University of Mannheim / Department of Economics
Working Paper Series
Platforms and network effects
Paul Belleflamme Martin Peitz
Working Paper 16-14
Platforms and network e§ects!
UniversitÈ catholique de Louvain
University of Mannheim
This version: September 2016
In many markets, user beneÖts depend on participation and usage decisions of other users
giving rise to network e§ects. Intermediaries manage these network e§ects and thus act as
platforms that bring users together. This paper reviews key Öndings from the literature on
network e§ects and two-sided platforms. It lays out the basic models of monopoly platforms
and platform competition, and elaborates on some routes taken by recent research.
Keywords: Network e§ects, digital platforms, two-sided markets, tipping, platform compe-
tition, intermediation, pricing, imperfect competition
JEL-ClassiÖcation: D43, L13, L86
!Chapter prepared for the forthcoming Handbook of Game Theory and Industrial Organization, edited by
Luis Corchon and Marco Marini, Edward Elgar. We thank Markus Reisinger for helpful comments. Martin Peitz
gratefully acknowledges Önancial support from the Deutsche Forschungsgemeinschaft (PE 813/2-2).
yUniversitÈ catholique de Louvain, CORE and Louvain School of Management, B-1348 Louvain la Neuve,
Belgium, Paul.Belleáamme@uclouvain.be. Other a¢litation: CESifo.
zDepartment of Economics and MaCCI, University of Mannheim, 68131 Mannheim, Germany, Mar-
tin.Peitz@gmail.com. Other a¢liations: CEPR, CESifo, and ZEW.
Over the last two decades, the fast penetration of the Internet and the digitization of information
products has led to the rise of electronic intermediaries such as Amazon, Google, and Facebook.
Some of these intermediaries have become darlings on the stock market reáecting the believe
that they have or will become central players in market economies. Of course, intermediaries
are not a phenomenon of the internet, but have been around since ancient times. Many of these
intermediaries play an important role because of supply-side or demand-side scale e§ects. Of
particular importance are the latter, as many intermediaries require a certain volume of usage
to attract additional users. This puts network e§ects at the core of intermediaries or, as we will
call them, platforms.
A platform brings together a typically large number of users who interact with each other. For
instance, the traditional telecommunications provider is such a platform that brings together
people who may want to engage in communicating with each other. The more users on the
network the more valuable the communication service. This is an example of direct network
On many platforms we can distinguish between distinct groups of users, whose activities
a§ect the well-being of those in another group. One example are software platforms: They bring
together application developers and end users. Here, everything else given, end users may not
care about the presence of other users, but only about the number and quality of application
developers, while developers only care about the number and demand of end users. In this case,
network e§ects are indirect, as end users care about participation and usage of other end users
only indirectly, as more end users attract more developers which is beneÖcial for each end user.
The platform managing the interaction among distinct groups of consumers is called two-sided.
Some platforms allows for the interaction of buyers and sellers. Shopping malls are an example,
as they o§er retail space to sellers and invite buyers to go shopping. Everything else given,
sellers prefer a shopping mall that attracts more buyers and buyers prefer a shopping mall that
hosts more sellers. Trade fairs, áee markets, auction houses, and yellow pages have similar
features. While some of these platforms have been around for centuries, platforms as a way to
organize market activities have arguably gained more prominence with the rise of the internet.
To enable consumers to choose among a myriad of o§erings, horizontal and vertical search
engines as well as price search engines, booking portals, online auction and retail places have
become commonplace. As these digital platform are not subject to physical capacity constraints
and can quickly guide a potential buyer to products of interest, they are able to manage huge
volumes of interactions between buyers and sellers.
Platforms try to manage user participation and volumes of interaction. They can use price
and non-price instruments for this purpose. In particular, they may court one particular group
of users, e.g., buyers, to extract revenues from another group of users, e.g. sellers, who see
the e§orts of the platform to attract the Örst group of users. Managing users participation
and volume of interaction thus depends on the ability of platforms to convince users about the
decisions taken by other users. Users can become convinced because of past decisions other users
have made (to the extent that they are not easily revised) or by a platformís actions such as
publicly observed prices applicable to other users, which in turn a§ect expectations about the
decisions these other users will make.
This chapter reviews some key contributions to the economics of network e§ects and two-
sided markets. In Section 2 we provide a discussion of network e§ects, criteria to classify
di§erent platform markets, and a number of examples. We explore the economics of markets
with network e§ects in Section 3, and of two-sided markets in Section 4. Our aim is not to
provide an exhaustive overview of the literatures on network e§ects and two-sided markets;
instead, we pick a few contributions and elaborate on some key Öndings in a number of stylized
In this section we give a deÖnition of network e§ects, drawing a distinction between direct
(within-group) and indirect (cross-group) e§ects. We also precise what we hear by ëplatformsí.
We then o§er some illustrations.
2.1 DeÖning network e§ects
Network e§ects are present if users care about participation and usage decisions of other users.
The group of users making the same usage decision is loosely called the ënetworkí. In the simplest
setting, users care from an ex ante perspective only about the size of the network; in a more
general setting, also the identity of the users of the network matters. For instance, social norms,
languages and communication devices clearly generate positive network e§ects: the more they
are adopted, the larger the utility they confer to their adopters. For a social norm, only the size
of the network matters; in contrast, for communication means (e.g., a particular language or an
instant messaging application), a user is primarily concerned with the decision of the subset of
users with whom she has regular interactions; hence, the identity of the users of the network
Network e§ects may emerge in a large variety of contexts and may be positive or negative
depending on the circumstances. Road congestion and tra¢c jams are the prototypical examples
of negative network e§ects: the more drivers choose a particular road at a particular moment, the
slower the tra¢c on that road at that moment and, thereby, the lower the utility of every driver.
Fashion and fads generate positive network e§ects for those individuals whose utility increases
when they conform with the choices of others. Yet, the exact opposite applies for snobs, who
value the idea of having di§erent tastes then the ëmassí: for them, having someone choosing
like them generates a negative network e§ect ñ see, e.g., Grilo, Shy and Thisse (2001). Another
example of positive network e§ects can be found in the choice of geographical locations by Örms.
1To th i s end, we d raw f rom and e xpan d u p o n C hapt e rs 20, 2 1 , and 22 of B e lleá a mme and P eitz ( 2 015) .
Following the seminal work of Marshall (1890, 1920), the economic geography literature explains
why Örms can beneÖt from locating close to one another; one explanation is that when more
Örms locate in the same region, more workers (or, more generally, input suppliers) are drawn
to this region, which in turn makes the region more attractive for Örms.2We see in this last
example that network e§ects do not arise directly from the Örms making the same choice but
indirectly through the induced decisions of another group of agents (i.e., workers and/or input
To clarify what we will discuss in this chapter, we Örst distinguish between the direct and
indirect sources of network e§ects; we then narrow our focus on so-called ëplatformsí, which
somehow manage network e§ects.
2.1.1 Within-group and cross-group external e§ects
The number of participants and the intensity of use may a§ect a population of agents, as argued
above. In many environments, we can distinguish di§erent groups. One example are trading
platforms on which buyers and sellers interact. In this case, positive cross-group external e§ects
are present because buyers are, everything else given, better o§ the more sellers are present
and vice versa. Another example are content platforms that carry advertising. Here, vertically
integrated content is o§ered to entice consumers to join the platform. Consumers then may
pay directly for participation or indirectly with their attention to advertising, which is bundled
with content. If consumersí utility is decreasing in the volume of advertising, advertisers exert a
negative cross-group external e§ect on consumers, while consumers exert a positive cross-group
external e§ect on advertisers. In addition to cross-group e§ects, also within group external e§ects
may be present. For instance, if the platform becomes congested if too many actors from one
group participate, there are negative within-group external e§ects. Also, increased competition
between sellers constitutes a negative within-group external e§ect on the seller side.
2.1.2 Markets with platforms
Markets with platforms can be broadly deÖned as markets where the interaction between at least
some participants is facilitated and managed by an intermediary. Managing this interaction can
take many forms; the most obvious ways are setting prices for participation or usage, or setting
participation levels. The intermediary may have other instruments at its disposal. For instance,
it may impose certain contractual terms, and it may provide monetary or non-monetary beneÖts
for certain actions.
Our deÖnition includes markets with a single group that exhibits within-group external
e§ects, as long as an intermediary has at least one instrument to a§ect these within-group
external e§ects and market outcomes. This sets our deÖnition apart from recent deÖnitions
entering the policy debate, which require platforms to feature at least two sides (or two groups,
2For a c o m preh e nsiv e text b o o k o n this t o pic, s e e Fujit a a nd Thi s se (20 1 3).
using our terminology).3We believe that the broader deÖnition serves the purposes of analyzing
platform strategies (and, in particular, platform dynamics) better. The reason is that the
property to feature only within-group external e§ects within a single group or also cross-group
external e§ects with a second group is in some cases endogenous, i.e., it is a decision by the
intermediary, as we illustrate below. Of course, there is no right or wrong deÖnition. However,
we believe that a broader deÖnition is more appropriate, as it allows us to apply some economic
mechanisms to this broad set of phenomena. Later on, it will be useful to dedicate particular
attention to two-sided platforms, where additional issues arise.
Amazon and Facebook were initially one-sided platforms. Even today, their attractiveness de-
pends to a large extent on the strength of positive within-group external e§ects. Amazon, by
establishing its MarketPlace, added another group, namely independent sellers, which generates
positive cross-group external e§ects between buyers and sellers. However, their main strength
vis-a-vis rivals comes, arguably, from the positive within-group external e§ects on the buyer side,
arising from its recommender and reputation systems (where the latter refers to the reviews and
grades about products, not sellers). Facebook built its business model on providing advertisers
the possibility of (targeted) advertising. To the extent that private users rather dislike advertis-
ing, Facebook is exploiting the user base it maintains because of the social networking beneÖts
(which are within-group external e§ects) to provide beneÖts to advertisers. It then charges ad-
vertisers for this service. Here, users exert a positive cross-group external e§ect on advertisers,
while advertisers exert a negative cross-group external e§ect on users. Clearly, advertising here
serves as a monetization device (which beneÖts from a large, interconnected user base, as it
allows for better targeting). However, the strength of Facebook in the market place is arguably
due to within-group external e§ects arising from social networking among users. An alternative
strategy by Facebook could have been to charge users for participation or usage. Using the
narrower deÖnition proposed elsewhere, Facebook would not be classiÖed as a platform, even
though the interaction between users would continue to be present.
Telecommunication networks also provide a nice illustration of the di¢culty to draw a clear
line between within- and cross-group external e§ects. Most of the economic literature on telecom-
munication networks assumes, for simplicity, uniform calling patterns, i.e., an equal likelihood
for each subscriber to call and be called by any other subscriber;4this assumption of fully
symmetric participants implied a single group exhibiting within-group external e§ects. Another
simplifying assumption would be to consider that some people only make calls, while others only
receive calls (e.g., restaurants and customers who want to order for delivery or make a reserva-
3For i n stan c e, the Eu r op ean Com m i ssio n ( 2015 , p .5) ga ve th e f oll owin g d e Önit ion: ì ëO n line p l atfo r mí ref e rs to
an undertaking operating in two (or multi)-sided markets, which uses the Internet to enable interactions between
two or more distinct but interdependent groups of users so as to generate value for at least one of the groups.î
See also Monopolkommission (2015) and House of Lords (2016).
4See, for instance, Armstrong (1998), La§ont, Rey, and Tirole (1998a, 1998b), and de Bijl and Peitz (2002).
tion); in that case, there would be two distinct groups, with only cross-group external e§ects.
The reality is naturally somewhere between these two extremes: subscribers are heterogeneous
in their propensity both to make calls and to receive calls. Moreover, calling patterns are largely
reported to be nonuniform: as indicated above, most subscribers have a ëcalling circleí, i.e., a
subset of subscribers with whom they interact more frequently than with others.5Seen from an
individual subscriberís perspective, external e§ects are then mostly within-group (i.e., inside the
calling circle), cross-group e§ects (i.e., outside the calling circle) being relatively limited; yet,
each subscriber makes a di§erent distinction between the two types of external e§ects as calling
Finally, even if a platform facilitates the interaction between two distinct groups of users,
some network e§ects may be jointly generated by all users, irrespective of the group they belong
to, which may further blur the distinction between cross- and within-group external e§ects. Take
the example of peer-to-peer marketplaces like Uber or Airbnb, which enable the interaction
between providers and consumers of services; clearly, each group exerts positive cross-group
external e§ects on the other group. Yet, the quality of the matching between peers from the
two groups increases with the volume and reliability of data that the platforms collect from
providers and consumers alike. Hence, a form of within-group external e§ects appears: the
larger the participation on both sides, the more data is generated (about feedbacks, reputation,
reviews, geo-localisation, etc.), which enhances the quality of the platformís service and, thereby,
the utility of all users.
In this section, we explore the economics of markets with network e§ects. We Örst focus on the
demand side and derive the demand for a good that exhibits network e§ects. We then turn to the
supply side, considering Örst markets with a single network good, provided either by a monopoly
or by perfectly competitive Örms. We move next to markets with several network goods supplied
by distinct Örms, where decisions about prices and capacities cannot be separated from decisions
3.1 Demand for a network good
To determine user demand in a market with network e§ects, we consider an intermediation mar-
ket with a large number of potential users.6The total number of potential users has measure !.
Each user "is characterized by some stand-alone utility #iof using the services of the interme-
diary irrespective of the number of users. Users are distributed according to some cumulative
distribution function $deÖned on an interval !#% #". The cumulative distribution function is
continuous on its support and takes values $##$%&and $##$%'.
5See, e.g., Hoernig, Inderst and Valletti (2014).
6Our exposition on the demand for network goods follows Belleáamme and Peitz (2015, Chapter 20).
For the sake of simplicity, we focus on the participation decision of each user and do not
look at the usage intensity. Each user "pays a price &ifor obtaining the possibility to interact
with other users via an intermediary. When interacting, we assume that each user obtains an
additional utility 'i#!$that depends on the measure of participating users !. We set 'i#&$ % &
for all users. In particular, if everybody with a stand-alone utility of #iand higher participates
we have !% #' "$##i$$. A user may decide not to participate and obtain an outside valuation
of (0%&or to be active and obtain valuation #i('i#!$"&i.
We consider two alternative speciÖcations of heterogeneity among users, one with respect to
#iand one with respect to 'i. Suppose Örst that all users have the same function '##$%'i##$and
face the same price &%&i. Consumers with a large #itend to participate, while those with a low
#itend to be more reluctant. Suppose that there is an interior user with )#$##% #$that satisÖes
)#('##'"$#)#$$$"&%&. Thus, when 'is downward sloping or not too strongly upward sloping,
there exists a stable user equilibrium such that all consumers with #i%)#participate, while all
users with #i))#do not, with )#(*('##'"$#)#(*$$$"&+&and )#"*('##' "$#)#"*$$$"&)&.
There is a stable user equilibrium such that all users participate if the slope of 'is positive and
su¢ciently large, and #"¬ too small such that #('#!$"&%&.
It is important to note that in the presence of positive network e§ects, multiple equilibria
may arise. That is, the same price may give rise to several equilibrium network sizes. For
instance, if #)&%)#('##' "$#)#$$$ )#('#!$, with )#$##% #$, then on top of the interior
equilbrium with a mass of &)'"$#)#$) ! users, no participation and full participation are also
equilibria. As an illustration, suppose that #iis uniformly distributed on !&%'", there is a mass
one of users, and '#!$%*!. Then, for &%+,*, the following three situations are equilibria:
(i) no user participate (as the user with the largest stand-alone utility is not willing to pay the
price when no other user participates: '(*&&)&%+,*); (ii) the set of users with #i%',*
participate (as #i%',*is the indi§erent user when the mass of participants is equal to ',*:
',*(*&',*%&%+,*); (iii) all users participate (as the user with the lowest stand-alone
utility is willing to pay the price when all other users participate: &(*&'+&%+,*).
A special case is the situation in which users are homogeneous and the stand-alone utility is
set equal to zero. Then, if users su§er from the presence of more users with the intermediary,
there will not be any interaction for any positive price. However, if users beneÖt from the
presence of other users such that 'is strictly increasing, there is a stable consumer equilibrium
such that all consumers participate if '#!$+&.
Suppose next that all users have the same stand-alone utility but di§er in their valuation of
the interaction with other users. That is, #i%#for all users, while 'i#!$'%'j#!$for "'%-.
As there are multiple ways in which the functions 'i##$may di§er, let us develop one speciÖc
example to show that multiple equilibria may arise as well under this alternative speciÖcation of
user heterogeneity. Suppose that there is a unit mass of users (!%'), identiÖed by a parameter
.that is uniformly distributed on !&%'", and let user .value the interaction with other users
according to the function '##!$%.!. That is, we assume that network beneÖts increase linearly
with the size of the network of participants but with a di§erent intensity for each user. If all
users face the same price &to interact through the intermediary, the indi§erent user is identiÖed
.such that #()
.! %&. As all users with .%)
.will participate, the mass of participants
is equal to !%'")
., which implies that )
.%'"!. It follows that the inverse demand for
participation can be written as &%#(!#' "!$, which is a bell-shaped function of !that
reaches a maximum at !%',*, where &%#(',,. So, for any price &$##% # (',,$, there
are two ëdemand levelsí, i.e., the two values of !that solve &%#(!#' "!$. Moreover, !%&
is also compatible with such a price as in this case (i.e., if no user participate), each user has
a negative net utility and Önds thus optimal not to participate (#(.&&"&)(
.). As above, we can thus Önd prices for which three equilibrium participation levels coexist.
They all stem from ëself-fulÖlling propheciesí insofar as they correspond to network sizes that
generate utilities such that the combined participation decisions of the users exactly generate
these network sizes.
3.2 Monopoly provision of a network good
Suppose that the network good is provided by a monopoly intermediary and that this interme-
diary is able to choose how many users to connect to the network. Suppose also that there is
a constant marginal cost, /%&, to connect an extra user to the network. By choosing the size
of the network, the intermediary has the potential to internalize network e§ects as it recognizes
that a larger network raises the usersí willingness to pay and, thereby, its revenues. We may then
wonder whether a monopoly intermediary does not have the incentive to extend the network
up to the size that would prevail under perfect competition. However, this is generally not the
case, as we now show in a simple example.
Consider a pure communication technology that generates only network beneÖts (the stand-
alone utility is equal to zero for all users). As above, suppose that there is a unit mass of users
(!%'), identiÖed by a parameter .that is uniformly distributed on !&%'"; user .values the
possibility to communicate with a mass !of other users according to the function '##!$%
.!. Following the methodology developed above, we compute the inverse demand for network
participation as &%!#' "!$. Note that the maximum price is reached for !%',*and is equal
to ',,; we therefore assume that /)',,to make the problem non trivial. The intermediary
chooses !to maximize !2#' "!$"!/. From the Örst- and second-order conditions, we Önd the
proÖt-maximizing network size as !m%#'(('"+/$,+.
If the network good was supplied by a competitive industry, the network size !c, would
be such that &%/; that is !c#' "!c$%/, which is equivalent to !c% #' ( ('",/$,*. It
is easily checked that !c+!
m, meaning that despite its ability to internalize network e§ects
(which competitive Örms lack), the monopoly intermediary still restricts the network size (i.e.,
the quantity) below the perfectly competitive level, thereby reducing welfare.
As far as welfare is concerned, it is important to note that in the presence of network e§ects,
perfect competition generally also fails to achieve the Örst best. In our example, social welfare
is maximized when all consumers join the network (i.e., !#%') but, for any /+&,!c)'.7
Network e§ects are the source of the market failure: when joining the network, users do not
internalize the positive consumption externality that they exert on the other users.
3.3 Provision of competing network goods
We consider now situations where users can choose among several substitutable network goods
that are provided by di§erent Örms. As we will now discuss, the degree of compatibility among
the various networks is of crucial importance since it conditions the network beneÖts that users
enjoy when joining a particular network. If networks are fully incompatible, each Örm makes up
its own network, insofar as network beneÖts are product-speciÖc. At the other extreme, when
networks are fully compatible, each user generates network beneÖts for any user of any network
alike; it is as though a single network existed. In between, there are situations of imperfect
compatibility where network beneÖts are stronger among users of the same network than among
users of di§erent networks. As an illustration, think of using your smartphone for its primary
function, that is to phone someone. If you use the regular phone lines, you do not need to worry
about which operating system (OS) your correspondant has on her smartphone: there is full
compatibility. In contrast, to make a video call through Facetime, users must both have an
iPhone because this application, developed by Apple, is not available to users of devices running
on other OSs than iOS; in this case, there is full incompatibility. As for partial compatibility,
most ëVoice over IPí applications (e.g., Skype, Whatsapp or Viber) are available for all major
OSs; yet, the performance of some apps may vary across OSs, which may cause problems (such
as synchronization issues) when calls are made between di§erent devices.
3.3.1 A duopoly model
To analyze the importance of compatibility, we develop the following model of competition
between two producers of network goods. We follow the seminal work of Katz and Shapiro
(1985) and its extension by CrÈmer, Rey and Tirole (2000).8As for the demand side of the
market, we assume that there are two types of users. On the one hand, a mass 0of users have
adopted the network good of Örm 1in the past and are now locked in by the contract they
have previously signed. These consumers, whom will be referred to as Örm 1ís ëinstalled baseí,
make no decision in the game but confer an advantage to Örm 1over its competitor (we will
therefore call Örm 1the ëbig Örmí). On the other hand, there is a continuum of unattached users
(of mass one) who are identiÖed by a parameter #, which is drawn from a uniform distribution
on the unit interval. These unattached (or ënewí) users choose whether to adopt the network
7Social welfare is computed as W(n)=R1
1$n#n d# !nc =1
W(1) !W(n)= 1
2(1 !n)(1!2c+n(1 !n)) >0.
8Relatedly, Doganoglu and Wright (2006) analyze compatibility choice in a model with price-setting Örms,
where consumers can overcome incompatibility by multihoming ñ i.e., by buying the two incompatible versions.
As they show, multihoming makes compatibility less attractive for Örms, while it may make compatibility socially
good of Örm 1or of Örm 2. Letting #measure the stand-alone beneÖt of any network good,
3imeasure the network beneÖt from good ", and &idenote the price of good ", we have that
a new user of type #$!&%'" obtains a net surplus from adopting the good of Örm "equal to
4i##$%#(3i"&i. The network beneÖts generated by the two goods are deÖned respectively
as 3A%5#6A(0(76B$and 3B%5#6B(76A(70$. In these formulations, the parameter 5
measures the strength of the network e§ects, 6iis the number of new users joining network ",
and the parameter 7measures the degree of compatibility between the two goods. We assume
that 5)',*to guarantee the existence of a stable equilibrium; as for 7, it is bounded below by
zero (full incompatibility) and above by one (full compatibility).
Regarding the supply side of the market, we assume that each good is produced at zero
marginal cost. The two Örms compete for the new users by choosing the capacity of their
network (i.e., they compete ‡ la Cournot). To derive the demand function facing each Örm, we
identify the new user #0who is exactly indi§erent between the three options of joining network
1, joining network 2or not joining any: #0(3A"&A%#0(3B"&B%&, which is equivalent to
#0%)&, where )&%&A"3A%&B"3Bis the common ëquality-adjusted priceí of the two Örms. As
all new users with #%#0decide to join one or the other network (and given our assumption of
a uniform distribution of #), we have that the total number of users is given by 6A(6B%'")&.
Using the deÖnitions of )&,3Aand 3B, we can then write the inverse demand functions as
!&A%'(50 "#' "5$6A"#' "75$6B%
&B%'(750 "#' "5$6B"#' "75$6A8
We observe that the intercepts of both demands increase with the size of Örm 1ís installed base
but to a lower extent for Örm 2than for Örm 1if compatibility is not complete (7)'); we
also observe that incompatibility introduces a form of horizontal di§erentiation among the two
network goods as the price of good "is more sensitive to a change in the capacity of network "
(6i) than of network -(6j).
Firm "chooses its capacity 6ito maximize its proÖt -i%&i6i. Solving for the system of
Örst-order conditions, one Önds the following equilibrium quantities:
From the Örst-order conditions, we easily Önd that -#
i$2. As expected, the installed
base 0provides Örm 1with a competitive advantage as it allows it to reach a larger capacity
than its rival at equilibrium: we check that
We observe that the larger 0, the larger the advantage. This does not mean, however, that a
su¢ciently large installed base would necessarily drive Örm 2out of the market. If the degree
of compatibility is large enough, more precisely, if 7+',#* "5$, then Örm 2ís users beneÖt
su¢ciently from the network e§ects generated by Örm 1ís installed base to guarantee that
B+&.9This also explains why enhanced compatibility reduces the di§erence between the
equilibrium capacities of the two Örms:
#' "#* "7$5$205 ) &8
As compatibility increases, the perceived quality di§erential between the two network goods is
reduced, which reduces Örm 1ís initial advantage.
The latter Önding suggests that the big Örm may prefer more incompatibility. However,
compatibility has an upside for both Örms as it enhances the usersí willingness to pay. It is
indeed clear that users are better o§ when compatibility is improved. To see this, note that the
consumer surplus is equal to 1
+"#* ( 7$5, with 9#6#
97 %* ( #+ "5$0
#+ "#* ( 7$5$25+&8
3.3.2 Ex ante vs. ex post standardization
Collecting the previous results, it appears that the small Örm (Örm 2here) unambiguously
prefers more compatibility as it reduces its disadvantage with respect to the big Örm and it
expands demand. As for the big Örm (Örm 1here), more compatibility is a mixed blessing: it
expands demand but it attenuates the initial competitive advantage conferred by the installed
base. As the latter e§ect depends on the size of the installed base, we expect the big Örm to
prefer full compatibility to full incompatibility if the installed base is not too large. We check
Suppose now for simplicity that compatibility is all or nothing and that it has to be agreed
upon by the two Örms to be achieved (i.e., it is technically or legally impossible to achieve
compatibility on a unilateral basis). Then, the previous analysis teaches us that if the installed
base of the big Örm is not too large (i.e., if 0+.0), both Örms will prefer compatibility. For
instance, they will agree to adopt the same speciÖcations for their network goods so as to
make them fully interoperable. In that case, ex ante (or ëde jureí) standardization prevails.
Typically, this form of standardization follows from negotiations among Örms that take place
within Standard-Setting Organizations (SSOs).10
In contrast, if the big Örm starts the game with a signiÖcant installed base (i.e., if 0+.0),
then it will not agree with the small Örm to achieve compatibility. Both Örms will then push
their own speciÖcation and let users choose between two incompatible network goods. A so-
9For valu e s of #<1=(2 !)),ÖrmBstays on the market provided that +"(1 !(2 !#)))=(v(1 !(2 !))#)).
10There exist a very large number of SSOs, working at di§erent levels: international (e.g., the International
Tel e comm unic atio n U nion , I TU), r egio n al (e . g., t h e Euro p e an Com m i tte e f or Sta ndar d izat ion, C E N ), or s e cto r ial
(e.g., the World Wide Web Consortium, W3C).
called ëstandards warí will ensue, with both Örms Öghting to become the ex post (or ëde factoí)
standard (i.e., the speciÖcation that will eventually gain widespread acceptance).11
3.3.3 Strategies in standards wars
As we have just seen, the existence of a larger installed base for some Örm does not only
confer a competitive advantage to this Örm but may also lead it to prefer incompatibility. If
incompatibility is the chosen course of action, one understands then that Örms may have an
incentive to build an installed base today so as to increase their chances to win the standards
war tomorrow. They may thus want to set lower prices today (so as to attract more users) with
the hope of being able to set higher prices tomorrow (as users will be willing to pay more to join
a larger network).
To analyze this kind of issues, we need to move away from the static framework that we
have used so far and adopt instead a dynamic approach, where Örms and users make decisions
over consecutive periods. Cabral (2011) develops a model of dynamic competition between
Örms producing incompatible network goods. Users are assumed to live for potentially many
periods (i.e., they die and are replaced with a constant hazard rate). As above, users derive
stand-alone and network beneÖts from the network good that they choose. The stand-alone
component is received once the user joins a network and its value is supposed to be the userís
private information; the network component is received each period that a consumer is still alive.
A newborn user chooses one of the two existing networks and stays with it until death. This
decision is assumed to be made in a rational, forward looking way. This means that users are
able to anticipate all future decisions so as to estimate correctly the evolution of network sizes.
In each period, the two Örms compete for new consumers to join their network by setting the
price of their network good (prices below marginal costs are allowed).
In this framework, a Örm with a larger network faces an interesting trade-o§ when setting
its prices: taking a short-term perspective, the big Örm should set higher prices as it is more
attractive to users (ëharvestingí e§ect); however, when taking future payo§s into account, the
Örm has an incentive to set lower prices because the gains from increasing network size are larger
for a big than for a small Örm (ëinvestmentí e§ect). It is a priori not clear which of these e§ects
will dominate, meaning that price functions may be increasing or decreasing in time.
Although the pricing equilibrium is symmetric, market shares are generally asymmetric be-
cause of the stochastic appearance of new users. There are then two questions of interest: (i)
Does the large network attract a new user with higher probability than the small network? (ii)
Does the large network increase its size in expected value? The answer to the Örst question
is yes. As for the second question, the answer is yes as well, as long as network e§ects are
su¢ciently strong and the big Örm is still shy of holding 100% of the market.
The previous results suggest that getting a head start may prove valuable in a standards wars.
However, the model abstracts away a number of factors that may reverse this statement. First,
11We ll- known st anda rds wars i n the c o nsu mer el ect ron ics se cto r are th e war s of fo rmat s for v ide otap e r eco rdin g
(with VHS winning over Betamax) and for high deÖnition optical discs (with Blu-ray discs supplanting HD DVD).
both network goods are assumed to have the same, Öxed, intrinsic quality (which is measured
by the value that users attach to the stand-alone beneÖts). This is a simplifying assumption
as product qualities result from Örmsí investments in R&D and are likely to improve over time
(because of knowledge spillovers and/or usersí feedback). Firms may then face the following
trade-o§ when deciding to enter early on the market: they may secure a head start but they
may also have to compromise on the quality of their product; a later entrant with a better
product would then be able to overcome an initial disadvantage.
A second issue concerns compatibility across periods: by posing that users can collect network
beneÖts during their whole life, the model implicitly assumes that new versions of the network
goods are backward compatible with old ones. In reality, Örms may decide against backward
compatibility. One reason is that Örms may want to force old users to buy a new version of the
network good (because, if they stick to the old version, they will not enjoy network beneÖts from
users of the new, incompatible, version). This form of ëplanned obsolescenceí is commonplace in
consumer electronics markets.12 Another reason, which is hard to disentangle from the previous
one, is that backward compatibility often constraints Örms in their e§orts to improve their
products; to reach the full potential of technological advancement, they may thus decide to
introduce a new version that is incompatible with the previous one. Abandoning backward
compatibility clearly modiÖes the incentives to build an early installed base of users.
Finally, when users are not as fully rational as they are supposed to be in the previous
framework, there may exist less costly ways to get a head start than through low introductory
prices. If users form their expectations in a non rational way, Örms have the potential to ináuence
these expectations in their favor, thereby creating self-fulÖlling prophecies: if users are made to
believe that a particular Örm will win the standards war, then they will adopt the product of
that Örm, thereby helping it to win the war and conÖrming the initial beliefs.
In this section, we turn to platform markets, where intermediaries facilitate the interaction
between separate groups of users. To understand how cross-group external e§ects shape the
strategies of intermediaries, we focus Örst on the decisions of a single intermediary. We then
consider competition among intermediaries in the presence of positive cross-group external ef-
fects; we do so in two distinct settings: when markets tip (i.e., when all users aggregate on a
single platform at equilibrium) and when they do not. We examine next media markets, where
one group of users may exert negative cross-group e§ects on the other. Finally, we explore a
number of issues that were abstracted away in teh previous analyses, namely advanced pricing
strategies, within-group external e§ects, and investment issues.
12Choi (1994) examines this issue in a monopoly framework.
4.1 Monopoly pricing by a two-sided platform
Consider a platform that invites two groups of agents, e.g. sellers and buyers, to interact via
the platform. Suppose that the platform charges a membership fee :son the seller side and a
membership fee of :bon the buyer side. Thus, the platformís proÖts are -%!s#:s"/s$(
!b#:b"/b$, where !kis the number of participants on side ;$-<% =.and /kis the cost per
participant on the respective side. In a two-sided market, participation on at least one side on
the market depends on participation on the other side. Thus, in general, demand on one side
depends on prices on both sides of the market.
Participants on side "obtain a stand-alone utility of #k(which corresponds to the intrinsic
willingness to pay, when participation on the other side is assumed to be nil) plus some utility
that depends on the number of participants on the other side. It may also depend on the
participation level on the own side (we will address this possibility in Section 4.5.2). Let us
postulate a positive cross-group external e§ect: a participant on side <beneÖts from more
participation on side =and vice versa. For the sake of simplicity, let us, for the moment, assume
that this relationship is linear. Thus the gross utility of a participant on side <is #s(>!band
of a participant on side =is #b('!s, where >and 'are the marginal external e§ect enjoyed by
sellers and buyers, respectively.
Participants have heterogeneous outside options; for simplicity, we assume that there is a
unit mass on each interval of length 1 in the range !&%?", where ?is su¢ciently large such
that some participants on each side do not participate in the solutions we are going to consider.
Hence, all those participant on side <with an outside option of less than (s%#s(>!b":s
will pay the membership fee if they expect a participation level of !b, and all those participant
on side =with an outside option of less than (b%#b('!s":bwill pay the membership fee if
they expect a participation level of !s. Hence, since buyers and sellers are uniformly distributed
as speciÖed above, we have !s%(sand !b%(b. For given membership fees, participants play
an anonymous game and we solve for the Nash equilibrium of this game; the expected number
of participants on each side has to be equal to the actual number. Hence, we solve the system
of two linear equations in two variables, !sand !band obtain
'">' and !b%#b('<s":b"':s
We assume that >' ) '(i.e., cross-group external e§ects are not too strong), so that the numbers
of agents registering on the platform are decreasing functions of the membership fees.
We can now solve the platformís maximization problem. The Örst-order conditions with
respect to :sand :bcan be written respectively as
We observe that the presence of positive cross-group e§ects a§ects the platformís choice of fees
in two ways. First, positive cross-group e§ects generate a negative relationship between the
two fees; because participations on the two sides are complementary to one another, lowering
the fee on one side drives the platform to raise the fee on the other side.13 Second, the cross-
group e§ects make the optimal fee on one side depend also on features (cost, willingness to
pay) pertaining to the other side. This is another consequence of the complementarity between
the two sides: the opportunity cost of attracting, say, an additional buyer is lower than the
marginal cost (/b) because this additional buyer will entice extra participation on the seller side
and hence, extra revenues and costs for the platform on that side (which depend on #sand /s).
In general, because of cross-group e§ects, revenues and costs cannot be easily allocated to one
side or the other.
We can now proceed by solving the previous system to obtain the optimal fees.14 To clarify
the intuition, we introduce the following notation: let @k,#k"/kdenote the di§erence between
the intrinsic willingness to pay (#k) and the marginal cost (/k) on side ;(;%<% =). We can then
We have expressed the optimal margins (i.e., the optimal fee minus the marginal cost) as the
sum of two terms. The Örst term is the margin that platforms would set absent cross-group
e§ects (i.e., for >%'%&); in that case, the margin on each side would only depend on costs
and willingness to pay on that side (it is as though the ëplatformí was selling two independent
services to two separate groups of customers). The second term depends on the intensity of
the cross-group e§ects and on parameters pertaining to both sides. Interestingly, in this linear
model, this second term vanishes in the special case where the marginal cross-group external
e§ects are equal across sides (>%'). In contrast, if >+', i.e., if sellers value more the
interaction with buyers than buyers value the interaction with sellers, we see that the platform
chooses to have a larger margin on the seller side and a lower margin on the buyer side (with
respect to what would prevail in the absence of cross-group e§ects). The intuition is simple: as
the two sides are complementary, the platform can attract more agents on one side by lowering
its fee on the other side; when, for instance, >+', this ëleverage strategyí is more e§ective
when applied on the buyer side; it is thus proÖtable for the platform to lower the buyer fee and
to capture the extra value created on the seller side by increasing the seller fee. This reasoning
may even lead the platform to set a negative margin on the buyer side: it can be checked that
for >+'and @b,@s)#>"'$,#* ">#>('$, we have :#
b. The exact reverse logic applies
in the case where '+>.
Finally, we compute the platformís proÖt at the optimum as
13In that regard, two-sided platforms bear some resemblance with multiproduct Örms. Yet, as Rochet and
Tirole (2003) point out, end users internalize the corresponding externalities in a multiproduct setting but not in
14To s atis fy the s e cond - orde r condi tion s f or pr o Öt maxi m iza t ion, w e need t o impose a m o r e str i ngent con d itio n
We see clearly in the latter expression that the platformís proÖt increases with the intensity of
the cross-group external e§ects (i.e., 'and >).
More generally, if the distribution of agentsí types is not linear, we obtain increasing and
monotone relationships !s%As#(s$and !b%Ab#(b$, respectively. As illustrated above for the
linear case, the platform maximizes As#(s$#:s"/s$(Ab#(b$#:b"/b$with respect to :sand
:b, where the participation levels As#(s$and Ab#(b$depend on membership fees.
4.2 Pricing under platform competition when markets tip
We now examine the competition between two platforms, which we note by 'and *.We
continue to consider two groups of agents, which we continue to call, for convenience, sellers
(group <) and buyers (group =). We focus here on situations where only one platform survives at
equilibrium; it is then said that the market ëtipsí in the sense that all agents end up interacting
on a single platform. Ingredients for such a result are positive and strong cross-group e§ects,
closely substitutable platforms, and singlehoming.
In the model we develop, adapted from Caillaud and Jullien (2003), the platforms o§er
exactly the same services and are thus seen, other things being equal, as perfect substitutes by
the agents. Both groups, sellers and buyers, are supposed to consist of a continuum of mass one.
There exist positive cross-group external e§ects between the two groups insofar as each agent
uses the matching services of one or the other platform to Önd the unique trading partner whom
she has in the other group. Hence, the probability to Önd oneís partner on a particular platform
increases with the number of agents of the other group that register with this platform. In
particular, if !i
ssellers (resp. !i
bbuyers) register with platform "("%'%*), then the probability
for a buyer (resp. a seller) to Önd her match on platform "is equal to B!i
Bis the probability that two matching partners Önd each other when they register with the
same platform. The gross gain from a successful match is equal to ',*for each agent (gains of
trade are normalized to one and are supposed to be equally shared among trading partners after
some e¢cient bargaining process). The net gain is then ',*%'"&i&, where &iis the transaction
fee that platform "charges.15 Given that platforms also set membership fees :i
k, the expected
utilities for a seller and for a buyer when registering with platform "along with !i
bbuyers are respectively equal to 4i
We analyze the following game. In the Örst stage, platforms set their price structure to
maximize their proÖt; that is, platform "chooses the triple %:i
i&to maximize -i%
b&i, where /k(;%<% =) is the constant cost a platform incurs
when providing services to one agent of type ;(with /s(/b)Bso that total gains from trade are
larger than total costs). In the second stage, agents choose which platform (if any) to register
with; we assume singlehoming on both sides (agents register with at most one platform) and we
normalize the outside option to zero.
15Because we assume constant gains from trade and e¢cient bargaining, transaction fees are non-distortionary.
As a result, it is only the total fee that matters (i.e., pi)andnotthewayitissplitbetweenthetradingpartners.
In this Bertrand competition, platforms resort to a ëdivide-and-conquerí strategy, which
consists in Örst in ëdividingí by subsidizing one group of agents to convince them to join and
next, in ëconqueringí the agents of the other group, who have no better option than to join the
platform as well. To be sure of attracting the group of, say, buyers, platform "must o§er them
a better deal than platform -, even in the worst-case scenario where buyers hold the pessimistic
belief that they will Önd no seller on platform "; that is, it must be that ":i
If this condition is met, buyers will join platform "and sellers will follow suit (whatever their
beliefs), thereby generating maximal aggregate surplus B"/s"/b, which platform "can capture
by setting the transaction fee at its maximal level (&i%'). Platform "Önds this strategy
proÖtable as long as B"/s"/b(:i
b. Yet, as platform -can act in exactly the same way,
competition through divide-an-conquer strategies will drive proÖts to zero and allow only one
platform to remain active. At equilibrium, the remaining platform subsidizes full participation,
charges the maximal transaction fee (&i%') and makes zero proÖt (:i
the presence of positive cross-side external e§ects makes it e¢cient to have all agents register
with the same platform, the equilibrium is socially desirable.
If one platform (the incumbent) could play before the other (the entrant), the same equi-
librium would prevail, with the incumbent deterring the entry but foregoing all proÖt. The
surviving platforms may, however, achieve positive proÖts at equilibrium when transaction fees
are not feasible (e.g., because agents, when matched, could bypass the platform to trade).16
4.3 Pricing under platform competition when markets do not tip
We now introduce some di§erentiation between the platforms, so as to add a ëdispersioní force
that can counterbalance the ëagglomerationí force exerted by the combination of the positive
cross-group e§ects. We also explicitly consider the possibility for the members of one group to
be active on two platforms at the same time (so-called multihoming). To this end, we follow the
approach proposed by Armstrong (2006) and Armstrong and Wright (2007).
4.3.1 Two-sided singlehoming
The basic ingredients are the same as in the model with two competing matching intermediaries:
a unit mass of sellers and a unit mass of buyers having to choose at most one platform on which
they will interact, with this interaction generating positive cross-group external e§ects. In
contrast with the previous model, we assume now that platforms only compete in membership
fees and that agents perceive them as horizontally di§erentiated. Horizontal di§erentiation is
modeled in the Hotelling fashion: platforms are located at the extreme points of the unit interval;
sellers and buyers are uniformly distributed on this unit interval and incur an opportunity
cost of visiting a platform that increases linearly in distance at rates Csand Cb, respectively;
participation is su¢ciently attractive to drive all buyers and sellers to be active on one platform
16See Caillaud and Jullien (2001, 2003).
(it follows that on each side, the total number of agents on the two platforms adds up to 1:
The nature of the interaction on the platform is the following: sellers o§er perfectly di§er-
entiated products and buyers purchase one unit from each seller active on the platform; each
seller makes a proÖt per buyer of >and each buyer derives utility 'per seller.17 Hence, seller
and buyer surpluses gross of any opportunity cost of visiting platform "are
sare the membership fees set by platform ", and #band #sare the stand-alone
beneÖts. The seller and the buyer who are indi§erent between the two platforms are respectively
located at Dsand Dbsuch that (1
s"Cs#' "Ds$and (1
It follows that !1
b%Db, and !2
b%'"Db. Combining these equations
together with the expressions of (i
b, we obtain the following expressions for the numbers
of buyers and sellers at the two platforms:
Solving the previous system, we derive the number of buyers and sellers as a function of the
where we assume that CbCs+ '>, i.e., that the transportation cost parameters Cband Cs(which
measure the perceived horizontal di§erentiation between the two platforms) are su¢ciently large
with respect to the gains from trade 'and >(which measure the cross-group external e§ects).
Under this assumption, the number of members of one group at one platform decreases not only
with the membership fee that they have to pay but also with the membership fee that the other
group has to pay on this platform.18
Platform "chooses :i
bto maximize -i%%:i
we assume as before that platforms incur costs /sand /bwhen they register additional sellers
and buyers. At the symmetric equilibrium (:1
b,:b), the Örst-order
conditions can be written as19
17It is assumed, quite realistically, that in a sellerñbuyer relationship, prices or terms of transaction are inde-
pendent of the membership fee that applies to buyers and sellers.
18Fo r stro n ger cro ss-g r oup ext e rnal e § e cts an d /or we ake r h o riz o nta l d i§er e nti a tion ( i .e., f o r u0 > 1 b1s), the
number of agents on one platform would b e an increasing function of their membership fee. The market would
then naturally tip as in the model of the previous section.
19The second-order conditions require 41s1b>(u+0)2,whichisamorerestrictiveconditionthan1b1s> u0.
The equilibrium membership fee for the sellers is equal to marginal costs plus the product-
di§erentiation term as in the standard Hotelling model, adjusted downward by the term u
:b"/b$. To understand this term, note from expression (1) that each additional seller attracts
',Cbadditional buyers. These additional buyers allow the intermediary to extract >per seller
without a§ecting the sellersí surplus. In addition, each of the additional ',C bbuyers generates
a margin of :b"/bto the platform. Thus u
/b#>(:b"/b$represents the value of an additional
buyer to the platform. The same holds on the buyersí side.
Solving the system of Örst-order conditions gives explicit expressions for equilibrium mem-
bership fees and platformsí proÖts:20
We observe that the equilibrium membership fee for one group is equal to the usual Hotelling
formulation (marginal cost plus transportation cost) adjusted downward by the cross-group
external e§ect that this group exerts on the other group. As for the platformsí equilibrium
proÖts, they are increasing in the degree of product di§erentiation on both sides of the market
(as in the Hotelling model) and decreasing in the buyersí and sellersí surplus for each transaction,
i.e., the magnitude of the cross-group external e§ects. The intuition for the latter result is the
following: as cross-group external e§ects increase, platforms compete more Öercely to attract
additional agents on each side as they become more valuable.
4.3.2 Multihoming on one side (competitive bottlenecks)
Suppose now that sellers have the possibility to multihome (i.e., to be active on both platforms
at the same time), while buyers continue to singlehome. Sellers are then divided into three
subintervals on the unit line: those sellers located ìon the leftî register with platform 1 only,
those located ìaround the middleî register with both platforms, and those located ìon the
rightî register with platform 2 only. At the boundaries between these intervals, we Önd the
sellers who are indi§erent between visiting platform 1 (resp. 2) and not visiting any platform;
their locations are found as, respectively, D10 such that #s(!1
s%CsD10, and D20 such that
s%Cs#' "D20$. We assume for now that &)D
10 )'(we provide necessary
and su¢cient conditions below), so that !1
s%D10 and !2
s%'"D20, with the multihoming sellers
being located between D20 and D10. As far as buyers are concerned, we have the same situation
as before. The number of buyers and sellers visiting each platform are thus respectively given
20We a ssu me tha t rb+u=2+0>c
s+1sto guarantee full participation on the two
Solving this system of four equations in four unknowns, we get
The maximization problems of the two platforms are the same as above. Platform 'ís best
responses are implicitly deÖned by the Örst-order conditions, which can be expressed as
b">' #>(/s(*#s$('C s/b(#>(*/s(*#s$CbCs
Second-order conditions require that /CbCs+>
2('2(0>'. This condition is also su¢cient
to have a unique and stable interior equilibrium. Solving the previous system of equations, we
Önd the equilibrium membership fees:
On the seller side, platforms have monopoly power. If the platform focused only on sellers, it
would charge a monopoly price equal to ##s(/s$,*(>,,(assuming that each seller would have
access to half of the buyers and, therefore, would have a gross willingness to pay equal to >,*).
We observe that this price is adjusted downward by ',,when the cross-group e§ect that sellers
exert on the buyer side is taken into account. Similarly, on the buyer side, platforms charge the
Hotelling price, /b(Cb, less a term that depends on the size of the cross-group e§ects and on
the parameters characterizing the seller side (#s,/s, and Cs).
It is useful to compare price changes in the competitive bottleneck model to those in the
two-sided singlehoming model. In equilibrium, we observe that the membership fee for sellers is
increasing in the strength of the cross-group e§ect (E:#
s,E> + &), whereas it is constant in the
two-sided singlehoming model. This is due to the monopoly pricing feature on the multihoming
side. Everything else equal, if sellers are multihoming, the platform operators directly appro-
priate part of the rent generated on the multihoming side by setting higher membership fees.
This is not the case in the singlehoming world, where the membership fee does not react to the
strength of the network e§ect on the same side since platforms compete for sellers (and buyers).
It follows that at equilibrium,
We still need to check under which conditions some (but not all) sellers multihome at equi-
librium. This is so provided that ',*)!
s)', which is equivalent to *Cs)>('(*##s"/s$)
,Cs. Under these conditions, the equilibrium net surplus of sellers and buyers (gross of trans-
portation cost and for one platform) are equal to:
Note that (#
sis the per platform sellerís surplus.21 We observe that (#
in the net gain of the other side and in the net gain of the own side. The intermediariesí
equilibrium proÖts are
4.3.3 Singlehoming vs. multihoming environments
How do sellers and buyers surpluses compare in the two environments?22 In the model in which
sellers multihome, platforms hold an exclusive access to their set of singlehoming buyers (the
ëbottleneckí), which makes buyers valuable to extract proÖts on the seller side. We expect thus
platforms to compete Öercely for buyers and, in return, to milk sellers. Hence, we may expect
lower prices on the buyer side and higher prices on the seller side when compared to the two-sided
singlehoming model. We call this the ëbottleneck e§ectí. However, this view can be challenged,
as under multihoming with !i#
s+',*, there are more sellers active on a platform than under
singlehoming, thus addings value to participation on the buyer side. We call this the ëexpansion
e§ectí. Moreover, multihoming sellers have access to all buyers (but pay twice the prices and
the transportation costs).
As we illustrate next, whether buyers and sellers prefer one or the other environment depends
on the parameter values. Using superscripts Fand Gto represent the competitive bottleneck
and two-sided singlehoming models, we can write the di§erences in surplus between the two
environments as follows. For buyers, we have (C
b&; for singlehoming
sellers, we have (C
s; as for multihoming sellers, we focus on the one located at
the middle of the Hotelling line for whom the surplus di§erence is equal to *(C
Developing the latter expression, we Önd that this seller is better o§ in the competitive bottleneck
environment than in the singlehoming environment if and only if Cs+'.
Comparing prices, we see that the bottleneck e§ect dominates if
Simple computations show that the two conditions are equivalent and boil down to H+,Cs"*',
with H,>('(* ##s"/s$. We recall from the analysis of the competitive bottleneck model that
21Sellers located between 1!ni!
smultihome and, therefore, earn a surplus of 2v!
sis the surplus earned by the sellers located between 0and 1!ni!
the sellers located between ni!
22This subsection closesly follows Belleáamme and Peitz (2016).
the following conditions are necessary for some (but not all) sellers to multihome: *Cs)H),Cs.
Note that ,Cs"*')*Cswhenever Cs)'.
We can thus distinguish between two cases. First, if Cs)', it is always true that sellers
pay higher and buyers lower fees in the model where sellers multihome than in the model
where they singlehome (the bottleneck e§ect dominates the expansion e§ect). It follows that
buyers are better o§ in the multihoming environment (as they beneÖt not only from larger seller
participation but also from lower fees), while sellers who singlehome in both environments are
worse o§ (as they pay higher fees for the same buyer participation); as for sellers who would
multihome in the competitive bottleneck environment, we have shown above that they prefer
the singlehoming environment when Cs)'. So, in this case, buyers and sellers have diverging
preferences regarding multihoming: sellers would prefer to be constrained to singlehome, while
buyers would prefer that sellers were allowed to multihome.
Second, if Cs+', we can Önd parameter values for which all agents are better o§ in the
competitive bottleneck environments. We already know that this is true for multihoming sellers
(whatever the value of H). It is also true for singlehoming sellers if *Cs)H ),Cs"*'(as
they pay lower fees in this case). Yet, in this region of parameters, buyers pay higher fees; the
beneÖt of interacting with more sellers must then be su¢ciently large to have (C
b; this is
so if H+*##*>('$Cs">'$,#>('$,H0, with H0),Cs"*'when Cs+'. In sum, if
0, then both groups prefer the situation where sellers are allowed to
We can conclude that without further information, we cannot decide whether allowing mul-
tihoming on one side (with the other side singlehoming) leads to higher or lower net surpluses on
either side. It is therefore a priori not possible to say whether the side that changes its behavior
from singlehoming to multihoming (or reverse) beneÖts or su§ers from this change of behavior.
4.4 Media markets
Media platforms are just one application of two-sided platforms where readers or viewers consti-
tute one group and advertisers the other. A feature of most media platforms is that cross-group
external e§ects go in opposite directions: while advertisers prefer a platform with more viewers,
everything else given, the reverse holds true for viewers, as they dislike a platform that carries
a lot of advertising. Thus, advertising is a nuisance. Many media platforms use a one-sided
revenue model: they charge advertisers for posting ads and give away content bundled with ad-
vertising for free to viewers. This applies to free-to-air television, radio broadcasting and many
media platforms on the internet.
According to the baseline media model developed by Anderson and Coate (2005), two media
platforms compete for viewers. While viewers are assumed to singlehome, advertisers can post
ads on multiple platforms. The key di§erence to the previous analysis is that media platforms
are purely advertising-Önanced. In particular, they are assumed to Öx the advertising space or,
equivalently, set the ad price per viewer.23
Here, we review the basics of the competitive bottleneck model with !media platforms.
Each media platform "provides program content to attract viewers and delivers these eyeballs
to advertisers.24 Advertising revenue is the sole source of Önance to platforms, and advertisers
are assumed to be price takers. Platform "ís proÖt is thus -i%IiJi,"%'% 888!, where Iiis the
price per ad and Jiis the number of ads posted.
Content is attractive to viewers, but viewers consider the embodied ads to be a nuisance.
Viewersí tastes over platform content is di§erentiated. Each viewer is assumed to singlehome,
i.e., she makes a discrete choice over which platform to visit. Denote by !i
of viewers (demand) for platform "as a function of its own ad level and the vector of ad levels,
a$i, of its competitors.
On the advertiser side, all ads on a platform are seen by the viewers. Advertisers have
di§erent willingness to pay for reaching viewers (impressions). Assume that the advertiserís
willingness-to-pay for advertising on each platform is a linear function of the number of viewers
on the platform. In other words, there are constant returns to reaching prospective customers.
This allows us to rank advertisers in terms of decreasing willingness to pay per eyeball, from large
to small, giving a downward-sloping function &##$. When platform "opens Jislots for advertising,
the price per ad per viewer is &%Ji&, so that the price per ad is Ii%&%Ji&!i
under these assumptions, we can write
where K%Ji&is the revenue per ad per viewer. The Örst-order condition (with ad levels as the
strategic variables) is written as
where E1is the partial dervative with respect to the Örst argument. This equations says that the
elasticity of revenue per viewer should equal the viewer demand elasticity. This expression relates
to the standard elasticity condition for oligopoly pricing.25 The left-hand side, K0#J$,K #J$,is
decreasing in Junder the condition that K#J$is log-concave. Hence, from (2), lower ad levels
result whenever the equilibrium value of "E1!i
b%Ji%a$i&increases in a change in
23This may approximate business practice. A standard practice in media markets is to report the CPM (cost
per thousand impressions) based on past experience. Advertisers are then compensated if actual participation
deviates from past participation.
24Our exposition follows Anderson, Foros, Kind, and Peitz (2012). For a general treatment see Anderson and
25Indeed, consider the (Bertrand) oligopoly problem of maxpi0i=(pi!ci)ni(pi;p$i)where now ni(pi;p$i)
is the demand addressed to Örm iand piis the price isets for its product, while ciis its marginal cost (and p$i
is the vector of other Örmsí prices). Then the Örst-order condition can be written as 1
in elasticity form, gives the inverse elasticity (Lerner) rule for pricing.
Consider the e§ects of platform entry at a symmetric equilibrium. For example, in the case
of the Salop circle model, the right-hand side takes the value !,C,where the transport parameter
Cmeasures how strongly platform content is di§erentiated. In the standard oligopoly context
with product di§erentiation, this means simply that increasing the number of competitors leads
to lower prices. In the present media economics context, this means that entry of a media
platform leads to lower equilibrium ad levels. Competition for viewers plays out as competition
in nuisance levels, and more competition leads to a lower nuisance level. For advertisers, the
lower level of ads implies a higher ad price per viewer.
While the prediction of entry (and, relatedly, mergers) are unambiguous in this competitive
bottleneck model, their empirical relevance may be questioned. Indeed, as Anderson, Foros,
Kind, and Peitz (2012) mention, at least two alternative mechanisms can overturn the result
that entry reduces ad levels and that a merger increases ad levels. A countervailing e§ect
arises when viewers spend some time on various media platforms and have limited attention for
advertising they are exposed to. This introduces strategic interaction among platforms on the
advertiser side. The attention of viewers becomes a common property resource that platforms
tend to overexploit. A larger platform has stronger incentives to internalize the associated
external e§ect on other ads and, therefore, opens fewer ad slots than a smaller platform. With
symmetric platforms, the entry of a platform tends to lead to higher ad levels, as the negative
e§ect of additional ads on existing advertisers is felt less strongly by each platform.26
An alternative explanation that entry can lead to higher ad levels (and a merger to lower ad
levels) rests purely on the e§ects of viewer multihoming on the exposure of viewers to ads. To
the extent that a single impression is all that an advertiser cares about, advertising on multiple
media carries the risk of wasting impressions because some viewers will be treated twice. This
has implications for the advertising strategy of media platforms (as well as the decision on
content of platforms).
Ambrus, Calvano, and Reisinger (2016) consider a setting with viewers who can multihome
and homogeneous advertisers who can post multiple ads.27 Start with a monopoly setting and
suppose that a second platform enters. There are two e§ects due to entry. First, there is a
duplication e§ect: as each multihoming consumer can now get informed about an advertiserís
product on both platforms, the single ad is worth less. Due to the duplication e§ect, the
advertising intensity tends to fall in duopoly. However, there is a countervailing e§ect, which
can be called the business-sharing e§ect. In monopoly, all consumers are exclusive consumers.
By contrast, in duopoly, some of the lost business due to increased advertising levels comes from
consumers active on both platforms. Here, the duopolist shares business with its rival. Losing
these consumers is less detrimental than losing exclusive consumers. Due to this business-
sharing e§ect, a duopolist tends to have a greater incentive to increase the advertising level than
a monopolist. The business-sharing e§ect possibly dominates the duplication e§ect, in which
26Fo r a form a l anal y sis of m e dia ma r ket s w ith li m ited v i ewe r a ttentio n f or ads , s ee And e rson an d Peit z (201 6 ) .
27Fo r a rela t ed contri b u tio n , s ee And e r son, Fo r os, and K i nd (20 1 5). Fo r a n ove rvie w o f the e§ e cts of v iewe r
multihoming in media markets, see Peitz and Reisinger (2015).
case advertising levels in duopoly are larger than in monopoly.
4.5 Further issues
The focus of our presentation in this section has been to determine features of the equilibrium
allocation of markets in which one or multiple platforms enable the interaction between two
groups of users restricting the analysis to stylized and simple settings. Here, we look at, in some
sense, richer settings: (i) richer price instruments by the platform, (ii) within-group external
e§ects and, in particular, competition among sellers, and (iii) investment incentives by sellers
4.5.1 Price instruments
In most of the analysis we postulated that platforms charge listing or access fees to the platform.
However, pricing strategies may be more involved. In particular, platforms may choose two-part
tari§s or o§er some insurance against unexpected drops in participation on the other side.
Two-part tari§s. Reisinger (2014) allows platforms to set two-part tari§s, adding a usage
(or per-transaction) fee to the membership (or subscription) fee that we considered so far. It is
indeed not rare that platforms charge two-part tari§s to at least one of the sides. For instance,
in software platforms, developers are charged a Öxed fee for getting access to source code of
the system and, in addition, pay royalties for the applications they sell to users. What are the
implications of this form of price discrimination on the proÖts of competing platforms and on
the welfare of the two sides?
The model is modiÖed as follows. The seller and buyer surpluses gross of any opportunity
cost of visiting a platform become
kis the transaction (usage) fee that platform "charges on side ;. As for platform "ís
proÖt, it is now expressed as
where 7is the constant per-transaction cost incurred by the platform. The game is the same as
before, except that platforms now have four strategic variables to choose.
A general result that emerges is that a continuum of equilibria exists when platforms compete
using two-part tari§s.28 The reason behind this multiplicity is that platforms only care about
the total price that agents pay but not about how it splits between the membership and the
transaction fees (i.e., di§erent combinations of the two fees yield the same proÖt); as a result,
platforms have a continuum of best responses to their rivalís tari§. For instance, in the two-sided
singlehoming game, the continuum of symmetric equilibria is characterized by platforms charging
28This was already pointed out by Armstrong (2006).
Mk%:k(Lk!k(;%<% =), where :s%/s(Cs"'(1
and &+:s+*>,&+:b+*'. The platformsí equilibrium proÖts are given by -%
As the previous expressions clearly show, these equilibria lead to di§erent proÖts for platforms
and di§erent surpluses for agents. An unwelcome consequence is that the models (with two-sided
singlehoming or with competitive bottlenecks) are deprived of any predictive power.
To solve this problem, Reisinger (2014) allows for heterogeneous trading behavior among
agents on both sides. In particular, there exist two types of agents on each side: the ënormalí
agents interact with all agents on the other side (as was assumed so far), while the ësmallí agents
only interact with a fraction of the agents on the other side (or interact with each of them only
with some probability); it is further assumed that the small agents are a minority and that
platforms are unable to price discriminate across types. In the competitive bottlenecks model
where sellers can multihome while buyers cannot, Reisinger shows that this formulation leads to
a unique equilibrium in the price game, even when the masses of small agents become inÖnitely
small (i.e., when heterogeneity disappears). Moreover, this equilibrium has many reasonable
properties; in particular, it is still the case that platforms price aggressively to the side that
exerts the larger cross-group external e§ect; the di§erence under two-part tari§s is that the
lower payment is only reáected in the transaction fee but not in the membership fee.29
The intuition for the unicity result is the following. The two types of agents react di§erently
to a particular combination of membership and transaction fees. To keep constant the utility of
a small agent (who trades less often), platforms have to balance an increase in the transaction
fee with a smaller reduction of the membership fee than they would do for a normal seller. It
follows that the e§ect on proÖt of a marginal change in platform "ís transaction fee is no longer
a constant multiple of the e§ect of a marginal change in "ís membership fee. This multiple
varies continuously as the fees change because the ratio of the two types that join platform "
also varies continuously. As a result, each platform has a unique optimal combination of the
fees as a reaction to the price quadruple of its rival.
Insulating tari§s. An alternative approach to platform pricing is to consider that platforms
set so-called ëinsulating tari§sí, i.e., tari§s such that after a deviation on one side, the pricing on
the other side is adjusted in order to insulate the demand e§ect on one side and leave demand
on the other side unchanged. This concept has been proposed and developed by Weyl (2010) in
a monopoly setting, and Weyl and White (2016) in a duopoly setting.
Suppose that platforms adapt their pricing to insure agents against any utility loss from low
participation on the other side of the market. Let us apply this alternative strategy space to
our previous model of two-sided singlehoming. Instead of setting the membership fees :i
b, platforms now set the net surpluses (i
b. This change of strategic variables removes
the feedback e§ect stemming from cross-group external e§ects. We should therefore expect a
weaker impact of cross-group e§ects on equilibrium prices. We check now that this conjecture
29Fo r a disc u ssio n o f the li n k b etw een he t erog e n eit y and pri ce ins t rumen ts, s e e Ro che t a nd Tir o le (20 0 6).
Recalling that (i
b, we can write membership fees
as a function of participation on the other side and utilities (i
b. If a platform sets an insulating tari§ (or o§ers a guarantee in utils) to all
agents, it Öxes (i
b, which means that agents are insured against any changes in platform
participation on the other side since the membership fee will be adjusted accordingly. We can
then write platform "ís proÖt as -i%##s(>!i
that the participation levels !i
bare obtained in the Hotelling fashion and can thus be
expressed as a function of the guarantees, we have
We can now characterize the Nash equilibrium of the game in which platforms simultaneously
choose guarantees. This allows us to determine the membership fees that platforms charge at
s%/s(Cs"',*, and :#
Recall that in the previous model in which platforms choose membership fees, equilibrium mem-
bership fees are equal to /s(Cs"'and /b(Cb">, which are lower than the ones obtained
here. This conÖrms our initial intuition that the impact of cross-side external e§ects on com-
petition is weaker when platforms o§er guarantees, as there are no feedback e§ects on platform
participation from one side to the other. The lesson to emerge from the comparison of these
two models is that platforms that can compensate for utility losses due to lower than promised
participation on the other side will compete less Öercely than platforms that only choose prices.
4.5.2 Within-group external e§ects
The analyses presented thus far considered exclusively cross-group external e§ects. Such focus
seems natural as cross-group e§ects directly stem from the desire of the two groups of agents
to interact and, thereby, give their raison díÍtre to two-sided platforms. However, there exist
a number of situations where platforms also have to factor in the existence of within-group
external e§ects when choosing their strategies. As explained in the introduction, these e§ects
describe the fact that the attractiveness of a platform for the members of one group depends
on the participation of the members of the very same group. Within-group e§ects are negative
when the members of one group compete with one another to interact with the other group (e.g.,
Uber drivers face a given set of passengers at any location and any point in time) or because of
30See Belleáamme and Peitz (2015, pp. 671-72) for the details. Note that in contrast with the model where
platforms set membership fees, an equilibrium with two active platforms exists here without having to impose
that platforms be su¢ciently di§erentiated.
some form of congestion (e.g., talking about Airbnb visitors, Slee (2016) reports that ìas their
numbers grow, they erode the very atmosphere in which they bask and threaten the livability
of the city for residents.î). In contrast, there also exist sources of positive within-group e§ects;
for instance, Belleáamme, Omrani and Peitz (2015) explain that a larger ëcrowdí of funders
on a crowdfunding platform increases the probability that any project will be realized, which
beneÖts all funders. In what follows, we focus on markets with competing platforms.31 We Örst
generalize the two-sided singlehoming model of Armstrong (2006) to allow for any type of cross-
and within-group external e§ects. We then examine the ináuence of within-group e§ects on the
coexistence of platforms.
A two-sided singlehoming model encompassing within-group e§ects. Following Belle-
áamme and Toulemonde (2016b), we deÖne the seller and buyer net surplus of visiting plat-
form "(gross of any opportunity cost) respectively as (i
b. The di§erence with the previous model is that we use now the general
s&to represent the net gains from trade for any seller and any
buyer on platform ". They both potentially depend on the number of buyers and on the number
of sellers who are present on the platform, meaning that any form of cross-group and of within-
group external e§ects are permitted. Both functions are supposed to be twice continuously
di§erentiable in their two arguments. We proceed as before by identifying the indi§erent seller
and buyer in the standard Hotelling fashion, which allows us to express the numbers of sellers
and buyers at platform "as:
We also introduce the following notation:
In words, the function 1u%!i
s&measures the di§erential in buyersí net gains from trade
between platforms "and -when there are !i
bbuyers and !i
ssellers on platform ". The derivatives
31Fo r an ana l ysis w i th wit h in-g r oup ext e rnal e § ects o n m onopoly p latf o rms, se e N o cke , Pei t z, and S t a hl (2 0 0 7).
They revisit the question of socially excessive or insu¢cient entry of monopolistically competitive sellers when the
monopoly platform can charge only sellers for their listing service, and compare their Öndings to what happens
with alternative governance structures of the platform.
smeasures the sensitivity of this di§erential to a change in the mass of, respectively,
buyers or sellers on platform "; the function 1-%!i
s&and derivatives 1-
accordingly for sellers.
The system of equations (3) implicitly determines the demand functions for platform ",
s$, which depend on the combination of the four fees.32
Using implicit di§erentiation and taking advantage of the fact that !1
at the symmetric equilibrium, it is then possible to show that the platforms set the following
membership fees at the symmetric equilibrium of the game:
s, and 1-
sare evaluated at %!i
2$. We observe that the equilibrium
membership fees depend on the nature of the within- and cross-group external e§ects. In the
complete absence of external e§ects within and across groups, fees would be as in the Hotelling
model. The presence of positive (resp. negative) external e§ects from, say, sellers to buyers leads
platforms to lower (resp. raise) the membership fee for sellers below (resp. above) the level that
would prevail absent any external e§ect. This is the standard result of Armstrong (2006). We
add here a new result related to the presence of external e§ects within groups. Positive (resp.
negative) external e§ects within groups leads platforms to lower (resp. raise) the membership
fee for the group below (resp. above) the level that would prevail absent any external e§ect.33
Coexistence of platforms. An interesting issue is the impact that within-side external e§ects
may have on the coexistence of competing two-sided platforms. We have seen above that positive
cross-group e§ects generate positive feedback loops that may lead to situations where only one
platform survives at equilibrium (ëwinner-takes-allí) unless competing platforms are su¢ciently
di§erentiated. We may conjecture, however, that negative within-side e§ects may contribute to
break the feedback loop and, thereby, facilitate the coexistence of competing platforms, even in
the absence of di§erentiation.
Karle, Peitz and Reisinger (2015) address this issue by examining how the degree of compe-
tition among sellers a§ect the possibility for non-di§erentiated platforms o§ering listing services
to coexist at equilibrium.34 Recall that Belleáamme and Toulemonde (2016b) deÖned the seller
32It is assumed that the functions uand 0are such that the system (3) leads to a unique solution "ni
iare decreasing functions of (mi
33To b e s u re, w e recove r t he pre v iou s r esul t s by s e ttin g 0"ni
e§ects are positive and linear and within-group e§ects are nil). We have then ,*=0"2ni
34Ellison and Fudenberg (2003) and Ellison, Fudenberg and Mˆbius (2004) also show that negative within-side
e§ects may contribute to the coexistence of two-sided platforms. Hagiu (2009) introduces competition among
sellers on a two-sided platform; competition stems from consumersí variable preference for variety over sellersí
products, which turns out to be a key factor determining the optimal pricing structure either of a monopoly
platform or of competing platforms.
and buyer net surplus of visiting platform "respectively as (i
b. A special case is that the proÖt per buyer depends on the number
competing sellers on the platform and platforms only charge sellers, (i
s'. Using these surplus functions, Karle, Peitz and Reisinger (2015) analyze the
two-sided singlehoming model in which buyers observe product o§ering on a platform only after
having visited the platform. If all sellers co-locate on the same platform then, in equilibrium, all
buyers will be active on this platform. Thus there is agglomeration in equilibrium and network
e§ects are fully exploited. In this equilibrium all proÖts are competed away and both platforms
make zero proÖts.
However, imperfect competition between sellers may lead to a di§erent equilibrium in which
both platforms have a positive number of users and make positive proÖts in equilibrium. Suppose
that there are two sellers that have to decide whether to join platform 1, join platform 2 or not
to participate at all. If they both join the same platform, they obtain duopoly proÖt >dper
buyer, which is less than monopoly proÖt >mper buyer they would obtain if they were the only
seller on the platform. If >mis su¢ciently large, there is an equilibrium in which sellers list on
di§erent platforms. Buyers are indi§erent between the two platforms. ProÖts are not competed
away; platforms can extract the full seller surplus in equilibrium.
Belleáamme and Toulemonde (2009) propose another way to address this issue. They ex-
amine the extent to which negative within-group e§ects among sellers may help a new platform
operator lure buyers and sellers away from an existing platform. In their model, only the new
platform can set membership fees; there is no price competition per se, but this is not a monopoly
setting either as the existing platform provides buyers and sellers with an outside option. As in
Caillaud and Jullien (2003), the new platform faces a ëchicken-and-eggí problem, which it tries
to solve by using a divide-and-conquer pricing strategy; that is, the platform must subsidize the
participation of one side (divide) and hope to recoup the loss through the membership fee it sets
on the other side (conquer). The question is whether the platform can make any proÖt with
such strategy. The answer is yes when the interaction among buyers and sellers only generates
(positive) cross-group external e§ects. However, the presence of negative within-group e§ects
among sellers (e.g., because they o§er substitutable products) blurs the picture. Competition
among sellers turns out to be a mixed blessing for the new platform. The upside is that the
sellersí willingness to pay to join the new platform increases if only a few of them make the move;
as a consequence, sellers are less sensitive to buyersí participation to the new platform, which
alleviates the ëchicken-and-eggí problem. Yet, the downside is that it will be more costly for the
new platform to attract buyers if only a small subset of the sellers join. The balance between
the two e§ects depends on the relative strength of the within-group e§ects (with respect to the
cross-group e§ects). There may be situations where entry is not proÖtable.
4.5.3 Investment issues
So far, the competition among platforms was studied in Öxed environments. All the models
we considered implicitly assumed that neither platforms nor their users are able to modify the
conditions under which the interaction among the two groups is taking place. We now relax this
assumption in two di§erent ways: Örst, we let sellers make ex-ante investments that a§ect the
gains from trade when they interact with buyers; second, we let platforms invest in reducing
their costs of registering agents.
Seller investment incentives on a platform. Belleáamme and Peitz (2010) analyze how
seller investment incentives are a§ected by the presence of competing for-proÖt platforms. Plat-
form competition is modeled as in Armstrong (2006) and Armstrong and Wright (2007); that is,
the Önal stages of the game are the ones we analyzed above with either both sides singlehoming,
or one or the other side being allowed to multihome. The novelty is to add an initial stage
where sellers have the possibility to make long-term investments, which may take the form of
cost reduction, quality improvement or marketing measures that facilitate price discrimination
or expand demand. These investments a§ect the surpluses, >and ', that sellers and buyers
obtain when they trade on the platforms. Typically, investments in cost reduction or quality
improvement increase both >and ', while investments in better price discrimination increase >
while decreasing '.
To assess the impact that for-proÖt platforms have on investment incentives, two trading
environments are contrasted. The Örst is the one we are considering in this section and is called
ëintermediated tradeí: trade takes place through for-proÖt intermediaries, which set membership
fees on both sides of the market. The other environment is called ënon-intermediated tradeí,
as trade is assumed to take place via open trading platforms, which can be accessed without
One could think that seller investment incentives would be weaker in the intermediated
trade environment, simply because the for-proÖt intermediaries capture part of the rents that
are available in the market. However, this reasoning is short-sighted because investments also
a§ect the membership fees that platforms set at equilibrium. Why? Because seller investments
modify the intensity of the cross-group external e§ects and, thereby, the competition between the
platforms. In particular, investments that increase buyer surplus lead platforms to lower their
fee on the seller side. As a consequence, sellers internalize changes in buyer surplus if products are
traded on for-proÖt platforms, whereas they do not in the context of open platforms. It follows
that investment incentives can be stronger with competing for-proÖt platforms than with open
platforms. The exact relationship between investment incentives and for-proÖt intermediation
depends on which side of the market singlehomes and on the nature of the investment e§ort.
In general, it can be said that as the intensity of competition for sellers increases, proprietary
platforms are more likely to provide better seller investment incentives than open platforms.
Indeed, this happens when the nature of platform competition moves from multihoming sellers
and singlehoming buyers to singlehoming sellers and buyers, and then to singlehoming sellers
and multihoming buyers.
Platform investment incentives. To study the investment incentives of competing plat-
forms, we need Örst to solve a game of price competition among asymmetric platforms. Belle-
áamme and Toulemonde (2016a) do so for the two-sided singlehoming model of Armstrong
(2006). We present here a simpliÖed version of their work by letting only marginal costs di§er
across platforms (all the other parameters, in particular, the cross-group external e§ects, remain
common to the two platforms). Let /i
bdenote the marginal cost for platform "of regis-
tering, respectively, an extra seller and an extra buyer; let also 7s,/i
("% - %'%*;"'%-). Repeating all the steps described in Section 4.3.1, they Önd the following
equilibrium membership fees for the two platforms ("% - %'%*;"'%-):
where N,2CsCb"#*>('$#>(*'$is positive.35
The equilibrium membership fees can be decomposed as the sum of four components: (i)
the Örst two terms (/i
k(Ck) are the classic Hotelling formula (marginal cost + transportation
cost); (ii) the third term was identiÖed by Armstrong (2006) as the price adjustment due to
cross-group external e§ects (the fee is decreased by the externality exerted on the other side);
(iii) the fourth term is the e§ect of cost di§erences across platforms; (iv) the last term results
from the interplay between cost di§erences and cross-group external e§ects. If platforms are
symmetric (7k%&), we Önd the same formulas as in Section 4.3.1. In the particular case where
cross-side external e§ects are the same on the two sides (>%', meaning here that the gains
from trade are equally split among buyer and seller), all terms but the last remain. The latter
result is reminiscent of what we already observed in the setting with a monopoly platform.
The equilibrium proÖt of platform "is computed as
with -j#being obtained by replacing 7sby "7s, and 7bby "7b.
Belleáamme and Toulemonde (2016a) then use this equilibrium proÖt function to estimate
platform "ís incentives to invest in cost reduction. Their main contribution is to show that
cross-group external e§ects a§ect incentives to invest in cost reduction through the strategic
e§ect of this investment. The strategic e§ect is the e§ect on one platformís proÖt that operates
through the modiÖcation of the other platformís equilibrium fees. Absent cross-group external
e§ects, we expect the strategic e§ect of a lower cost to be negative if Örms compete in prices
over substitutable services: a lower cost for Örm 1leads this Örm to decrease its price, which
leads the rival Örm to decrease its price as well (because of strategic complementarity); this, in
35This follows from the second-order condition 41s1b>(0+u)2.
turn, reduces Örm 1ís proÖt, which contributes to attenuate the direct positive impact of cost
reduction on proÖts.
The presence of cross-group external e§ects challenges the previous results in two major
ways. First, cross-group external e§ects may decrease the strategic e§ect and they may do so
to such an extent that the strategic e§ect outweighs the positive direct e§ect; it follows that
the net e§ect of lower costs on proÖt becomes negative. In that case, platforms would be better
o§ if they could increase, rather than decrease, their costs. Second, in complete contrast with
the previous case, external e§ects may increase the strategic e§ect, even up to a point where
it becomes positive; in the latter case, platforms would have a twofold incentive to invest in
cost reduction as it would beneÖt them Örst directly and next, indirectly, through the upward
adjustment of the rival platformís equilibrium prices. It is shown that for either of these extreme
cases to arise, cross-group external e§ects must be large relatively to the intensity of competition
on the two sides.
In intermediate microeconomics, students learn that market power, asymmetric information,
and externalities are sources of market failure. Classic oligopoly theory has focused on market
power; modern industrial organization has, in addition, incorporated asymmetric information
into the analysis of markets. While markets with network e§ects have also been investigated
by industrial organization economists at least since the beginning 1980s, more recent e§orts
investigate decision making by intermediaries in markets characterized by imperfect competition
and external e§ects. A particular focus lies on market environments in which a platform caters
to multiple audiences, which are distinct, but connected.
A number of theoretical insights have been derived, some of which are reviewed in this
chapter. This chapter has been silent regarding the empirical literature on network e§ects and
platforms. An early guide to empirical work on two-sided platforms is Rysman (2009). Quite
some work has been done in the context of media markets; we refer the reader to the overviews
by Chandra and Kaiser (2015), Crawford (2015), and Sweeting (2015).
This chapter has ignored recent work on competition policy issues regarding two-sided plat-
forms. A case in point is the analysis of price-partity clauses imposed by platforms (see, in
particular, Edelman and Wright, 2015). Also, there is a well-developed literature on payment
systems in which the two-sidedness of the payment system is a critical feature. Finally, the two-
sidedness of the business model features prominently in the economic analysis of net neutrality
(see Greenstein, Peitz, and Valletti, 2016). We expect the analysis of two-sided platforms to be
one of the most active areas of research in industrial organization theory in the years to come.
 Ambrus, A., E. Calvano and M. Reisinger (2016). Either or Both Competition: A "Two-
Sided" Theory of Advertising with Overlapping Viewerships. American Economic Journal:
Microeconomics 8, 189ñ222.
 Anderson, S. and S. Coate (2005). Market Provision of Broadcasting: A Welfare Analysis.
Review of Economic Studies 72, 947ñ972.
 Anderson, S., O. Foros and H.-J. Kind (2015). Competition for advertisers and for viewers
in media markets. CEPR Discussion Paper 10608.
 Anderson, S., O. Foros, H.-J. Kind, and M. Peitz (2012). Media Market Concentration,
Advertising Levels, and Ad Prices. International Journal of Industrial Organization 30,
 Anderson, S. and M. Peitz (2015). Media See-saws: Winner and Losers on Media Platforms,
University of Mannheim Working Paper 15-16.
 Anderson, S. and M. Peitz (2016). Advertising Congestion in Media Markets. Unpublished
 Armstrong, M. (1998). Network Interconnection in Telecommunications. Economic Journal
 Armstrong, M. (2006). Competition in Two-sided Markets. Rand Journal of Economics 37,
 Armstrong, M. and J. Wright (2007). Two-sided Markets, Competitive Bottlenecks and
Exclusive Contracts. Economic Theory 32, 353ñ380.
 Belleáamme, P., N. Omrani and M. Peitz (2015). The Economics of Crowdfunding Plat-
forms. Information Economics and Policy 33, 11ñ28.
 Belleáamme, P. and M. Peitz (2010). Platform Competition and Seller Investment Incen-
tives. European Economic Review 54, 1059ñ1076.
 Belleáamme, P. and M. Peitz (2015). Industrial Organization: Markets and Strategies. 2nd
edition. Cambridge: Cambridge University Press.
 Belleáamme, P. and M. Peitz (2016). Platform Competition: Who BeneÖts from Multihom-
ing? Unpublished manuscript, University of Mannheim.
 Belleáamme, P. and E. Toulemonde (2009). Negative Intra-Group Externalities in Two-
Sided Markets. International Economic Review 50, 245ñ272.
 Belleáamme, P. and E. Toulemonde (2016a). Tax Incidence on Competing Two-Sided Plat-
forms: Lucky Break or Double Jeopardy. CORE Discussion Paper 2016/12.
 Belleáamme, P. and E. Toulemonde (2016b). Who BeneÖts from Increased Competition
among Sellers on B2C Platforms? Forthcoming in Research in Economics.
 Cabral, L. (2011). Dynamic Price Competition with Network E§ects. Review of Economic
Studies 78, 83ñ111.
 Caillaud, B. and B. Jullien (2001). Competing cybermediaries. European Economic Review
(Papers and Proceedings) 45, 797ñ808.
 Caillaud, B. and B. Jullien (2003). Chicken and Egg: Competition among Intermediation
Service Providers. Rand Journal of Economics 34, 521ñ552.
 Chandra, A. and U. Kaiser (2015). Newspapers and Magazines, in: Anderson, S.P.,
Stromberg, D., Waldfogel, J. (eds.), Handbook of Media Economics, vol 1A, Amsterdam:
Elsevier, pages 397ñ444.
 Choi, J.P. (1994). Network Externality, Compatibility Choice, and Planned Obsolescence.
Journal of Industrial Economics 42, 167ñ182.
 Crawford, G. (2015). The Economics of Television and Online Video Markets, in: Ander-
son, S.P., Stromberg, D., Waldfogel, J. (eds.), Handbook of Media Economics, vol 1A,
Amsterdam: Elsevier, pages 267ñ340.
 CrÈmer, J., P. Rey and J. Tirole (2000). Connectivity in the Commercial Internet. Journal
of Industrial Economics 48, 433ñ472.
 de Bijl, P. and M. Peitz (2002). Regulation and Entry into Telecommunications Markets.
Cambridge: Cambridge University Press.
 Doganoglu, T. and J. Wright (2015). Multihoming and Compatibility, International Journal
of Industrial Organization 24, 45ñ67
 Edelman, B. and J. Wright (2015). Price Coherence and Excessive Intermediation, Quar-
terly Journal of Economics 130, 1283ñ1328.
 Ellison, G. and D. Fudenberg (2003). Knife-Edge or Plateau: When Do Market Models
Tip? Quarterly Journal of Economics 118, 1249ñ1278.
 Ellison, G., D. Fudenberg and M. Mˆbius (2004). Competing Auctions. Journal of the
European Economic Association 2, 30ñ66.
 European Commission (2015). Public Consultation on the Regulatory Environment for Plat-
forms, Online Intermediaries, Data and Cloud Computing and the Collaborative Economy.
Published September 2015.
 Fujita, M., and J.F. Thisse (2013). Economics of Agglomeration. Cities, Industrial Location,
and Globalization (2nd Edition). Cambridge: Cambridge University Press.
 Greenstein, S., M. Peitz and T. Valletti (2016). Net Neutrality: A Fast Lane to Understand-
ing the Trade-o§s. Journal of Economic Perspectives 30, 127ñ149.
 Grilo, I., O. Shy and J.F. Thisse (2001). Price Competition when Consumer Behavior is
Characterized by Conformity or Vanity, Journal of Public Economics 80, 385ñ408.
 Hagiu, A. (2009). Two-Sided Platforms: Product Variety and Pricing Structures. Journal
of Economics and Management Strategy 18,1011ñ1043.
 Hoernig, S., R. Inderst and T. Valletti (2014). Calling Circles: Network Competition with
Nonuniform Calling Patterns, Rand Journal of Economics 45, 155ñ175.
 House of Lords (2016). Online Platforms and the Digital Single Market. Report published
April 20, 2016.
 Karle, H., M. Peitz and M. Reisinger (2016). Segmentation versus Agglomeration: Compe-
tition between Platforms with Competitive Sellers. Mimeo.
 Katz, M. and C. Shapiro (1985). Network Externalities, Competition and Compatibility.
American Economic Review 75: 424ñ440.
 La§ont, J.-J., P. Rey, and J. Tirole (1998a). Network Competition: I. Overview and Nondis-
criminatory Pricing. Rand Journal of Economics 29, 1ñ37.
 La§ont, J.-J., P. Rey, and J. Tirole (1998b). Network Competition: II. Price Discrimination.
Rand Journal of Economics 29, 38ñ56.
 Marshall, A. (1890). Principles of Economics, Macmillan, London (8th ed. published in
 Monopolkommission (2015). Competition Policy: The Challenge of Digital Markets. Special
Report by the Monopolies Commission, 1 June 2015.
 Nocke, V., M. Peitz and K. Stahl (2007). Platform Ownership, Journal of the European
Economic Association 5, 1130-1160.
 Peitz, M. and M. Reisinger (2015). Media Economics of the Internet, in: Anderson, S.P.,
Stromberg, D., Waldfogel, J. (eds.), Handbook of Media Economics, vol 1A, Amsterdam:
Elsevier, pages 445ñ530.
 Reisinger, M. (2014). Two-Part Tari§ Competition between Two-Sided Platforms. Euro-
pean Economic Review 68, 168ñ180.
 Rochet, J.-C. and J. Tirole (2003). Platform Competition in Two-sided Markets, Journal
of the European Economic Association 1, 990ñ1024.
 Rochet, J.-C. and J. Tirole (2006). Two-sided Markets: A Progress Report. Rand Journal
of Economics 37, 645ñ667.
 Rysman, M. (2009). The Economics of Two-Sided Markets. Journal of Economic Perspec-
tives 23, 125ñ143.
 Slee, T. (2016). Airbnb Is Facing an Existential Expansion Problem. Harvard Busi-
ness Review. Online July 11, 2016 (https://hbr.org/2016/07/airbnb-is-facing-an-existential-
 Sweeting, A. (2015). Radio, in: Anderson, S.P., Stromberg, D., Waldfogel, J. (eds.), Hand-
book of Media Economics, vol 1A, Amsterdam: Elsevier, pages 341ñ396.
 Weyl, E. G. (2010). A Price Theory of Multi-sided Platforms. American Economic Review
 Weyl, E.G. and A. White (2016). Insulated Platform Competition. Mimeo. Available at