With regard to the timing, platforms first select their advertising levels and advertisers
then submit their conditional demands specifying which types of users they wish to be
matched with for a given advertising price. For example, advertisers announce that they
want to have ads displayed only to users who show an interest with a probability higher
than some cutoff level. After users have decided which media platform to consume from,
advertisers observe the signals and update their valuations. The advertising price then
adjusts to clear the market. Afterward, consumers enjoy the media content, observe
the advertisement, and make purchases.
In this setting,
Rutt (2012) shows that targeting increases the advertising price and
thereby allows platforms to receive higher profits. This effect is the stronger the more
competitive is the market and the more users are averse to ads. This is because in markets
satisfying these conditions, the equilibrium features only few advertisements, implying
that the marginal advertiser received a particularly high signal. As a consequence,
improved targeting increases this advertiser’s willingness-to-pay, resulting in a strong
increase in the advertising price. In the case of free entry of platforms (at some setup
costs), improved targeting may exacerbate excessive entry and leads to insufficient adver-
tising due to high prices. Overall, the model therefore predicts a heterogeneous effect of
targeting in different media markets.
The models of
Athey and Gans (2010) and Bergemann and Bonatti (2011) share the
assumption that offline and online advertising are substitutes.
55,56
If a potential consumer
can be contacted via an advertisement through one channel, a conditional contact
through the other channel is worth less.
Goldfarb and Tucker (2011a) provide empirical
support. They use data on estimated advertising prices paid by lawyers to contact potential
clients with recent personal injuries.
Goldfarb and Tucker (2011a) exploit state-level var-
iation in the ability of lawyers to solicit those customers. In particular, this “ambulance-
chasing” behavior is regulated in some states by the state bar associations, which forbid
written communication (including direct electronic communication via email) with
potential clients for 30–45 days after the accident.
Goldfarb and Tucker (2011a) use data
on estimated auction prices of 139 Google search terms on personal injury in 195 regional
city markets to analyze the effects of these regulations.
57
They find that, compared to the
prices for personal injury keywords in non-regulated states, in states with solicitation
55
Rutt (2012) considers only online platforms. When interpreting some platforms as online and others as
offline media outlets in his model, online and offline advertising would be independent. This is because
users only visit a single platform and so advertisers can reach each particular user exclusively via online or
via offline advertising.
56
More generally, there is a link between offline and online in the sense that advertising offline affects con-
sumer purchase online. In particular,
Liaukonyte et al. (2015) study the effect of television commercials on
actual purchase behavior on the Internet. They find a positive effect, even immediately after a viewer is
exposed to the commercial.
57
Search terms on personal injury can be identified objectively because there is a precise legal definition by
the bar association.
504
Handbook of Media Economics
restrictions, such keyword prices are between 5% and 7% higher. Therefore, when offline
marketing communication is not possible, firms appear to switch to online advertising.
This suggests that there is substitution between the two forms of advertising. In addition,
Goldfarb and Tucker (2011a) demonstrate that this effect is larger in locations with a small
number of potential clients. One interpretation is that mass-media advertising may not be
cost-effective when consumers are hard to reach. Therefore, the possibility of direct off-
line advertising is particularly valuable. If this advertising channel is closed, online adver-
tising prices rise. This implies that in markets with fewer customers, online advertising
allows firms to reach the hardest-to-find customers, thereby lending support to the
“long tail” hypothesis in Internet advertising (
Anderson, 2006).
We also note that targeting on the Internet can take many different forms beyond
geographical or contextual targeting. For example, social targeting has become increas-
ingly popular. Socially targeted ads, when displaying the ad to a particular user, refer to
another user (e.g., a friend on Facebook). More precisely, a social ad is an online ad that
“incorporates user interactions that the consumer has agreed to display and be shared.
The resulting ads display these interactions along with the user’s persona (picture and/
or name) within the ad’s content” (
IAB, 2009). Tucker (2012) compares the effectiveness
of socially targeted ads to that of conventionally (demographically) targeted ads and
non-targeted ads. She conducted a field experiment on Facebook involving a non-profit
charity organization that provides educational scholarships for girls to attend high school
in East Africa. The non-profit organization launched a standard advertising campaign and
a social variant of it. In this social variant, the recipients of the ads were only Facebook
users who are friends of existing fans of the charity.
Tucker (2012) finds that these socially
targeted ads were more effective than regular display advertising. This holds both for
randomly selected users and users who previously expressed their interest in either charity
or education. For example, the aggregate click-through-rate of the socially targeted ads
was around twice as large as that of non-socially targeted ads.
10.5.2 Keyword Advertising
A very effective form of targeted advertising is keyword advertising. Keyword targeting
refers to any form of advertising that is linked to specific words or phrases and is displayed
when a user is looking for information. Therefore, such advertising is not necessarily a
nuisance to users and is wasted with a lower probability than, say, TV advertising or ban-
ner ads, as the ad is relevant to the user’s query and therefore valuable. Nowadays, almost
all search engines offer keyword advertising. The most well-known form is probably
Google AdWords.
58
It also engages in contextual advertising; that is, Google’s system
scans the text on the websites that are most relevant to the search query and displays
ads based on the keywords found in the respective texts. A main question is whether this
form of targeting is welfare-enhancing.
58
The study by Goldfarb and Tucker (2011a) discussed above uses keyword advertising by Google.
505
The Economics of Internet Media
In this subsection, we focus on keyword advertising on search engines. The question
has been raised whether search engines have an incentive to present the most relevant ads
according to the keyword entered by the user. To address this question,
de Cornie
`
re
(2013)
proposes a model with a single search engine that matches potential consumers
and producers. Consumers are uniformly distributed along a circle with circumference
1. Each consumer is described by a two-dimensional vector: first, by the consumer’s
location on the circle, which describes her favorite product denoted by ω 2 0; 1½; and
second, by her willingness-to-pay, denoted by θ 2 0,
θ
½
. In particular, in each location,
there is a continuum of mass 1 of consumers whose reservation price θ is distributed with
c.d.f. F. Both variables ω and θ are private information. Each consumer buys, at most, one
unit and obtains a utility of
v θ, d , P
ðÞ
¼θ cd
ðÞ
P;
where P is the price of the product and d ¼d χ, ω
ðÞ
measures the distance between the
product’s location χ and the consumer’s location ω. The function c(d), therefore mea-
sures the mismatch costs and is assumed to be increasing and weakly convex.
Products are continuously distributed on the circle. There is a continuum of entrants
for each product. Each product can be described by a keyword, which is denoted by
χ 2 0; 1½; that is, the keyword is identical to the location of the product. The parameter
χ is private information to the producer, implying that consumers know neither the posi-
tion of a firm on the circle nor the price; hence, they need to use the search engine. When
a firm wants to advertise on the search engine, it incurs a fixed cost of C to launch an
advertising campaign. (This cost is not a payment to the search engine.)
The search process works as follows: Firms select a set of keywords that they want to target.
Thesetisassumedtobesymmetricaroundχ—that is, Σ χðÞ¼χ D
χ
;χ + D
χ
,where
the meaning of D
χ
becomes clear below. Consumers enter the keyword of their prefer-
red product ω. After entry of the keyword ω, the search engine randomly selects a firm χ
such that ω 2Σ χðÞ. Once a consumer has decided to use the search engine, she incurs
search costs of s and learns the price and location of the firm selected by the search engine.
The firm then pays an amount p to the search engine; therefore, p represents the price-
per-click. The consumer can then buy the product or not buy it and stop searching, or
she can hold the offer and continue searching. That is, recall is costless. For each additional
search, the consumer again incurs costs of s.
The timing of the game is as follows: In the first stage, the search engine chooses the
per-click-price p,
59
which is public information to producers and consumers. In the
59
For a model of click fraud, in which publishers affiliated with the search engine’s advertising network or
competing advertisers artificially drive up clicks (e.g., by impersonating consumers) without increasing
sales, see
Wilbur and Zhu (2009).
506
Handbook of Media Economics
second stage, producers make their decisions. They first decide whether to be active on
the search engine; if so, they incur the fixed cost C . Each active firm located at χ then
chooses a price P
χ
for its product and an advertising strategy D
χ
. The mass of active firms
is denoted by h. Finally, consumers decide whether to use the search engine. If they do so,
they incur search costs s, type in the keyword of the most preferred product ω, and start a
sequential search among firms d χ, ωðÞD
χ
. The search engine draws firms in the respec-
tive range with equal probability.
What is the perfect Bayesian equilibrium with free entry of firms for this game? Once
a consumer has decided to use the search engine and has entered the keyword ω, she
obtains a search result showing the link to a firm in the support ω D
;ω + D
½. Suppose
that all firms charge an equilibrium price of P
. Then, the expected utility that a
consumer gets from this search is
ð
ω + D
ωD
v θ,d χ, ωðÞ,P
ðÞ
2D
dχ ¼
ð
D
0
v θ, x, P
ðÞ
D
dx:
The consumer’s optimal search behavior is a cutoff rule. That is, the consumer will buy
the product of a firm χ if the distance d(ω, χ) is lower than or equal to a reservation dis-
tance, denoted by R. If a consumer decides not to buy the product, she can improve her
utility only by finding a firm that is located closer to her most preferred product because
all firms charge the equilibrium price P
. Therefore, for R
to be an equilibrium
reservation distance, the consumer must be indifferent between buying the product
and continue searching; that is,
ð
R
0
vu, x, P
ðÞvu, R
, P
ðÞ
D
dx ¼
ð
R
0
cR
ðÞcxðÞ
D
dx ¼s:
The left-hand side is the expected gain from continuing to search, and the right-hand side
represents the search costs. By totally differentiating this expression, one can show that
R
increases with s and D
.
To determine the firms’ optimal targeting strategy, note that a consumer will never
come back to a firm if she does not buy from this firm immediately. This is because the
consumer’s stopping rule is stationary, and she will keep on searching as long as her match
is at (weakly) lower distance than her reservation distance. This implies that the condi-
tional purchase probability of a consumer after clicking on a firm’s link is either 0 or 1.
Because the firm has to pay only for consumers who click on the link, the optimal target-
ing strategy is simple and equal to R
¼D
. Therefore, in equilibrium, consumers will
not search more than once. This allows us to deduce the consumers’ participation deci-
sions regarding whether or not to use the search engine. The cutoff reservation value,
such that a consumer is indifferent between using the search engine or not, is given
by θ P
s Ec dðÞjd R
½¼0.
507The Economics of Internet Media
We now turn to a firm’s optimal pricing decision. If a firm charges price P different
from the candidate equilibrium price P
, it will optimally also change its targeting strat-
egy. In particular, it will target consumers located at a distance smaller than the new res-
ervation distance R(P, P
, D
), taking into account that all other firms follow the
candidate equilibrium strategy P
and D
. Given this new strategy, the firm faces a mass
of 2R(P,P
, D
)h
competitors in equilibrium. This is because every consumer within
distance R(P, P
, D
) is targeted by exactly this mass of firms. Since all consumers buy
without searching a second time, the firm’s profit function per consumer is
π P, P
, pðÞ¼P pðÞ
RP, P
, D
ðÞ
RP
, P
, D
ðÞh
:
The equilibrium price P
is given by the first-order condition of this expression with
respect to P. In equilibrium, the mass of participating consumers is given by 1 F(θ
).
Therefore, a firm’s expected profit is (1 F(θ
))π(P, P
, p). Since all firms charge the same
price in equilibrium, the free-entry condition determines the mass of entering firms.
Explicitly accounting for the dependence of P
, θ
, and h
on p, we can write the
free-entry condition as
P
pðÞpðÞ
1 F θ
pðÞðÞ
h
pðÞ
¼C:
Finally, we turn to the profit of the search engine. Since every consumer searches only
once, the profit of the search engine is given by π
SE
pðÞ¼p 1 F θ
pðÞðÞðÞ. This profit
function shows the tradeoff faced by the search engine. Everything else being equal,
the search engine obtains a higher revenue when increasing p. However, such an increase
in the per-click fee leads to a higher price on the product market. Since consumers antic-
ipate this, fewer of them will use the search engine. Maximizing π
SE
( p), we obtain that
the optimal per-click fee is implicitly given by p
¼
1F θ
p
ðÞðÞ
f θ
p
ðÞðÞ
. It is evident that the search
engine sets a positive p
. Instead, the socially optimal fee equals zero. Hence, the equi-
librium implies a distortion, as consumer participation is too low and product prices are
too high.
Within
de Cornie
`
re’s (2013) model, we can now analyze the effects of targeting.
Suppose that targeting is not possible. In the model, this corresponds to the case in which
D ¼1=2 for all products. The optimal reservation distance for consumers, R
, is then
implicitly given by
Ð
R
0
cR
ðÞcxðÞ
1=2
dx ¼s. Therefore, consumers may search more than
once, and targeting reduces the expected number of clicks. Since the reservation distance
is increasing in D, the expected mismatch costs are also lower with targeting. As a con-
sequence, targeting reduces the search frictions.
Targeting has more subtle effects on the price of the final good. First, targeting
changes the pool of firms from which consumers sample. In particular, with targeting,
this pool is composed of firms that are expected to be a better match for consumers. This
508 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.