the attention economy of
DESCRIPTION
The Attention Economy of Search and Web Advertisement. Alexander White. Tsinghua University, School of Economics and Management (joint with Kamal Jain, Microsoft Research). October 28, 2011. MoscowTRANSCRIPT
![Page 1: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/1.jpg)
The Attention Economy of Search and Web Advertisement
Alexander White Tsinghua University, School of Economics and Management
(joint with Kamal Jain, Microsoft Research)
October 28, 2011 Moscow
![Page 2: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/2.jpg)
People Surf the Web
![Page 3: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/3.jpg)
People Surf the Web Website Search % Bounce % Minutes on site
BBC.co.uk 11.8 35.9 6.6
Cooks.com 56.5 28.8 2.5
Facebook.com 6.7 12.8 32.2
HuffingtonPost.com 13.9 47.3 5.7
IMDB.com 30.2 31.1 4.6
MySpace.com 16.4 31.3 9.6
WebMD.com 36.5 45.0 4.2
Wikipedia.org 50.6 49.4 5.1
Yelp.com 42.1 47.8 3.8
Youtube.com 14.9 23.4 20.2
Source: Alexa.com
![Page 4: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/4.jpg)
![Page 5: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/5.jpg)
![Page 6: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/6.jpg)
Typical Complement Sellers’ Problem:
Software
Hardware
Single Demand
PH PS
Double Marginalization: PH+PS > P*
• Cournot 1838, ch. IX
![Page 7: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/7.jpg)
Typical Solutions: One Price Setter
Single Demand P = P*
With advertisement, however, there are two effects at play • Different websites have different advertising technologies
![Page 8: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/8.jpg)
A Model
Ui =vi − δse +δcw( ), if visits both sites
0, otherwise
⎧⎨⎪
⎩⎪
A “Search Engine” (SE), a “Content Website” (CW), and “Users”
vi ~ v,v⎡⎣ ⎤⎦
![Page 9: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/9.jpg)
SE and CW Profits
Πse = (ase − cse)D δse(ase)+δcw(acw)( )
Πcw = (acw − ccw)D δse(ase)+δcw(acw)( )
![Page 10: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/10.jpg)
A Model
Timing 1. SE and CW set advertising levels 2. Users decide whether to search and visit content site
![Page 11: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/11.jpg)
“Industry” Optimum
maxase ,acw{ }
(ase +acw − cse − ccw)D δse(ase)+δcw(acw)( )Necessary conditions at optimum:
aseΠ +acw
Π − cse − ccw =D
−D '·1
δcw'
These imply:
δse' (ase
Π ) = δcw' (acw
Π )
aseΠ +acw
Π − cse − ccw =D
−D '·1
δse'
![Page 12: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/12.jpg)
Equilibrium
maxa j(a j − c j )D δ j (a j )+δ− j (a− j )( )
a j* − c j =
D
−D '·1
δ j'
Each site solves: j = se,cw
Necessary condition for each site:
![Page 13: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/13.jpg)
Two Distortions
ase* +acw
* − cse − ccw =D
−D '·1
δse'
+1
δcw'
⎛
⎝⎜
⎞
⎠⎟
Equilibrium • Double Marginalization • “Mis-marginalization”
aseΠ +acw
Π − cse − ccw =D
−D '·1
δse'
Industry Optimum
δse' = δcw
'
![Page 14: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/14.jpg)
Comparison
Until Now • One Search Engine • One Content Website
Let’s Compare With • One Search Engine • Perfectly Competitive Content Websites
Search Engine
Search Engine
Content Website
Content Websites
![Page 15: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/15.jpg)
Comparison
Until Now • One Search Engine • One Content Website
Lets Compare With • One Search Engine • Perfectly Competitive Content Websites
Search Engine
Search Engine
Content Website
Content Websites
Key point Adding competition can reduce both
• Industry profits • Total welfare
![Page 16: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/16.jpg)
Example
δse(ase) = ase2Advertising Technologies:
δcw(acw) =acw2
γ
vi ~U 0,v[ ]Users’ Valuations:
Zero Marginal Costs
![Page 17: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/17.jpg)
Case 1: One SE, One CW
Choosing Advertising Levels—SE and CW solve:
maxase
ase 1−ase2 + acw
γ⎛⎝⎜
⎞⎠⎟2
v
⎡
⎣
⎢⎢⎢
⎤
⎦
⎥⎥⎥ max
acwacw 1−
ase2 + acw
γ⎛⎝⎜
⎞⎠⎟2
v
⎡
⎣
⎢⎢⎢
⎤
⎦
⎥⎥⎥
(Same Users)
![Page 18: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/18.jpg)
Case 2: One SE, Competitive CWs Here, only search engine sets positive advertising
Choosing Advertising Levels—SE solves:
maxase
ase 1−ase2 + 0
γ⎛⎝⎜
⎞⎠⎟2
v
⎡
⎣
⎢⎢⎢
⎤
⎦
⎥⎥⎥
![Page 19: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/19.jpg)
Tradeoff: Double Marginalization versus Mis-marginalization
Total Profits
0.5 1.0 1.5 2.0 2.5
0.4
0.5
0.6
0.7
0.8
γ
One SE, Competing CWs
One SE, One CW
![Page 20: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/20.jpg)
Tradeoff: Double Marginalization versus Mis-marginalization
Total Welfare
0.5 1.0 1.5 2.0 2.5 3.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
γ
v large
v small
One SE, Competing CWs
One SE, One CW
![Page 21: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/21.jpg)
Proposition
a) Starting from the Nash Equilibrium outcome with a single SE and a single CW, if the CW advertising level, acw , is exogenously decreased, and the SE responds optimally, then total industry profits increase
b) Starting from an outcome featuring competitive CW advertising, if the CW advertising level, acw, is exogenously increased, then total industry profits increase, so long as
δcw' (ccw) <δse
' (ase* (δcw(ccw)))
![Page 22: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/22.jpg)
What Can We Take from This?
• When complementary websites advertise differently, there are (at least) two sources of economic distortion:
• Level of distraction • Payoff from distracting
• Empirically, payoff from showing ads seems to be highly variable: In 2007, the CPM paid by advertisers varied from less than $1 to more than $100 (source: Evans ‘08)
• When evaluating conduct of firms in this industry, both of these issues should be taken into account
• Potential benefits of allowing sites to share information about users?
![Page 23: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/23.jpg)
In the Paper
General, price theoretic treatment of the problem
• Start off with one site, examine different advertising technologies
a
!(a," )
a
!(a," )!a" '' large !a" '' small
![Page 24: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/24.jpg)
In the Paper
General, price theoretic treatment of the problem
• Start off with one site, examine different advertising technologies
0.2 0.4 0.6 0.8 1.0 1.2
0.5
1.0
1.5
2.0
![Page 25: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/25.jpg)
In the Paper
General, price theoretic treatment of the problem
• Start off with one site, examine different advertising technologies • Analyze problem with arbitrary number of sites
• Two fundamental distortions • Double marginalization • Mis-marginalization
![Page 26: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/26.jpg)
In the Paper
General, price theoretic treatment of the problem
• Start off with one site, examine different advertising technologies • Analyze problem with arbitrary number of sites
• Two fundamental distortions • Double marginalization • Mis-marginalization
• Salop model: 1 search engine, n content websites • Study effects of differentiation, incentives for entry • Surprising result: In equilibrium, users benefit from more differentiation/ less entry by content websites
![Page 27: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/27.jpg)
Ongoing Work
• Relate “non-transferability of utility” of this setting to two-sided platform competition
• Incorporate more flexible treatment of websites’ complementarity • Better understand relation to Cournot with
asymmetric costs (e.g. Schwartz ’89)
![Page 28: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/28.jpg)
Search
Searchers
aw as + aw
Content
Direct Visitors
Partial Complementarity
Searcher benefit direct visitors, and direct visitors harm searchers
![Page 29: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/29.jpg)
Unreliable Content Sites
λvi − as2 − aw ,1
2
(1− λ)λvi − as2 − aw ,12 − aw ,22
1 2
![Page 30: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/30.jpg)
Related Literature • Surveys:
• Evans (RNE ’08), (JEP ‘09) • Link Structure:
• Katona & Sarvary (Marketing Science ‘08) • Dellarocas, Katona & Rand (NET ‘10)
• Competition Among Complement Producers: • Casadesus-Masanell, Nalebuff & Yoffie • Cheng & Nahm (RJE ‘07) • Lerner & Tirole (AER ’04)
• Advertising on Platforms: • Anderson & Coate (RES ’05) • Choi (IEP ’06) • Crampes, Hartichbalet & Jullien (JIE ’09)
• Search Engine as a Platform • Athey-Ellison (QJE ‘11) • Gomes (WP ‘11) • White (WP ’09)
![Page 31: The attention economy of](https://reader033.vdocuments.mx/reader033/viewer/2022052900/555da465d8b42a45788b56bb/html5/thumbnails/31.jpg)
Conclusion
1. Multiple websites are often complements 2. They use very different methods to turn user attention into revenue This combination of factors leads to two separate coordination problems 1. Double Marginalization: too much nuisance 2. Mis-marginalization: inefficient nuisance For websites, there is a tradeoff between solving one and solving the other