Competition in the news industry: ghting aggregators

with versions and links

+
Joan Calzada
y
and Guillem Ordez,
z
October 3, 2013
Abstract
We analyze the linking and versioning strategies of a media publisher when facing
competition from blogs, search engines and news aggregators. We show that when the
publisher competes against a blog it is less likely to release a ghting version if this
generates signicant spillovers for its competitor. When the publisher competes against a
search engine, it wants to be linked when it receives enough indirect trac from it, and
in this case it uses the search engine as its own low quality version. We also demonstrate
that when the publisher is integrated with a search engine it might link other publishers
to obtain more indirect visitors if the contents are suciently dierentiated.
Keywords: Product segmentation, versioning, linking, media market, search engines,
news aggregators, Internet.
JEL: D83, D85, L12, L22, L86, M31

We thank Christian M. Bender, Luis Collado, Martin Peitz, Joan Santal and Tommaso Valletti for helpful
comments. We also thank participants at the 3rd Workshop on the Economics of ICTs (Porto), EARIE
(Rome), JIE (Murcia), and at semininars at the Universitat de Barcelona, Universitat Autonoma de Barcelona
and Universidad Complutense de Madrid. We gratefully acknowledge nantial support from the NET Institute
(www.NETinst.org).
y
Departament de Poltica Econmica, Universitat de Barcelona. calzada@ub.edu.
z
Department of Economics, University of Warwick, g.ordonez-cala@warwick.ac.uk
1 Introduction
The newspaper industry is undergoing a major transformation.
1
Internet has signicantly
reduced the costs of accessing information and of distributing the contents globally and this
has favoured the creation of many news services that compete with traditional newspapers.
In this new scenario, publishers seek out new competitive strategies and business models to
retain consumers and advertisers. This paper analyzes how the changes in the industry are
modifying the linking and versioning strategies of media publishers.
Some of the most successful new players in the media market are search engines and news
aggregators, which have risen to occupy the top positions in audience ranking (Nielsen 2011).
Search engines link the pieces of news of other on line newspapers and display them in a single
site accompanied by a title or an excerpt. This represents a considerable saving to consumers
in terms of time and eort as they can simultaneously check several information sources for
updates.
2
Search engines use algorithms that index and group stories according to diverse
criteria including the originality of information, their immediate interest for readers and their
contagious capacity. News search engines such as Google News, Bing News, and Summify
neither hang advertisements nor set user charges. Their commercial objective is essentially to
bring visitors to other Internet services on the same platform, such as common searches that
sell advertising space or social platforms.
News aggregators, by contrast, provide news articles that are created with information
obtained from newspapers and press agencies, and they do not usually include links to the
original sources. Some of the best known aggregators are Yahoo! News, Drudge Report, and
The Hungton Post. These usually reach licensing agreements with information suppliers to
avoid copyright infringements.
The emergence of these new services has created a trade-o for traditional newspapers,
who must decide between ghting the entrants or accommodating them. On the one hand,
entrants create a business-stealing eect since they use information from traditional newspapers
1
Figures for 2002 show that the newspaper industry amounted to around $50 billion business and employed
around 400,000 people. However, today it is facing a severe crisis in part due to the migration of readers from
printed newspapers to on-line news sources (PEJ, 2011).
2
According to this argument, U.S. legislation considers that these entrants make a transformative use of
contents and, as a consequence, do not infringe copyright law (Isbell, 2010).
1
to create their own contents and attact visitors.
3
On the other hand, they create a market
expansion eect because they reduce consumer search costs by making great variety of contents
widely available from one single web site. This situation benets traditional newspapers, which
receive indirect trac from the visitors of the search engines that click their links.
4
The relationship between search engines and news agencies is even more complex, because
the main activity of agencies is precisely that of feeding newspapers with their content. In order
to deal with this situation, in 2004 Google News reached an agreement with the Associated
Press (AP), permitting Google to host AP contents in exchange for compensation.
5
Similarly,
in 2005 Agence France Press (AFP) led a lawsuit against Google News for removal of copyright
management information and hot news misappropriation. After two years of litigation, AFP
and Google News settled the case by signing a licensing agreement, according to which Google
was granted the right to post AFP content. In spite of this, other press associations are still
pressing public authorities to protect their content,
6
and some regulators such as the Federal
Trade Commission have suggested various solutions to deal with this problem, which range
from extending copyright legislation to diverse antitrust exemptions for news organizations
(FTC, 2010).
Traditional newspapers have modied their commercial strategies to defend themselves
against new competitors. Some publishers believe that they still have a signicant demand
from readers who prize high quality journalism, editorial guidance and the opinions of experts.
Thus, for example, The Times launched a pay wall in 2010 and The Wall Street Journal has
3
Some traditional newspapers consider aggregators to be free-riders reselling the information they have
gathered at vast expense (Frijters and Velamuri, 2010). R. Murdoch has declared: To aggregate stories is not
fair use. To be impolite, it is theft. See The Guardian, 1/12/2009, Theres no such thing as a free news
story, http://www.guardian.co.uk/media/2009/dec/01/rupert-murdoch-no-free-news.
4
Evidence suggests that 44% of Google News! users in 2010 didnt click through to the original articles.
http://techcrunch.com/2010/01/19/outsell-google-news/.
5
This agreement was broken at the end of 2009 and the search engine stopped oering APs links. In spite of
this, in early 2010, AP reached a new agreement with Google and another with Yahoo! News. See Associated
Press Returns to Google News, http://mashable.com/2010/02/09/ap-google-news.
6
See, for instance, the lawsuit of Federazione Italiana Editori Giornali against Google in 2009; the demands
of the Federation of German Newspaper Publishers for a new copyright in 2009; the Hamburg Declaration on
Intellectual Property Rights, signed by the World Association of Newspapers and News Publishers in 2009; or
the case that the Newspaper Licensing Agency won in the UK High Court involving the main newspapers and
news aggregators in 2010.
2
recently launched freemium, by which general news remain free but premium contents and
some blogs are locked behind a pay wall. Similarly, the Financial Times oers subscriptions
to premium and standard articles at dierent prices.
7
Some publishers have tried to increase
their audiences by buying news aggregators. For example, in 2005 Gannet bought Topix.net,
and in 2011 AOL bought The Hungton Post. Other publishers have even created their own
news aggregators with high quality contents.
8
The objective of this paper is to analyze how new business models in the media market are
modifying the commercial strategy of media publishers. We present a model that examines
the linking and versioning strategies of one publisher that can compete with one blog, search
engine or news aggregator. By so doing, we identify when it wants to ght the entrant by
releasing a low quality version and when it prefers to accept the links of the search engine. We
show that links allow the publishers to use the web sites of their competitors as an alternative
distribution channel for their contents.
Our paper contributes to the recent literature that analyzes the impact of news aggre-
gators in the managerial strategies of publishers. Katona and Sarvary (2008) were the rst
to theoretically analyze strategic linking between web sites in a market for advertising links.
Dellarocas et al. (2010) analyze competition among news sites that generate revenue from user
visits and who compete with each other as well as with alternative media (e.g. TV, blogs,
Twitter feeds) for user attention. In their model, news sites decide how much to invest in
original content and how many links to include. The authors show that linking has dierent
implications depending on the level of dierentiation of the web sites. They also show that the
presence of an outside alternative may benet incumbent content sites, since they increase the
total trac that ows in the content ecosystem and divert most of it to the best content sites.
George and Hogendorn (2012) adapt a two-sided market model to digital media markets to
7
Other examples are the plans of the New York Times to erect a paywall or those of News Limited to
charge for access to certain parts of Daily Telegraph and Herald Sun. See The Conversation, 8/6/2011,
http://theconversation.edu.au/news-ltd-announces-pay-wall-plan-as-newspapers-struggle-nline-1743.
8
In 2009, Murdochs News Corporation launched the Project Alesia and in 2010 Gannett, the New
York Times Co. and the Washington Post Co. created Ongo. The two aggregators had to oer high
quality contents behind a paywall but both ceased operations short after their release. See A. LaFrance,
Ongo, an attempt at a pan-media paywalled aggregator, is closing; Nieman Journalism Lab, 8/5/2012,
http://www.niemanlab.org/2012/05/ongo-an-attmpt-at-a-pan-media-paywalled-aggregator-is-closing/.
3
examine how search technology and aggregators can alter both market participation and the
number of sites visited. They explain that both aggregators and improved search technology
tend to increase viewer multi-homing. But unlike search, aggregators may not expand the
market. Jeon and Nasr (2012) show how the presence of a news aggregator aects competition
among newspapers. They nd that competition with a news aggregator can lead to the special-
ization of newspapers when their advertising revenue increases substantially with an increase
in quality. With specialization, the presence of the aggregator increases the average quality of
newspapers, which in turn increases consumer surplus.
Our paper also contributes to the economic literature on versioning. The seminal papers
of Mussa and Rosen (1978) and Moorthy (1984) showed that the introduction of a new version
by a monopolist might be protable when the benets of expanding the market overcome the
cannibalization eect on the original product. Stokey (1979) and Salant (1989) later showed
that versioning essentially depends on the form of the cost function. In particular, versioning
is optimal when the marginal cost function of improving the quality is suciently convex.
More recently, a number of papers have sought to clarify the circumstances under which
versioning is convenient for a monopolist. Bhargava and Choundary (2008) nd that version-
ing depends on the relation between the optimal market share of the low and high quality
versions of one product when they are oered alone. Anderson and Dana (2009) show that
versioning requires that the relative change in overall surplus associated with a product qual-
ity improvement is increasing in consumers willingness to pay. Calzada and Valletti (2012)
demonstrate that versioning might arise when consumers are allowed to buy two versions of
the same product. They examine the particular case of the movie industry and explain that
versioning is optimal when the theater and the DVD versions are not merely substitutes. In
this case, while some consumers buy both versions, others only use the theater or the DVD
version. This situation is closely related to that discussed here, since the main characteristic
of news aggregators is to allow reading the contents of several newspapers.
Other studies have analyzed versioning in the presence of competition. Wu, Chen and
Anandalingam (2003) show that versioning can be a very eective and protable instrument
in the ght against piracy.
9
Valletti and Szymanski (2006) analyze parallel trade for products
9
An early analysis on this issue can be found in Johson and Myatt (2003).
4
protected by intellectual property rights such as the export of pharmaceuticals. They show that
pharmaceutical rms only nd it optimal to release a lower quality version when they compete
with a generic product. Our paper is related to their model, since we analyze the versioning
strategy of a publisher who faces competition from other sites. A remarkable dierence is,
however, that the contents distributed by publishers through Internet can be immediately used
by other news sites and can be linked. We show that these situations can modify importantly
the versioning strategy of rms.
Our paper extends the traditional models of Mussa and Rosen (1978) and Moorthy and
Png (1992) to examine the versioning policy of a publisher that competes with other news sites.
The publisher can commercialize two news sites of dierent qualities and/or dierent editorial
approaches, as is the case of the News Corporation or The New York Times Co.
10
Taking this
into account, we rst analyze the eects of the entry of a blog (or a news site) that uses the
information released by the publisher to generate its own contents, but that doesnt set links
to the original sources of information. We show that a competitor of this type reduces the
publishers incentives to release a second news site. While in Valletti and Szymanski (2006)
the incumbent release a ghting version to defend itself against the competitor, in our model
the ghting version generates positive spillovers that increases the quality of the competitor.
As a result, it is more likely that the publisher accommodates the blog instead of releasing a
second version.
The second part of the paper analyzes the case when the publisher competes against a
search engine. The search engine benets from the news sites of the publisher when it links
them. We show that the publisher accepts the links of the search engine when it receives
sucient indirect trac from it. The reason is that the link can generate more revenues than
a low quality version. In fact, for the publisher the link plays a similar role than a low quality
version. When the indirect trac generated by the link is even higher the publisher accepts
the links but releases a second version in order to expand the audience of the search engine.
Although the new version cannivalizes its high quality news site it generates enough indirect
trac to compensate this eect. As in Calzada and Valletti (2012), the publisher releases two
versions because some consumers buy both of them. Finally, we examine the case where the
10
Alternatively, we could think in one newspaper that commercializes high and low quality pieces of news
that cover the same event, as in the freemium pricing model.
5
quality of the search engine can be higher than the quality of the news sites oered by the
publisher, and we show that in this case linking and versioning are more likely.
The last part of the paper considers an integrated publisher that owns a newspaper and a
search engine. We explain that the publisher can be interested in linking its competitors when
their news sites are suciently complementary and the indirect trac generated by the search
engine is not too high. We also show that the publisher can maintain the search engine active
but without links to its competitors because it constitutes a protable distribution channel.
The rest of the paper is structured as follows. Section 2 introduces the model. Section 3
analyzes the versioning strategy of a publisher that competes with a blog. Section 4 examines
the linking and versioning strategies of a publisher that competes with a high and a low quality
search engine. Section 5 considers an integrated publisher that owns a news site and a search
engine. Section 6 concludes.
2 The Model
We analyze the versioning and linking strategies of a publisher that can have dierent competi-
tors such as a blog, a search engine or a news aggregator. The publisher can oer one or two on
line newspapers with dierent editorial approaches. It always oers the news site H of quality
n
1
= n and it might also release the site 1 of a quality n
J
= cn, where c < 1. The quality
of each site is exogenously determined and their content can be related. For example, 1 may
be a degraded version of site H and consumers might not obtain any additional utility from
it after having visited H. Alternatively, H and 1 can be complementary sites. For example,
H can be a site with general information and 1 a site specialized in sports or a site with the
blogs of experts.
The competitor oers the news site , which can include contents obtained from H and
1, or links to these products. When the publisher only releases H the quality of is n
.
= ,n,
where , < c reects that doesnt oer original contents and doesnt follow any editorial
policy. On the other hand, when the publisher releases H and 1 , the quality of is n
.
=
n[,+c(1:)], where : (0, 1) measures the positive spillovers that the release of 1 generates on
the consumers that are already consuming H. Notice that n
.
n
1
when : < (, + c 1),c.
This occurs when consumers prefer contrasting the information of several news sites than
6
visiting just one newspaper with a particular editorial policy.
In order to simplify the model and to make it more realistic we assume that the consumers
searching costs are so high that they can only visit one site (the anchor site). In spite of this,
when consumers visit a search engine or a news aggregator they can click through to the
original contents of all newspapers (see Purcell et al. 2010 and Dellarocas et al. 2012). This
assumption highlights the key feature of search engines and news aggregators, which is to
group the contents of dierent newspapers on a single web site, thereby signicantly reducing
consumers search costs.
Consumers are characterized by the intensity of their preferences for news sites. Imag-
ine that they are uniformly distributed in the segment [0, 1] and dene as 0 the consumers
willingness to pay. Taking this into account, the consumers net utility when they visit the
publishers version i is 0n
I
j
I
, with i H, 1, , where j
I
is the price of the news sites or the
consumers costs when they visit the entrant. An alternative way of interpreting our model is
to imagine that j
I
is the number of advertisements inserted on the news site i when the price
of the advertisements is exogenously set and normalized to one. Under this interpretation the
publisher would choose the number of advertisements that maximizes its prots, taking into
account that advertisements reduce the utility of consumers.
Finally, in order to focus our attention on the rms linking and versioning strategy, we
assume that the costs associated with producing each version are sunk, and that the marginal
costs of the versions are zero. This assumption means that it could be expensive to generate
contents for the newspapers, but that they can be distributed at no cost through the Internet.
3 Competing with blogs: no exclusion option
This section examines the commercial policy of a publisher competing with a blog, or a news
site using third party content. Internet allows bloggers to publish opinions and pieces of news
that are largely based on information produced by traditional on line newspapers. As a result,
when a publisher releases new contents it feeds these sites with materials that improve their
quality. In this section, we assume that the publisher is not able to prevent the use of its
contents by others, as it happens with bloggers.
The following set of indierent consumers describes the segmentation of consumers under
7
two possible scenarios. If the publisher only releases newspaper H the consumer indierent
to visiting the newspaper and the blog is 0
1.
= (j
1
j
.
),n(1 ,), while the consumer
indierent to visiting the blog and not visiting any news site is 0
.0
= j
.
,,n. In this case, the
publishers prot is
1.(1)
= j
1
(1 0
1.
). On the other hand, if the publisher releases H
and 1 the consumer indierent to visiting H and 1 is 0
1J
= (j
1
j
J
),n(1c), the consumer
indierent to visiting 1 and is 0
J.
= (j
J
j
.
),[n(c (, + c(1 :))], and the consumer
indierent to visiting and not visiting any news site is 0
.0
= j
.
,[n(, + c(1 :))]. In this
case, the publishers prot is
1J.(1J)
= j
1
(1 0
1J
) + j
J
(0
1J
0
J.
). In this section, we
assume for simplicity that n
J
n
.
, which means that the quality of the blog can never be
larger than the quality of 1.
The timing of the game played by the publisher and the blog is as follows: rst, the
publisher decides how many versions it oers; second, it sets the prices and releases its news
sites; third, the blog creates its own contents using the information generated by the publisher.
Finally, consumers choose their preferred news site.
The following proposition shows when the publisher only releases H and accommodates
the blog, and when it releases a ghting version. It also describes the resulting segmentation
of the consumers.
Proposition 1. In the presence of a blog the publisher reacts as follows:
1) If , < ,
1
its optimal versioning strategy is:
When 0 _ : _ :, the rm only oers H, and the blog is active;
When : < : _ 1, the rm supplies H and 1, and the blog is active;
2) If , _ ,
1
its optimal versioning strategy is:
When 0 _ : _ :
2
, the rm only supplies H, and the blog is active;
When :
2
< : _ :
1
, the rm supplies H and 1, but the blog is not active;
When :
1
< : _ 1, the rm supplies H and 1, and the blog is active;
Proof. See the Appendix.
8
The rst part of this proposition presents the strategy of the publisher when the quality
of is low in comparison with that of 1 (, < ,
1
). In this case, the rm releases 1 when the
spillovers generated on are small (: :), because the market expansion eect obtained with
1 osets the cannibalization eect of 1 on H (Figure 1). However, when spillovers are large
(: _ :) the release of 1 generate little revenues and the publisher accommodates instead of
releasing 1. Also notice that for : : there is a complete segmentation of consumers: those
with a high type visit H, those with an intermediate type visit 1, those with a low type visit
, and those with an even lower type dont visit any news sites at all (region C in the two
panels of Figure 1).
Figure 1: Competition with a blog (, = 0.45 in Panel I and , = 0.5 in Panel II; c = 0.75; n = 1;
j
.
= 0.12). The gure shows the consumers segmentation as a function of :. In Panell II, for : < :
2
the publisher only oers H: low type consumers dont visit any site, intermediate consumers visit ,
and high type consumers visit H. For : :
2
the publisher oers 1 and H. If :
2
< : _ :
1
, j
J
is
low and the blog is not active; if :
1
< : _ 1, j
J
is higher and some consumers visit .
The second part of the proposition considers the case where and 1 have similar qualities
(,
1
_ , < c). The intuition of the result is the same than before, but now when : takes an
intermediate value (:
2
< : _ :
1
) the publisher sets a low j
J
and fails to attract consumers
(region B Panell II in Figure 1).
This proposition can be directly related to the model of Valletti and Szymanski (2006) who
show that under the presence of a competitor an incumbent will release a ghting brand to
9
retain part of its prots. Proposition 1 reaches the same conclusion when : is close to 1, that
is, when the blog gains few spillovers from the ghting newspaper. However, when the new
site strengthens signicantly the blog, the publisher only oers H.
The model reveals that the publishers versioning strategy depends on the externalities
generated by 1. The next result shows that when the publisher releases 1 it would prefer to
oer a close substitute to H in order to minimize the spillover eect.
Corollary 1. Suppose the publishers cost of horizontally dierentiate 1 from H is c(:) =
1,/:. Then when it releases 1 it prefers the lowest possible level of dierentiation, : = 1.
Proof. See the Appendix.
The key element to understand this result is that consumers can only access the information
oered in both H and 1 when they visit the blog. In fact, the publisher doesnt has any site
that aggregates the contents of the versions and for this reason it can not benet of their
complementarity. For this reason, it if was able to determine the quality of 1 it will just create
a degraded version of H (i.e. : 1). For example, 1 could summarize the main information
in H or could reproduce part of the contents. The reduction of : reduces the utility of and
allows the publisher to increase its prices. In spite of this, some switch from to 1.
The main result of this section has been to show that when publishers cant avoid its con-
tents being used by others versioning is less likely than what was previously considered in the
literature. The reason is that a ghting versions improves the quality of the blog. In spite
of this, if the publisher was able to control the use of its contents it could also accomodate
the entrant in exchange of an appropiate compensation. This is the case of several publishers
and news agencies such as Reuters, BBC News, or USA Today which have reached an agree-
ment with Yahoo! News. In this case, the negotiation gives the publishers the possibility of
internalizing the complementarity generated by their versions.
11
The next section examines
an alternative mechanism that allows the publisher to benet of the complementarity between
the versions, which is the establishment of links in the aggregators web sites.
11
Calzada and Ordoez (2012) analyze the negotiation of a revenue sharing agreement between a publisher
and a news aggregator.
10
4 Competing with search engines: spillovers and links
Next we extend our model to examine the versioning strategy of a publisher that competes
against a search engine. Imagine that the publisher can release one or two newspapers and
that the search engine can link them in its web site. The visitors of the search engine read the
headlines of the news articles (or excerpts of them) and click through to them with a probability
c, which we assume to be the same for H and 1. For example, c = 1 means that the visitors
of the search engine always click the links and generate indirect trac for the publisher, and
c = 0 means that they read the headlines but never click through to any news articles.
Both the publisher and the search engine must agree in the establishment of the links. The
publisher will accept the links to its newspaper if this increases its prots; and the search engine
links the newspapers if this increases its audience. We consider that consumers searching costs
are so high that the search engine is the only available mechanism for accessing the contents of
two newspapers. In fact, the main feature of search engines is that they aggregate and classify
the contents of several newspapers and reduce consumers searching cost.
The quality of the search engines site, which we call , depends on the links that it
contains. When it reaches an agreement with the publisher its quality is n
.
= n, if the publisher
releases H and n
.
= n[, + c(1 :)] if the publisher releases both H and 1. Contrary, when
there is no agreement its quality is n
.
= nr, with r < ,. This implies that consumers value
less than 1 when the search engine doesnt link the publisher.
Taking this into account, the net utility of consumers that directly visit H is 0n j
1
and
that of those who directly visit 1 is 0cn j
J
. The net utility of consumers that visit when
this links H and 1 is 0n[, +c(1 :)] c(j
1
+j
J
), when it links H is 0n, cj
1
, and when it
doesnt link any newspaper is 0nr j
.
. Notice that the search engine consumers only pay for
the contents when they click through to the links. This reects the present situation of some
search engines like Google News!, which neither charge any price to consumers nor include
advertisements in their sites.
When the search engine links H and 1 it has a quality that can be higher or lower than those
of H, depending on the complementarity between H and 1. To account for this possibility,
we rst analyze the case of a low quality search engine (n
1
n
J
n
.
) and afterwards we
11
examine the case of a high quality one (n
.
n
1
n
J
).
12
The following proposition describes the linking and versioning strategies of a publisher that
compete with a low quality search engine. This is the case where users derive greater utility
from directly accessing one news site than from visiting one search engine that links several
news sites (e.g. they like the editorial policy of the newspapers and value their reputation).
The timing of the game is as follows: rst, the publisher decides how many versions to release;
second, the publisher and the search engine negotiate about setting links on the publishers
sites, and taking into account the results of the negotiation the publisher set the prices; and
third, the news sites are released and consumers choose their preferred service.
Proposition 2. The linking and versioning strategies of a publisher and a low quality
search engine ( : ,,c) are:
1) If 0 _ c _ c
1
the publisher releases H and 1, but it doesnt accept the links. H, 1 and
receive direct visitors;
2) If c
1
< c _ c
2
the publisher releases H and both rms agree on linking it. H and receive
direct visitors;
3) If c
2
< c _ c
3
the publisher releases H and 1 and both rms agree on linking them. Only
H and receive direct visitors;
4) If c
3
< c _ c
4
the publisher releases H and both rms agree on linking it. H and receive
direct visitors;
5) If c
4
< c _ 1 the publisher releases H and 1, but the search engine is not interested in
linking them. H, 1 and receive direct visitors;
The existence and size of these intervals depend on :.
Proof. See the Appendix.
The rst part of the proposition shows that the publisher doesnt accept to be linked when
it receives a small amount of trac from the search engine (0 _ c < c
1
). In spite of this, the
publisher releases the ghting version 1 because it doesnt create spillovers on (region A in
the two Panels of Figure 2). As a consequence, there is a complete market segmentation of
12
We omit the case where &
H
&
A
&
L
for simplicity.
12
Figure 2: Competition with a search engine (, = 0.4, c = 0.8, n = 1, r = 0.3, j
.
= 0.05).The
gures show the consumers segmentation as a function of c. Each panel considers a dierent
level of spillovers and is divided in regions according to the parts of Proposition 2. In region
A the publisher doesnt want to be linked. In region B the publisher is linked and releases H.
In regions C the publisher is linked and releases H and 1. In region D the publisher is linked
and releases H. Finally, in region E the search engine doesnt want to link the publisher.
consumers: lowest-type consumers do not visit any site and the rest of consumers visit H, 1
or according to their preference for quality. It is interesting to mention that c
1
is increasing
in j
.
, which means that when the search engine becomes more unattractive to consumers the
publisher is less interested in reaching an agreement with it. Regarding the search engine, in
this interval it wants to link the H and 1. Indeed, although this implies an increase in its price
c(j
1
+ j
J
) the increase in quality would compensate this situation and it will attract more
visitors.
The second and third parts of the proposition show that when the publisher receives a
larger amount of trac from the search engine it accepts to be linked and uses as its own
low quality version. If c
1
< c _ c
2
, the publisher only releases H. If it also releases 1 this
would generate positive spillovers on that would reduce its prots. Particularly, when c and :
are small the market expansion eect generated by 1 does not compensate the cannibalization
eect on H.
13
Note that in this interval the link of H transforms in the low quality version
13
If c is close to 1 the publisher always releases 1 and region B vanishes (see Panel I in Figure 2).
13
of the publisher. Thus, the publisher prefers to accept the link and to use as an alternative
distribution channel for H than to release 1.
If c
2
_ c _ c
3
, the publisher releases 1. Like in the previous case, it transforms the search
engine in its low quality version. Specically, it segments the market in such a way that high
type consumers visit H and a group of intermediate consumers visit H and 1 indirectly via
the search engine (see region C). In fact, the publisher releases 1 to increase the audience of
and to obtain indirect visits from both versions. If the level of complementarity between H
and 1 is not too high, this strategy does not create a large cannibalization eect on H.
The last two parts of the proposition show that the search engine doesnt want to link
the publishers newspapers when c is high. When c
3
< c _ c
4
, the publisher only releases H
because if it releases H and 1 the search engine would not link any site. The reason is that
the price of the seach engine, c(j
1
+ j
J
) would be very high (see region D).
Finally, when c is suciently large (c
4
< c _ 1) the search engine never links the publisher.
The links would increase importantly its price and reduce its audience. In this case, consumers
are segmented as in the rst part of the proposition (see region E).
In our model the complementarity between H and 1 plays a key role in the strategy of
the rms, but it is introduced exogenously. The following result considers how the publisher
would set : once it oers the two versions and these are linked.
Corollary 2. Suppose that the publisher releases H and 1 and that these are linked by a
low quality search engine. If the cost of horizontally dierentiate 1 from H is c(:) = 1,/:,
then the publisher will choose the lowest possible level of product dierentiation, : = 1.
Proof. See the Appendix.
Therefore, even when the publisher accommodates the search engine and releases the two
newspapers it prefers to increase : in order to reduce spillovers as much as possible.
The strategy of the publisher changes signicantly when the quality of the search engine
is larger than the quality of H and 1 consumed separately, n
.
n
1
n
J
. This situation
reects the case where consumers prefer product variety to limiting themselves to one high
quality newspaper. The following proposition describes the commercial strategy of rms in
this situation.
14
Proposition 3. The linking and versioning strategy when the publisher competes against
a high quality search engine ( : < (c + , 1),c) is:
1. If 0 < c _ c
1
, the publisher releases H and 1 and accepts the links. Only receives
direct visits;
2. If c
1
< c _ c
2
, the publisher releases H and 1 and accepts the links. , H and 1 receive
direct visits;
3. If c
2
< c _ 1, the publiser releases H and 1, but either the search engine (for c
2
< c _
c
3
) or both rms (for c
3
< c) are not interested in the links.
Proof. See the Appendix.
The rst part of the proposition focuses on the case in which c is small (0 _ c _ c
1
). In
such a situation, the publisher releases H and 1 and allows the links of the search engine. As c
is small and n
.
_ n
1
only receives direct visitors. In spite of this, the publisher maximizes
prots with this segmentation of the market because it can set high prices for the two services.
Indeed, it obtains more prots that if it was a monopolist. Hence, the presence of one search
engine that aggregates contents makes both rms better o because consumers indirectly visit
both H and 1.
The second part of the proposition shows that when the users of the search engine make
more visits to the newspapers (c
1
_ c < c
2
) prices become too high for some consumers to
visit the two versions. As a result, there is a complete segmentation of the market: high type
consumers visit , intermediate types visit H, low types visit 1, and very low types dont visit
any site. Notice that in this scenario, version 1 cannibalizes H in two ways. First, some low
type consumers who could consume H actually visit 1. And second, some high type consumers
who could visit H now visit . In spite of this, as c is high the cannibalization over H is oset
by the direct and indirect visits to 1 and the indirect visits to H.
The last part of the proposition presents the case for the highest values of c (c
2
< c _ 1).
In this case, both rms lose incentives to reach an agreement. When c c
2
the high price
of will oset the increase in the quality of generated with the links. As a result, the
search engine prefers not to link the publisher. To illustrate this situation, imagine that : = 1
15
Figure 3: Consumers segmentation when the publisher competes with a high quality search engine
(,=0.4, c=0.8, n=1, r=0.2, j
.
=0.05). Each panel considers a dierent level of spillovers : and is
divided in regions according to Proposition 3. In region A the publisher releases two versions that are
linked, but only the seach engine is active. In region B the publisher releases two versions that are
linked, and all products are active. In region C the publisher releases two versions that are not linked
and the three sites are active.
and n
.
= n
1
. In this case, if consumers access H directly they pay j
1
, but if they use the
search engine and click through to the two newspapers they pay c(j
1
+ j
J
). Therefore, if
c j
1
,(j
1
+ j
J
) consumers are better o buying only H and the search engine does not
receive visits. Similarly, for c c
3
(c
3
c
2
is not shown in Figure 2) the publisher doesnt
accept to be linked and releases a ghting version. This is because it faces a trade o when
setting the prices. If they are too high, becomes very expensive and loses audience, so H
and 1 receive less indirect visitors. And if they are too low totally cannibalizes H and 1.
Finally, notice that the size of the regions identied in the proposition crucially depends
on the values taken by :. When : increases the region in which both companies are willing to
cooperate increases. In Figure 3 this corresponds to the areas A and B. Thus, spillovers make
more likely the establishment of links.
The next corollary shows how the publisher would like to modify the complementarity level
between H and 1 once these have been released and linked.
16
Corollary 3. Suppose that the publisher releases H and 1 and that these are linked by a
high quality search engine. If the cost of dierentiating 1 from H is c(:) = 1,/: the publisher
will increase the complementarity of the versions as long as / /

.
Proof. See the Appendix.
While in the case of a low quality search engine the publisher prefers the higher possible level
of sustitutability between the versions, in this case it prefers to increase the complementarity
of the versions to be able to increase the prices. This result reveals that the publisher does
not see the search enigne as a rival, but as an opportunity to receive a higher number of
visitors. Consumers now can visit the two versions and their total willingness to pay for the
two products decreases with :.
To sum up, in this section we have shown that the distribution strategies of the publisher
and the search engine crucially depend on the amount of indirect trac that can generate the
search engine and on the complementarity of the contents. Proposition 2 shows that when
n
.
< n
J
, the publisher wants to reach an agreement with the search engine if this generates
sucient trac for its sites. In this case, the publisher uses the search engine as its low quality
version. Furthermore, the publisher can release a second site to strengthen the quality of the
search engine.
Proposition 3 examines the case where spillovers are larger and n
.
n
1
. In this situation,
if the search engine redirects a large number of consumers to the publisher the rms dont
want to reach an agreement. Indeed, the search engine loses audience and the publisher has to
set low prices. However, if smaller volumes of trac are redirected to the publisher both rms
are interested in reaching an agreement. The results of the two propositions are in line with
Calzada and Valletti (2012), who show that a monopolist might release two versions when a
part of the consumers can buy both of them. Interestingly, in our model it is precisely the
search engine that permits this possibility.
5 Vertical integration and linking strategy
In the last years, several news publishers have acquired or created news aggregators with the
aim to grow and engage their audiences through forums and RSS feeds. For example, in
17
2005 Gannet (the USA Todays publisher), Knight-Rider and Tribune acquired the 75% of
Topix.net, in 2011 AOL acquired The Hungton Post, and in 2012 Gannet bought Quickish,
a sports aggregator. With these investments traditional media groups look for alternative
revenue sources and separate their activities of production and distribution of contents. Other
recent experiences have been less successful. For example, in 2009 News Corporation planned
to launch the news aggregator Alesia, but the project was canceled due to the trouble of
nding other publishers interested. In 2010, several leader publishers in the US market such
as Gannett, The New York Times Company, The Washington Post Company, Reuters, Los
Angeles Times or Chicago Tribune created a news aggregator named Ongo, but this site closed
two years later.
14
One distinctive feature of Alesia and Ongo is that they were created from
scratch with high quality pieces of news of their promoters and behind a pay ball. In the
acquisitions of The Hungton Post and Topix, by contrast, the publishers were interested in
buying an alternative and successful distribution channel for their contents.
The vertical integration between publishers and aggregators rises novel economic questions
regarding its eects in the linking strategies of rms and in competition. Publishers want to
integrate aggregators to obtain indirect trac and referrals, but they can use their position to
eliminate the links of their competitors or to give an advantage to their own contents.
To address this case, we next analyze the linking and versioning strategy of a vertically
integrated publisher that commercializes both a high quality newspaper H and an aggregator
.
15
The publisher competes with an independent publisher that oers a low-quality newspaper
1. For simplicity, we assume that the price of the independent publisher, j
J
, is set exogenously.
The integrated and the independent publisher can agree to link 1. When this happens,
the level of complementarity between H and 1 determines the quality of . When H and
1 are close substitutes becomes a low quality search engine and when their contents are
more complementary is transformed in a high quality aggregator. Notice that the quality of
the aggregator inuences the strategies of the rms. Indeed, in the rst case we obtain that
n
1
n
J
n
.
which means that the independent publisher has a stronger competitor "from
14
One of the explanations of the failure of Ongo is that the managers established a bad pricing strategy. See
http://www.niemanlab.org/2012/05/ongo-an-attempt-at-a-pan-media-paywalled-aggregator-is-closing/
15
De Cornire and Taylor (2013) analyze the eects of the integration between search engines and publishers
of contents in the internet search engine markets.
18
below", and in the later case the link implies that n
.
n
1
n
J
which means that H has a
competitor "from above".
The next proposition analyzes the linking strategies of the publishers when the contents
are close substitutes. The timing of the game is as follows: rst, the two publishers negotiate
about setting a link to 1; once they have taken a decision the integrated publisher sets j
1
;
and nally consumers decide which news site to visit.
Proposition 4. The linking strategy of an integrated and an independent publisher that
commercializes close substitute contents ( : ,,c) is:
1. When ,,c < : _ :, the integrated publisher wants to link 1 when c < c
1
whereas the
independent publisher is never interested.
2. When : < : _ 1, the integrated publisher wants to link 1 when c < c
2
whereas the
independent publisher wants when c c
3
. Only when , ,

there is a small region

c
3
< c < c
2
in which the two publishers reach an agreement.Proof. See Proof in the
Appendix.
The proposition shows that the integrated publisher is interested in linking 1 when c and
: are suciently small; when a small percentage of s visitors click the link and the two
newspapers are suciently complementary. Only in this case the link expands the market
share of and increases the indirect visits to H. When the publisher links 1 the quality
of increases in nc(1 :) and its price in cj
J
. Therefore, the market expansion eect on
depends on the relation between c and :. In the case of the independent publisher, the
protability of the link depends on two opposite eects. On the one hand, the link generates a
market expansion eect because all the visitors of click through to 1 with probability c. On
the other hand, it creates a cannibalization eect, because competes with 1 from below and
it reduces part of its direct. visitors. Only when c and : are suciently high the rst eect
compensates the second and the publisher accepts the link
The rst part of the proposition shows that when : is small (: < :) the integrated publisher
wants to link 1 if c < c
1
but its competitor is not interested because the cannibalization over 1
is too important (Figure 4). The second part shows that when : is large (: :) the integrated
19
publisher wants to link 1 when c is small (c < c
2
) and that it becomes less and less interested
as : increases. Indeed, for larger values of c and : the rm has to reduce signicantly j
1
in
order to generate a market expansion eect. On the other hand, the independent publisher
wants to be linked when c is suciently large and becomes more and more interested in the
agreement as : increases. Taking this into account, only in the small region c
3
_ c < c
2
the
two publishers agree on linking 1, but this region only exists when , is close to c.
Figure 4: Linking strategy of the integrated and the independent publishers (,=0.4, c=0.8,
n=1, j
.
=0.05)
The linking strategies of the rms described above determine the segmentation of con-
sumers, which is characterized below.
Corollary 4. In the presence of an integrated and an independent publisher that commer-
cialize close substitutes contents ( : ,,c) consumers are segmented as follows:
1. When ,,c < : _ : and c < c

the news sites H, 1,and receive direct visits and

links H. For c _ c

is not active.
2. When : < : _ 1 and c < c

the news sites H, 1,and receive direct visits and links

20
H, except when , ,

, in which case it links both H and 1 in the interval c

3
< c < c
2
.
For c _ c

is not active.
Proof. See Proof of Proposition 4 in the Appendix.
The rst part of the corollary shows that for c < c

the integrated publisher maintains

active as a low quality version because this generates indirect visitors. But for c _ c

the
price of , cj
1
, is too high and it would receive few visitors. For this reason, the publisher
eliminates it and sets a higher j
1
. The second part of the corollary presents the same results,
but as we have seen before in Proposition 4 in the interval c
3
< c < c
2
the rms reach an
agreement and the search engine links both H and 1.
The main insight of these results is that an integrated publisher might maintain active its
low quality search engine, even when it doesnt link other sites. The search engine is used like a
ghting version, although in this case the publisher can only use j
1
to implement its strategy.
Next we examine the strategies of the rms in the case of a high quality search engine.
Recall that in this case the linking decisions of the rms can change importantly since the link
implies that competes with H "from above". The timing of the model is as before.
Proposition 5. The linking strategy of an integrated and an independent publisher that
oer complementary contents ( : <
o+o1
o
) is:
1. When 0 _ : < :, the integrated publisher wants to link 1 when 0 _ c < c
1
, whereas the
independent publisher always wants to be linked.
2. When : < : _
o+o1
o
, the integrated publisher is never interested in linking 1, whereas
the independent publisher always wants to be linked.
Proof. See the Appendix.
As in Proposition 4 these results show that the integrated publisher only links 1 when both
: and c are small. When this occurs, it creates a high quality news aggregator that completely
cannibalizes H (j
1
is so high that nobody directly visits H) and that directly competes with
1. The market expansion eect of compensates the cannivalization eect on H as long as
c < c
1
. For larger values of c the publisher doesnt link 1 because the search engine would
21
become more expensive and it would need to reduce j
1
signicantly in order to maintain its
market share. On the other hand, contrary to what happens in Proposition 4, the independent
publisher always want to be linked: although directly competes with 1 from above, it obtains
many indirect consumers that compensate this eect.
The second part of the proposition shows that for higher values of : the market expansion
eect generated when links 1 does not compensate the cannibalization eect on H. Again, in
order to incentive consumers to switch from 1 to the rm would need to reduce signicantly
j
1
, which is not protable.
The proposition shows that the integrated publisher only wants to link its competitor when
: is small and it obtains a signicant increase in the market share of . And when this happens
the independent publisher can accept to be linked because this expands its market share with
indirect visitors.
Figure 5: Linking strategy of the integrated and the independent publishers (,=0.4, c=0.8,
n=1, j
.
=0.05).
Finally, we examine how the linking decisions of rms inuence consumers segmentation.
22
Corollary 5. In the presence of an integrated and an independent publisher that commer-
cialize complementary contents ( : <
o+o1
o
) consumers are segmented as follows:
1. When 0 < : < : and c < c
1
news sites and 1 receive direct visits and links H and
1. For c _ c
1
H and 1 receive direct visits and is not active.
2. When : < : _
o+o1
o
and c <

c
1
the news sites H, 1 and receive direct visits and
only links H. For c _

c
1
H and 1 receive direct visits and is not active.
Proof. See Proof of Proposition 5.
The corollary shows how c (indirect visitors to H) and : (complementarity of the sites)
determine the versioning strategy of the publishers. The rst part shows that when 0 < : < :
and c < c
1
the integrated publisher maintains active. As a consequence, high type consumers
visit (which links H and 1 ) and low type consumers visit 1. However, when c _ c
1
the
publisher eliminates because it has a high price c(j
1
+j
J
) and receives few visitors. Hence,
only H and 1 are active and j
1
is higher. The second part of the corollary shows that for
: : the integrated publisher doesnt link 1, although for c <

c
1
it maintains active because
it attract low type consumers that can click through to H. Finally, when c _

c
1
the publisher
eliminates and increases j
1
.
6 Conclusions
In recent years, blogs, search engines, and news aggregators have reached the top positions
in audience ranking of news sites. While traditional publishers accuse these news sites of
stealing their contents and revenues, entrants argue that they are in fact expanding the
market by improving accessibility to newspapers and their contents. The reason for this is
that they reduce consumer search costs by oering links to many news sites and/or by editing
the contents originated by other rms. In this paper we have examined the product line
response of publishers to this type of entrants.
Our rst contribution has been to show that when publishers cant avoid its contents being
used by others versioning is less likely than what was previously considered in the literature.
The reason is that ghting versions improve the quality of competitors. We have also ana-
23
lyzed the linking and versioning strategies of a publisher in competition with a search engine.
We have demonstrated that the publisher can accept to be linked when it receives sucient
trac from the search engine. When this occurs, the search engine becomes an additional
distribution channel of the publisher that helps to weaken the eects of competition. In the
case of an integrated publisher, the search engine can set links to its competitor to benet of
the complementarity of the contents and increase its audience. To the best of our knowledge,
this results represents a novel nding in the literature on versioning, which hitherto has not
considered the possibility of links substituting ghting versions.
All in all, our results show that the disputes between publishers and search engines are
essentially dependent on the amount of trac that search engines send to the news sites and
on the complementarity of the aggregated contents. It is therefore an empirical question to
examine whether publishers obtain sucient trac from search engines to justify their links,
and whether search engines maximize their audience with their linking policies. The recent
papers of Chiou and Tucker (2010) and Athey and Mobius (2012) show evidence that Google
News generate signicant indirect trac, but it is still unclear which publishers are beneting
of search engines and how they adjust their marketing strategies to accomodate them.
16
Our analysis can be extended in several directions in order to understand recent develop-
ments in the media market. Here, we have focused on the optimal versioning strategy of a
publisher that can create news sites of dierent qualities; however, it would be interesting to
study how publishers accept the links to some pieces of news to attract visitors to their web
sites and to promote their premium contents (freemium model).
17
On the other hand, our
model could be extended to examine the role of news agencies in the media market. As the
recent conicts between Google News and some press associations illustrates, news agencies
provide fundamental information to traditional newspapers and they are very concerned about
their use by news aggregators.
16
Chiou and Tucker (2010) investigate the eects of the breakdown of the agreement between Google News
and AP in 2009, and show that the event was correlated with a decline in the demand for content from AP
sources. Athey and Mobius (2012) analyze the eect of introducing a local news feature in the French version
of Google News in 2009, whereby users were able to see news from local outlets on the web site.
17
Rutt (2012) considers how aggregators aect the pricing decisions of rms and the quality of their contents.
He assumes that rms randomize between providing the article for free and charging a price, and shows that
this behavior creates a mixture of advertiser and subscription business models.
24
7 Appendix
Proof of Proposition 1. When the publisher only oers H the indierent types are 0
1.
=
(j
1
j
.
),(1 ,)n and 0
.0
= j
.
,,n. The rm sets j
1
to maximize
1.(1)
= j
1
(1 0
1.
)
and as a result the price and the prots are:
j
1
=
j
.
+ n(1 ,)
2
; (1)

1.(1)
=
n(1 ,)
4
+
j
.
4
(2 +
j
.
n(1 ,)
). (2)
With this price, it is veried that 0
1.
=
1
2
_
1

A
u(1o)
_
0
.0
=

A
ou
as long as j
.
<
ou(1o)
2o
.
To ensure the participation of the blog we consider that this condition is always satised.
Imagine now that the publisher releases H and 1. In this case, the indierent types are
0
1J
= (j
1
j
J
),n(1 c); 0
J.
= (j
J
j
.
), [n(c (, + c(1 :))] ; and 0
.0
= j
.
,[n(, +
c(1 :))]. The rm sets j
1
and j
J
to maximize
1J.(1J)
= j
1
(1 0
1J
) +j
J
(0
1J
0
J.
).
Solving the rms problem we obtain:
j
1
=
1
2
[j
.
+ n(c: , + 1 c)]; j
J
=
1
2
[j
.
+ n(c: ,)]; (3)

1J.(1J)
=
n
4
[c: , + 1 c] +
j
.
4
[2 +
j
.
n(c: ,)
]; (4)
To ensure the participation of the blog now we need that : :
1
=
(o+2o)u
A

2
A
6o
A
u+o
2
u
2
2ou
.
If this is satised, we obtain that 0
1J
=
1
2
0
J.
=
1
2
_
1

A
(oso)u
_
0
.0
=

A
u(o+o(1s))
. On
the other hand, when : 6 :
1
the ranking is not preserved and the publisher sets j
J
=
o
A
o+o(1s)
to ensure that 0
J.
= 0
.0
= 0
J0
. Taking this into account, it sets j
1
to maximize
1J.(1J)
=
j
1
(1 0
1J
) + j
J
(0
1J
0
J0
). The prices and the associated prot are:
j
1
=
1
2
[(1 c)n +
2cj
.
, + c(1 :)
]; j
J
=
cj
.
, + c(1 :)
; (5)

1J.(1J)
=
n
4

c[(, + c(1 :))n 2j
.
]
2
4n(, + c(1 :))
2
. (6)
Moreover, the indierent types are: 0
1J
=
1
2
0
J0
=

A
(o+o(1s))u
.
Next, we analyze the publishers optimal versioning policy. Note that for : :
1
it is
satised that
1J.(1J)

1.(1)
for : :, where
25
: =
cj
2
.
+ c(1 ,)(c + ,)n
2
+
_
c
2
_
j
4
.
+ 2 (1 ,) j
2
.
n
2
(c + , 2) + (c ,)
2
(, 1)
2
n
4
_
2c
2
(1 ,) n
2
.
(7)
Therefore, the publisher releases H and 1 for : < : _ 1 and only H for : _ :. Notice that
this segmentation only emerges when : :
1
, which occurs when , < ,
1
, where ,
1
is the value
that satises : = :
1
(the expression for ,
1
is long and we dont report it for simplicity). For
, ,
1
, then : < :
1
and the publisher oers H and 1 for :
1
_ : _ 1 and the blog is always
active.
Consider now the case for : < :
1
. If the rm oers the two versions the relevant prices and
prot functions are those in (5) and (6), respectively. In this case, we obtain that
1J.(1J)

1.(1)
for : :
2
, where
:
2
=
2
_
c
2
(, 1)j
2
.
(j
2
.
2 (, 1) j
.
n + (, 1) ,n
2
) + c((c + ,)j
2
.
2(1 ,),j
.
n (c ,)(1 ,)(c + ,)n
2
c
2
[2 (, 1) j
.
n j
2
.
+ (c ,) (, 1)n
2
]
.
(8)
Hence, for :
2
< : < :
1
the publisher oers the two versions and the blog is not active. And
for : < :
2
it only releases H and the blog is active.Q.E.D.
Proof of Corollary 1. Consider the prot function of the publisher in (4) when it releases
the two versions and the cost of dierentiation is c(:) = 1,/:. For : :
1
the three news sites
have positive market share and the publishers prot is:

1J.(1J)
=
n
4
[c: , + 1 c] +
j
.
4
[2 +
j
.
n(c: ,)
]
1
/:
; (9)
The FOC of
1J.(1J)
with respect to : is
0
1J.(1J)
0:
=
1
/:
2
+
nc
4
_
1
j
2
.
(,n c:n)
2
_
.
This expression can only be negative if j
2
.
(,n c:n)
2
, which happens for :
_
uo
A
uo
,
uo+
A
uo
_
.
In spite of this, notice that :
1

uo+
A
uo
for j
.
0. Therefore, it must be that
Jt
HLA(HL)
Js
0.
Finally, for : _ :
1
the publisher sets the prices in (5) and the prots are those in (6).
Taking this into account, the FOC of the prot with respect to : is:
0
1J.(1J)
0:
=
1
/:
2

c
2
[2j
.
n(c(1 :) + ,)]j
.
n(c(1 :) + ,)
3
26
A sucient condition for this to be positive is n[c(1 :) + ,] 2j
.
. This is always satised
for j
.
<
uo(1o)
(2o)
, which has been previously assumed.Q.E.D.
Proof of Proposition 2. Imagine that the publisher releases both H and 1 and that
these are not linked by the search engine. In this case, the consumer indierent to H and 1 is
0
1J
= (j
1
j
J
),n(1c), the consumer indierent to 1 and is 0
J.
= (j
1
j
.
),n(cr), and
the consumer indierent between visiting and not visiting any site is 0
.0
= j
.
,nr. Taking
this into account, the rm sets j
1
and j
J
to maximize
1J.
= j
1
(1 0
1J
)+j
J
(0
1J
0
J.
),
which yields the following prices and prots:
j
1
=
1
2
[j
.
+ n(1 r)] ; j
J
=
1
2
[j
.
+ n(c r)] ; (10)

1J.
=
1
4
_
2j
.
+ n(1 r) +
j
2
.
n(c r)
_
<
n
4
.
Moreover, the audience for the search engine is
1J.
= 0
J.
0
.0
=
1
2


A
(2o:)
2(o:)u:
.
Notice that the publisher always releases 1 since in this case it doesnt create spillovers on
.
18
On the other hand, with the above prices we obtain 0
1J
=
1
2
0
J.
=
_
1
2


A
2u(o:)
_

0
.0
=

A
u:
as long as j
.
< j
.
=
u:(o:)
2o:
, which we assume for simplicity.
Imagine next that the publisher only releases H and that this is linked by the search
engine. In this case, the consumer indierent to H and is 0
1.
= j
1
(1 c),n(1 ,), and
the indierent to and not visiting any site is 0
1.
= cj
1
,n,. Taking this into account, the
publisher sets j
1
to maximize
1.(1)
= j
1
(1 0
1.
) + j
1
c (0
1.
0
.0
). As a result, the
price and the prots are:
j
1
=
,n(1 ,)
2(, 2,c + c
2
)
;
1.
=
,n(1 ,)
4(, 2,c + c
2
)
. (11)
Moreover, the search engines audience is
1.(1)
= 0
1.
0
.0
=
oo
2(o2oo+o
2
)
and it is
satised that 0
1.
=
o(1o)
2(o2oo+o
2
)
0
.0
=
o(1o)
2(o2oo+o
2
)
if c < ,.
When the publisher oers both H and 1 and these are linked the consumer indierent to H
and 1 is 0
1J
= (j
1
j
J
),n(1c), the indierent to 1 and is 0
J.
= (j
J
(1c)cj
1
),n(c:
18
When the publisher only releases 1 and there are not links the results are j
H
=
p
A
+(1r)u
2
and
HA
=
[p
A
+(1r)u]
2
4u(1r)
. It can be veried that
HLA

HA
for j
A
0.
27
,), and the indierent to and not visiting any site is 0
.0
= c(j
1
+j
J
),n[, +c(1 :)]. The
publisher sets j
1
and j
J
to maximize
1J.(1J)
= j
1
(1 0
1J
) + j
J
(0
1J
0
J.
) + (j
1
+
j
J
)c (0
J.
0
.0
). The solution of its problem is:
j
1
=
[c(1 c)(2, 1 + c) + ,(, 1 + 2c) + c(1 2, 2c): + c
2
(c 1 + :)
2
]n
2[4,c , c
2
c(1 : + c(3c + 4(: 1)))]
; (12)
j
J
=
[,(, c) c
2
(c
2
c(1 :) + :(1 :)) + c(,(1 + c 2:) c(1 c :))]n
2[4,c , c
2
c(1 : + c(3c + 4(: 1)))]
.
With these prices we observe that 0
1J
0
J.
if c

c = [, + c(1 :)],(1 + c), and
0
J.
0
.0
if c <

c. This implies that the publisher cant get a positive market share for
the two news sites. As an alternative, for c <

c the publisher can set j
J
=

H
(o(1o)oos)
o1+o+o(1os)
to guarantee that 0
1J
= 0
J.
= 0
1J.
, where now 0
1J.
is the consumer indierent between
visiting H and visiting when this links H and 1. Taking this into account, the rm sets
j
1
to maximize
1J.(1J)
= j
1
(1 0
1J.
) + c(j
1
+ j
J
) (0
1J.
0
.0
). The price and the
associated prots are:
j
1
=
[, + c(1 :)][, 1 + c + c(1 c :)]
2
n
2[,(1 ,) + 4(, 1),c + c
2
+ c(1 2,)(1 : 2c(2 c 2:)) + c
2
(c + : 1)(1 : + c(4: 3))]
; (13)

1J.(1J)
=
[, + c(1 :)][, 1 + c + c(1 c :)]
2
n
4[,(1 ,) + 4(, 1),c + c
2
+ c(1 2,)(1 : 2c(2 c 2:)) + c
2
(c + : 1)(1 : + c(4: 3))]
.
The audience for the search engine is
1J.
= 0
1J.
0
.0
, where:
0
1J.
=
(2c 1)(, + c(1 :))(, 1 + c + c(1 c :))
2[,(1 ,) + 4(, 1),c + c
2
+ c(1 2,)(1 : 2c(2 c 2:)) + c
2
(c + : 1)(1 : + c(4: 3))]
;
0
.0
=
c(1 , c + c(c + : 1))(1 2, + c(2: 1))
2[,(1 ,) + 4(, 1),c + c
2
+ c(1 2,)(1 : 2c(2 c 2:)) + c
2
(c + : 1)(1 : + c(4: 3))]
.
It can be checked that now 0
1J.
0
.0
for c <

c. On the other hand, for

c < c < ,
the publisher sets j
J
=

H
(oo+o(1+os))
o+oo(o+s1)
to guarantee that 0
1J.
= 0
.0
= 0
1J.0
, where
0
1J.0
is the consumer indierent between H and not visiting any site when the publisher
also releases 1 at j
J
and the price of is c(j
1
+ j
J
). As a result, the publisher maximizes

10(J.)
= j
1
(1 0
1J.0
) by setting j
1
=
u(o+oo(o+s1))
4o
and obtains 0
1J.0
= 1,2 and

10(J.)
=
u(o+oo(o+s1))
8o
. Notice that if : _ (1 ,) then , _

c, so the strategy described

for c
_

c, ,
_
is not relevant. Finally, for c , nobody wants to visit . Hence, the publisher
releases H, sets j
1
= n,2 and obtains the monopoly prot
1
= n,4.
28
Taking into account the previous results, next we examine the linking and versioning policy
of the rms. Consider rst the publisher when it releases H and 1 and compare its prots
with and without links. For c
_
0,

c
_
, we nd

c
1
such that
1J.

1J.(1J)
for c <

c
1
and

1J.
_
1J.(1J)
for

c
1
_ c <

c. The value of

c
1
is not shown for simplicity. Analogously,
in the region c
_

c, ,
_
there is

c
2
that satises
1J.(1J)
_
1J.
if c _

c
2
and
1J.

1J.(1J)
for

c
2
_ c < ,.
19
The expression for

c
2
is:

c
2
=
(c r)(, + c(1 :))n
2
2j
2
.
+ 4j
.
(c r)n + (1 + c 2r)(c r)n
2
.
Now compare these results with those that will obtain the publisher if it oers H and this
is linked. First, observe that in the interval c
_
0,

c
1
_
there is c
1
such that
1J.

1.(1)
for c < c
1
and
1.(1)
_
1J.
for c _ c
1
. The value for c
1
is:
c
1
= ,
_
(, 1),[j
2
.
+ 2j
.
(c r)n + (1 r)(c r)n
2
][j
2
.
+ 2j
.
(c r)n r(c r)n
2
]
[j
2
.
+ 2j
.
(c r)n + (1 r)(c r)n
2
]
;
Next, it is veried that in the interval c
_

c
1
,

c
_
there exists c
2
satisfying
1.(1)

1J.(1J)
for c < c
2
and
1J.(1J)
_
1.(1)
for c
2
_ c <

c. Also note that in the interval
c
_

c,

c
2
_
there is a threshold

c
3
that satises
10(J.)

1.(1)
for c <

c
3
. The expressions
for c
2
and

c
3
are not shown for simplicity. Furthermore, it is conrmed that c
1
<

c
1
< c
2
as
long as : < : with : = [c + , c(1 + c)] ,(1 +c).
20
Similarly, one can show that c
2
<

c <

c
3
if : (1 ,).
21
Finally, for values c
_

c
2
, ,
_
we nd that
1.(1)

1J.
.
Now consider the linking strategy of the search engine, which maximizes its audience. Let
c
3
and c
4
be the value of c that satisfy
1J.(1J)
=
1J.
and
1.(1)
=
1J.
, respectively.
The expression for c
3
is long and we dont present it
22
, and the expression for c
4
is:
c
4
=
_
_
_
c(4,j
.
+ nr(1 2,)) r(2,j
.
nr(1 2,))

_
[2,j
.
(2c r) + nr(1 2,)(c r)]
2
+ 4,j
.
(2c r) [r(j
.
rn) + c(rn 2j
.
)]
_
_
_
4cj
.
2rj
.
2crn + 2r
2
n
19
Recall that for c
b
c if two versions are released and linked j
H
is high and nobody visits .
20
When c = b c then c
1
=
b
c
1
= c
2
. For c b c it holds that c
1

b
c
1
and c
2
<
b
c
1
. This implies that the regions
c 2 [c
1
,
b
c
1
) and c 2 [
b
c
1
, c
2
) vanish and the relevant threshold is
b
c
1
.
21
When c = (1 o) then c
2
=
b
c =
b
c
3
. For c < (1 o) it holds that c
2

b
c and
b
c
3
<
b
c. This implies that the
regions c 2 [c
2
,
b
c) and c 2 [
b
c,
b
c
3
) vanish.
22
Note that c
3
only exists in the interval
h
0,
b
c

as
HLA(HL)
= 0 when c
b
c. If
HLA(HL)

HLA
for
the whole interval c 2
h
0,
b
c

then
b
c applies instead of c
3
.
29
One can show that c
3
_ c
4
if r _ r. We use this restriction in order to reduce the number
of possible cases and to focus on the interesting situations. Taking this into account, when
c [0, c
3
) it holds that
1J.
< 'i:
1J.(1J)
,
1.(1)
so the search engine accepts to link
the publishers newspapers regardless of the number of versions that are released. If c [c
3
, c
4
)
then
1.(1)

1J.
_
1J.
, implying that the search engine accepts to link H but not
both H and 1. Finally, for c [c
4
, 1] it is satised that
1J.
_ 'ar
1J.(1J)
,
1.(1)
so
it never links the publishers newspapers because this increases too much its price and reduces
the audience.
Last, we contrast the strategies of the two rms to derive the market outcome established
in the proposition. Note that for c < c
1
the publisher does not want to be linked by the search
engine in any situation, whereas for c c
4
is the search engine that does not want to link any
newspaper. In such cases the publisher releases both H and 1, and H, 1, and receive direct
visits. However, recall that when : _ : then c
2
< c
1
and the relevant threshold is

c
1
. In this
case, for c <

c
1
the publisher doesnt accepts the links (Panel I in Figure 2).
Consider now the interval c [c
1
, c
4
) in which the rms can reach an agreement. In this
case when :

< : < : it is veried that c

1
< c
2
< c
3
< c
4
.
23
Hence, when c [c
1
, c
2
) the
publisher only releases H, when c [c
2
, c
3
) it releases H and 1 (only H and (H1) are active)
and when c [c
3
, c
4
) it only releases H. Finally, for : : the relevant threshold is

c
1
. In this
case when c
_

c
1
, c
3
_
the publisher releases H and 1 (only H and (H1) are active), and
when c [c
3
, c
4
) the publisher only releases H. Q.E.D.
Proof of Corollary 2. Suppose that the publisher oers H and 1 and accepts the links
of the search engine. From Proposition 2 we know that when c <

c the rm sets the price
in (13) and obtains the corresponding prots
1J.(1J)
. Recall that for c <

c only H and
(H1) are active. Moreover, for c

c the search engine doesnt link the newspapers because
only H is active at the equilibrium price. Next, notice that 0
1J.(1J)
,0: _ 0 for c _

c,
implying that for c <

c the rm sets : = 1. Finally, notice that this result is veried for any
dierentiation cost c(:) = 1,/:.Q.E.D.
Proof of Proposition 3. When the publisher and the search engine dont reach an
agreement, the publisher releases H and 1 and its prices and prots
1J.
are those in (10).
23
For c

= c then c
2
= c
3
= c
4
and the intervals [c
2
, c
3
) and [c
3
, c
4
) shrink.
30
On the other hand, if only H is released and it is linked then j
1
and
1.(1)
are those in
(11).
Consider next that the publisher releases H and 1 and that the rms reach an agreement.
As a result, n
.
n
1
because n
.
= n[, + c(1 :)] and : < (, + c 1),c. In this case, the
indiferent consumers are 0
.1
= [c(j
1
+j
J
)j
1
],n[,+c(1:)1], 0
1J
= (j
1
j
J
),n(1c),
and 0
J0
= j
J
,nc. The rm sets the prices j
1
and j
J
to maximize
.(1J)1J
= c(j
1
+
j
J
) (1 0
.1
) + j
1
(0
.1
0
1J
) + j
J
(0
1J
0
J0
). Taking this into account, the prices and
the associated prots are:
j
1
=
c[, 1 + c(, + c : c(c + : 1))]n
2[, + c(c 2) + c(1 + c(3c 2) :)]
; j
J
=
cc[2, 1 c + c(1 + c 2:)]n
2[, + c(c 2) + c(1 + c(3c 2) :)]
(14)

.(1J)1J
=
c
2
[, 1 + c(3, 1 : c(3: 2))]n
4[, + c(c 2) + c(1 + c(3c 2) :)]
. (15)
With these prices, the audience for the search engine is
.(1J)1J
= 1 0
.1
. Moreover,
it is guaranteed that 0
.1
0
1J
0
J0
as long as c c
1
, where c
1
= [, +c(1 :)],(1 +c):
24
0
.1
=
c[c 1 + c(3c 1)]
2[, + c(c 2) + c(1 + c(3c 2) :)]
;
0
1J
=
c[, 1 + c(2c :)]
2[, + c(c 2) + c(1 + c(3c 2) :)]
; (16)
0
J0
=
c[2, 1 c + c(1 + c 2:)]
2[, + c(c 2) + c(1 + c(3c 2) :)]
.
Imagine rst that c c
1
. Then there is a value c
3
for which
.(1J)1J
=
1J.
. Then,
for : < (, + c 1),c it holds that
.(1J)1J
_ 'ar
_

1.(1)
,
1J.
_
when c [c
1
, c
3
) and

1J.
_ 'ar
_

1.(1)
,
.(1J)1J
_
for c [c
3
, 1].
25
Consider next that c _ c
1
. In this case, with the prices in (14) it is satised that 0
.1
_ 0
J0
,
which implies that only has direct visits. Taking this into account, the consumer indiferent
between and not visiting any news site is 0
.0
=
o(
H
+
L
)
u(o+o(1s))
. In order to ensure that
0
.0
= 0
J0
the publisher sets j
J
=

H
oo
o+o(1so)
. Taking this into account, it sets j
1
to
24
The expression for c
1
is the same than in the proof of Proposition 2. Indeed, this is the value of c for which
the three indiferent types are the same, which has to be the same c in the two propositions.
25
Recall from the prof of Proposition 2 that for c c
1
it is satised that
HLA

HA(H)
and c
1
c
3
holds
if v < v, which is assumed.
31
maximize
.(1J)
= cj
1
(1 0
.0
) + cj
J
(1 0
.0
). This yields:
j
1
=
, + c(1 : c)n
2c
; j
J
=
cn
2
; (17)

.(1J)
=
[, + c(1 :)]n
4

n
4
. (18)
Note that
.(1J)

u
4

1J.
. Furthremore, in the interval c [0, c
1
) it is satised that

.(1J)
_ 'ar
_

1.(1)
,
1J.
_
for the relevant values of :.
Consider now the linking strategy of the search engine. It can be checked that for c [0, c
1
)
it holds
1J.
_ 'i:
_

1.(1)
,
.(1J)
_
so the search engine wishes to link the publishers
content independently of the number of versions released. Furthermore, when c [c
1
, c
3
)
it holds that
.(1J)

1J.
_
1.(1)
and when c [c
1
, c
2
) that
.(1J)1J
_
1J.

1(.)1
, where c
2
is the value that satises
.(1J)1J
=
1J.
and c
1
_ c
2
.
26
Hence, when c
[c
1
, c
2
) the search engine only links the content maker when this releases the two versions. Last,
it can be shown that in the interval c [c
2
, 1] we obtain
1J.
_ 'ar
1.(1)
,
.(1J)1J
,
which implies that the search engine doesnt link the publisher.
Finally, we compare the linking strategies of the two rms. Taking into account that
c
2
< c
3
the publisher releases both H and 1 and these are linked when c [0, c
2
). On the
other hand, the two versions are releases but are not linked for c [c
2
, 1].Q.E.D.
Proof of Corollary 3. The proof of Proposition 3 has shown that in the interval c
1
_
c < c
2
the publisher releases the two versions and that these are linked by the search engine.
Taking into account the prots in (14) and the costs c(:) =
1
|s
the FOC of the prot with
respect to : is:
0
_

.(1J)1J

1
|s

0:
=
1
/:
2

cc
2
[c + c(3c 1) 1]
2
n
4[, c(2 c) + c(1 c(2 3c) :)]
2
.
Dene :

as the value of : that satises 0

_

.(1J)1J

1
|s

,0: = 0. The expression for

:

is not shown for simplicity but it can be shown that 0

_

.(1J)1J

1
|s

,0: 0 for : < :

and that :

is a decreasing function of /. Taking this into account, we can dene / such that
if / < / then :

(o+o1)
o
and / / such that :

<
o+oo(1+o)
o
, where:
27
/ =
4c[1 + c (c + c(3c 2) 2)]
2
(c + , 1)
2
c
2
[c + c(3c 1) 1]
2
n
; / =
4c
[c + , (1 + c)c]
2
n
;
26
When j
A
!0 then
HLA
!
A(HL)
= 12 and as a consequence c
2
! c
1
. Graphically, this implies that
in Figure 3 region B vanishes.
27
Note that c <
+(1+)

is equivalent to c < c
1
. Hence, it denes the domain of our analysis.
32
Suppose now that c < c
1
and therefore :
_
0,
o+oo(1+o)
o
_
. In this case the publisher
sets the prices in (17) and obtains the prots in (18). Taking this into account and the cost
function, we nd that the FOC of the prot with respecto to : is:
0
_

.(1J)

1
|s

0:
=
1
/:
2

cn
4
It is easy to see that, 0
_

.(1J)

1
|s

,0: < 0 when : :

=
2
p
o
p
|
p
u
. Moreover, as :

is decreasing in / we can nd a value /

such that when / < /

then :

o+oo(1+o)
o
so
the publisher always wants a high degree of dierentiation. Oppositely, when / then
:

0 and the publisher always wants the versions to be perfect substitutes. Finally, one
can show that /

= /.Q.E.D.
Proof of Proposition 4.
Consider rst the case where the integrated publisher doesnt link 1. The consumer in-
dierent to H and 1 is 0
1J
= (j
1
j
J
) ,n(1 c), the consumer indierent to 1 and
is 0
J.
= (j
J
cj
1
) ,n(c ,), and the consumer indierent to and not visiting any site
is 0
.0
= cj
1
,,n. Taking this into account, the integrated publisher sets j
1
to maximize

1
1J.(1)
= j
1
(1 0
1J
) +cj
1
(0
J.
0
.0
). On the other hand, the prot of the independent
publisher is
2
1J.(1)
= j
J
(0
1J
0
J.
). Solving the rms problem we get:
j
1
=
,
_
cj
J
nc
2
,(n + j
J
) + c[n(1 + ,) + j
J
(1 c)]

2
_
c(, + c
2
) ,
2
c
2
c
2
; (19)

1
1J.(1)
=
, [(c , + c cc) j
J
+ (1 c)(c ,)n]
2
4(1 c)(c ,)
_
(c ,), + (1 c)cc
2

n
, (20)

2
1J.(1)
= j
J
_
_
_
_
2(1 c),(c ,)c + (1 c)(, c(2 ,))c
2
(c ,) ,(2 c ,)
_
j
J
(1 c)(c ,),(, c(1 c) c)n
_
_
_
2(1 c)(c ,)
_
(c ,), + (1 c)cc
2

n
. (21)
With this price it holds that:
0
1J
=
1
n(1 c)
_
,j
J
(c , + c cc) + (1 c)(c ,),n
2(1 c)cc
2
+ 2,(c ,)
j
J
_
0
J.
=
1
n(c ,)
_
j
J

,c [j
J
(c , + c cc) + (1 c)(c ,)n]
2(1 c)cc
2
+ 2,(c ,)
_
0
.0
=
cj
J
(c(1 c) ,) + (1 c)(c ,)cn
2
_
(c ,), + (1 c)cc
2

n
.
33
In order to ensure the participation of the independent publisher we assume j
J
<
(1o)(oso)u
2o(2s)o
.
28
Note that the above prices are valid as long as 0
1J
0
J.
0
.0
. However, for c _

c
1
we nd
that 0
.0
_ 0
J.
, where

c
1
=
c
2
(j
J
+ n(1 + ,)) nc
3
c,(n + j
J
)
_
c(c ,)
_
c(c ,) (j
J
+ n(1 c))
2
8(1 c),j
2
J
_
2(1 c)cj
J
.
(22)
Therefore, when c _

c
1
doesnt receive visits and the relevant indierent consumers are
0
1J
= (j
1
j
J
) ,n(1 c) and 0
J0
= j
J
,nc. In this case, the prots are
1
1J
= j
1
(10
1J
)
and
2
1J
= j
J
(0
1J
0
J0
). Solving for j
1
yields:
j
1
=
n(1 c) + j
J
2
; (23)

1
1J
=
[n(1 c) + j
J
]
2
4(1 c)n
;
2
1J
=
j
J
2
_
1
(2 c)j
J
(1 c)cn
_
. (24)
With this price it is satised that 0
1J
=
1
2
_
1

L
u(1o)
_
0
J0
=

L
ou
. Notice, however,
that
1
1J

1
1J.(1)
for c _ c

, where
c

=
2(c ,),j
J
(j
J
+ n cn)
(c
2
,)j
2
J
+ 2(1 c)c(c ,)j
J
n(c 1)
2
c(c ,)n
2
<

c
1
.
Therefore, the integrated publisher oers for c < c

and it eliminates it otherwise.

Consider now the case where the aggregator links 1. The quality of is n
.
= n[, + c(1 :)]
and the indierent types are 0
1J
= (j
1
j
J
),n(1 c), 0
J.
= [j
J
c (j
1
+ j
J
)] ,n(c: ,)
and 0
.0
= c (j
1
+ j
J
) ,n[, + c(1 :)]. The integrated publisher sets j
1
to maximize
1
1J.(1J)
=
j
1
(1 0
1J
) + cj
1
(0
J.
0
.0
) and the prot of the independent publisher is
2
1J.(1J)
=
j
J
(0
1J
0
J.
) + cj
J
(0
J.
0
.0
). Solving the problem yields:
j
1
=
j
J
_
,(c ,) + c
2
(c
2
c(1 :) + :(1 :)) c(,(1 + c 2:) + c(c + : 1))

+
2
2t
; (25)

1
1J.(1J)
=
_
j
J
_
,(c ,) + c
2
(c
2
c(1 :) + :(1 :)) c(,(1 + c 2:) + c(c + : 1))

2
4t
. (26)
where t =
_
c(2,: + c
2
,) ,
2
c
2
(c
2
(1 :):)

and = (c 1) (c: ,) (c + , c:) n.

28
This is a general condition. When c = 1 it simplies to j
L
<
(1)()u
2
.
34
We dont show
2
1J.(1J)
for simplicity. The price in (25) implies that:
0
1J
=
j
J
_
,(c + ,) + c
2
(3c
2
c(1 :) :(1 :)) + c(c c(3c + :) ,(c + 2: 1))

+
2(1 c)nt
;
0
J.
=
j
J
_
_
_
c
_
(1 + : c)c
2
+ ,(4: + c(3 + c 6:) 2)

,
_
,(2 3c) + c
2

c
2
_
3c(1 :): + c
2
(1 + :) c
3
2(1 :)

_
_
_
+
2(c: ,)nt
;
0
.0
=
cj
J
_
,(c 3,) + c(c(1 + c :) ,(3 + c 6:)) c
2
(c + c
2
c: 3(1 :):)

+
2 [c(1 :) + ,] nt
.
The relation 0
1J
0
J.
0
.0
is satised for c <

c
2
(we dont show

c
2
because it is a
long expression). When c _

c
2
it holds that 0
J.
_ 0
.0
so nobody visits . In this case, the
independent publisher sets the price in (23) and obtains the prots in (24).
Next, we examine the linking strategy of both publishers taking into account the prot
functions dened above. In the case of the integrated publisher, notice that in the interval
c [0, c

) it is veried that
1
1J.(1J)

1
1J.(1)
. On the other hand, we dene as c
1
the value
of c for which
1
1J.(1J)
=
1
1J
. As a result, for c [0, c
1
) we obtain that
1
1J.(1J)

1
1J
and for c [c
1
, 1) that
1
1J
_
1
1J.(1J)
. Finally, notice that c
1
is a decreasing function in
: and that c

= c
1
for : = :. Hence, for : (0, :] the integrated publisher wants to link 1 if
c [0, c
1
) and only oers H if c [c
1
, 1).
On the other hand, in order to analyze the region : [ :, 1) we dene as c
2
the value of c
for which
1
1J.(1J)
=
1
1J.(1)
. Therefore, for c < c
2
we obtain that
1
1J.(1J)

1
1J.(1)
and for c [c
2
, 1) that
1
1J.(1)
_
1
1J.(1J)
. Also notice that c
2
= c

= c
1
for : = : and
c
2
= 0 for : = 1. Taking this into account, in the region : [ :, 1) the integrated publisher
wants to link 1 if c [0, c
2
), maintains without linking 1 if c [c
2
, c

], and only oers H if

c [c

, 1).
To examine the preferences of the independent publisher we dene as c
3
the values of c for
which
2
1J.(1)
=
2
1J.(1J)
. In the space c, : this denes a set of parameter combinations
for which
2
1J.(1J)
_
2
1J.(1)
. In the rest of cases the publisher dont accept to be linked.
Finally, we identify in which situations the two publishers will reach an agreement over
1. When , ,

it can be shown that c

2
intersects c
3
twice in the c, : space: call these
roots (

c
1
, :
1
) and (

c
2
, :
2
), where :
1
< :
2
. In particular, when : :
1
and : < :
2
it is satised
that c
2
c
3
and therefore the two rms agree on setting the link to 1. For :
1
_ : _ :
2
the
35
agreement is not protable for one of the publishers. Q.E.D.
Proof of Proposition 5. The results for the case in which the integrated publisher
doesnt link 1 are shown in the proof of Proposition 4. Consider now the case where links
1 and : <
o+o1
o
. In this case it is satised that n
.
n
1
n
J
and the relevant indierent
consumers are: 0
.1
= [c (j
1
+ j
J
) j
1
] ,n[, + c(1 :) 1], 0
1J
= (j
1
j
J
) ,n(1 c),
and 0
J0
= j
J
,cn. According to this, the integrated publisher sets j
1
to maximize
1
.(1J)1J
=
cj
1
(1 0
.1
) + j
1
(0
.1
0
1J
). The resulting price and prots are:
j
1
=
j
J
[, + c(1 c) + c(1 c(1 c) :) 1] (1 c) c [1 , c(1 :)] n
2
_
, 2(1 c)c + (1 c)c
2
c:
; (27)

1
.(1J)1J
=
[j
J
[, + c(1 c) + c(1 c(1 c) :) 1] (1 c) c [1 , c(1 :)] n]
2
4(1 c) [1 , c(1 :)] [, (1 c)(2 c)c + c:] n
. (28)
The prots of the independent publisher
2
.(1J)1J
are not shown for simplicity. The price
in (27) leads to the following expressions for the indierent consumers:
0
.1
=
_
_
_
j
J
[1 + c(3, c(2 c)) , 2c c(1 : c(2 + (2 c)c 3:))]
(1 c)(1 c)c(, + c(1 :) 1)n
_
_
_
2 [1 , c(1 :)] [c: + (1 c)(1 c)c ,] n
; (29)
0
1J
=
j
J
_
5c 3c
2
1 , + c(1 + c(3c 5) + :)

+ (1 c)c(, + c(1 :) 1)n

2(1 c) [, (1 c)(2 c)c c:] n
;
0
J0
=
j
J
cn
.
With the above prices, it can be shown that 0
.1
< 0
1J
when : < :
1
, implying that any
consumer visits H directly. The value :
1
that satises that 0
.1
= 0
1J
is:
:
1
=
_
_
_
[c + 2, 1 2(1 c)c + 3(1 c)c
2
]j
J
(1 c)
_
(1 c)
2
(1 c)
2
c
2
n
2
2(1 c)(1 c)c[1 + c(5c 4)]nj
J
+ [1 c(8 c(18 c(16 9c)))] j
2
J
_
_
_
2c[j
J
(1 c)cn]
Notice that for : < :
1
the relevant indierent consumers are 0
.J
= [c (j
1
+ j
J
) j
J
] ,n(, c:)
and 0
J0
= j
J
,cn. As a result, the publisher sets j
1
to maximize
1
.(1J)J
= cj
1
(1 0
.J
),
which leads to the following results:
36
j
1
=
j
J
(1 c) + (, c:)n
2c
; (30)

1
.(1J)J
=
[j
J
(1 c) + (, c:)n]
2
4(, c:)n
; (31)

2
.(1J)J
=
j
J
_
c(1 + c)(, c:)n j
J
_
2, + c
_
(1 c)
2
2:
___
2c(, c:)n
.
With this price we also observe that 0
.J
=
1
2
_
1
(1o)
L
(oos)u
_
0
J0
=

L
ou
.
Suppose now that : < :
1
. Then, it can be shown that
1
.(1J)J

1
1J.(1)
for c <

c
1
,
where

c
1
is dened as in (22). On the other hand,
1
.(1J)J
_
1
1J
for c _ c
1
, where
c
1
= 1 +
(1 c)j
J
(, c:)
_
(1 c)j
2
J
(1 c + j
J
)
2
(, c:)
(1 c)j
2
J


c
1
.
Taking this into account, if : < :
1
the aggregator is never active if c c
1
and it wants to
link 1 if c < c
1
. In the later case, high type consumers visit and intermediate consumers 1.
Next, dene as :
2
the value of : for which indierent consumers in (??) satisfy 0
.1
= 0
J0
.
:
2
=
_
2, (c + , 1)c 4c + (3 + c)cc + (2 c c
2
)c
2

j
J
(1 c)(c + , 1)ccn
c[(2 c)j
J
(1 c)ccn]
Notice that :
1
intersects :
2
once at the c, : space: call this root (

c, :). For (

c, :) it is
veried that 0
.1
= 0
1J
= 0
J0
. Moreover, for : : and : < :
2
we nd that 0
1J
< 0
J0
and
for : < : and : < :
1
that 0
.1
< 0
1J
.
For : < :
2
1 does not receive direct visits and the relevant indierent consumers are
0
.1
= [c (j
1
+ j
J
) j
1
] ,n(, + c(1 :) 1) and 0
10
= j
1
,n. Thus, the publisher sets j
1
to maximize
1
.(1J)1
= cj
1
(1 0
.1
) + cj
1
(0
.1
0
10
), which yields:
j
1
=
c[(c 1)j
J
(, + c(1 :) 1)n]
2[c: (c + , + (c 2)c)]
; (32)

1
.(1J)1
=
c
2
[(c 1)j
J
(, + c(1 :) 1)n]
2
4(, + c(1 :) 1)[c + , + (c 2)c c:]
; (33)

2
.(1J)1
= cj
J
[1
c[j
J
(2, 1 + (c 2)c + 2c(1 :)) + (c 1)(, + c(1 :) 1)n]
2(, + c(1 :) 1)[c + , + (c 2)c c:]
]. (34)
With this price we obtain that 0
.1
0
10
if j
J
is suciently high.
Finally, let :
3
be the value of : for which the indierent consumers in (29) satisfy 0
.1
= 1.
:
3
=
1
4nc
2
_
_
_
c[2(c + 2, 1)n j
J
(3c 1) (1 c)(3 c)cn]

_
c
2
_
((3c 1)
2
j
2
J
+ 2(1 c)(1 c)(2 c(3 + c))j
J
n + (1 c)
2
(2 3c + c
2
)
2
n
2

_
_
_
37
This implies that for : :
3
is never active and consumers only visit H or 1.
Taking this into account we now determine the linking strategy of the publishers. For : < :
and : :
1
and for : : and :
2
< : < :
3
all , H and 1 are active when links 1, and
the price and prots are those in (27) and (28) (see Figure 5). In this region, it is satised
that
1
1J.(1)

1
.(1J)1J
for c <

c
1
and
1
1J

1
.(1J)1J
for c _

c
1
.
29
Therefore, the
integrated publisher does not want to link its competitor. For : : and : < :
2
only and H
receive direct visits when links 1, and the price and prots are those in (32) and (33). In
this interval, it can be found that
1
.(1J)1
<
1
1J.(1)
for c <

c
1
and that
1
1J

1
.(1J)1
for c

c
1
.
30
As before, the integrated publisher does not want to link 1. Finally, for : :
and : _ :
3
no consumer visits . As a consequence, the optimal price and prots are those in
(19) and (20) if c <

c
1
, or the ones in (23) and (24) otherwise.
Consider now the linking strategy of the independent publisher. For : _ :
1
it holds that

2
.(1)J

2
1J.(1)
when c <

c
1
if j
J
0 and
2
.(1)J

2
1J
for c _

c
1
if j
J
0. These
results imply that both publishers agree to set a link on 1 for c < c
1
. When : :
1
the
integrated publisher is not interested on linking 1. Q.E.D.
References
[1] Anderson, E. and J. Dana (2009), When is Price Discrimination Protable?, Management
Science, Vol. 55, 6, 980-989.
[2] Athey, S. and M. Mobius (2012), The Impact of News Aggregators on Internet News
Consumption: The Case of Localization, mimeo.
[3] Argentesi E. and L. Filistrucchi (2007), Estimating market power in a two-sided market:
The case of newspapers, Journal of Applied Econometrics, 22, 7, 1247-1266.
[4] Bhargava, H.K. and V. Choudhary (2001), Information Goods and Vertical Dierentiation,
Journal of Management Information Systems, 18 (2), 85-102.
[5] Bhargava, H.K. and V. Choudhary (2008), When is Versioning Optimal for Information
Goods?, Management Science, 54 (5), 1029-1035.
29
Notice that
1
HL
=
1
A(HL)HL
for c c
3.
30
Notice that
1
HL
=
1
A(HL)H
for c c
2.
38
[6] Calzada, J. and T. Valletti (2012), Intertemporal Movie Distribution: Versioning When
Customers can Buy Both Versions, Marketing Science, 31, 649-667.
[7] Chiou, L. and Tucker, C.E. (2010), Copyright, Digitalization and Aggregators, Mimeo.
[8] De Cornire, A. and G. Taylor (2013) Integration and search engine bias, Mimeo.
[9] Dellarocas, C., Katona, Z. and Rand W. (2010), Media, Aggregators and the Link Econ-
omy: Strategic Hyperlink Formation in Content Networks. NET Institute, WP #10-13.
[10] Dertouzos J. and Trautman W. (1990), Economic Eects of Media Concentration: Esti-
mates from a Model of the Newspaper Firm, Journal of Industrial Economics, Vol 39, 1,
1-14.
[11] Jeon, D.S. and N. Nasr (2012), News Aggregators and Competition Among Newspapers
in the Internet, mimeo.
[12] Johnson, J. and D. Myatt (2003), Multiproduct Quality Competition: Fighting Brand
and Product Line Pruning, American Economic Review, Vol 93 (3), 748-774.
[13] Fritjers P. and M. Velamuri (2010), Is the Internet Bad News? The Online News Era and
the Market for High-Quality News, Review of Network Economics, 9 (2), Article 2.
[14] FTC (2010), Potential Policy Recommendations to Support the Reinvention of Journalism.
Federal Trade Comission, Sta Discussion Draft:
[15] Gentzkow M. and J. Shapiro (2008), Competition and Truth in the Market for News,
Journal of Economic Perspectives, 22 (2), 133-154.
[16] Gentzkow M. (2007), Valuing New Goods in a Model with Complementarity: Online
Newspapers, American Economic Review, 97 (3), 713-744.
[17] George L. (2008), The Internet and the Market for Daily Newspapers, B.E. Journal of
Economic Analysis & Policy, 8 (1), Article 26.
[18] George, L. and C. Hogendorn (2012), Aggregators, search and the economics of new media
institutions, Information Economics and Policy, 24, 40-51.
39
[19] Google Inc. (2010), Comments on Federal Trade Commissions News Media Workshop and
Sta Discussion Draft on Potential Policy Recommendations to Support the Reinvention
of Journalism. 20 July 2010.
[20] Isbell K. (2010), The Rise of the News Aggregator: Legal Implications and Best Practices.
The Berkman Center for Internet & Society, Harvard University, Research Publication
Series, No 2010-10.
[21] Kaiser U. and J. Wright (2006), Price structure in two-sided markets: Evidence from the
magazine industry, International Journal of Industrial Organization, 24, 1-28.
[22] Katona, Z. and M. Sarvary (2008), Network Formation and The Structure of the Com-
mercial World Wild Web, Marketing Science, 27 (5), 764-778.
[23] Moorthy, K. (1984), Marketing Segmentation, Self-Selection, and Product Line Design,
Marketing Science, 3(4), 288-307.
[24] Moorthy, K. and I. Png (1992), Market Segmentation, Cannibalization, and the Timing
of Product Introductions, Management Science, 38 (3), 345-359.
[25] Mussa, M. and S. Rosen (1978), Monopoly and Product Quality, Journal of Economic
Theory, 18 (2), 301-317.
[26] Newspapers Association of America (2011), Trends and Numbers.
http://www.naa.org/Trends-and-Numbers.aspx [Accessed 7th July 2011].
[27] Project for Excellence in Journalism (2006-2011) The State of the News Media. Annual
Report on American Journalism. Pew Research Center, 2006-2011.
[28] Reddaway W. (1963), The Economics of Newspapers, Economic Journal, 73, 201-218.
[29] Rosse J. (1967), Daily Newspapers, Monopolistic Competition and Economies of Scale,
American Economic Review, Papers and Proceedings of the Seventy-ninth Annual Meeting
of the American Economic Association, 57 (2), 522-533.
[30] Rosse, J. (1970), Estimating Cost Function Parameters Without Using Cost Data: Illus-
trated Methodology, Econometrica, 28 (2), 256-275.
40
[31] Rutt, J. (2012), Aggregators and the News Industry: Charging for Access to Content,
mimeo.
[32] Salant, S.W. (1989), When is Inducing Self-Selection Suboptimal for a Monopolist?, Quar-
terly Journal of Economics, 104, 391-397.
[33] Stokey, N. (1979), Intertemporal Price Discrimination, Quarterly Journal of Economics,
93, 355-371.
[34] The Economist (2006), Who Killed the Newspapers?, 24th August 2006. Available at
http://www.economist.com/node/7830218 [Accessed 9th July 2011].
[35] Thompson R.S. (1984), Structure and Conduct in Local Advertising Markets: the Case
of Irish Newspapers, Journal of Industrial Economics, 33(2), 241-249.
[36] U.S. Census Bureau. Industry Statistics Sampler, Newspapers Pub-
lishers, NAICS 511110 [Internet]. 4th May 2011. Available at
http://www.census.gov/econ/industry/current/c511110.htm [Accessed 9th July 2011].
[37] Valletti, T. and S. Szymanski (2006), Parallel Trade, International Exhaustion and In-
tellectual Property Rights: A Welfare Analysis, The Journal of Industrial Economics, 54
(4), 499-526.
[38] Wu, S., P. Chen, and G. Anandalingam (2003), Fighting Information Good Piracy with
Versioning, ICIS Proceedings. Paper 51.
41