Bergemann and Bonatti (2015) characterize the demand for cookies for a given price
P
c
. An advertiser may buy from a single interval that includes either the lowest-value or
the highest-value consumer. Alternatively, he may want to buy from two intervals, one
of which includes the lowest-value consumer and the other the highest-value consumer.
Bergemann and Bonatti (2015) then determine the optimal pricing by the data provider.
The data provider may limit the amount of data being bought in equilibrium, as the data
provider maximizes its profits. Hence, consumers benefit if the data provider enjoys some
market power because, absent market power of the data provider, advertisers would
become fully informed about the match value.
To put
Bergemann and Bonatti (2015) into broader perspective, we note that the
publisher may be vertically integrated with the data provider. In this case, the publisher
has three potential sources of revenues: First, the publisher can charge consumers directly
for its content services. Second, the publisher can charge advertisers for offering adver-
tising space (to the extent that consumers dislike advertising, this constitutes an indirect
charge to consumers). These two revenue sources are well established in media econom-
ics. Third,
Bergemann and Bonatti (2015) formalize that the publisher may also offer
advertisers data services for which it can charge a fee. As this allows advertisers to better
extract surplus from consumers, providing this information constitutes another indirect
charge to consumers.
Tracking allows advertisers to use a sophisticated targeting strategy. While the above
models provide relevant insights into the role of the tracking technology, they are not
embedded in a dynamic environment. In particular, consumers may search within a
product category on a particular website but do not close their session with a purchase.
Possible reasons are that they are still in doubt or that they decide that the offer is dom-
inated by the outside option, which may include the possibility of searching again in the
future. Due to tracking, advertisers become aware of the identity of such consumers.
They may therefore “redouble” their efforts and retarget these consumers via advertising
on this website or specific offerings on another website when consumers are visiting a
website that is affiliated with the ad network selling ads to advertisers. For instance, Face-
book has introduced retargeted ads in its users’ newsfeed. With dynamic retargeting, the
ad network identifies people with the help of the individual cookie profile and recalls the
exact product a consumer has looked at before. With this information, it targets the con-
sumer by displaying the same or a related product offered by the firm visited before. This
is in contrast to generic retargeting, which uses cookie information only to select the firm
of which an ad is shown (see
Lambrecht and Tucker, 2013). It has been claimed that
dynamic retargeting is four times more effective than generic retargeting and six times
more effective than generic advertising using banner ads (
Criteo, 2010).
Lambrecht and Tucker (2013) provide a detailed assessment of the effectiveness of
dynamic retargeting, based on a field experiment with data from a travel website that sells
hotel accommodations and vacation packages. In the field experiment, consumers were
521The Economics of Internet Media
randomly assigned generic and dynamic retargeted ads when they visited an external
website affiliated with the ad network. The generic ad showed a generic brand ad for
the travel website, while the dynamic retargeted ad showed the hotel the consumer
had looked at before, plus a few similar offerings. Perhaps surprisingly, the authors find
that, on average, dynamic retargeting is not more effective, where effectiveness is mea-
sured by the probability that the consumer makes a purchase on the travel website within
a specified time interval.
A possible explanation for the failure of dynamic retargeting vis-à-vis generic retar-
geting is that consumers may have had only a loose idea of what they were looking for (as
argued by
Lambrecht and Tucker, 2013). Thus, a generic ad may be more effective since
consumers may have figured out that the specific offer they previously looked at was
actually not what they wanted. However, such behavior seems less likely if consumers
dedicate quite some effort on their search and make further inquiries. In this case, it
appears likely that consumers have well-specified ideas of what they liked and may be
inclined to revisit their previous searches.
Lambrecht and Tucker (2013) try to proxy this
by identifying consumers who used a review site. The empirical result supports the view
that dynamic retargeting is more effective for those consumers. These findings suggest a
rather complex effect of targeting on purchase intent. It suggests that simple and universal
messages of how (re)targeting will affect market structure may be too much to expect.
10.6.2 Advertising Congestion and Limited Attention
Advertisers post ads to leave an impression on consumers and to make a profit by selling a
product or service or by being able to sell at a higher price. To achieve this, advertisers
have to overcome several hurdles. First, consumers must not block advertising; second,
consumers have to remember the ad (remember the advertised product, product features,
or the ad experience as a complement) when making a purchase decision; and third,
advertisers must be able to make money from such a consumer. We touched upon
the first hurdle in
Section 10.6.1 in the context of targeting ( Johnson, 2013). The third
hurdle arises, for example, with competing advertisers in a Bertrand world. If a consumer
receives two competing ads, the ad is essentially a wasted impression for the advertiser
(
Taylor, 2013b). The second issue is one of limited attention, which we did not elaborate
on in the context of advertising. Advertising congestion can be seen as a mismatch
between advertisers and consumers, since high-value advertisers may not capture con-
sumers’ attention and thus are not matched, whereas some lower-value advertisers do
manage to gain the consumer’s attention.
Anderson and de Palma (2009) formalize information congestion, postulating that
consumer attention spans are limited (for a formal presentation of the model in a different
context, see
Section 10.3.2). Under “open access” to attention (for example, through
billboards or bulk mail), attention is a common property resource to which all consumers
522 Handbook of Media Economics
have access. This common property resource tends to be excessively exploited, resulting
in congestion in equilibrium. By contrast, a monopoly gatekeeper prices out congestion.
Anderson and Peitz (2014a) show how this approach can be integrated into a model of
competing media platforms. Here, we illustrate their setting, closely following
Anderson
et al. (2012)
. To see how advertising congestion changes the nature of media competi-
tion, consider, first, the situation with a fixed amount of time spent by a representative
user on each of the n media platforms. Website quality maps into usage time. The idea
here is that consumers surf the web, spend more time on higher-quality websites and less
on lower-quality ones. Advertisers are supposed to offer totally differentiated products.
Thus, advertisers are monopolists vis-à-vis consumers. Suppose, furthermore, that adver-
tisers extract the full expected surplus from consumers.
Consumers access pages of the different websites in random order. If they visit more
pages of a website, they will be exposed to more ads from that website. Websites are
assumed to benefit from industry-wide perfect tracking. If the value of the marginal
advertiser is larger than the expected value of a repeated impression of a given ad (which
holds if congestion is not too severe), all consumers will see an ad only once. Ads will be
placed randomly. Even though advertising does not affect media consumption, the web-
site will not post an unlimited number of ads. This is so because it cannot discriminate
between different types of advertisers and thus has to lower the ad price as it takes in more
advertisers.
Each media website decides how many ads to place at an initial stage since this is
mostly a question of how to design the website or how to structure the bundle of content
and advertising. Denote by σ
i
the amount of time spent on website i ¼1, … , n, and by
σ
0
the time spent not using the Internet (the outside option), and normalize the total time
available to 1. If website i shows a
i
ads and a consumer spends all her time on this site, the
consumer’s total exposure is a
i
. To make a match between an advertiser and any given
consumer, this consumer must be exposed to the corresponding ad and she must recall the
ad. Each consumer will see Γ ¼
P
n
i¼1
a
i
σ
i
ads in total. However, if the (fixed) attention
span φ of a consumer is less than this number—i.e., φ < Γ—the consumer will not
remember some of the ads. Hence, for these ads no match is formed. Advertisers are
ranked according to the willingness-to-pay to contact potential consumers. Hence,
the a-th advertiser is willing to pay p(a) to attract the attention of a consumer. With con-
gestion, the willingness-to-pay of the a-th advertiser reduces to p(a)φ/Γ. With a
i
ads on
platform i, the ad price conditional on making an impression must be equal to the
willingness-to-pay of the marginal advertiser—i.e., p(a
i
)φ/Γ. Since the ad makes an
impression with probability σ
i
, each ad generates revenue σ
i
p(a
i
)φ/Γ. Thus, website i
maximizes its profit as the product of number of ads and revenue per ad,
σ
i
a
i
pa
i
ðÞφ
Γ
¼Aa
i
ðÞ
σ
i
φ
Γ
;
523The Economics of Internet Media
with respect to the ad space a
i
, where A(a
i
) is the revenue per ad per viewer. Strategic
interaction between websites arises because of advertising congestion and the fact that
consumers have a limited attention span for all ads combined across different websites.
Thus, each website has access to the common property resource, which is the attention
span of each consumer. Because of the free-rider problem, advertising in a market with
strictly more than one website may result in congestion. In the monopoly case, the result
of
Anderson and de Palma (2009) applies, and the monopoly website always sets its ad
level such that congestion does not arise. If the attention span is sufficiently large (i.e.,
φ Γ), each website i chooses a
i
so as to maximize revenue per ad per viewer—i.e.,
a
i
solves A
0
a
i
ðÞ¼0 and all websites choose the same ad level. With congestion, when
consumers spend the same amount of time on each website, the first-order condition
of profit maximization becomes
A
0
aðÞ
φ
Γ
AaðÞ
φ
Γ
2
σ ¼0
after using symmetry—i.e., σ ¼σ
i
and a ¼a
i
. This implicitly defines the equilibrium
advertising level a
. Since total advertising is Γ ¼nσa, the equilibrium ad level is given
by the solution to
aA
0
aðÞ
AaðÞ
¼
1
n
:
If p(a) is log-concave, the left-hand side is decreasing in a. This implies that a larger num-
ber of websites leads to an increase in the advertising level of each website and thus to
more severe congestion.
78
The intuition is that with more alternative websites around,
each website internalizes to a smaller extent the negative effect that an increase in its ad
level has on the resulting advertising price. In an asymmetric market,
Anderson and Peitz
(2014a)
show that the larger website chooses a lower ad level a
i
. This is intuitive, as a
larger platform internalizes the congestion externality to a larger extent since more infra-
marginal ad slots are affected by the price decrease resulting from a larger ad volume.
79
The general lesson that emerges from the ad congestion model is that congestion
generates interdependence among websites due to price effects on the advertiser side.
To the extent that consumers consult a large number of ad-financed websites, the analysis
suggests that congestion issues may be important drivers of competition between
websites.
78
A merger between two websites is beneficial because it leads to less congestion. For a detailed discussion of
the merger effects in media markets and the regulatory environment, see
Chapter 6 in this volume.
79
Anderson and Peitz (2014a) extend their analysis to include advertising as a nuisance. This makes the
advertising market two-sided, as consumers now care about the number of ads carried on each platform.
Then, the higher-quality website still features a lower ad level. This is the opposite finding one would get
if the consumers’ attention span was unlimited.
524
Handbook of Media Economics
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.