1. Ad networks allow digital publishers to outsource the sale of advertising space to centralized networks like Google AdSense. This allows for tracking of consumers across multiple publisher websites.
2. Tracking benefits advertisers by reducing repeated impressions, but some consumers see it as a violation of privacy. Ad networks internalize the effects of consumer multi-homing but privacy preferences are a challenge.
3. The document discusses research questions around the role of ad networks, how tracking affects supply and whether supply is higher when outsourced to a network versus direct publisher competition. It presents a model to analyze these issues and the effects of privacy policies on welfare and consumer surplus.
1. 1
Ad networks, consumer tracking, and privacy
Florence, March 25th 2017
EUI Media Conference 2017
Anna DβAnnunzio
Telenor Research
Antonio Russo
ETH Zurich
2. Ad networks
β’ Digital publishers can outsource sale of ads to ad networks
β’ Google ad network (AdSense) reaches:
β’ ~40% of (Alexa) top 500 sites (Roesner et al., 2012)
β’ ~94% of total internet audience (Comscore, 2016)
2
Publisher 1
Publisher 2
AN Advertisers
3. Tracking and multi-homing
β’ Advertisers and consumers multi-home
β’ Consumers spread attention over several websites
β’ Problem: no tracking across publishers, higher chance of
repeated impressions
β’ Ad inventories lose value with multi-homing (Ambrus et
al., 2016)
β’ Ad networks:
β’ Centralize sale of ads, internalize external effects due to
multi-homing
β’ Track consumers across publishers, reduce repetition 3
4. Privacy and tracking
β’ Some consumers see tracking as a violation of their privacy
β’ 84% of U.S. respondents does not want advertising tailored on
their behavior on the internet (Turow et al., 2009)
β’ A variety of techniques allows consumers to avoid being
tracked (e.g. browser plug-ins, anonymizing apps)
β’ Idea: model the between link privacy preferences, tracking
and advertising market
β’ EU and US authorities are discussing do-not-track policies
β’ E.g., FTC βdo not track proposalβ
4
5. 5
Research questions
β’ What is the role of ad networks in the online advertising
market?
β’ How does tracking affect the supply of ads?
β’ Does supply of ads increase with an ad network compared
to the case where publishers compete?
β’ Which privacy policy is desirable when the effects on the
advertising market are considered?
β’ Effect of tracking and consumersβ choice to block it on
welfare and consumer surplus
β’ Market failures and regulation
6. 6
Previous literature
β’ Analysis on supply of ads with single-homing
consumers: Anderson and Coate, 2005
β’ Multi-homing consumers: Ambrus et al., 2016;
Anderson et al., forthcoming; Athey et al., forthcoming
β’ We add tracking by ad networks and privacy-related
choices
β’ Targeting and ad-avoidance: Johnson, 2013
β’ Ad-avoidance β cookies blocking
β’ Advertising and privacy regulation: Campbell et al.,
2015; Goldfarb and Tucker, 2011
β’ No effect on supply of ads and no welfare analysis
7. Model overview
β’ Publishers: free for consumers and advertising-financed
β’ Ad network can track consumers
β’ Two scenarios:
β’ Outsourcing to AN
β’ Competition among publishers without AN
7
Publisher 1
Publisher 2
ANConsumers Advertisers
8. Consumers
β’ Unit mass, π’", π’$ ~β π’", π’$
π·"$ = ππ π’" β πΏπ" β₯ 0; π’$ β πΏπ$ β₯ 0
π·2 = ππ π’2 β πΏπ2 β₯ 0; π’3 β πΏπ3 < 0 π, π = 1,2; π β π
π·: = 1 β π·" β π·$ β π·"$
π2: ad level on π; πΏ: nuisance cost of ads
β’ Property:
;<=>
;?@
< 0;
;<=>
;?@
= β
;<A
;?@
β’ Also: consumers choose to block/allow tracking (e.g. third
party cookies)
β’ I will discuss this in the second part
8
9. 9
Advertisers
β’ Unit mass, homogeneous
β’ Advertiser surplus: π·" π" + π·$ π$ + π·"$ π"$
C
β’ Normalize return from informing a consumer to 1
β’ π2 π2 : prob. informing a SH (π2
E
> 0, π2
EE
< 0)
β’ π"$
C
π", π$, π½ : prob. informing a MH
;H=>
I
;J@
> 0,
;>H=>
I
;J@
> < 0,
;>H=>
I
;J@ ;JA
< 0
β’ π½: cross-publishers tracking effectiveness
;H=>
I
;K
> 0 and
;>H=>
I
;?@ ;K
> 0
10. 10
Timing
0. Ad Network deals with publishers (outsourcing of ad
inventory)
1. Publishers choose ad level π2.
2. Consumers choose whether to block tracking which
publisher to visit
3. AN sells impressions to the advertisers
4. Consumers get utility, get informed, payoffs realized
11. 11
Ad network: contracts
β’ Advertisers:
β’ Ad network unique gatekeeper to consumers
β’ Offer to each advertiser π", π$ impressions per consumer
[at eq. π2 = π2], in exchange for lump-sum transfer πNO
πNO = π" π·" + π$ π·$ + π"$
C
π·"$
β’ Publishers
β’ AN acquires right to sell ad inventory of publisher i, offers
per impression price π₯2 π2
β’ In equilibrium, π2
NC
: πππ₯ ?@,?A
πNO
12. Effect of tracking on advertising levels
β’ Assuming symmetry:
ππ2
NC
ππ½
> 0 βΊ π·"$
π$
π"$
C
ππ2 ππ½
+
ππ·"$
ππ2
ππ"$
C
ππ½
> 0
β’ π·"$
;>H=>
I
;?@;K
> 0: tracking increases revenues on infra-marginal
multi-homers
β’
;<=>
;?@
;H=>
I
;K
< 0: tracking increases opportunity cost of losing
marginal multi-homer (
;<=>
;?@
< 0 and
;H=>
I
;K
> 0)
12
13. 13
Competing publishers, no outsourcing
β’ Differences wrt AN model:
β’ Revenue per MH changes without tracking (π½ = 0)
π"$
C
> π"$
_C
> 0
;H=>
I
;?@
>
;H=>
`I
;?@
> 0
β’ Publishers offer contracts π2, π2
π2 = π2 π·2 + π3 π·3 + π"$
_C
π·"$ β π3 π·3 + πΜ3 π·"$
= π2 π·2 + π"$
_C
β πΜ3 π·"$ π = 1,2
β’ Publishers decide ad levels π2
b
: πππ₯?@
π2
Total advertiser surplus Advertiser surplus on π
Value SH Incremental value MH
14. Competition Vs AN
β’ π2
NC
> π2
b
iff
π·"$
ππ"$
C
ππ2
β
ππ"$
_C
ππ2
+
ππ·"$
ππ2
π"$
C
β π"$
_C
+ πΜ3 β π3 c
?@d?@
e
,2d",$
> 0
β’ Tracking increases MR on infra-marginal MHs and increases
opportunity cost of losing marginal MH
β’ Joint control:
β’ opportunity cost for AN of marginal ad on π: π"$
f
β π3
β’ opportunity cost for publisher π : π"$
f
β πΜ3
Tracking effect Joint control effect
14
15. Consumersβ blocking choice
β’ Disutility from being tracked: π~πΉ 0, πΜ , π β₯ π’2, π’3
β’ Consumers block iff π β₯ π (cost of blocking)
β’ ANβs ability to track determined by consumers
β’ share of consumers tracked (i.e. not blocking) is π½
β’ Privacy has only intrinsic value (Acquisti et al. 2014)
β’ Extension: blocking affects disutility from ads (privacy
intermediate good)
15
17. Second best (SB) tracking
β’ Regulator controls the tracking, publishers decide ad levels
ππ
ππ½
= β
ππΆπp
ππ½
+
ππΆπo
ππ2
ππ2
NC
ππ½
+
ππΆπo
ππ3
ππ3
NC
ππ½
+
ππ΄π
ππ½
= 0
β’ Tracking at equilibrium may be too low compared to SB
β’ Always the case if
;?@
|I
;K
< 0
β’ Tracking may be too low even if objective is consumer
surplus
17
Externality on other
consumers
Externality on
advertisers
Internalized
β β +Β± Β±
18. Policy implications
β’ Promoting cookies blocking (e.g. FTCβs βdo not trackβ) may
not be beneficial for welfare and not even for consumers
β’ It may be desirable
β’ to reduce consumersβ disutility from being tracked (e.g.,
forbid intrusive third-party cookies)
β’ to increase perceived cost of blocking tracking (e.g., opt-out)
β’ Reducing disutility from being tracked or cost of blocking
reduce the net disutility from privacy losses πΆπp β¦
β’ β¦ but have opposite effects on advertising market
18
19. 19
Conclusion
β’ More efficient tracking technologies by ad networks may
reduce the advertising consumers are exposed online
β’ Policy-makers should consider the effect on the supply of
ads when designing digital privacy regulation
β’ Encouraging the use of tools that prevent firms from
collecting data about consumers to deliver advertising
may have adverse effects on both consumers and society
20. Extension: privacy as intermediate good
β’ Farrell (2012): privacy is a final and an intermediate good
β’ Consumersβ choice to block cookies change the ads they
see, hence their disutility from advertising
β’ Formally:
β’
β πΏ β π§ π2 β π if not block cookies
β πΏ π2 β π if block cookies
β’ The marginal multi-homer blocking cookies depend on π2
β’ Preliminary analysis: results of the main model confirmed
20
21. Welfare analysis
β’ Effect of AN
β’ If π2
NC
< π2
b
consumer and advertiser surplus are higher when
advertising is outsourced
β’ Otherwise, conflicting interests: consumer surplus is higher
when publishers compete directly, and advertiser surplus
lower
21
22. Micro-foundation (1/4)
β’ Interpretation: an advertiser buys ads to reach most
interested consumers
β’ π advertisers, each two ad messages
β’ Informing a consumer is worth:
β’ first message 1 (mainstream product)
β’ second message π¦ < 1 (e.g., niche product)
β’ π₯2
Λ
: probability a consumer registers an ad by advertiser π
when exposed to it on publisher π = 1,2
β’ π₯2
Λ
~π 0,1 : i.i.d. for all consumers, publishers, advertisers
β’ Assumption: each advertiser buys at most one impression
on a consumer on a given publisher
22
23. Micro-foundation (2/4)
β’ Publishers send ads to consumers with largest π₯2
Λ
β’ Advertiser π buys π2
Λ
ad impressions on π
π2
Λ
= 1 β π₯Μ 2
Λ
π·2 + π·"$
π₯Μ 2
Λ
: consumer with higher π₯2
Λ
receives an ad on π from π
Perfect internal tracking: no repeated internal impressions
β’ Hence:
π₯Μ 2
Λ
= 1 β
π2
Λ
π·2 + π·"$
β’ Expected prob. consumer on π registers an ad by π:
z π₯ππ₯
"
Ε Μ @
βΉ
=
1
2
1 β π₯Μ 2
Λ $
23
24. Micro-foundation (3/4)
β’ Revenues from SH:
π2
Λ
= π·2 πΌ2 z π₯ππ₯
"
Ε Μ @
βΉ
+ π·"$ πΎ2 z π₯ππ₯
"
Ε Μ @
βΉ
πΌ2 (resp., πΎ2) probability that a SH (resp., MH) watches an ad
when impressed on π
β’ Revenues from MH without tracking:
π"$
Λ_OC
= π·" πΌ" z π₯ππ₯
"
Ε Μ =
βΉ
+ π·$ πΌ$ z π₯ππ₯
"
Ε Μ >
βΉ
+ π·"$ πΎ" z π₯ππ₯
"
Ε Μ =
βΉ
+ πΎ$ z π₯ππ₯
"
Ε Μ >
βΉ
β πΎ" z π₯ππ₯
"
Ε Μ =
βΉ
πΎ$ z π₯ππ₯
"
Ε Μ >
βΉ
24
25. Micro-foundation (4/4)
β’ Revenues from MH with tracking:
β’ AN observes π₯2
Λ
for a consumer, and decides whether to
impress her. Next, AN observes with prob. π½ whether the
selected consumer has already been informed by π's
message on π. Then, AN decides whether to send the first or
the second message from π
π"$
Λ_C
= π·" πΌ" z π₯ππ₯
"
Ε Μ =
βΉ
+ π·$ πΌ$ z π₯ππ₯
"
Ε Μ >
βΉ
+ π·"$ πΎ" z π₯ππ₯
"
Ε Μ =
βΉ
+ πΎ$ z π₯ππ₯
"
Ε Μ >
βΉ
β πΎ" z π₯ππ₯
"
Ε Μ =
βΉ
πΎ$ z π₯ππ₯
"
Ε Μ >
βΉ
+ π·"$ π½π¦ πΎ" z π₯ππ₯
"
Ε Μ βΉ
πΎ$ z π₯ππ₯
"
Ε Μ βΉ
25