Essays on Mobile Advertising and Commerce

Preparing to load PDF file. please wait...

0 of 0
100%
Essays on Mobile Advertising and Commerce

Transcript Of Essays on Mobile Advertising and Commerce

Essays on Mobile Advertising and Commerce
A dissertation presented
by
David Yu-Chung Chen
to
The Committee for the Ph.D. in Science, Technology and Management
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy in the subject of
Science, Technology and Management
Harvard University Cambridge, Massachusetts
May 2009

c 2009 – David Yu-Chung Chen All rights reserved.

Professor David C. Parkes Professor Peter A. Coles Dissertation advisors
Professor Benjamin G. Edelman Committee member

David Yu-Chung Chen Ph.D. candidate

Essays on Mobile Advertising and Commerce
Abstract
The mobile industry holds the promise of increasing connectivity, productivity, and entertainment as mobile devices become ever more ubiquitous and powerful. Many of the market structures that enable mobile commerce are still under rapid development and afford us opportunities to ask new questions about market design. This thesis examines three such markets and mechanisms that drive advertising platforms and application stores for mobile commerce.
In the first essay, I analyze a mobile web advertising auction that employs a proportional allocation rule in which advertisements are shown with frequencies proportional to the bids. Proportional allocation is used to address the space constraints of the mobile environment and the accompanying ad fatigue. I show that the second-price
iii

rule currently used in real-world auctions admits no pure-strategy Nash equilibrium. I propose the use of a first-price rule and prove the existence of a unique pure-strategy Nash equilibrium. This reverses the sponsored search result in which the second-price auction has pure-strategy Nash equilibria while the first-price auction does not. I also show that by tuning a single parameter in the allocation rule, the auctioneer can make trade-offs between revenue and efficiency.
In the second essay, I examine an optimize-and-dispatch scheme for delivering pay-per-impression advertisements in online and mobile advertising. Using traffic predictions based on historical traffic patterns, the platform provider seeks to allocate future inventory to advertisers such that commitments are fulfilled in expectation, and no single advertiser bears too much of the burden if actual traffic diverges from predicted traffic. I propose a maximum entropy approach and provide theoretical analysis and simulation to show how it accomplishes these goals.
In the final essay, I analyze the market used in mobile device application stores that enable the widespread distribution of third-party software. I characterize the conditions under which an application store seeking to maximize revenue should rank applications based on revenue versus download volume. I also present empirical data from the Apple iPhone App Store to illustrate some of the general features of this type of market.
iv

Contents

Abstract

iii

Acknowledgments

vii

1 Proportional Allocation Share Auctions

1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.1.1 Related Literature: Proportional Share Mechanisms . . . . . . 4

1.2 First-Price Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2.1 First-Price Model . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.2.2 Best Response Strategies, First-Price Rule . . . . . . . . . . . 7

1.2.3 Nash Equilibrium, First-Price . . . . . . . . . . . . . . . . . . 12

1.3 Second-Price Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

1.3.1 Second-Price Model . . . . . . . . . . . . . . . . . . . . . . . . 22

1.3.2 Nash Equilibrium, Second-Price . . . . . . . . . . . . . . . . . 23

1.4 Ad Fatigue Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

1.4.1 Best Response Strategies, Ad Fatigue Extension . . . . . . . . 25

1.4.2 Symmetric Nash Equilibrium, Ad Fatigue Extension . . . . . . 27

1.5 Empirical Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.5.1 Efficiency of FPP . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.5.2 FPP Revenue Versus VCG Revenue . . . . . . . . . . . . . . . 30

1.5.3 FPP Value: Magnitude and Distribution . . . . . . . . . . . . 32

1.5.4 Dynamics of Realizing the FPP Nash Equilibrium . . . . . . . 33

1.6 Auction Design Discussion . . . . . . . . . . . . . . . . . . . . . . . . 36

1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

1.8 Appendix to Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2 Maximum Entropy Banner Allocation

42

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.1.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.2 Banner Allocation Problem . . . . . . . . . . . . . . . . . . . . . . . 46

v

2.2.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 2.2.2 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 2.2.3 Banner Commitment Problem . . . . . . . . . . . . . . . . . . 48 2.2.4 Banner Delivery Problem: Maximum Entropy Formulation . . 49 2.3 Empirical Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 2.3.1 Commitment Violations . . . . . . . . . . . . . . . . . . . . . 54 2.3.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 2.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 2.5 Appendix to Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3 Market Design for Mobile Application Stores

62

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.1.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

3.2 App Store Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.2.1 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.2.2 Demand Functions . . . . . . . . . . . . . . . . . . . . . . . . 68

3.2.3 Ranking Functions . . . . . . . . . . . . . . . . . . . . . . . . 68

3.2.4 Slot Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.2.5 Application Store Revenue Share . . . . . . . . . . . . . . . . 70

3.2.6 Bidder Utility . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.2.7 Platform Revenue . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.2.8 Bidder Strategies . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3.4 Characterizing Nash Equilibria . . . . . . . . . . . . . . . . . . . . . 74

3.5 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

3.6 Empirical Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

3.6.1 iPhone App Store Background . . . . . . . . . . . . . . . . . . 85

3.6.2 App Store Data . . . . . . . . . . . . . . . . . . . . . . . . . . 86

3.6.3 App Store Prices . . . . . . . . . . . . . . . . . . . . . . . . . 87

3.6.4 Application Popularity and the Slot Effect . . . . . . . . . . . 88

3.6.5 Data limitations . . . . . . . . . . . . . . . . . . . . . . . . . . 90

3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

3.8 Appendix to Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . 92

Bibliography

95

vi

Acknowledgments
Graduate school has been an incredible learning experience that would not have been possible without the help of many. First and foremost, I thank the members of my dissertation committee: David Parkes, Peter Coles, and Ben Edelman. David has generously met with me weekly for years, and has watched me mature as a researcher since the beginning of my program. He has been the most significant influence on how I view research, and his insights never cease to amaze me. Peter has spent many an afternoon helping me brainstorm and work through challenging problems. Ben has an incredible knack for seeing the big picture and helping me connect my work with real-world problems. Without their collective guidance, I could not have made it this far.
I am grateful for the countless faculty members at the Harvard Business School and the School of Engineering and Applied Sciences who have helped me along the way. A partial list could not exclude Carliss Baldwin, Yiling Chen, Tom Eisenmann, Lee Fleming, Andrei Hagiu, Marco Iansiti, HT Kung, and Jan Rivkin. I have also learned a tremendous amount from my graduate school friends and colleagues including Florin Constantin, Jacomo Corbo, Adam Juda, Richard Lai, Robin Lee, Merlina Manocaran, Sven Seuken, Marcin Strojwas, Allan Sumiyama, Mark Szigety, Jason
vii

Woodard, Haoqi Zhang, and Feng Zhu. Members of the EconCS research group also deserve mention as they have provided crucial and insightful feedback on all of my work, often from the earliest stages.
These final two years have been incredibly productive and intense and have, I suspect, caused a great deal more stress for my partner Jane than even myself. For all her loving encouragement and patience, I will always be grateful. Finally, I am deeply thankful to my parents Sam and Theresa and my sister Anna for a lifetime of love and faith in me.
viii

Chapter 1
Proportional Allocation Share
Auctions
I analyze a mobile web advertising auction that employs a proportional allocation rule in which advertisements are shown with frequencies proportional to the bids. Proportional allocation is used to address the space constraints of the mobile environment and the accompanying ad fatigue.
I show that the second-price rule currently used in real-world auctions admits no pure-strategy Nash equilibrium. I propose the use of the first-price proportional allocation auction (FPP), and show the existence of a unique pure-strategy Nash equilibrium in a one-shot complete information game. In a dynamic game of incomplete information, I demonstrate in simulation that bids converge to this Nash equilibrium quickly. This reverses the sponsored search result in which the second-price auction has pure-strategy Nash equilibria while the first-price auction does not.
I also show that by tuning a single parameter in the FPP allocation rule, the
1

auctioneer can make trade-offs between revenue and efficiency.
1.1 Introduction
Like their counterparts in the online world, many mobile web companies depend on advertising as a major source of revenue. A number of established players like Google and Yahoo, along with upstarts such as AdMob or Quattro Wireless, have begun acting as advertising intermediaries, connecting advertisers with mobile websites. Due to the highly-constrained environment, most mobile websites choose to only display a single advertisement on each page. The Generalized Second-Price slot auction (GSP), which has emerged as the standard auction for selling advertising in the online world, does not necessarily translate naturally into the mobile setting where there is only one slot for sale.
GSP maximizes expected revenue to the advertising platform and prices with a second-price rule. Research on sponsored search auctions1 has uncovered a number of factors that influence expected revenue, including budget constraints (Abrams et al. (2007), Feldman et al. (2007)), the need for exploration and exploitation (Gonen and Pavlov (2007), Pandey and Olston (2006)), and ad fatigue (Abrams and Vee (2007)). While all these factors remain important in mobile web advertising, the issue of ad fatigue is especially relevant for a number of reasons. First, users can quickly tire of seeing the same ads if there is only one slot per page. Fatigue can be a concern of similar severity with online contextual ads (e.g. Google AdSense), but another characteristic of the mobile ad environment is that contextual matching is
1For a good overview of results from the sponsored search literature, see Lahaie et al. (2007).
2
RevenueAdvertisingCommerceBidsAllocation Rule