Major broadcasting networks usually focus on the taste of the masses to increase their adver-
tising revenues. For example, sports channels do not devote much air time to niche sports,
such as shot put or weight lifting. However, followers of these sports can access reports
online and watch them at any time. This allows small businesses whose target groups are
people interested in such niche sports to access these advertising markets.
Bergemann and Bonatti (2011) examine these effects and their implications for offline
versus online media. Suppose that there is a continuum of products and a continuum of
advertising markets. A product is denoted by y and is produced by a single firm y with
y 2 0, 1½Þ. Similarly, advertising markets are denoted by z 2 0, 1½Þ. There is a contin-
uum of buyers with a mass of one. Each buyer has a preference for a particular product
and is located in one advertising market. The joint distribution of consumers across
advertising markets and product markets is F(z, y) with density f(z, y). The fraction of
consumers interested in product y can be written as fyðÞ¼
ð
1
0
fz
0
, yðÞdz
0
, when integrat-
ing over all advertising markets. Similarly, the size of advertising market z can be written
as zðÞ¼
ð
1
0
fz,y
0
ðÞdy
0
, integrating over all products. The conditional distribution of
advertising markets for a given product y is fzjyðÞ¼fz, yðÞ=fyðÞ. Product differences
can be expressed by differences in the size f(y) to distinguish mass from niche products.
Each firm y can inform consumers about its product by sending a number of adver-
tising messages a
z,y
in advertising market z. Each message reaches a random consumer
with a uniform probability, as in the model of
Butters (1977): With probability
pr a
z,y
, fzðÞ
¼1 e
a
z,y
=fzðÞ
;
a given consumer in advertising market z of size f(z) becomes aware of product y.
In each advertising market z, the supply of messages M
z
is fixed. This supply is pro-
portional to the size f(z) of the advertising market—that is, M
z
¼fzðÞM. Here, M can be
interpreted as the average time a consumer spends on advertising messages. In each adver-
tising market, there are a large number of media outlets. Outlets act as price takers, imply-
ing that a firm y can purchase messages at a price p
z
in each market. The profit of firm y
can then be written as
51
π
y
¼
ð
1
0
fz, y
ðÞ
pr a
z,y
, fz
ðÞ
p
z
a
z,y
dz:
To easily distinguish between mass and niche products,
Bergemann and Bonatti (2011)
impose that fyðÞ¼αe
αy
. Here, a larger parameter value α represents a more concen-
trated product market. With this formulation, firms can be ranked in decreasing order
of market size—i.e., a firm with a higher index y is smaller in the sense that fewer
consumers are interested in its product.
51
The value of informing a consumer is normalized to 1.
499
The Economics of Internet Media
The advertising markets are also ranked according to the mass of consumers interested
in the market. Advertising market 0 is a large market in which all advertisers are inter-
ested, and advertising markets become smaller and more specialized with an increasing
index. To formalize this, suppose that firm y is interested only in consumers in markets
with z y. For each firm y, the advertising market z ¼y is the one with the highest den-
sity of consumers, conditional on market size. The conditional distribution of consumers
with interest in product y over advertising markets z is given by the following truncated
exponential distribution:
fz, yðÞ
fyðÞ
¼
βe
β y zðÞ
if 0 < z y,
0ify < z < 1;
for all advertising markets z > 0. There is a mass point at z ¼0 and the conditional dis-
tribution is fz, yðÞ=fyðÞ¼e
βy
if z ¼0. The parameter β measures the concentration of
consumers in advertising markets. We will explain below how the possibility of targeting
consumers can be measured by β. Combining the definition of market size with the con-
ditional distribution gives the unconditional distribution
fz, y
ðÞ
¼
αβe
α + βðÞy
e
βz
if 0 < z y,
0ify < z < 1;
with a mass point at z ¼0, where the unconditional distribution is fz, yðÞ¼αe
α + βðÞy
.
The market size can then be calculated by integrating over the population shares.
Since consumers who are potential buyers of product y are present in all advertising
markets z y but not in advertising markets z > y, we have
fz> 0ðÞ¼
ð
1
z
αβe
α + β
ðÞ
y
e
βz
dy ¼
αβ
α + β
e
αz
and
fz¼0ðÞ
ð
1
0
αe
α + βðÞy
dy ¼
α
α + β
:
The distribution of consumers across product and advertising space has a natural inter-
pretation in terms of specialization of preferences and audiences. First, a product with
a larger index is a more specialized product in the sense that there are fewer potential
buyers. Similarly, an advertising market with a higher index z is a market with a more
narrow audience. Second, potential consumers of larger firms are distributed over a smal-
ler number of advertising markets. This can be seen by the assumption above that f(z, y)is
positive only for z y. For example, potential buyers of product y ¼0 are concentrated
in the advertising market z ¼0. Interpreting advertising markets as media outlets,
this implies that a consumer interested in a mass product does not visit a website with
advertisements for niche products.
500 Handbook of Media Economics
Third, the variable β, ranging from 0 to 1, captures in a simple way the ability of
firms to target consumers. For example, as β !0, all consumers are concentrated in
advertising market 0, implying that there is a single advertising market. By contrast, as
β !1, then all potential buyers of product y are in advertising market y, and so there
is perfect targeting. In general, an increase in β implies that consumers are spread over
more advertising markets and can be better targeted by firms. Overall, the highest
conditional density of potential consumers of firm y is in advertising market z ¼y.As
β gets larger, more consumers move away from the large advertising markets (near
z ¼0) to the smaller advertising markets (near z ¼yÞ.
To illustrate how the model works, let us look at the benchmark case in which all
consumers are present in a single advertising market z ¼0. We solve for the equilibrium
amount of advertising and the equilibrium price. Since there is a single advertising mar-
ket, we drop the subscript z in the notation. The profit function of firm y is then
π
y
¼fyðÞpr a
y
pa
y
. Determining the first-order conditions and using the definition
of f(y) yields
a
y
¼
ln fyðÞ=pðÞif fyðÞp,
0iffyðÞ< p:
(10.3)
As is evident, firms with a larger market size optimally choose a larger amount of adver-
tising. Therefore, in equilibrium, only firms with the largest market size find it optimal to
advertise. Let M be the total number of advertising messages and denote by Y the mar-
ginal advertiser. The market-clearing condition is given by
Ð
Y
0
a
y
dy ¼M. Using demand
for ads given by
(10.3) and fyðÞ¼αe
αy
yields
ð
Y
0
ln
α
p
αy
dy ¼M:
Using a
Y
¼0 together with the last equation, we can solve for the equilibrium price and
the marginal advertiser. This gives p
¼αe
ffiffiffiffiffiffiffi
2αM
p
and Y
¼α
ffiffiffiffiffiffiffiffiffiffiffiffi
2M=α
p
. Inserting back
into the demand function of advertiser y yields
a
y
¼
α
ffiffiffiffiffiffiffiffiffiffiffiffi
2M=α
p
αy if y Y
,
0ify > Y
:
(10.4)
Therefore, only the largest firms advertise, and the equilibrium number of advertising
messages is linearly decreasing in the rank y of the firm. As the concentration in the prod-
uct market measured by α increases, fewer advertising messages are wasted, leading to an
increase in social welfare. In particular, the allocation adjusts to firms with a larger market
size, implying that fewer firms advertise as α increases.
To analyze the effect of targeting,
Bergemann and Bonatti (2011) examine the situation
with a continuum of advertising markets and a positive targeting parameter β 2 0, 1
ðÞ
.
The allocation of advertising messages is then given by a generalization of
(10.4):
501The Economics of Internet Media
a
z
,y
¼
αβe
αz
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2M= α + βðÞ
p
y zðÞ
if z > 0,
α
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2M= α + βðÞ
p
αy if z ¼0:
(
Does targeting improve social welfare and do firms benefit from targeting?
Bergemann
and Bonatti (2011)
show that the social value of advertising is increasing in the targeting
ability β. The intuition is that targeting increases the value of advertising for a firm y in its
“natural” advertising markets z y. This leads to an increased volume of matches
between firms and potential consumers, which improves social welfare.
However, looking at the cross-sectional implications of targeting, not all firms benefit
from improved targeting. In particular, only the small firms that are not active in the mass
advertising market z ¼0 and the largest firms, which are primarily active in that market,
benefit. By contrast, medium-sized firms, which are active in the mass market and also in
several other markets 0 < z y, are hurt. To grasp the intuition behind this result,
observe, first, that for small firms (those not active in market z ¼0), the mass of potential
buyers in their natural advertising markets z y increases, allowing them to reach a larger
fraction of consumers. A similar effect is present for large firms. Their customers are con-
centrated in a small number of markets, and an increase in the targeting ability increases
the chances of achieving a match. By contrast, medium-sized firms are hurt by the
decrease in consumers participating in market z ¼0, and this decline cannot be compen-
sated by the rise in participation in their natural markets z y .
The model can be used to analyze the implications of targeting for “online” versus
“offline” media. In the offline medium, there is only a single advertising market, whereas
there is a continuum of advertising markets in the online medium. For simplicity, suppose
that the online medium allows for perfect targeting of advertising messages to consumers.
Consumers are dual-homing and spend a total amount of M
1
on the offline medium and
M
2
on a single online market z. More specifically, f(z)M
2
is the supply of advertising mes-
sages in each targeted market z. So, the online medium consists of a continuum of spe-
cialized websites that display firms’ advertisements. There is competition between the
two media because each firm views the advertising messages sent online and offline as
substitutes due to the risk of duplication.
Bergemann and Bonatti (2011) show that
the price for offline advertising decreases in M
2
, reflecting the decreased willingness-
to-pay for regular ads if a better-targeted market is present. The price for online adver-
tising decreases in M
1
only on those websites that carry advertisements of firms that are
also active offline. However, advertising markets with a high index z carry only the adver-
tisements of niche firms, which are not affected by the allocation in the offline medium
because they do not advertise there. This implies that online advertising reaches new
consumer segments that are distinct from the audience reached by offline advertising.
Suppose, in addition, that each consumer is endowed with an amount of time equal
to M and allocates a fraction σ of this time to the online medium. This implies that M
1
¼
1 σðÞM and M
2
¼σM. It is now possible to analyze what happens when consumers
spend more time online—that is, when σ increases. The effect on the offline advertising
502 Handbook of Media Economics
price is then non-monotonic. If σ is low (i.e., online exposure is low), the marginal
willingness-to-pay for offline advertising falls because online advertising is more efficient.
This induces a decrease in the offline advertising price, although the supply of offline
advertising messages decreases. However, as σ increases further, the composition of firms
active in offline advertising changes. In particular, only the largest firms display advertis-
ing messages offline, implying that the marginal advertiser has a high willingness-to-pay.
This leads to an increase in the offline advertising price with σ. With regard to firm rev-
enue, this implies that if consumers spend more time online, firms that are active solely in
the online market unambiguously benefit. These are rather small firms. The effect on
large firms is ambiguous: Since they are active on the offline medium, they may pay a
larger advertising price, which reduces their profits.
In summary,
Bergemann and Bonatti (2011) show that targeting on the
Internet allows platforms to split up a single advertising market into multiple ones. This
allows producers of niche products, who are not active in the single large advertising mar-
ket, to advertise, thereby increasing advertising efficiency. Small and also large firms ben-
efit from targeting. By contrast, medium-sized firms are worse off because attention of
consumers migrates to smaller advertising markets.
Rutt (2012) proposes a different formalization of targeting. He considers a model with
n platforms which are distributed equidistantly on a circle, single-homing users, and
multi-homing advertisers. A user’s valuation for an advertiser’s product is binary, namely
either of high or of low valuation (with the low valuation being set equal to 0).
52
Advertisers are uncertain about the true valuation. In particular, advertiser j does not
know consumer i’s valuation for her product with certainty but only has an expectation.
Each advertiser receives an informative signal about a consumer’s true valuation. The
realization of the signal induces an advertiser to update the expectation. The targeting
technology can now be modeled as a change in the informativeness of the signal.
53
In
the extreme case that targeting is impossible and signals are pure noise, the updated
expectation equals the prior expectation. By contrast, when the signal is perfect, the
advertiser knows the consumer’s valuation with certainty. As a result, an increase in
the informativeness leads some advertisers to revise their beliefs upward, while others
revise them downward. Advertisers receiving a positive signal become more optimistic,
whereas advertisers with a negative signal become more pessimistic.
54
52
Pan and Yang (2015) consider a similar market structure with two platforms to analyze the effects of tar-
geting; they specify user demand and thus improved targeting differently.
53
A simple example of a signal structure which fits this description is the truth-or-noise information struc-
ture: suppose that the prior expectation about consumer i having a high valuation for the product is X, the
signal is Y, and the signal reveals the true consumer valuation with probability Z. Then, the posterior
expectation is ZY +1ZðÞX.
54
In this respect, the information structure is similar to the demand rotation considered in Johnson and
Myatt (2006)
.
503
The Economics of Internet Media
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.