insert Table 5 about here --

Conclusion Policy Implications and Directions for Further Research

The absence of meaningful competition in providing broadband access to consumers in many

areas of the United States makes the broadband service provider a de facto monopolist, and therefore the

sole gatekeeper in determining (a) the content that gets across to the end users and (b) in what fashion.

Therefore the debate about net neutrality assumes tremendous importance to a policymaker. This research

aims to answer two issues therein in a stylized framework. We find that if the principle of net neutrality is

abolished, the ISP definitely stands to gain from the arrangement, as a result of extracting the preferential

delivery charge from the content providers. The content providers are thus left worse off, mirroring the

stances of the two sides in the debate. Depending on the parameter values in our framework, consumer

surplus either does not change or is higher in the short-run, and in the latter case, while a majority of

jaminb720Highlight

31

consumers are better off, a minority is left worse off with larger wait times to access their preferred

content. Social welfare in the short-run increases when compared to the baseline net neutrality case when

one content provider pays for preferential treatment, but remains unchanged when both content providers

pay. The crucial parameter that determines the nature of the equilibrium is the relative magnitude of the

revenue generation capabilities of the two content providers: if they differ significantly, the consumers of

the less effective (i.e. in terms of revenue generation) content provider, who are a minority, are left

worse off.

The incentive for the broadband service provider to expand under net neutrality is mostly higher

than the incentive to expand when the principle of net neutrality is abolished. The exception to this

outcome occurs when the ISPs profit accrues mostly from the consumers and only one content provider

has the incentive to pay the priority delivery fees. Similarly, for most of the parameter space, the ISPs

optimal capacity choice under net neutrality (NN) is higher than that under the no net neutrality (NNN)

regime. In fact, the experience in broadband markets around the world indicates that there might be some

other forces in play that account for the infrastructure capacity expansions in other countries. In Japan, for

example, fierce competition among broadband service providers has led to the introduction of download

bandwidth speeds in excess of 100 Mbps as far back as in 2004 (Yang, et al. 2004), with prices for the

consumers significantly lower than that in the United States (Turner 2005).

A final finding that should be of interest to policymakers is that under net neutrality, the ISP

invests in broadband infrastructure to reach the socially optimal level, but when there is no net neutrality,

the ISP either under- or over-invests in infrastructure.

Some immediate areas of future research include relaxing the assumption that the market is

covered under both NN and NNN, an assumption made in part to capture the broadband service

providers claim to not abandon its customers to win support for NNN. The broadband provider might

decide to not pursue all the current subscribers after all, if that scenario can ensure higher profits. When

the full market coverage assumption is relaxed, we postulate that under net neutrality the broadband

service provider will choose to cover the market if customers valuation of the broadband service is

jaminb720Highlight

jaminb720Highlight

jaminb720Highlight

jaminb720Highlight

jaminb720Highlight

32

higher than a threshold value, and only part of the market will be served by the broadband provider if

customers valuation is low. This threshold value is likely to depend on both the fit cost t and the delay

cost parameter d. With NNN, the broadband providers decision as to whether to serve the whole

consumer market should be similar. Intuitively, the threshold value under NNN should be lower than that

of NN, since the broadband provider has the option to subsidize the end users from the payments

collected from content providers. That is, the broadband provider could cover more of the market under

NNN.

It is important to note that in order to make a meaningful comparison between the two regimes,

the content providers must follow the same revenue model under both NN and NNN, and the issue

therefore is to choose a revenue model that accurately captures the incentives of the content providers.

Thanks to content providers like Google or Yahoo!, the overwhelmingly popular revenue model for

content providers in the online world is the advertisement-assisted model. In this framework, consumers

get full access to all the content from the online providers. The reason for the popularity of this model is

that free content brings about a large number of visitors, who in turn generate revenue by clicking on the

advertisements. For several years, many content providers aimed for a hybrid model, a prominent

example of which being the online edition of the New York Times, whereby some content, which was

advertisement-assisted, was available for free, but other, ostensibly higher-quality content, was made

available only to paying subscribers. The New York Times abandoned this revenue model in late 2007,

after discovering that the free, advertisement-supported content that was viewed by a large number of

users was more profitable than a limited audience of paying subscribers. In fact, Sydell (2007, audio

broadcast) reports that this might be the way that everything is going on the Web, and that the lone

example of a mainstream content provider that relies on a subscription model, the online version of the

Wall Street Journal, is also expected to make most of its content available for free before the end of 2008

(Anderson 2008).

While it has been argued that in the future a significant amount of content would have to be paid

for, that right now that remains a conjecture, with a growing body of evidence that the advertisement-

jaminb720Highlight

jaminb720Highlight

jaminb720Highlight

33

generated model is becoming more popular with time (Sydell 2007), even with many content providers

that would have been expected to follow a subscriber revenue model. A new online music company,

SpiralFrog, for example, is based on the advertisement-supported model, whereby users get to download

legal copies of music for free, and part of the advertisement-generated revenue goes towards paying the

royalty to the music publishers and labels that provide the music. Another example is Qtrax, a new

business offering free and legal music downloads with 25 million songs in its inventory. Thus, under the

current state of affairs, where the hybrid quality-differentiated model has been all but abandoned in favor

of the purely advertisement-assisted model, advertisement-supported free content from the content

providers seems to mirror the ground realities.

In fact, the prominent technology journalist and author Chris Anderson has argued that this new

revenue model based around free is here to stay (Anderson 2008), pointing out that today online content

generates revenue from banner advertisements, affiliate revenues, rental of subscription lists, sale of

aggregate information, licensing, live events, listing, paid inclusion, cost per install, getting users to create

content for free, streaming audio and video advertising, and API fees, to name a few (Wilson 2008).

Anderson (2008) argues that this is possible today, as the Web has made it possible to monetize two

scarcities that are valuable, reputation and attention, and coupled with the fact that the marginal cost of

the content is almost negligible, it has been possible for an increasing number of companies to generate

more revenue from the free content (e.g. a free album by Radiohead) than they could have by charging for

that content on traditional media (charging for that album on a compact disc).

In this paper, we do not consider the capacity allocation issue. One interesting extension would be

to study whether the ISP will find it optimal to partition the capacity and whether such capacity

partitioning will change the ISPs incentive to invest in expanding the infrastructure capacity. A major

reason for the ISP to invest less under NNN than under NN is that capacity expansion reduces the

attractiveness of priority delivery of packets for content providers. A partitioned capacity may enhance

the content providers willingness to pay for the priority charge under NNN, which in turn may make

34

ISPs capacity expansion more desirable under NNN. 10 Similarly, it is of interest to examine the

implication of net neutrality on the ability of a content provider to provide premium services (e.g., real-

time video or remote medical supervision) that require dedicated bandwidth.

Another direction of research would be to consider the effect of the broadband service provider as

a potential competitor to the content (or other service) providers. Such a situation already exists today

(albeit with limited success so far) with broadband service providers like Comcast building their own

modest Internet portals, or with providers like AT&T or Comcast offering VoIP digital phone services

(Krim 2005). Policymakers would then like to assure that the monopolist broadband service provider does

not enjoy unfair competitive advantage, and look for guiding principles for ensuring fair competition

under NN and NNN. The issue gets more complicated and interesting when a service provider like AT&T

might end up cannibalizing its own traditional phone service by offering the new VoIP product.

Acknowledgement: We gratefully acknowledge the very useful comments from the associate editor and three anonymous reviewers, and seminar participants of Arizona State University, Bocconi University (Milan, Italy), National Tsing Hua University (Taiwan, R.O.C.), Purdue University, University of Arizona, University of British Columbia (Canada), University of Florida, University of Illinois at Urbana Champaign, University of Notre Dame, University of Utah, University of Texas at Austin, and University of Washington (Seattle). Any remaining error belongs to the authors.

10 We thank an anonymous reviewer for raising this issue.

35

Figure 1: Schematic of the model

Figure 2: The content providers and their share of consumers

Figure 3: The sequence of events in the game

The broadband provider announces F and p

Content providers choose Pay or Not Pay

Stage 1 Stage 2

Consumers choose content provider Y or content provider G

Stage 3

Content Provider #1 (e.g., Company Y)

Internet Backbone

Content Provider #2 (e.g., Company G)

End Consumers Broadband Service Provider at local loop

0 (Y) 1 (G)x, marginal consumer

36

Figure 4: Graphical representation of the regions for arriving at different equilibria of the game when G Yr r>

Figure 5: Generalized representation of the regions for arriving at different equilibria of the game

G Yr r=

Region A

( )32 1 2G Y tr r x= +

Region B

Gr

Yr

Gr

Yr

G Yr r=

( )22 1 2Y G tr r x= +

Region A

( )32 1 2G Y tr r x= +

Region B

Region D

Region C

37

Figure 6: Graphical representation of the region for arriving at a different equilibrium of the long-run game when capacity is expanded from 1 to 2

Notes: In the shaded region, the ISP has higher incentive to expand capacity and the optimal capacity level is higher under NN. In the white region, the ISP has lower incentive to expand capacity and the optimal capacity level is lower under NN.

Figure 7: ISPs incentive to expand capacity

Region A ( )23,2 1 2G Y tr r x = +

Yr

G Yr r=

( )13,2 1 2G Y tr r x = +

Region B

Gr

0

( )13,1 2t x ( )23,1 2t x

( )32 1 2G Y tr r x= + G Yr r=

Region D ( )22 1 2Y G tr r x= +

Region A

Region B

Gr

Yr

Region C

( )31 2t x

( )32 1 2t x

( )21 2t x ( )22 1 2t x

0

38

Table 1: Delays under No Net Neutrality G pays G does not pay Y pays

11

Yw = , 11

Gw = ( ) ( )3 31Yw x

=

,

( )3 31

1Gw

x =

Y does not pay 2 2

1Y

xw

= ,

( )( )2 2Gw x

=

41

Yw = , 41

Gw =

Table 2: Content Providers Payoffs G does not pay G pays Y does not pay 1

12Y Y

r = , 112G G

r =

3 3Y Yx r = , ( ) ( )3 3 31 1G Gx r x p = Y pays 2 2 2Y Yx r x p = ,

( )2 21G Gx r = 41 12 2Y Y

r p = , 41 12 2G G

r p =

Table 3: Summary of results of the game

Case A: ( )32 1 2G Y tr r x + ( ) ( ) ( )* *

3 33

11dF F V t x

x

= =

* * 33

3

1 21 G

xp p rx

= =

( ) ( ) ( )* *

3 3 33

111 2 GdV t x x r

x

= = +

Case B: ( )32 1 2G Y tr r x< + * *

4 ( ) 2t dF F V

= =

( )* *4 31 2 Yp p x r= = ( )* *4 3( ) 1 22 Y

t dV x r = = +

39

Table 4: Comparison of various economic outcomes of interest under NN and NNN (The bold and italicized text shows how those economic outcomes change when moving from NN to NNN)

NN (Benchmark) NNN (Case A: Only G pays) NNN (Case B: Both Y and G pay)

F ( )NN 2t dF V

= ( ) ( ) ( )NNN_A 3 3 NN31 1

dF F V t x Fx

= =

Better off

( ) ( )NNN_B_ISP 4

3 NN1 22 Yt dV x r

=

= + >

Better off Content Provider Ys Profit

NN_Y12 Y

r = NNN_A_Y 3 NN_YYx r = < Worse off

NNN_B_Y 3 NN_YYx r = < Worse off

Content Provider Gs Profit

NN_G12 G

r = NNN_A_G NN_G12 G

r = = Unchanged

( )NNN_B_G 3 NN_G1 1 22 G Yr x r = <

Worse off Consumer Surplus NNCS 4

t=

2NNN_A 3 3 NN

1CS CS2

t x x

= + >

Better off NNN_B NNCS CS4

t= =

Unchanged

Social Welfare ( )

NNSW1 1

4 2 2Y Gt dV r r

= + +

( ) ( ) ( )NNN_A

23 3 3

3

NN

SW1 12 1

>SW

Y GdV t x x r x r

x

= + +

Increased

( )NNN_B

NN

SW1 1

4 2 2SW

Y Gt dV r r

= + +=

Unchanged

40

Table 5: Comparative statics with respect to capacity (Legend: + : increasing in ; : decreasing in ; 0 : independent of ; ? : depends on parameter values.)

NN (Benchmark) NNN (Case A: Only G pays) NNN (Case B: Both Y and G pay)

F ( )NN 2

t dF V =

+

( ) ( ) ( )NNN_A 3 3 31 1dF F V t x

x

= =

+

( )NNN_B 4 2t dF F V

= =

+

p N/A

3NNN_A 3

3

1 21 G

xp p r

x

= =

( )NNN_B 4 31 2 Yp p x r= =

ISPs Revenue

( )NN_ISP 1 2t dF V

= =

+

( ) ( ) ( )

NNN_A_ISP 3

3 33

111 2 GdV t x x r

x

=

= +

?

( ) ( )NNN_B_ISP 4

31 22 Yt dV x r

=

= +

?

Content Provider Ys Profit

NN_Y12 Y

r =

0

NNN_A_Y 3 Yx r = +

NNN_B_Y 3 Yx r = +

Content Provider Gs Profit

NN_G12 G

r =

0

NNN_A_G12 G

r =

0

( )NNN_B_G 31 1 22 G Yr x r = +

Consumer Surplus NNCS 4

t= 0 2NNN_A 3 3

1CS2

t x x

= +

NNN_BCS 4t

= 0

Social Welfare

( )NN 1 1SW 4 2 2Y Gt dV r r

= + +

+

( ) ( ) ( )NNN_A

23 3 3

3

SW

1 12 1 Y G

dV t x x r x rx

= + +

?

( )NNN_BSW

1 14 2 2Y Gt dV r r

= + +

+

41

References: House of Representatives, To establish broadband policy and direct the Federal Communications Commission to conduct a proceeding and public broadband summits to assess competition, consumer protection, and consumer choice issues relating to broadband Internet access services, and for other purposes, 2008, H.R. 5353, 2d 2d, February 12, R. E. J. Markeys R. E. J. Markey,at http://thomas.loc.gov/cgi-bin/query/z?c110:H.R.5353:.

Anderson, Chris, "Free! Why $0.00 is the Future of Business," Wired, 16, February 25 2008, 140-194.

Armstrong, M., "Competition in two-sided markets," RAND Journal of Economics, 37, 3, (2006), 668-691.

Armstrong, M. and J. Wright, "Two-sided Markets, Competitive Bottlenecks and Exclusive Contracts," Economic Theory, 32, 2, (2007), 353-380.

Bandyopadhyay, Subhajyoti and Hsing Kenneth Cheng, "Liquid Pricing For Digital Infrastructure Services," International Journal of Electronic Commerce, 10, 4, (2006), 47-72.

Berners-Lee, Tim, "Neutrality on the Net," (2006), http://dig.csail.mit.edu/breadcrumbs/node/132, last accessed on February 13, 2009.

Bhargava, H. K. and D. Sun, "Quality-Contingent Pricing for Broadband Services," Proceedings of the Hawaii International Conference on Systems Sciences, Hawaii, 2005, 211b-220b.

Brander, J.A. and B.J. Spencer, "Strategic Commitment with R&D: The Symmetric Case," The Bell Journal of Economics, 14, 1, (1983), 225-235.

Carr, Nicholas, The Big Switch: Rewiring the World, from Edison to Google, W. W. Norton, 2008.

Dyson, Esther, "Intellectual Property on the Net," Edventure Holdings Inc., 28 December 1994.

Economides, N. and S. C. Salop, "Competition and Integration Among Complements, and Network Market Structure," The Journal of Industrial Economics, 40, 1, (1992), 105-123.

Economides, Nicholas, "Nonbanks in the Payments System: Vertical Integration Issues," (2007), NET Institute Working Paper #07-06, Stern School of Business, New York University, mimeo.

Economides, Nicholas, "'Net Neutrality,' Non-Discrimination and Digital Distribution of Content through the Internet," I/S: A Journal of Law and Policy for the Information Society, 4, 2, (2008), 209-233.

Economides, Nicholas and Joacim Tag, "Net Neutrality on the Internet: A Two-sided Market Analysis," (2007), NET Institute Working Paper #07-45, Stern School of Business, New York University, mimeo.

Edell, Richard J. and Pravin P. Varaiya, "Providing Internet Access: What We Learn from the INDEX Trial," INDEX Project Report #99-010W, University of California at Berkeley, 1999.

42

Gupta, A., Dale O. Stahl and A. B. Whinston, "An Economic Approach to Network Computing with Priority Classes," Journal of Organizational Computing and Electronic Commerce, 6, 1, (1996), 71-95.

Gupta, Alok, Dale O. Stahl and Andrew B. Whinston, "A Stochastic Equilibrium Model of Internet Pricing," Journal of Economic Dynamics and Control, 21, (1997), 697-722.

Hahn, Robert and Scott Wallsten, "The Economics of Net Neutrality," The Berkeley Economic Press Economists' Voice 3, 6, (2006), 1-7.

Hausman, J. A., J. G. Sidak and H. J. Singer, "Residential Demand for Broadband Telecommunications and Consumer Access to Unaffiliated Internet Content Providers," Yale Journal on Regulation, 18, 1, (2001), 129-173.

Helm, Burt, "Tech Giants' Internet Battles," (2006), http://www.businessweek.com/technology/content/apr2006/tc20060426_553893.htm?chan=technology_te