treat prices per ad per viewer as the “strategic” variable. Alternatives, such as choosing ad
prices per se, are discussed below.
The first-order condition for the ad-finance game is then readily expressed (by max-
imizing ln π
i
) as the equality between two elasticities (Anderson and Gabszewicz, 2006),
those of revenue per viewer and viewer demand. Equivalently, letting a prime denote a
derivative with respect to own advertising,
R
0
R
¼γ
N
0
i
N
i
; (2.1)
where N
0
i
< 0 denotes the derivative of the demand function with respect to its first
argument (i.e., the full price). This equation underscores a crucial distinction between
cases when ads are a nuisance and when they are desirable to viewers. To see it, assume
first that viewers are ad-neutral, so they are indifferent between having an extra ad or not.
Then platforms set ad levels such that marginal revenue per viewer (R
0
) is zero. When ad
levels do not affect viewer levels, platforms simply extract maximal revenue from their
viewer bases. Otherwise, there is a two-sided market effect, and platforms internalize
the ad effect on viewer participation. For γ > 0 (ad nuisance), they restrain ad levels
below the level at which marginal revenue is zero. This entails ad prices above the
“monopoly” level even when there is competition among platforms. Conversely, for
γ < 0, they sacrifice some revenue per viewer by expanding ad levels in order to entice
viewers and so deliver more of them and consequently charge advertisers less per ad. That
is, the ads themselves are used as part of the attraction to the platform, though the plat-
form does not expand ad levels indefinitely because then the revenue per viewer would
fall too much as more marginal willingness-to-pay advertisers would have to be attracted.
The RHS of
(2.1) can readily be evaluated for standard symmetric oligopoly models
with n platforms, to deliver some characteristic properties of the solution. For the
Vickrey
(1964)
10
and Salop (1979) circle model, it is γ(n/t)(Anderson and Gabszewicz, 2006;
Choi, 2006
), where t is the “transport cost” to viewers. For the logit model
(
Anderson et al., 1992), it is γ n 1ðÞ=μnðÞ, where μ is the degree of product heteroge-
neity.
11
In both cases (as long as R
0
(a)/R(a) is decreasing, as implied by the marginal ad
revenue decreasing), a higher ad nuisance causes lower ad levels per platform. More pref-
erence heterogeneity (t or μ respectively) raises ad levels as platforms have more market
power over their viewers. Increasing the number of platforms, n, decreases the ad level.
The analogy is that advertising is a “price” to viewers, so naturally such prices go down
with more competition. Because the advertising demand curve slopes down, this means
that the ad price per viewer exacted on the advertiser side actually rises with competition
10
The relevant analysis is republished in Vickrey et al. (1999).
11
Similar properties hold for other discrete-choice models with i.i.d. log-concave match densities in that the
corresponding expression is increasing in n: see
Anderson et al. (1995).
48
Handbook of Media Economics