Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Internet Finance in China

Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

This book is about Internet finance, a concept coined by the authors in 2012.
Internet finance deals specifically with the impacts of Internet-based technologies,
such as mobile payments, social networks, search engines, cloud computation, and
big data, on the financial sector. Major types of Internet finance include third-party
payments and mobile payments, Internet currency, P2P lending, crowdfunding, and
the use of big data in financial activities.
Internet finance is highly popular and heavily discussed in China. Chinese Premier
Li Keqiang made the healthy development of Internet finance a policy priority in 2014
state-of-union address. This book, as a detailed report on Internet finance in China,
will help readers understand the status quo and development of China’s financial
system and will serve as a guide for readers doing financial businesses in China.

Dr. Ping XIE has been executive vice president of China Investment Corporation
since 2007. He is responsible for private equity and direct investment. He is currently
the Vice Chairman of CF40 Executive Council. From 2005–2007, he served as the
President of Central Huijin Investment Corporation, which is now the domestic
arm of CIC. He played an important role in the reform of China’s big five state
owned banks. Prior to that, he worked at the People’s Bank of China for 20 years,
where he held a number of senior positions, including the Head of the Financial
Stability Department and Head of the Research Department. He has conducted
extensive research in monetary theory and policy, comparative studies of financial
systems, financial markets, rural finance, and financial regulation. He is a three-time
winner of the Sun Ye Fang Economic Prize (1995, 2000, and 2005).

Dr. Chuanwei ZOU works at China Investment Corporation. He received his

BS in statistics and MA in economics from Peking University, and his PhD in
economics from the Graduate School of People’s Bank of China (now PBC School
of Finance, Tsinghua University). His research work is focused on credit markets,
risk management, and financial regulation.

Dr. Haier LIU currently works at Guangdong University of Finance. He received

his MA and PhD in economics from Southwestern University of Finance and
Economics. His research work is focused on mobile banking and rural finance.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

The Editorial Board

Directors
TU Guangshao Executive Vice Mayor of Shanghai
WAN Jianhua Chairman of E-Capital Transfer Co., Ltd
QIAN Yingyi Director, Shanghai Finance Institute

Deputy Directors
WENG Zuliang Chief, CPC Huangpu District, Shanghai
XU Zhen Chairman, Shanghai Clearing House

Chief Editors
LI Xunlei Vice President & Chief Economist, Haitong Securities
LIAN Ping Chief Economist, Bank of Communications
LIAO Min Director, Shanghai Office, China Banking Regulatory
Commission
MIAO Jianmin President, China Life Insurance (Group) Company
WANG Qing President of Shanghai Chongyang Investment Management
Co., Ltd
ZHANG Chun Executive Dean, Shanghai Advanced Institute of Finance,
Shanghai Jiaotong University
ZHENG Yang Director, Shanghai Financial Services Office
ZHONG Wei Deputy Director, Shanghai Finance Institute

Executive Editor
WANG Haiming Executive Deputy Director, Shanghai Finance Institute

Chief Translators
Martin Chorzempa
John Lin

The “New Finance Book Series” was created by the Shanghai Finance Institute
(SFI). The book series traces developments in new finance, explores new trends,
pursues solutions to novel problems, and inspires new knowledge.

Founded on July 14, 2011, the Shanghai Finance Institute (SFI) is a leading non-
governmental, non-profit institute dedicated to professional academic financial
research. SFI is operated by the China Finance 40 Forum (CF40) and has a strategic
cooperation with the Shanghai Huangpu District government.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Ping Xie

Haier Liu
Chuanwei Zou
Introduction and Practical Approaches
Internet Finance in China
 First published 2016
by Routledge
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
and by Routledge
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

711 Third Avenue, New York, NY 10017

Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2016 Ping Xie, Chuanwei Zou, and Haier Liu
The right of Ping Xie, Chuanwei Zou, and Haier Liu to be identified
as authors of this work has been asserted by them in accordance with
sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or
utilised in any form or by any electronic, mechanical, or other means, now
known or hereafter invented, including photocopying and recording, or in
any information storage or retrieval system, without permission in writing
from the publishers.
Trademark notice: Product or corporate names may be trademarks or
registered trademarks, and are used only for identification and explanation
without intent to infringe.
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloging in Publication Data
Xie, Ping, 1955- | Zou, Chuanwei. | Liu, Haier.
Internet finance in China : introduction and practical approaches /
Ping Xie, Chuanwei Zou, Haier Liu.
Description: 1 Edition. | New York, NY : Routledge, 2016.
Identifiers: LCCN 2015037695| ISBN 9781138195080 (hardcover : alk. paper) |
ISBN 9781315637921 (ebk)
Subjects: LCSH: Internet banking—China. | Mobile commerce—China. | Big
data—China.
Classification: LCC HG1708.7 X54 2016 | DDC 332.0285/4678—dc23
LC record available at http://lccn.loc.gov/2015037695

ISBN: 978-1-138-19508-0 (hbk)

ISBN: 978-1-315-63792-1 (ebk)

Typeset in Bembo
by diacriTech, Chennai
 Contents
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

List of figures xi
List of tables xiii

1 Introduction 1
Section 1: Defining Internet finance 1
1 Internet finance is a forward looking concept 1
2 The changeable and unchangeable of Internet finance 2
Section 2: The development of Internet finance 5
1 The rise of Internet finance 5
2 The scope of Internet finance 6
3 Governmental attitudes toward Internet finance 7
Section 3: Structure of this book 8

2 The theory of Internet finance 10

Section 1: Introduction 10
Section 2: Payment under Internet finance 11
Section 3: Information processing under Internet finance 12
1 General views 12
2 Model 13
Section 4: Resource allocation under Internet finance 17
1 General views 17
2 “Transaction possibility set” 18
Section 5: Concluding remarks 20
Notes 21
 vi Contents
3 The internetization of finance 23
Section 1: The internetization of finance 23
1 Network banks 23
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2 Mobile banks 25
Section 2: Network securities companies 32
1 Business models worldwide 32
2 Chinese models 33
Section 3: Network insurance companies 34
1 Development 34
2 Major types 35
Section 4: Network financial trading platforms 36
1 SecondMarket 36
2 SharesPost 37
Section 5: Distribution networks for financial products 38
Notes 39

4 Mobile and third-party payments 40

Section 1: Basic concept and development 40
1 Mobile payments 40
2 Third-party payments 41
3 Third-party mobile payments 41
Section 2: Basic principles and the account system 42
1 Mobile payments: Forms and principles 42
2 The principle of third-party payments 43
3 The function and categories of accounts 44
Section 3: Properties of financial products and monetary control 45
1 Payments and electronic currency 45
2 Payments and financial product properties 48
3 Payment and monetary control 48
Section 4: Analysis of Tencent’s Wechat payment 49
1 A brief history 49
2 Operating principles 50
3 Risk and risk management 50
Section 5: A brief analysis of Yu’E Bao 51
1 Creation and development 51
2 The fundamentals of Yu’E Bao 51
3 Risks 52
 Contents vii
4 Financial products as instruments of payment 53
5 Main models 53
Notes 55
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

5 Internet currency 56
Section 1: The concept of Internet currency 56
Section 2: The economics of Internet currency 57
1 Online communities and network economics 57
2 The new form of currency: Facilitating network payments 58
3 The risks of Internet currency 59
Section 3: Bitcoin 60
1 Working mechanism 60
2 Issuance mechanism 63
3 Bitcoin’s effect 64
4 Innovation and insufficiency 65
5 Risk and regulation 66
Notes 68

6 Big data 70
Section 1: Big data: Concepts and main types 70
1 Basic concept 70
2 Recorded data 71
3 Data based on graphs 72
4 Data in order 72
Section 2: Primary tasks for big data analysis 73
1 Classification 73
2 Regression 75
3 Association analysis 76
4 Cluster analysis 77
5 Recommender system 77
6 Anomaly detection 78
7 Link analysis 79
Section 3: Comparison of big data analysis and econometrics 81
1 Introduction to econometrics 81
2 Differences between big data analysis and econometrics 83
3 Inherent contact between big data analysis and econometrics 84
Notes 89
 viii Contents
7 Big data-based credit and Internet lending 90
Section 1: Big data-based credit 90
1 Basic concepts 90
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2 Brief introduction of the credit rating center of People’s Bank of China 91

3 Brief introduction of the Alibaba credit system 93
4 Main credit evaluation approaches 94
Section 2: Big data-based Internet lending 99
1 Analysis on Kabbage 99
2 Ali (Alibaba) small loan analysis 101
Notes 106

8 P2P network loans 107

Section 1: Analysis of Lending Club 107
1 Operational framework of Lending Club 107
2 Borrowers 109
3 Investors 112
4 Supervision of Lending Club in the U.S. 113
Section 2: Economics of P2P network loans 115
1 Comparison with private finance 115
2 Comparison with bank deposit 118
3 Core technology 118
Notes 122

9 Crowdfunding 123
Section 1: About Kickstarter 123
1 The establishment of Kickstarter 123
2 The operating model of Kickstarter 123
3 The pricing structure of Kickstarter 124
4 Projects on the Kickstarter platform 124
5 Kickstarter guidelines 124
Section 2: Operational principle and development 125
1 Relation between crowdfunding and crowdsourcing 125
2 Operational principles and project classification 125
3 Development overview 128
Section 3: The economics of crowdfunding 131
1 Common characteristics 131
2 Incentives 131
3 Crowdfunding risks 133
 Contents ix
4 Market design 134
5 Several open questions 136
Notes 138
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

10 The regulation of Internet finance 139

Section 1: Should Internet finance be regulated? 139
1 Necessities of regulation 139
2 Specialties of regulation 141
Section 2: Functional regulation of Internet finance 143
1 Prudential regulation 144
2 Behavioral regulation 145
3 Protection for consumers of financial products 146
Section 3: Organizational regulation of Internet finance 147
1 Current regulatory framework 148
2 Regulation of P2P network loans 149
3 Public finance regulation 152
Notes 153

11 Internet exchange economy 154

Section 1: Analysis of the sharing economy 154
1 Definition of sharing economy 154
2 Case analyses 155
Section 2: Principles of the Internet-based exchange economy 157
1 Basic framework 157
2 Logistics and payment 160
3 Information processing 161
4 Resource allocation 162
Section 3: The relationship between the Internet-based exchange
economy and Internet finance 167
1 Internet finance: a special case 167
2 Internet finance derives from the Internet-based exchange economy 167
3 The Internet-based exchange economy serves Internet finance 167
Notes 168

12 Issues requiring further research 169

Section 1: Utilizing big data for securities investment 169
1 The Black–Litterman model 169
2 The fundamental law of active management 174
3 How is big data implemented? 177
 x Contents
Section 2: Big data in actuarial science 180
1 An introduction to life contingency 180
2 An introduction to non-life contingency 183
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

3 How is big data implemented? 185

Notes 189

Appendix 1 Overview of SFI 190

Appendix 2 Organizational structure 191
 List of figures
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

1.1 How to understand a state without intermediaries and markets.

Arrows denote fund flows 3
1.2 Three pillars of Internet finance 4
1.3 Structure of this book 9
4.1 The market structure of online third-party payments 41
4.2 The market shares for mobile third-party payment 42
4.3 The demand curve for mobile payments 47
4.4 The process of Wechat payment 50
4.5 The operational structure of Yu’E Bao 52
5.1 A decentralized payment system 60
5.2 An illustration of Bitcoin transaction 61
5.3 The full Bitcoin transaction chain 62
5.4 Bitcoin’s volume of issuance (by the end of January 2014, in millions) 64
5.5 The “mining pool” distribution mode 65
5.6 Bitcoin’s price volatility in 2013 65
6.1 ROC curve 75
7.1 Business model of the credit rating center 91
7.2 Data processing framework of the credit rating center 93
7.3 Products of the credit rating center 93
7.4 Merton model 95
7.5 Business model of Kabbage 100
7.6 Main categories of loans by Ali Small Loan 102
7.7 Business model of Ali Small Loan 103
7.8 The hydrological transaction forecast model of Ali Small Loan 104
7.9 The PD model of Ali Small Loan 104
8.1 Credit transactions facilitated by Lending Club 108
8.2 Operational framework of Lending Club 108
8.3 Purpose of loans of Lending Club 110
8.4 Portfolio building tool of Lending Club 113
8.5 Risk diversification effect of loan investment 114
8.6 Supervision on Lending Club in the United States 115
 xii List of figures
8.7 ROC curve of Lending Club 119
9.1 Complexity variation of corporate finance tools 126
9.2 The basic process of crowdfunding 127
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

11.1 Edgeworth box 164

11.2 Markowitz mean-variance model 166
12.1 The general steps of active management 178
 List of tables
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

3.1 Information technology and commercial bank innovation 24

3.2 The forms of mobile banking 25
3.3 Overview of African national mobile banking 26
3.4 M-PESA fees (Kenyan shillings) 28
5.1 Balance on Bitcoin accounts 66
6.1 Confusion matrix 74
7.1 Sources of Alibaba credit system 94
7.2 Data sources of Kabbage 100
8.1 Credit rating method of Lending Club 110
8.2 Loan pricing mechanism of Lending Club 111
8.3 Loan service fee rate 112
8.4 Characteristics of risk and return in Lending Club by rating 119
9.1 Classification matrix of crowdfunding projects 127
9.2 Ten selected crowdfunding platforms’ performance data 129
12.1 Asset types and symbols 171
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

This page intentionally left blank

 1 Introduction
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

SECTION 1: DEFINING INTERNET FINANCE

The concept of Internet finance originated in 2012. It includes all financial

transactions and organizational structures under the influence of the Internet and
ranges from traditional financial intermediaries and markets such as banks, securi-
ties, insurance, and exchanges all the way to those corresponding with Walrasian
equilibrium.

1 Internet finance is a forward looking concept

It takes imagination to understand the concept of Internet finance. Today, all levels
of institutions are involved. E-commerce companies, IT companies and mobile
network operators now join the typical players such as banks, securities, insurance,
and funds. The latter group is especially active in the evolution of various business
models. This then blurs the line between the financial and non-financial industries.
Even so, at the time of writing, the development of Internet finance is far
from complete. We optimistically estimate that it will take at least twenty years for
Internet finance to fully develop. We draw this conclusion mainly based on the
following two considerations. First, the speed of Internet finance’s development
largely depends on that of Internet technology, rather than that of finance itself. We
predict that in twenty years, Internet technology will further decrease transaction
costs and information asymmetry in financial transactions. Second, after twenty
years, a generation that has grown up with the Internet will be the pillar of our
society. Their habit of Internet use will influence financial transactions and organi-
zational structures across the economy.
In such circumstances, academic research must look forward to examine future
prospects. Internet finance is not merely a summary of history or the current sit-
uation; it is a forward-looking concept that emphasizes the future. It embodies
three “rational anchors.” First, Internet finance is grounded in reality. Forms of
Internet finance that have already arisen serve as a starting point for our considera-
tion. Second, Internet finance corresponds with basic principles of economics and
finance, just like moving objects follow basic physical principles. We believe that
 2 Introduction
the present theories of economics and finance provide sufficient analytical tools
to explain forms already in existence and predict the prospects of Internet finance.
This approach guides our thinking for this book. Third, the ultimate culmination
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

of Internet finance research is the unintermediated market corresponding with

Walras’ General Equilibrium. Walras’ General Equilibrium is a theoretical corner-
stone of economics. It demonstrates that with a series of ideal assumptions, a per-
fectly competitive market will reach an equilibrium state, in which supply of and
demand for all commodities are equivalent and resource allocation is Pareto optimal
(making someone better off without making anyone else worse off). In Walras’
General Equilibrium, there are no financial intermediaries, and markets and cur-
rency become unnecessary. Influential factors such as asymmetric information and
transaction costs lead to the existence of financial intermediaries and markets in
reality. Asymmetric information and transaction costs have markedly decreased due
to the Internet. Internet finance will gradually reach an unintermediated state for
finance and markets (Figure 1.1).

2 The changeable and unchangeable of Internet finance

2.1 The unchangeable
First, the core function of Internet finance is unchangeable. It allocates resources
spatially and temporally in an uncertain environment and thus serves the real econ-
omy. Specifically, it includes: payment and settlement, fund allocation, channels
for transferring economic resources, risk management, information provision, and
incentive management.
Second, the connotation of financial contracts such as equity rights, creditor
rights, insurance, and trust funds is unchangeable. The nature of financial contracts
is to specify rights and obligations, which mainly aims at the future cash flows of
each party in an uncertain environment in the future. For example, equity rights
denote residual claims or interests in the assets of all shareholders. Creditor rights
permit creditors to charge principal and interest for a fixed period. Not long ago,
financial contracts mainly existed in physical form (like the earliest A shares in
China), but now are generally in electronic form. Moreover, it sets up a mechanism
concerning trusteeships, transactions, and clearing. The connotation of financial
contracts is unchangeable no matter the form. In Internet finance, all contracts are
digitized and constitute the concept’s foundations.
Third, the meaning of some concepts such as financial risk and externality are
unchangeable. In Internet finance, risk means the possibility of suffering future
loss. Concepts and analytical frameworks of market risk, credit risk, liquidity risk,
operational risk, reputational risk, and legal compliance risk always apply. Problems
such as misleading consumers, exaggerated advertising, and fraud also exist in
Internet finance.Therefore, basic concepts for the regulation of Internet finance are
unchangeable, and the main types of regulation like prudential regulation, behavio-
ral regulation, and investor protection also apply. However, it differs from traditional
finance for specific regulatory measures (see Chapter 10).
 Introduction 3

Financial
intermediaries
(such as banks)
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Depositors Borrowers

Financial
markets (such as
exchanges)

With financial intermediaries and

markets

Depositors Borrowers

Internet

Without intermediaries and

markets

Figure 1.1 How to understand a state without intermediaries and markets. Arrows denote
fund flows.

2.2 The changeable

The changes of Internet finance are mainly manifest in the penetration of the Internet
into finance. First, we have the influence of Internet technology, mobile and third-
party payments, big data, social networks, search engines, and cloud computing.
 4 Introduction
Internet finance

Information
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Internet processing Resource

Payment currency (analyzing allocation
big data)

Figure 1.2 Three pillars of Internet finance.

The Internet can significantly decrease transaction costs and asymmetric informa-
tion, increase efficient management and pricing of risk, and expand the boundaries
of possible transactions. It provides a platform for suppliers and demanders of capital
to transact directly, which influences financial transactions and organizational forms.
Second, we have the influence of the Internet “spirit.”The core of Internet spirit
lies in openness, sharing, decentralization, equality, freedom of choice, usefulness,
and democracy. On the other hand, traditional finance is more elitist. It stresses
professional qualifications and entry thresholds, which manifest in the fact that not
everyone can access financial services. Internet finance reflects the emergence of
the individual organization and platform model in finance. Finance will be even
more useful when financial distribution and specialization are weakened.

2.3 Three pillars of Internet finance (Figure 1.2)

The first pillar is payments. Payments are the infrastructure of finance and flow
through all financial activities. In Internet finance, payments are based on mobile
and third-party payments. It is out of the range of traditional payment and settle-
ment systems, which are predominantly led by banks, and it reduces transaction
costs. In Internet finance, payment links financial products, promoting richer busi-
ness models (see Chapter 3 for the analysis of Yu’e Bao). Finally, due to the close
connection of payment and currency, Internet currency has emerged.
The second pillar is information processing. Information is the core of finance.
It constitutes the foundation for the allocation of financial resources. In Internet
finance, big data is widely used in information processing (reflected in algorithms
and automatic, high-speed calculations). It improves the pricing of risk and dramat-
ically decreases asymmetric information. Internet finance’s information processing
method is the most significant difference between indirect finance through com-
mercial banks and direct finance through capital markets.
The third pillar is resource allocation. Allocation of financial resources refers to
the way they are transferred from suppliers to demanders. The fundamental goal of
financial activities is resource allocation, thus the efficiency of resource allocation
in Internet finance is the basis of its existence. In Internet finance, financial prod-
ucts are closely connected with the real economy. Boundaries for possible transac-
tions are significantly expanded, and the supply and demand of funds can reach
 Introduction 5
equilibrium without traditional financial intermediaries and markets, such as banks,
securities companies, and exchanges.
We believe that any kind of financial transaction and contract form possessing
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

at least one of the characteristics in the three pillars should be classified as Internet
finance. This can be viewed as a constructive definition of Internet finance that
can cover its main types. For example, if commercial banks use big data to conduct
credit assessment and lending businesses, they are using Internet finance.

SECTION 2: THE DEVELOPMENT OF INTERNET FINANCE

1 The rise of Internet finance

First, the Internet exerts its strongest influence on industries that do not need
physical logistics, and finance is no exception. Over the past decade, it has exerted
a revolutionary influence on industries such as telecommunication, journalism,
books, publishing, television, music, and retail, followed by the film, education, and
advertising industries. One clear example is that traditional handwritten letters have
virtually disappeared after the rise of e-mail. Finance is essentially figures, (fixed
assets account for a relatively low percentage of financial assets) and it has the same
numerical characteristics as the Internet. All financial products can be viewed as a
combination of figures and all financial activities as a movement of figures on the
Internet.
Second, society is moving toward digitalization. About 70% of societal informa-
tion has already been digitized. In the future, sensory equipment will be increasingly
common. For example, smartphones integrate complicated sensors and programs.
Activities such as shopping, consumption, reading, and others will move online.
After the popularization of 3D printing, even manufacturing will move online.The
Internet will bring about complicated new ways to communicate, cooperate, and
divide labor. Under these conditions, it is possible that 90% of societal informa-
tion could be digitized. This creates the conditions under which big data can be
employed in finance. If individuals and companies put the majority of their infor-
mation online, one could make accurate estimations of their credit quality and the
outlook for their profitability (see Chapter 6 for a discussion of big data).
Third, risk-management tools and big data, which are accumulated by enter-
prises operating in the real economy, can also be applied in financial activities, typi-
cal cases of which are e-commerce companies like Alibaba and Jingdong. Internet
finance thus naturally connects up with e-commerce and the sharing economy.
Fourth, some distortions and inefficient aspects of the Chinese financial system
create space for the development of Internet finance.

• Formal finance has not always been able to meet the financial demands of
small- and medium-sized enterprises (SMEs) and farmers, and private finance
(or informal finance) has intrinsic limitation and risks.
• Economic restructuring has created spending and credit demands beyond
those which can be met by formal finance.
 6 Introduction
• Banks are hugely profitable when the interest spreads between deposits and
loans are protected, which is currently the case in China due to interest rate
controls on deposits and lending. Capital thus has a strong incentive to enter
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

the banking industry.

• The capped deposit interest rate, which has recently been at or below inflation,
a stock market that has been weak for years and the recent restrictive policy
on home purchases create barriers to effectively fulfilling the financial services
demands of the population.
• Under the current IPO management system, channels for equity financing are
impeded.
• Sales of securities, funds, and insurance are restricted to banks, so they are moti-
vated to expand network sales.

In this context, Internet finance in China currently aims to meet the credit
financing needs of individuals and SMEs, the equity financing needs of some crea-
tive projects, the investing and financing needs of ordinary people, and financial
product sales through non-bank channels. It thus can also be considered inclusive
finance. Internet finance will not have as large of an impact on big enterprises or
big projects, but the share of such enterprises in the economy will decline over
time. Moreover, financial resources in China have been allocated mostly to the cen-
tral government and state-owned sectors for a long time.We predict that in the next
ten years, large amounts of financial resources will be distributed from the central
to local and state-owned to private sectors. This profound change in the distribu-
tion of financial resources will contribute to the development of Internet finance.

2 The scope of Internet finance

One typical example of the concept of spectrum is that of light. Sunlight can
be divided into several continuous spectra according to their frequency. One end
of the spectrum of Internet finance includes traditional financial intermediaries
and markets such as banks, securities, insurance, and exchanges; the other end is
a state without intermediaries or markets, corresponding to the Walrasia General
Equilibrium. All other kinds of financial transactions and organizational forms
between the two ends are on the range of Internet finance.
We divide Internet finance into six main types depending on the three pillars:
payment, information processing, and resource allocation.

1 The internetization of finance

Finance internetization reflects that the Internet is replacing manual provision
of services by financial intermediaries and markets. This includes network and
mobile banks, such as by ING Direct in Europe and M-Pesa in Kenya. It also
comprises network stock exchanges, such as Charles Schwab in the United
States. Network insurance companies such as Esurance provide insurance
 Introduction 7
products over the Internet, and network platforms such as SecondMarket,
SharePost, and the Chinese Qianhai equity exchange handle financial trans-
actions. Finally, network sales of financial products such as Yu’e Bao, Baidu
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Finance and Rong 360 are growing parts of the Internet finance landscape in
China and elsewhere.
2 Mobile and third-party payments
Mobile and third-party payments reflect the influence that the Internet has
on the payments industry. The main players are Paypal in the USA, Alibaba,
Caifutong, and Tencent’s Wechat payment in China.
3 Internet currency
Internet currency reflects the influence that Internet has on currency. It
includes Bitcoin, Q-coin, and Amazon Coins.
4 Using big data for credit scoring and network loans
Credit bureaus are key for loan provision, so we discuss credit investiga-
tion together with network loans. ZestFinance in the USA and Kreditech
in Germany already do credit scoring based on big data, while Kabbage in
the USA and Alibaba microcredit in China give out network loans based on
big data.
5 Peer-to-Peer (P2P) network loans
P2P network loans include personal debit and credit on the Internet, which
is represented by Prosper, Lending Club, Zopa, Credit Ease, Lufax, Paipai
Lending, and P2P Lending.
6 Crowdfunding
Crowdfunding is financing over the Internet in which many investors are able
to contribute small sums for a project or company. Current crowdfunding plat-
forms include Kickstarter (which provides returns through means other than
equity) and AngelCrunch.

We must clarify that there are no clear lines between so many different forms of
Internet finance. They are all constantly changing. For example, usage-based insur-
ance came into being in the insurance industry; equity research found that Twitter
activity can predict future share prices; and the combination of big data with actu-
arial studies and portfolio investment will contribute to many new business models
(see Chapter 12). Thus, the six main types of Internet finance in this book are nei-
ther mutually exclusive nor collectively exhaustive.

3 Governmental attitudes toward Internet finance

Generally speaking, the Chinese government has taken a positive attitude toward
Internet finance. In April 2013, the Chinese State Council set up nineteen key
financial projects, including “the development and regulation of Internet finance.”
The research group is composed of the People’s Bank of China (PBoC), the China
Banking Regulatory Commission (CBRC), the China Securities Regulatory
 8 Introduction
Commission (CSRC), the China Insurance Regulatory Commission (CIRC), the
Ministry of Industry and Information Technology, the Ministry of Public Security,
and the office of Legal Affairs. The results of this research will have far-reaching
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

implications for the development of Internet finance in China.

In August 2013, Internet finance was written into two important State Council
documents. In “Suggestions to Promote the Development of Small- and Medium-
Sized Enterprises” by the State Council, the Council proposes making the most of
new technologies and tools to create innovate network financial service models.
In “Several Opinions on Promoting Information Consumption and Expanding
Domestic Demand” by the State Council, the government proposes promoting
Internet financial innovation and standardizing Internet financial services.
The PBoC gave a positive evaluation of Internet finance in its Q2 2013 execu-
tive report of currency policy. It deemed that Internet finance has advantages in its
high transparency, inclusiveness, low transaction costs, convenient payments, rich
credit information, and high efficiency of information processing. This was the
first time that Internet finance has been written into such authoritative official
documents.
In December, 2013, the Payment and Clearing Association of China established a
specialized committee for Internet finance and introduced a self-regulatory system.
Local governments are also very enthusiastic about Internet finance. For example,
Zhongguancun and Shijingshan in Beijing and the Huangpu district in Shanghai all
view Internet finance as an important emerging industry and have created incen-
tives to promote the development of Internet finance in their jurisdictions.

SECTION 3: STRUCTURE OF THIS BOOK

This book has 12 chapters divided into four parts. Figure 1.3 shows the structure.

1 Chapter 1 is a brief introduction of the book.

2 Chapters 2–10 is the main body of this book, discussing Internet finance.
Chapter 2 discusses the principles of Internet finance.
Chapters 3–9 discuss six main types of Internet finance: finance internetization
(Chapter 3), mobile and third-party payments (Chapter 4), Internet currency
(Chapter 5), credit bureaus and network loans based on big data (Chapter 7),
P2P network loans (Chapter 8), and crowdfunding (Chapter 9). Before
Chapter 7 (credit investigation and network loans based on big data), we will
have a general discussion about big data to serve as a technical base. Chapters 4
and 5 mainly discuss the first pillar, Chapters 6 and 7 the second, and Chapters
8 and 9 the third. Chapter 10 discusses the regulation of Internet finance.
3 Chapter 11 discusses the Internet exchange economy.
4 Chapter 12 points out questions that need further research, mainly the
application of big data for portfolio investment and actuarial studies.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Chapter 1
Introduction

Chapter 2 Principles of Internet

finance

Chapter 11
Chapter 4 Chapter 7
Chapter 6 Chapter 8 Internet
Chapter 3 Mobile Chapter 5 Credit Chapter 9
Generally p2p exchage
Internet and Internet Investigation Crowdfunding
discuss network economy
webification third-party Currency and network financing
big data loans
payments loans

Chapter 10 Regulation of Internet

finance

Chapter 11 Further research

needed
Introduction

Figure 1.3 Structure of this book.

9
 2 The theory of Internet finance
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

SECTION 1: INTRODUCTION

The basic function of a financial sector is to intermediate funds between those in

need of money and those with extra money. However, financial intermediaries do
not exist in the classic version of general equilibrium theory (Mas-Colell et al.).
Mishkin suggests two reasons why financial intermediaries do exist in the real
world.1 First, financial intermediaries have special technologies and economies of
scale to reduce transaction costs. Second, financial intermediaries process infor-
mation to alleviate adverse selection and moral hazard caused by information
asymmetry.
Traditionally, there are two types of financial intermediaries: commercial banks
(indirect financing) and securities markets (direct financing).They play an important
role in resource allocation and promote economic growth, but also incur large costs.
Internet-based technologies such as mobile payments, social networks, search
engines, and cloud computing will lead to a paradigm shift in the financial sector.
Beside indirect financing via commercial banks and direct financing through
securities markets, a third way to conduct financial activities will emerge, which we
call “Internet finance.” Over the past ten years, similar paradigm shifts driven by the
Internet have occurred in areas such as bookstores, music, and retail sales.2
Under Internet finance, mobile payments of individuals will be efficiently
integrated with central payment systems managed by central banks. Information
processing and risk assessment will be conducted on the Internet, and information
will be highly transparent. Costs of maturity matching and risk sharing will be so
low that financial intermediaries will be unnecessary, and the issuance and trading
of stocks, bonds, and loans will be carried out smoothly. Markets will be so efficient
that they highly resemble the world without financial intermediaries described by
general equilibrium theory. Furthermore, Internet finance will be able to achieve
the same allocative efficiency while greatly reducing transaction costs.
More importantly, under Internet finance, the division of labor of the modern
financial industry will be obsolete and replaced by Internet-related technologies.
Companies and ordinary people will be able to conduct financial transactions on
the Internet. Complicated tasks such as risk sharing and maturity matching will
be simpler and more user friendly. Financial services will be available to all, and
 The theory of Internet finance 11
everyone will enjoy the benefits. In this way, Internet finance is more democratic
than a finance controlled by professional elites.
Internet finance is still in the nascent stage of development. Currently, the most
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

prominent examples of Internet finance include mobile banking and peer-to-peer

(P2P) lending. Internet finance will generate great opportunities and challenges.
Governments can employ Internet finance to address the problem of small- and
medium-sized enterprise (SME) financing, promote the transparency and safety
of informal financing, and increase the availability of financial services. However,
Internet finance will also bring about regulatory challenges.3 For financial and
IT companies, Internet finance will generate tremendous opportunities but also
induce fierce competition. For academia, the payments revolution will shed new
light on monetary policy.
We are the first to introduce the concept of Internet finance and present an
in-depth analysis of this new concept based on case studies, economic modeling,
sociological studies, and IT know-how. Our study focuses on three pillars of
Internet finance.The first pillar is payment. Payment is the most important piece of
the financial system’s infrastructure and influences the form of financial activities.
The second pillar is information processing. Information is at the heart of finan-
cial activities and lays the foundation for resource allocation. For Internet finance,
information processing marks the greatest divergence from commercial banks and
securities markets.The third pillar is resource allocation. Internet finance is justified
by its allocative efficiency.

SECTION 2: PAYMENT UNDER INTERNET FINANCE

Under Internet finance, money and securities will be transferred through mobile
communication networks.
The foundation for mobile payments is the development of mobile commu-
nication technologies, especially the high penetration of smart phones and tablets.
According to Goldman Sachs, the volume of mobile payments reached US$105.9
billion in 2011 and was expected to grow at 42% annually in the coming five years
to reach US$616.9 billion by 2016. The share of mobile payments in the global
payment market was 1% in 2011 and will reach 2.2% by 2015.4
Backed by Wi-Fi and 4G technologies, Internet and mobile communication
networks are increasingly integrated, and cable telephone, radio, and TV networks
will be incorporated in the near future. Mobile payments will then be further
combined with credit cards, online banking, and e-payments to become more
available, convenient, and user-friendly. With the development of security software
such as identity authentication and digital signatures, mobile payments will be
applied not only to small sum payments in daily life, but also to large payments
between companies. It could completely replace payment instruments such as cash
and checks.
Cloud computing ensures the storage and computing capabilities necessary for
mobile payments. Despite becoming increasingly smart, mobile communication
 12 The theory of Internet finance
devices cannot match personal computers in terms of storage capacity and
computing speed due to requirements for portability and size. Cloud computation
can overcome such shortcomings through a transfer of storage and computation
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

from mobile communication terminals to cloud computation servers.

Under Internet finance, the payment system will have the following
characteristics. First, all individuals and institutions will open accounts at the
payment center of their central bank for registration of deposits and securities.
Second, transfers of money and securities will be conducted through mobile
Internet networks. Third, payment and settlement will be electronic. The need
for cash circulation will be greatly reduced. Fourth, the division of labor between
commercial banks and central banks (or the two-tier banking system) will cease
to exist. If deposit accounts are all located at central banks, the theory and prac-
tice of monetary policy will be fundamentally changed.5 For instance, current
deposits reserved for payment will diminish and the proportion of time deposits
will increase. However, such a payment system will not challenge central banks’
role in money supply. The relationship between money and price of goods will
not fundamentally change either.

SECTION 3: INFORMATION PROCESSING UNDER

INTERNET FINANCE

1 General views
Information about capital suppliers and, more essentially, capital demanders, is
imperative in financial activities. Mishkin points out that there are two types of
information processing under direct and indirect financing.6 The first type is the
production and sale of information by private entities. Many specialized institu-
tions produce information that can differentiate the quality of capital demanders,
such as credit rating agencies and research teams at investment banks. Commercial
banks are information producers and users at the same time, so they also belong to
this category. The second type is information disclosure required or encouraged by
governments, such as financial statements of listed companies.
Compared with commercial banks and securities markets, Internet finance will
differ most in its information processing. First, social networks will generate and
spread information, especially information without disclosure obligations. Second,
search engines will structure, sequence, and index information to alleviate the over-
load problem. Third, cloud computation will ensure rapid processing capabilities
for mass information. Thus, the overall picture is that, with the help of cloud com-
putation, information from capital suppliers and demanders will be revealed and
spread through social networks, concentrated and standardized by search engines
to produce a dynamic information sequence. With such processed information,
risk assessment of any capital demander will be carried out at a low cost. Thus
information needs in financial activities will be satisfied in a way similar to the
credit default swap (CDS) market. According to Xie and Zou, CDS creates a time
series of default probability via transaction mechanisms similar to social networks
 The theory of Internet finance 13
and search engines.7 It is even more effective than credit rating agencies. The
following paragraphs discuss the roles of social networks, search engines, and cloud
computation, respectively.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

First, social networks digitize and map social relationships on the Internet and
serve as platforms to release, spread, and share information. Social networks are
based on two foundations. First, networking behaviors are intrinsic to human beings
and characterized by the attributes of exchangeability,8 consistency,9 contagious-
ness,10 and transmissibility.11 Second, the development of the Internet and other
communication technologies has reduced individuals’ costs to release information
and contact with strangers, leading to new types of collaboration such as Chinese
phenomenon of “Cyber Manhunt” and editing of Wikipedia.12 Social networks also
contain lots of relational data, that is information about contact, connection, com-
munity attachment, and gathering.13
Second, from mass information, search engines can identify contents that best
match the needs of information users. Integration between search engines and social
networks is inevitable,14 which is embodied in the development of social search.
Third, with the performance of integrated circuit (IC) approaching physical
limits, cloud computation employs a large number of PCs to share computational
tasks with great extendibility, fault tolerance and consistency of multiple backup
data, producing huge computational capabilities and storage space. Cloud com-
putation thus facilitates the processing of mass information and is instrumental
in the development of search engines.15 The financial sector, as one of the biggest
users of computation power, will also be influenced by the development of cloud
computation.
We use a simple example to demonstrate information processing under
Internet finance. Individuals (or institution) have many stakeholders, who all have
some information about their wealth, employment status, personality, and so on.
If all stakeholders’ information is released and pooled on social networks, and
inaccurate information is disputed or filtered through social networks and search
engines, we will get a reliable picture of their creditworthiness. Social networks
also enable the accumulation of “social capital” among people, with which costs
of financial activities will drop considerably and opportunistic behaviors will be
constrained.

2 Model
2.1 Assumptions
Suppose there are n persons who express their views on default probability of a per-
son or entity (“reference entity”) by trading a financial product similar to CDS.This
financial product is essentially a two-period contract with two types of participants,
sellers and buyers. For one unit of financial product, in the first period, the buyer
pays to the seller a premium of s (also the financial product’s price); in the second
period, if the reference entity defaults, the seller compensates the buyer an amount
of l. Suppose l is determined beforehand and s by market equilibrium. Our target
is to explore the information content of s.
 14 The theory of Internet finance
Suppose all persons have an initial wealth in the first period, which exists in the
form of risk-free bonds with zero risk-free rate. In the first period, every person
decides whether to buy or sell the financial product and by what amount based on
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

his information, wealth and risk preference. In the second period, if the reference
entity defaults, compensation between buyers and sellers is triggered. Suppose that
the utility of all persons is a CARA function of second-period wealth with absolute
risk aversion coefficient α , that is:

u ( w ) = −α exp ( −α ⋅ w ) (2.1)

We use Y to denote the fundamentals of the reference entity, such as credit

record, income, liabilities, and so on. Suppose the default of the reference entity
is characterized by a Logistic model. If Y + e > 0, default occurs; otherwise, there
is no default. e is random disturbance term and follows a Logit distribution with
exp ( e )
cumulative probability distribution function F ( e ) = . Hence, the default
probability of the reference entity is: 1 + exp ( e )

exp (Y )
P = Pr (Y + e > 0 ) = 1 − Pr ( e ≤ −Y ) = (2.2)
1 + exp (Y )

Suppose there are two types of information in Y .The first is public information,
denoted by X . The second is private information acquired by every person, with
private information of no. i person denoted by Z i . We introduce five assumptions
on information structure:
n
1 Y = X + ∑ Z , that
i =1
i is simple linear addition between public and private
information;
2 For any i, E (Z i ) = 0;
3 For any i ≠ j, E (Z j Z i ) = 0, that is private information of different persons
is uncorrelated;
4 For any i, E (Z i X ) = 0, that is public and private information is uncorrelated;
5 Assumptions I–IV are public knowledge.

2.2 Model solution

2.2.1 Representative person’s utility maximization problem

Take person i as an example. Based on his estimation of default probability, he
decides how to buy or sell the financial product in the first period to maximize
expected utility.
First, person i has public information X and private information Z i and his estima-
tion of the fundamentals of the reference entity is Yi = E ⎡⎣Y X , Z i ⎤⎦. Based on the
above-mentioned assumptions,Yi = X + Z i, his estimation of default probability is:
exp ( X + Z i )
Pi = Pr (Yi + e > 0 ) = (2.3)
1 + exp ( X + Z i )
 The theory of Internet finance 15
Second, we use w i1 to denote the initial wealth of person i and θi to denote the
financial product bought in the first period, with θi > 0 meaning a purchase and θi < 0
meaning sales. Hence, second-period wealth is w i 2 = w i 1 − θi ⋅ s + θi ⋅ l ⋅ 1{default},
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

where 1{default} indicates whether reference entity defaults.

Therefore, the utility maximization problem for person i is:

max Ei ⎡⎣U ( w i 2 ) ⎤⎦ (2.4)

θi

s.t. w i 2 = w i 1 − θ i ⋅ s + θ i ⋅ l ⋅ 1{default}

where Ei denotes conditional expectation based on the information set of

person i.
FOC is :
Pi ⋅ exp ( −α ( w i 1 − θ i ⋅ s + θ i ⋅ l )) ⋅ (l − s ) − (1 − Pi ) ⋅ exp ( −α ( w i 1 − θ i ⋅ s )) ⋅ s = 0,
thus

1 ⎛ Pi ⎛ l ⎞⎞
θi = ln ⎜ − 1⎟⎠ ⎟ (2.5)
α l ⎜⎝ 1 − Pi ⎝ s ⎠

⎛l ⎞
Monotonic increasing transformation S = − ln ⎜ − 1⎟ is introduced (or equiv-
exp (S ) ⎝ s ⎠
alently s = l . Since S has the same information content as s,16 we will
1 + exp (S )
focus on S in the following analysis. Based on Equation (2.3), θi can be equivalently
expressed as:

X + Zi − S (2.6)
θi =
αl

2.2.2 Model equilibrium

At equilibrium, the market clears, that is the buy and sell orders of the financial
product offset each other.
n

∑θ
i =1
i =0 (2.7)

Based on Equations (2.6) and (2.7), the equilibrium price of the financial
product is:
1 n
S = X + ∑ Zi
n i =1
(2.8)
 16 The theory of Internet finance
2.2.3 The information content of the equilibrium price
The equilibrium price (Equation 2.8) embodies major attributes of information
processing under Internet finance. First, private information is reflected and
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

concentrated in the equilibrium price through the channel of Z i → Pi → θi → S.

Second, in the real world, much of private information may be classified as soft
information and is difficult to transfer to others without distortion.17 However, after
every person converts his private information into the purchase or sale of the finan-
cial product, it is clear whether the information was positive or negative. Thus, soft
information turns into hard information that is understandable to others.These two
points mainly demonstrate the role of social networks in information processing.
Third, there is a relationship between the equilibrium price and the fundamen-
tals of the reference entity: Y = X + n (S − X ). Clearly,18

E [Y S, X ] = Y (2.9)

Therefore, the fundamentals of the reference entity can be correctly deduced

based on public information and the equilibrium price, leading to accurate estima-
tion of default probability based on Equation (2.2).19 Hence, the equilibrium price
can reflect all available information. This mainly demonstrates the role of search
engines in information processing, which produce indicators that contain informa-
tion in a condensed and effective way (like “sufficient statistics”).

2.2.4 The spread of information in social networks

Suppose that during a certain period, the risk aversion coefficient and private and
public information all remain constant. Consider the scenario that one person
spreads private information in social networks. For simplicity, assume no. i person
spreads his private information Z i .
Suppose that at a certain time t , the proportion of individuals who know Z i
(“the informed”) is vt ∈ ( 0,1) and who are unaware of Z i (“the uninformed”) is
1 − vt . Suppose that during a short period of time with a length of dt , the informed
increase by the following proportion:

dvt = λ vt (1 − vt ) dt (2.10)

λ reflects the interconnectedness of social networks.With other conditions con-

stant, the higher interconnectedness of social networks is (higher value of λ ), the
faster information spreads. Based on Equation (2.10), we get

v 0 exp ( λt ) (2.11)
vt =
1 − v 0 + v 0 exp ( λt )

Where v 0 is the initial proportion of the informed.When t → ∞, vt → 1, that is

after a sufficiently long time, almost everyone will become informed.
 The theory of Internet finance 17
Based on Equations (2.8) and (2.11), the relationship between equilibrium price
and time is:
1 n
St = X + Z i ⋅ v t + ∑ Zi
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

(2.12)
n i =1
1 n
Obviously, when t → ∞, St → X + Z i + ∑ Z i. So the spread of information
n i =1
is essentially a process of private information becoming public. This demonstrates
the sharing and communication of information in social networks.

SECTION 4: RESOURCE ALLOCATION UNDER

INTERNET FINANCE

1 General views
Under Internet finance, capital supply and demand information will be released on
the Internet and smoothly matched. Capital suppliers and demanders will contact
each other directly and conduct transactions without the help of any financial
intermediary.
An example is P2P lending such as Lending Club, which provides deposit and
loan services similar to commercial banks. Lending Club is a US company founded
in 2007. For qualified loan applications, Lending Club assigns internal credit ratings.
Different ratings are associated with different loan interest rates. The lower the
rating is, the higher the loan interest rate is. Lending Club refers to each loan as
a note and publishes information of the loan and the borrower on its website
for potential investors to select. For each note, the minimum amount an investor
can buy is US$25, which is small enough to ensure risk diversification. Lending
Club provides instruments for investors to construct loan portfolios and trade loans.
Lending Club is also responsible for loan administration, such as receiving principal
and interest payments from borrowers and transferring them to investors, handling
payment delay or default, and so on.
Another example is crowdfunding such as Kickstarter, which functions similar
to securities markets. Kickstarter is a US company founded in 2009. It helps creative
projects to raise funds through an innovative online platform. Return to investors is
in the form of project products, such as music CDs and movie posters. Investors can
also recommend projects to their friends on Facebook. In April 2012, the United
States passed the Jumpstart Our Business Startups Act (JOBS Act), allowing small
companies to raise equity through crowdfunding.
To better explain resource allocation under Internet finance, we compare P2P
lending represented by Lending Club with Rotating Savings and Credit Association
(ROSCA) in the following paragraphs.
ROSCA is an informal financial organization that exists almost worldwide.
Typically, an originator invites several friends or relatives to participate and meet
every month or quarter for mutual assistance. For example, in coastal regions of
 18 The theory of Internet finance
southeast China, the number of participants is usually around 30. At every meeting,
participants lend a certain amount of money to one person on a rotational basis.
According to Zhang and Zou, ROSCA can be considered as a collection of P2P
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

lending between participants who receive funds early and participants who receive
funds later.20 Although many researchers find that ROSCA plays an import role in
promoting credit availability, crashes of ROSCA do occur. Zhang and Zou point
out that ROSCA mainly relies on networks of acquaintances and has a safety fron-
tier21. Once ROCSA expands beyond the circle of friends or relatives, it becomes
very difficult to control participants’ moral hazard, usually in the form of arbitrage
among different ROSCAs. ROSCA has multiple rounds and it is almost impossible
for participants to transfer their shares to others or withdraw early. When ROSCA
encounters any problem, participants’ utility maximizing behaviors usually lead to
“fallacy of composition.”
We can arrive at two conclusions. First, essentially both P2P lending and ROSCA
are direct lending between two individuals. In fact, according to SmartMoney
Magazine,22 the founder of Prosper, the first P2P lending company, was inspired
by ROSCA. Hence, P2P lending can be considered as an integration between
Internet-based technologies and informal financial organizations.
Second, in P2P lending, an investor may extend loans as small as a few dozen
dollars to hundreds of borrowers, which would be unimaginable in ROSCA. This
is made possible by two factors. First, in P2P lending, borrowers’ credit risk is evalu-
ated by independent third parties.This greatly alleviates the information asymmetry
problem and enables transactions between strangers. Second, investment and loan
administration are carried out by modern technologies, which reduce transaction
costs substantially.
By an extension of the above logic, we believe that, driven by Internet-based
technologies such as mobile payments, social networks, search engines and cloud
computing, direct financing among people (the oldest type of financial activi-
ties in human society) will reach beyond the traditional frontier of safety and
commercial viability. With little information asymmetry and low transaction
costs, Internet finance will generate a sufficiently large “transaction possibility
set” where bilateral or multilateral transactions can be carried out simultaneously,
quickly, and efficiently. Resource allocation under Internet finance will maximize
social welfare and promote social equality. Everyone will have transparent and fair
opportunities to invest or raise money. People who have never met will become
acquaintances through Internet finance, which will facilitate their cooperation in
other activities.
A key concept here is “transaction possibility set.” Below is an explanation of this
concept and how it is influenced by information asymmetry and transaction costs.

2 “Transaction possibility set”

We define “Transaction possibility set” as a set of one or multiple pairs of capital
demander and supplier where among each pair, the highest funding cost affordable
to the capital demander is higher than the minimum investment yield required by
 The theory of Internet finance 19
the capital supplier. “Transaction possibility set” emphasizes that based on the price of
financing, the capital demander and supplier have the possibility to reach deals. But in
the real world, capital suppliers usually have budgetary constraints and multiple invest-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

ment opportunities. Whether a deal can be reached with a certain capital demander
depends on complex conditions, which are not covered by “transaction possibility set.”

2.1 Highest funding cost affordable to capital demander

Let’s use I to denote the set of capital demanders. Suppose all capital demanders
are risk neutral and consider a representative capital demander i ∈ I. Assume he has
an initial wealth of Ei and needs a loan23 of L i to start an investment project. The
project has an expected yield of μi , a success probability of θi and revenue in case
of success (1 + μi )( Ei + L i ) and 0 in case of failure. Assume that when the capital
θi
demander does not apply for a loan, his wealth will stay at Ei . Let’s use f i to denote
L
loan interest rate and li = i to denote the capital demander’s debt-to-equity ratio.
Ei
Whenever the expected net profit of the project (1 + μi )( Ei + L i ) − θ i (1 + f i ) L i
exceeds Ei , the capital demander will apply for a loan. This is equivalent to:

1 + μ i + μ i / li
1 + fi ≤ (2.13).
θi

Equation (2.13) gives the highest funding cost affordable to the capital demander.
The higher expected yield is (higher μi ), the higher project risk is (lower θi ) or the
lower leverage ratio is (lower li ), the higher funding cost the capital demander can
afford.

2.2 Minimum investment yield required by capital supplier

We use J to denote the set of capital suppliers. Assume that all capital suppliers
are risk neutral and consider a representative capital supplier j ∈ J. Suppose his
capital cost (or opportunity cost) is r j . Assume that there exists transaction costs and
information asymmetry between capital supplier j and capital demander i. Assume
transaction costs (credit assessment cost included) equal c ij times the loan amount,
where c ij > 0 and higher c ij indicates higher transaction costs. Assume that even
after credit assessment, capital supplier j still cannot accurately evaluate the suc-
cess probability of capital demander i and underestimate it to be (1 − λ ij )θi , where
λij ∈ ( 0,1) and higher λij indicates higher degree of information asymmetry.
The capital supplier’s condition to extend a loan is that, loan yield (adjusted
for possible loss caused by loan default) (1 − λ ij )θi (1 + f i ) − c ij is higher than his
opportunity cost, which is equivalent to:

c ij + 1 + r j
1 + fi ≥ (2.14)
(1 − λ )θ
ij i
 20 The theory of Internet finance
Equation (2.14) gives the minimum investment yield required by the capital
supplier. It needs to compensate for capital cost, transaction costs, credit risk of the
capital demander and information asymmetry.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2.3 Expression of “transaction possibility set”

The necessary condition for a pair of capital demander and supplier to reach a deal
is that the highest funding cost affordable to the capital demander is higher than
the minimum investment yield required by the capital supplier, which according to
Equation (2.13) and (2.14) is equivalent to:

c ij + (1 + μi + μi /li ) λij ≤ μ i + μ i /li − r j (2.15)

In Equation (2.15), only c ij and λij are determined by the relationship between
the capital demander and supplier. With other parameters constant, the lower
transaction costs are, or the lower the degree of information asymmetry becomes,
Equation (2.15) is more likely to be satisfied. So the “transaction possibility set” is:

{(i, j ) i ∈ I , j ∈ J , c ij + (1 + μi + μi /li ) λij ≤ μ i + μ i /li − r j } (2.16)

The “transaction possibility set” has the following characteristics. First, it is

determined by transaction costs and information asymmetry. Different levels of
transaction costs or information asymmetry correspond to different “transaction
possibility sets.”
Second, with other conditions held constant, lower transaction costs or informa-
tion asymmetry are associated with larger “transaction possibility sets.” Under such
circumstances, capital suppliers and demanders are more likely to reach deals, which
is “financial deepening” in some sense.
Third, assuming transaction costs and information asymmetry cease to exist
(c ij → 0, λij → 0), the “transaction possibility set” will approach:

{(i, j ) i ∈ I , j ∈ J , μ i + μ i /li ≥ r j } (2.17)

Under this scenario, as long as the expected return of a capital demander

(adjusted for leverage) exceeds the opportunity cost of a capital supplier, a deal
between them is possible.

SECTION 5: CONCLUDING REMARKS

We have introduced the concept of Internet finance and discussed its payment,
information processing and resource allocation.We believe that Internet finance can
promote economic growth and generate considerable social benefits by increasing
resource allocation efficiency, reducing transaction costs, and promoting availability
of financial services.
 The theory of Internet finance 21
Currently, China has made the following progresses in Internet finance. First,
the central bank (People’s Bank of China) issues third-party payment licenses to
the top three mobile operators. Second, several P2P lending companies have been
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

established. Third, some institutions such as Alibaba have employed information on

social networks to facilitate SME financing. However, many problems have also
emerged and the banking regulator (China Banking Regulatory Commission)
alerted banks of P2P lending’s risk in 2011.
We believe Internet finance will become more and more important in the future.
It is inevitable that Internet finance will face many technological, commercial, and
regulatory challenges in early stages of development. But we should never deny,
overlook, or underestimate its potential.

Notes
1 Mishkin, Frederic. 1995. “The Economics of Money, Banking, and Financial Markets,”
Harper Collins College Publishers.
2 For instance, physical bookstores such as Borders have gone bankrupt under competi-
tion from electronic books and online bookstores. MP3 and music sharing websites
have reshaped the business model of music industry. Amazon and Taobao have seriously
eroded the traditional retail industry.
3 Under Internet finance, prudential regulation on financial institutions (such as commer-
cial banks, securities firms, and insurance firms) may cease to exist and is to an extent
replaced by behavioral regulation and protection of financial consumers.
4 Goldman Sachs. 2012. “Mobile Monetization: Does the Shift in Traffic Pay?”
5 Xie, Ping, and Long Yin. 2001. “The Financial Theory and the Financial Governance
Under Internet Economy.” Journal of Economic Research, no. 4.
6 Mishkin, Frederic. 1995. “The Economics of Money, Banking, and Financial Markets,”
Harper Collins College Publishers.
7 Xie, Ping, and Chuanwei Zou. 2011. “The Irreplaceable Functions of CDS.” Review of
Financial Development, no. 1.
8 Conditions for people to establish relationships and access precious resources, that is the
concepts, “courtesy demands reciprocity” and “return a favor with a favor.”
9 People have the tendency to establish communication networks with others who have
similar traits, that is “birds of a feather flock together.”
10 How ideas, information, and views are exchanged among people in a communication
network, that is “if you live with a lame person, you will learn to limp.”
11 Monge, Peter R., and Noshir S. Contractor. 2003. “Theories of Communication
Networks,” Oxford University Press, Inc. If individual A has a relationship with indi-
vidual B and individual B has a relationship with individual C, then individual A has a
relationship with individual C, that is “a friend’s friend is a friend and an enemy’s enemy
is a friend.”
12 Shirky, Clay. 2008. “Here Comes Everybody: The Power of Organizing Without
Organizations,” Penguin Press.
13 Scott, John. 2000. “Social Network Analysis: A Handbook,” Sage Publications, Inc.
14 Technically speaking, processing of relational data has always been a major component
of search engines. For instance, the “crawler” algorithm for capturing web pages and link
analysis method for web page sequencing have all employed the linkage between web
pages that belong to relational data. Recently, Facebook has launched Graph Search.
15 For instance, real-time search involves a tremendous quantity of computation. Google is
a forerunner in the development of cloud computation.
 22 The theory of Internet finance
16 Strictly speaking, due to the one-to-one correspondence between S and s, σ-algebras
induced by S and s are the same.
17 Petersen, Mitchell A. 2004. “Information: Hard and Soft,”Working Paper, Kellogg School
of Management.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

18 Under current model setting, people may deduce others’ private information from
equilibrium price and adjust their estimation of default probability and trade decision,
thus affecting market equilibrium.This scenario belongs to rational expected equilibrium.
19 Huang, Chi-fu, and Robert H. Litzenberger. 1988. “Foundations for Financial
Economics,” Elsevier Science Publishing Co., Inc.
20 Zhang, Xiang, and Chuanwei Zou. 2007.“The Mechanism of Systemic Bidding ROSCA
Default.” Journal of Financial Research, no. 11.
21 Zou, Chuanwei, and Xiang Zhang. 2011. “Arbitrage and Systemic Bidding ROSCA
Default.” Journal of Financial Research, no. 9.
22 “Global Lessons for Better Savings Habits,” SmartMoney, Nov 18 2011.
23 In other words, financing takes the form of loans. However, similar logic also applies to
other forms of financing such as preferred stocks, ordinary stocks, and convertible bonds.
 3 The internetization of finance
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

SECTION 1: THE INTERNETIZATION OF FINANCE

Internetization (financing activity moving online) is an unavoidable trend. More

and more people are getting used to the Internet, and many no longer have the
time to make it to a bank teller due to the accelerating pace of life. Thus, financial
institutions are using the Internet to adapt to customers’ needs. Additionally, the
rapid development of Internet technology helps reduce these institutions’ transac-
tion costs. This chapter successively discusses network banking, mobile banking,
network securities companies, network insurance companies, network financial
trading platforms, and the online marketing of financial products.

1 Network banks
According toYin Long,1 network banking has two meanings.The first touches upon
the meaning of the term and identifies the nature of network banking. “Network”
has a meaning beyond Local Area Networks (LAN), the Internet, and other open
electronic networks. It also includes internal bank networks, money transfer net-
works, payment and clearing networks, and even telecommunication networks. If it
serves as a carrier for financial information, products, or services, it is seen as a new
channel for the banking industry and fits under the scope of network banking. The
second meaning touches upon the identification of the network banking business
and the function of banking. Network banks are recognized as possessing an inde-
pendent website used to provide certain customer services (here network means
Internet). Network banking is not only an evolution in banks’ business model; it is
the internetization of banking itself.
The evolution of network banking has experienced three stages: business process
digitization, business management digitization, and reengineering of the bank
(see Table 3.1)
In the first stage, banks mainly use information and communication technologies
to assist and support business development, such as data storage and centralized
processing for transactions. In this stage, banks basically automated office processes,
but at that time information communication technology was not mature.The bank
information systems were dispersed and closed.
 24 The internetization of finance
Table 3.1 Information technology and commercial bank innovation2

Time Innovative themes Related technologies

Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

1950 Credit cards Magnetic stripe

Early-1960 Autopay Telephone

Early-1960 Check processor Magnets

1969 ATM Mechatronics

1970 POS Computer and Communications

1970 Credit scoring model Database Technology

1970 CHIPS Communication

1973 Automatic payment Communications, computer

technology

1977 SWIFT system Communication

1980 Derivatives High-speed operation of com-

puter and information communi-
cation technology

1982 Household bank Computer and information

communication technology

Mid-1980 Enterprise bank Computer and information

communication technology

1988 Electronic Data Communications, security control

Interchange (EDI)

1990 Customer relationship Database Technology

management (CRM)

1990 Credit separation model Database Technology

1990 Credit scoring model Information and communication

technologies and the Internet

In the second stage, the rapid development of information technology and the
reduction of cost provided favorable conditions for the extensive application of
information technology networks in banks. Banks widely use networked real-time
trading, developed internal networks, and rolled out both point of sale (POS) and
ATM machines.
The third stage is marked by the birth of the first network bank, Security
First Network Bank, in October 1995. With this, a new service channel emerged.
Network banking, telephone banking, mobile banking, and television banking
helped customers get more convenient service. The innovation of this phase has
 The internetization of finance 25
revolutionized the banking business system, breaking the sub-sectoral restrictions
of banking, insurance, and securities, thereby continuously integrating the financial
sector. Business development in turn increased the demand for information and
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

communication technology, which has been increasingly outsourced.

2 Mobile banks
Mobile banking refers to the use of mobile phones, PDAs and other mobile devices
to connect customers and financial institutions. The first mobile bank was built in
the Czech Republic in the late 1990s by the Expandia Bank and mobile operator
Radiomobile. Since then mobile banking has appeared in a variety of modes and
a lot of cases. In December 2013, Zhou Xiaochuan, the governor of the PBoC
said, “We should learn from international experience through mobile banking to
provide basic financial services to rural areas, remote areas and poor areas” in an
interview with the Financial Times.

2.1 The main forms of mobile banking

Table 3.2 summarizes four major mobile banking types. Most mobile banks are
bank-led, so mobile operators only supply the operating platform. This model was
the first to appear and is still the mainstream in developed countries.
Second, there have been a large number of mobile banking innovations in
African countries, and non-bank institutions such as mobile operators and third-
party payment companies play an important role. For example, Kenya’s mobile
banking M-PESA, which is owned by a mobile operator, has become the world’s
most used mobile payment system, and the remittance business of M-PESA has
exceeded that of all other financial institutions in Kenya. African countries’ financial

Table 3.2 The forms of mobile banking

Bank based Partnership Non-bank based Non-bank-

business sponsored

Who hold Bank Bank Bank Operators or

accounts or other non-bank
deposits institutions

Withdraw Bank Bank Bank or agents Operators

agency or other
institutions

Who executes Any operator Specific Specific Specific

payment operator operator operator
instructions

Typical examples Most mobile MTN Mobile M-PESA, Globe, Celpay

banks Money, Smart Wizzit
 26 The internetization of finance
Table 3.3 Overview of African national mobile banking3

Celpay M-PESA MTN Mobile Wizzit

Money
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Whether they No Yes Yes Yes

target financial
gaps

Security Funds Funds Bank account Bank account

deposited in deposited in required required
banks banks

Withdrawal Withdraw cash Agent ATM;bank ATM;bank

method not allowed branches branches

Transfers Yes Yes Yes, any bank Yes, any bank

allowed? account account

Special hardware Yes No 32k SIM No

requirements

systems are underdeveloped and find it difficult to meet basic needs for financial
services, especially lacking physical outlets, which leads to a huge space for develop-
ment of these emerging mobile banking models (see Table 3.3, analysis focused on
the following cases).

2.2 Requirements for mobile banking

Mobile banking requires a specific infrastructure, such as mobile terminals and
information communication technology. Economic conditions are even more
important. Surprisingly, mobile banking mainly exists in less developed countries
such as some African countries, but why? In the current circumstances, mobile
banking acts more as a substitute than a complement for online banking and bank-
ing networks. Most African countries do not have access to basic financial services,
yet they have strong financial needs. Mobile banking does not need a sales net-
work, which grants access to basic financial services. Mobile banking providers also
significantly reduce transaction costs in order to make profits. This provides the
necessary conditions for mobile banking to solve the problem of financial inclusion.
In addition, African countries’ financial regulation allows mobile operators with
a comparative advantage in providing mobile banking. In short, the matching of
supply and demand are the basic conditions for the existence of mobile banking.
Why is the penetration rate of mobile banking in Europe and other developed
countries not as high? This is due to the widespread availability of bank branches.
People can get financial services at any time, and the small mobile phone screen
makes them less likely to use their phones for mobile banking.
Why has mobile banking developed so slowly in rural China and other countries?
The development of mobile banking in underdeveloped areas needs a “Coca-Cola
 The internetization of finance 27
method,” namely to design a business model that can be promoted at a large scale
to achieve low cost and high-density sales. The higher the marketization level of
banks and mobile communication institutions, the more it can take the initiative to
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

meet customer demand rather than make customers adapt to providers. However,
if the provider is an absolute monopoly, there is no incentive to build and promote
the Coca-Cola model.
Most people now believe security is the main obstacle to the promotion of
mobile banking. For example, security concerns make mobile banking unpopular
in developed countries, while people in underdeveloped countries may not be
cognizant of the security problems of mobile banking. In fact, the poor are more
concerned about safety, because deposits are essential to them. In the future, the
financial model will change. There will be more branchless services and the supply
and demand sides of funds through mobile phones and other mobile terminals can
be directly matched. At that time, product pricing, risk management, and informa-
tion processing can be completed in everyone’s hand. There will be no need for
bank intermediaries. At that time, mobile banking will replace physical financial
institutions and gain popularity in both underdeveloped and developed regions.

2.3 Mobile banking case studies

2.3.1 M-PESA (Kenya)4

Safaricom,Vodafone’s5 partner in Kenya, launched M-PESA in 2007. Its initial pur-
pose was to meet the remittance demands of the poor, but it later developed to
complete services like transfers, remittances, cash withdrawal, prepaid recharges,
bill payment, wage payment, and repayment of loans by phone. People can remit
through M-PESA both home and abroad. One important factor in the success
of M-PESA is to achieve access to cash businesses. M-PESA introduced the post
office, pharmacy, supermarket, and others as agents to provide cash services under
the aegis of M-PESA.
M-PESA charges transfer fees, but the account registration, deposits and pre-
paid recharge M-PESA are all free. M-PESA has also opened up channels between
mobile operators and banks to transfer between M-PESA accounts and bank
accounts, and M-PESA users can withdraw cash at bank ATMs (Table 3.4).
The design of M-PESA’s virtual account means its bank activities do not fall
under Kenyan banking laws. Therefore, Safaricom can choose agents according to
its own business judgment and is not liable for agent misconduct. Other than ensur-
ing that customer-stored value funds are not deposited into only one bank, there
are no strict regulations for those branchless banking services dominated by the
non-bank institutions such as M-PESA.
In the beginning, M-PESA only had 52,000 users, 355 agents. By the end of
April 2011, M-PESA customers grew to about 14 million, and agents to nearly
30,000. M-PESA’s success has attracted interest from imitators in emerging markets
such as Tanzania, South Africa, Afghanistan, India, and Egypt.
 28 The internetization of finance
Table 3.4 M-PESA fees (Kenyan shillings6)

Transfer sum interval Transfer to other Transfer to non- Withdrawal from

M-PESA users M-PESA users M-PESA agents
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

10 49 3 N/A N/A

50 100 5 N/A 10

101 500 25 60 25

501 1, 000 30 60 25

1, 001 1, 500 30 60 25

1, 501 2, 500 30 60 25

2, 501 3, 500 30 80 45

3, 501 5, 000 30 95 60

5, 001 7, 500 50 130 75

7, 501 10, 000 50 155 100

10, 001 15, 000 50 200 145

15, 001 20, 000 50 215 160

20, 001 25, 000 75 250 170

25, 001 30, 000 75 250 170

30, 001 35, 000 75 250 170

35, 001 40, 000 75 N/A 250

40, 001 45, 000 75 N/A 250

45, 001 50, 000 100 N/A 250

50, 001 70, 000 100 N/A 300

The amount interval ATM withdrawal fee

200 2, 500 30

2, 501 5, 000 60

5, 001 10, 000 100

10, 001 20, 000 175

Note: The maximum M-PESA account balance is 100,000 Shillings and transfers cannot exceed
140,000 Shillings. Daily transfers and per transfer amounts cannot exceed 70,000 Shillings, M-PESA
agents do not accept cash under 50 Shillings.
 The internetization of finance 29
2.3.2 Wizzit (South Africa)7
Wizzit was launched in November 2004 by a joint venture between mobile operator
MTN and Standard Bank. It is a virtual bank with no branches.Target users are the
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

16 million strong low-income group in South Africa (48% of the adult population).
In addition to financial services such as transfer, payment, prepaid recharge, wage
payment, and account query by phone, Wizzit also offers a Maestro debit card.
Users can use the card for withdrawals at ATMs and deposits at bank branches.
It uses more than 2,000 WIZZ Kids8 to advertise and serve as agents (including
registration for the service) rather than placing advertising in conventional media.
Wizzit has no minimum balance requirement or fixed costs, which makes it
attractive to groups not normally covered by financial services. Its greatest advantage
is its low transaction costs, which are sustained through high volumes. Despite
being founded partially by a telecom company,Wizzit is compatible with all mobile
operators. Like similar services, it monitors user accounts and suspends transactions
that cross certain thresholds for transaction amount or total account volume.

2.3.3 GCash (Philippines)9

The Philippine mobile operator Globe Telecom founded GCash in 2004. Users
register with text messages, then deposit or withdraw funds through agents.10
GCash can make transfers between bank accounts, but does not function as a bank.
Banks and other financial institutions act only as agents in their model.
At present, GCash’s has more than 1.2 million users and its transactions total
more than 60 billion Philippine Pesos, charging only for withdrawals.11 The
Philippine central bank has taken a relatively relaxed stance in the regulation of
GCash. Requirements focus on information disclosure, client fund ring-fencing,
transaction limits, and anti-money laundering.

2.4 The risk and regulation of mobile banking

Banks faces two types of risk when they use mobile phones for banking in poor areas.
The first risk is related to agents, who often have a shortage of trained employees
and insufficiently developed security systems. Banks are thus subject to risks such
as theft, fraud, and improper transactions. Policymakers and regulators are seriously
considering how to deal with these risks emanating from agency arrangements.12
For example, Brazil has facilitated branchless banking development by opening
up the agency function to all retail stores with a POS machine, but has mitigated
risk by holding the banks responsible for agents, requiring central bank approval
for new agents, mandating data collection, and placing limits on clearing time for
transactions. The result is to both create incentives and give the proper tools for
banks to properly manage their agents. In contrast, India’s central bank placed severe
restrictions on which entities can serve as agents.This is to a certain extent justified
 30 The internetization of finance
by the poor track record of many retail entities, but has impeded the development
of branchless banking services in India. This is in stark contrast to Kenya, in which
operators like M-PESA both freely choose and are not responsible for their agents.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

The second type of risks relates to misconduct or mistakes committed by opera-

tors. For example, client funds may be misappropriated for use in risky investments,
which can lead to client losses. Electronic currency must be regulated. If a bank
issues electronic currency, regulators monitor the flows of stored value or unpaid
funds. For example, Smart Money in the Philippines records its funds on coop-
erating banks’ books as accounts payable rather than savings to reduce regulatory
costs, but customers get a lower degree of protection. In non-bank mode, however,
mobile operators open virtual accounts for the customer and directly establish a
contractual relationship between clients and mobile operators rather than opening
a contractual relationship with the bank. In this case, the electronic currency is less
regulated. Once a risk occurs, customers can put in a claim to the provider, but
not the bank. Therefore, there is a need to formulate special regulations to ensure
the provider has enough capital to deal with and prioritize customer claims for
compensation.

2.5 Mobile banking in China

2.5.1 Developments
Some banks in China have launched mobile banking, including big state-owned
banks such as Bank of China (BOC), Agricultural Bank of China (ABC), Industrial
and Commercial Bank of China (ICBC), Construction Bank of China (CBC), and
Bank of Communications (BCM); nationwide joint-stock banks; city commercial
banks and rural commercial banks; rural cooperative banks; new-type rural financial
institutions; and rural credit cooperatives. Mobile banking through regional banks
is basically network banking on the phone. Mobile banking is relevant to rural
finance when it includes cardless cash withdrawal, small loans to farmers, on-site
remittance, and mobile finance.
BCMs first launched mobile banking with cardless cash withdrawal, then
China Guangfa Bank (GDB), Shenzhen Development Bank (SDB) and ICBC
launched similar services. Cardholders prearrange the withdrawal on the platform
and then can withdraw funds without their card. This can protect the client from
card theft, allow emergency withdrawals, and permit others to remotely complete
the withdrawal for them (if they keep the money, this becomes a quasi-transfer).
The ATM requirement has led to this service expanding more in cities than the
countryside.
Small loans to farmers was first launched by ABC’s Guangxi and Henan branches
as pilot programs. They provide six basic functions including self-help borrowing,
self-help payment, reimbursement trial, loan contracts, information queries, and
reimbursement queries.This service is a win-win; farmers can obtain loans without
leaving home, ABC expands the rural market with lower transaction costs, and the
government promotes financial inclusion.
 The internetization of finance 31
Postal Savings Bank of China (PSBC) was the first to launch site-based remit-
tance services.This service allows farmers to remit funds by delivering the notice of
withdrawal with a listed beneficiary.The farmers with no bank card in remote areas
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

can complete the remittance by address.

Mobile banking can also merge other financial services. The mobile bank of
Chongqing Rural Commercial Bank (CRCB) highly integrates basic financial
services including: a mobile payment application, cross-industry mobile payments,
remote payments, online shopping, prepaid phone cards, game cards, ticket and
hotel booking, and many others.

2.5.2 Improving financial services with mobile banking

Rural financial institution branches in China cannot satisfy the increasing demand
for rural financial services. At the same time, regulators have realized the impor-
tance of mobile banking to bridge the financial gap in villages and towns. In 2011,
the China Banking Regulatory Commission (CBRC) issued a policy pronounce-
ment titled “continue providing basic financial services for the absence of finance in villages
and towns” to push financial institutions in this direction.
Cash access is the key for mobile banking to solve the general inclusion problem,
but cash business generally requires an agent. The specific operation is as follows:
mobile bank users can use the card provided by a bank as a mobile virtual bank
account provided by mobile operators and agents install POS equipment or phones
with mobile banking functions. If customers want to make a deposit with the agent,
the bank will automatically deduct the amount from the agent’s account as the
customers’ deposit funding. The customer cash offsets the deduction to the bank/
mobile operator’s account. If the customer wants to take out cash, the opposite
occurs. Clients can thus obtain important financial services without commuting to
a specific bank branch.
The conditions are ripe for promoting mobile banking in rural China. First,
mobile phones are on the rise. At the end of October 2012, China’s mobile phone
penetration rate reached 80.6%. By the end of 2011, the accelerated development
of wireless mobile communication networks will cover all counties and most of the
villages and towns throughout the country. At present, 3G network transmission
speeds reach 2 mbps and support encryption.
Second, mobile banking does not require a network. Without additional equip-
ment and personnel, mobile banking saves on transaction costs. Empirically, bank teller
transactions outside China cost an average of US$1.07, but mobile banks only require
US$0.16. At Chinese bank counters, transaction costs an average of RMB 4 yuan,
compared with mobile banking transactions at only RMB 0.6 yuan.
Third, branches of existing rural financial institutions cannot cover vast rural
areas, so the absence of rural financial institutions provides the market demand
for mobile banking development. In addition, rural income level, education
level, and consumption habits support the promotion of mobile banking in
rural China.
 32 The internetization of finance
Fourth, the rural income level has increased. Since China’s reform and opening
up, China’s rural residents’ per capita net income has increased from US$133.6 in
1978 to RMB 5,153.2 in 2009, up by a factor of 38. This laid a solid economic
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

foundation for the promotion of mobile banking.

Fifth, the rural education level has increased. Generally speaking, a high school
education is sufficient to use mobile banking. Those in rural areas who have
completed middle schools and high school degrees has risen steadily from 34.7% of
the rural population in 1985 to 64.4% in 2009.13
Sixth, rural consumption habits are changing. It remains unclear whether
Chinese rural areas will change from the current cash transaction model into a
mobile payment mode. As incomes, educational attainment, and culture change, we
may see very different consumption patterns in the future, patterns that will also be
influenced by mobile banking.

SECTION 2: NETWORK SECURITIES COMPANIES

The network securities business refers to investors taking advantage of resources

including the Internet, LANs, private networks, wireless Internet, and other
electronic means to transmit information, data, and deals relating to securities and
exchange. It includes a series of activities such as real-time price acquisition, market
information, investment consultation, and online trust products.14
Once, the stock exchange was full of gesticulating traders in red vests racing
against time to place orders over the phone. The online trading system has made
this history. The trading floor of securities firms is also gradually being replaced by
online transactions. E-Trade launched online securities trading in the United States
in 1992, and from then on the online securities trading business has boomed. Now
most customers are accustomed to online trading. All one needs to do is open an
account and download a simple trading software, then it is possible to trade stocks
from home or even from one’s cell phone, anywhere in the world.
The impact of information and communication technologies on China’s
securities markets has been profound. China has experienced successive stages of
centralized trading, online securities trading, and mobile securities trading.The first
stage was centralized trading, marked with the establishment of Shanghai stock
exchange in 1990 and the Shenzhen stock exchange in 1991. The second stage was
online trading, marked with an online trading system launched by Huarong Trust
Investment Company in March 1997.The third stage was mobile securities trading,
meaning securities trading entered the mobile era.

1 Business models worldwide

Based on the depth of the use of the Internet, we divide network securities
companies into three models: the first is a pure Internet securities brokerage com-
pany such as E-Trade and TD Ameritrade, the second is a comprehensive securities
 The internetization of finance 33
brokerage firm like Charles Schwab and Fidelity, and the third is a traditional
securities brokerage such as Merrill Lynch and A.G. Edwards.
E-Trade was formally established in 1992. It first provided investors with online
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

securities services through AOL, and then established its own online trading site
www.etrade.com in 1996. At present, E-Trade is the world’s largest personal online
investment services site, with customers in over 100 countries. Its most salient
advantage is low transaction costs. E-Trade has strong technical capabilities and a
convenient online trading channel, all without physical locations. Thus, they are
able to offer trades with a commission average of only about US$10. The disadvan-
tage of E-trade lies in its lack of experienced investment advisers.
Charles Schwab has become one of America’s personal financial services market
leaders. In the mid-1990s, Charles Schwab made a major breakthrough when it
launched an online financial services platform. Charles Schwab offers investors a
relatively inexpensive service mainly through telephone, fax, and Internet trading.
Unlike E-trade, Schwab is not a pure Internet securities company because it has a
few physical locations for customer service. The Schwab mode has low cost, but its
research and development ability is weak.
Merrill Lynch was one of the world’s leading financial management and con-
sulting companies. Unlike Schwab, Merrill Lynch focused on high-end customers,
providing customers with face-to-face, comprehensive asset investment advisory
services. Merrill lynch had strong investment research and portfolio consulting, but
it had a limited potential customer base due to the cost of providing personalized
service. Its use of the Internet was not nearly as deep as the other two modes.

2 Chinese models
Internet securities companies in China mainly provide information and financial
services. We divide them into three forms: brokerage website mode, independent
third-party site, and brokerage and bank cooperation.
The brokerage website mode is common for the securities companies, such
as GF Securities Guotai Junan, CITIC Securities, and Haitong Securities. These
trading and service sites are part of the company’s inner service centers. Their
services are not limited to the traditional brokerage business. Customers can also
buy information products and funds on the platform. The advantage of this model
is that the securities company can directly provide traditional market services
for online clients through the website, and brokerage service advantages can be
fully displayed. The problem for small and medium-sized brokerages is that special
website construction requires significant funds.
In independent third-party site mode, the Internet services company, consulting
firm, and software system developers build websites to provide consultancy services
for customers, and brokerages are in the background providing online securities
trading services to customers. This model is an open platform. If the client needs
securities trading, he/she can just open “add broker” on the software. Flash and Great
Wisdo are some of the leaders in this area for the Chinese market. Some websites
 34 The internetization of finance
only provide consulting services, such as Oriental Wealth. The advantage of this
mode is that technology and information advantages can be fully reflected. The
disadvantage is that the content of the securities service and professional level trust
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

by clients need a time to develop.

Bank-Securities Link exemplifies a new type of financial services business between
brokerage and bank. On the basis of banks’ network with securities firms, investors
can directly use the current savings account provided by bank branches as a securi-
ties margin account.They can buy and sell securities through the bank’s trust system
or through securities firm commissioned systems. The model permits commercial
banks to get involved in the securities market business.The advantage is convenience,
low fees, and protection from margin appropriation risks.The drawback is that there
is legal risk; for example Bank-Securities Link shut its doors for this reason in 2006.

SECTION 3: NETWORK INSURANCE COMPANIES

Online insurance refers to insurance companies or other intermediary institutions,

which use the Internet to facilitate their business. There are narrow and broad
definitions. In the narrow sense, online insurance refers to insurance companies
or other intermediaries that provide information about insurance products and
services over the Internet. They may also directly complete the sale of insurance
products and services online. More broadly, online insurance includes web-based
internal management activities and transactions, information exchange activities,
insurance regulation, taxation, and management institutions.15

1 Development
In the insurance industry, information and communication technology originally
was used for electronic insurance products. Marketing is also increasingly electronic.
With the development of IT, e-commerce platforms gradually developed, through
which customers can self-service, obtain quotes, and buy products.
The first insurance company to promote its insurance products entirely through
the Internet was founded in 1999 in Japan as a joint venture between AFLAC and
Japan Telecom. It focuses on customers under the age of 40. INSWEB (US) is the
world’s largest and most respected insurance e-commerce site. Forbes once called
it the best site on the Internet. The site covers insurance of just about everything,
including cars, housing, medical treatment, personal life, and pets.
China’s online insurance is still at a rudimentary stage. Most insurance companies
just set up their own web portal with little comparative information. In June 2012,
“Rest Assured Insurance” put B2B and B2C trading online. It is also a third-party
insurance sales platform. In 2013, Alibaba, China’s Ping An Insurance, and Tencent
jointly established an online property insurance company that will disrupt the exist-
ing Chinese insurance marketing model.
 The internetization of finance 35
2 Major types
The online insurance businesses can provide insurance services online, use spe-
cialized companies, or collaborate together online.16 The first type operates sales
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

through the company’s website. This helps promote the company and its products
through new channels. The insurance company can then manage customer data to
provide other value-added services, such as providing free SMS and personalized
e-mail subscriptions. The disadvantage of this scheme for its consumers is a lack
of comparability. It only shows one company’s products and has high technical
requirements. The eCoverage (US) was the first company to provide customers
with service from quotation to claims service on the Internet.
Another model uses a network platform to put all related insurance company
products information online. Users search for and choose insurance products,
then are linked to a fitting insurance company. These platforms can charge com-
paratively lower commission and fees. In this mode, users can quickly find what
they need by comparing different insurance companies. Huize.com was the first
third-party insurance electronic commerce platform in China to provide product
comparisons, vertical transactions, purchases, and professional insurance consult-
ing interaction in one place. It combined with several large insurance companies
to provide real-time online insurance. INSWEB is a typical example in foreign
countries; it inked agreements with more than 50 insurance companies all over the
world and cooperates with more than 180 websites. There are three main profit
points for this platform mode. The first is the intermediary fee paid by consum-
ers, the second is a “finder’s fee” paid by the agent, and the third is advertising and
other fees.
In the Taobao network insurance mode, the insurance site does not list insur-
ance products or information. It just provides a platform for insurance suppliers and
demanders to match themselves. The core of this website is to provide help both
sides make an independent choice and provide some “soft” information for the
insurance market. Many insurance companies are now on Taobao, such as China’s
Ping An Insurance and China Life Insurance, who now do online sales of insurance,
car insurance, health insurance, and other products.
A network insurance support platform does not directly provide insurance prod-
ucts. It provides information and technical support for insurers, and is generally
founded by a non-insurance agency. These generally have a very deep industry
background with strong information superiority and social credibility. Firms such
as China Insurance Network provide theory and policy, member communication,
real-time news, data, information, training information, and information related to
insurance companies, insurance agents and brokers, and other information for insur-
ance practitioners and consumers. Network insurance technical support provides
information technology specifically for insurance companies. Typical representa-
tives such as the Yi-Bao network only provide technical guarantees and services for
insurance companies.
 36 The internetization of finance
SECTION 4: NETWORK FINANCIAL TRADING PLATFORMS

1 SecondMarket
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

SecondMarket was established in 2004 to provide online trading services for prod-
ucts such as equity of non-listed companies, fixed income securities, bankruptcy
claims, warrants, and alternative assets.

1.1 Basic information

Barry Silbert, the founder of SecondMarket, once worked at a Wall Street invest-
ment bank with a focus on financial restructuring, mergers and acquisitions, and
corporate finance. He was involved in major events such as the bankruptcies of
Enron and WorldCom. He found that in a restructuring project, creditors often get
shares of the new company, but there was no effective channel to resell these shares.
This gap inspired Silbert’s idea to establish a centralized and transparent market for
trading alternative assets.
SecondMarket initially traded restricted stock, warrants, and convertible
bonds of public companies. Gradually, its business extended to include fixed
income securities, bankruptcy claims, and equity of non-public companies. In
early 2009, there were only 2,500 investors, but this reached 75,000 in 2011.
Now, equity in more than 50 companies, including Internet startups such as
Facebook and Twitter, trades on SecondMarket. In 2008, SecondMarket reached
an annual trading volume of US$30 million, but grew to US$100 million in
2009. In 2011, it reached US$558 million. Thanks to its broad prospects for
development, SecondMarket has attracted investments from New Enterprise
Associates, the Li KaShing, and Temasek Holdings. Its market valuation is about
US$200 million.
SecondMarket has now been registered as a trade broker by SEC and as a mem-
ber of FINRA. It also operates an alternative trading system (ATS).

1.2 Policies
1 The willingness of companies to trade their equity is a precondition for trans-
actions. Non-public companies vary widely. Some companies value the control
and operating flexibility, while others may pay more attention to liquidity and
enterprise value. Therefore, SecondMarket only provides the trading services
for non-public companies after obtaining their consent. It also allows the com-
pany to set restrictions, including trade restrictions and investor qualifications.
Some companies only allow former employees to trade, while others only
allow existing shareholders to purchase. A majority of companies are opposed
to excessively frequent trading.
2 Disclosure of certain information is required to participate in SecondMarket.
It requires non-public companies to disclose financial information to its
registered members on the online trading platform, including the audited
 The internetization of finance 37
annual reports of last two years and other risk factors. SecondMarket strictly
controls the database to prevent leaks.
3 Individual investors must have net assets of more than US$1 million or
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

annual income of more than US$200,000 to invest in SecondMarket’s traded

companies.
4 The trading mechanism involves sellers who post assets for sale on the website’s
board. The system automatically finds proper buyers in the database according
to the members’ interests or past transactions. SecondMarket will then notify
both parties over the phone. After the buyer and seller reach an agreement,
SecondMarket processes legal, settlement and payment issues of transactions. It
charges a transaction commission of 2%–4%.

1.3 Statistical data

Based on the data from completed transactions in 2011, approximately 27.2% of the
shares were purchased by individual investors, 72.8% by institutional investors such
as asset management companies, hedge funds, private equity funds, venture capital,
and mutual funds. Here the majority are asset management institutions. The main
source of equities is former employees of the companies whose shares are up for
sale. These account for 79.3% of the total trading volume. Current employees are
the next largest source of shares, at 11.1%. Investors and founders’ shares make up
most of the rest, at 10%.

2 SharesPost
SharesPost was founded in 2009. It focuses on the market of trading non-listed
companies’ equity. SharesPost also provides services such as financing through
private equity, index preparation, and third-party research reports.

2.1 Background
In 2009, Greg Brogger and Scott Painter founded the SharesPost website to establish
an exchange market for equity of non-listed companies. Brogger worked as a
securities lawyer after earning a JD and MBA from the University of Pennsylvania.
The other founder started his own business in 1993. In 1998, they started a company
together, zag.com, that provided an online platform for purchasing cars. It disrupted
the traditional sales model of cars and quickly swept the nation. Subsequently, they
found that private equity transactions were still realized through telephone, which
followed the traditional brokerage model pervasive in the 1930s.They believed that
its disruption would be a tremendous opportunity, so they founded SharesPost.
SharesPost has already attracted 83,000 investors. There are more than 150
stocks traded on the platform, with a total value of up to US$1 billion. The most
frequently traded stocks include hot stocks such as Facebook and Yelp. Even a single
quarter (Q4 2011) has seen a trading volume up to US$180 million in Sharespost’s
secondary market. It shows both momentum and a large potential for growth.
 38 The internetization of finance
2.2 Main policies
1 Threshold for investment: Although SharesPost is an open market, in order to
conform with securities regulations, SharesPost sets the criteria for qualified
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

investors along the lines of those for private placement: individual investors
must have net assets of more than US$1 million or have annual income of
more than US$200,000. Institutional investors must possess net assets of at least
US$100 million.
2 Private equity transfer: SharesPost is similar to a BBS or forum. Buyers and
sellers on SharesPost register as members and trade equities through the web-
site platform. The two parties agree on a time to discuss the transaction, and
then process the negotiation. Finally, the deal reached between the parties is
submitted to the registered broker of SharesPost for an audit. In contrast with
SecondMarket, equity transfers do not need the approval or authorization of
the company.
3 Private placement financing: In addition to private equity transfers, SharesPost
has a private placement market. Start-ups and their agent investment banks can
post offers on SharesPost’s bulletin board.The buyers and sells can then contact
each other directly.
4 Creation of indices: SharesPost created the first private equity index in US
history—the SharesPost Index. The Index consists of thirty representative
companies traded on SharePost’s platform. It is reported in real time during
trading hours.
5 Providing third-party research reports: To find the intrinsic value of the stocks,
SharesPost offers up to 450 reports provided by nine third-party research
institutions. The reports cover many well-known venture enterprises such as
Twitter,Yelp, Facebook, Zynga, and so on.
6 Providing transaction information: Currently, through the Bloomberg termi-
nal, it is possible to browse real-time quote and transaction information on
SharesPost. Historical transaction data is also available.
7 Cost mode: SharesPost charges US$34 to each party in completed transactions,
not distinguishing between private or institutional investors.

SECTION 5: DISTRIBUTION NETWORKS FOR

FINANCIAL PRODUCTS

Essentially, Internet sales of financial products match the suppliers and demanders of
financial products through Internet channels. The demanders of financial products
play a dominating role in the matching process. They search for financial products
and conduct asset allocation. Financial product providers aim at maximizing the
probability and amount of their financial products that are selected into demanders’
“allocation bracket” (a concept of collection of trading possibility). To achieve this
 The internetization of finance 39
goal, they focus on demanders’ preference, disclose the risk-return characteristics of
their products, and conduct certain promotional activities. Some financial products
of low-volatility and high mobility are often linked to payment. It not only meets
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

demanders’ needs of investment needs, but also their needs for payment.
Next, we introduce the main model of Internet sales of financial products, then
analyze Yu’E Bao.Yu’E Bao significantly and unexpectedly changed the structure of
China’s fund industry when it started in 2013. Finally, we discuss the principles of
economics supporting financial products as payment instruments.

Notes
1 Yin, Long. 2012. “Internet Banking and Monetary Electronics - Network Theory of
Finance,” Southwestern University of Finance and Economics doctoral thesis.
2 Jiang, Jianqing. 2000. “Financial Development of High Technology and In-depth Impact
Study,” China Financial Publishing House.
3 Porteous, David. 2006. “The Enabling Environment for Mobile Banking in
Africa,”working paper.
4 Safaricom. 2013. www.safaricom.co.ke/.
5 Vodafone is a multinational mobile operator. Headquartered in Newbury, UK and
Dusseldorf, Germany, is one of the world's largest mobile communications operators, its
network covering 26 countries directly and providing network services together with
partners in another 31 countries.
6 At year-end December 2013, one Yuan equaled about 14 Shillings.
7 Wizzit. 2013. www.wizzit.co.za/.
8 WizzKids are generally young people who sign up as agents.
9 Globe. 2013. http://gcash.globe.com.ph/.
10 Retail stores, banks, Globe Telecom locations, and myriad other locations serve as agents.
11 At year-end 2013, 1 RMB = 6.6 pesos.
12 CGAP (Consultative Group to Assist the Poor). 2006. “Use of Agents in Branchless
Banking for the Poor: Rewards, Risks, and Regulation.”
13 Rural Social and Economic Investigation Department of National Bureau of Statistics
of China. 2010. “China’s Rural Household Survey Yearbook 2010,” China Statistical
Publishing House.
14 Zhang, Jin, and Yao Zhiguo. 2002. “The Network Finance,” Peking University Publishing
House.
15 Zhang, Jinsong. 2007. “The Theory and Practice of Network Finance,” Zhejiang Science
and Technology Publishing House.
16 Zhang, Jinsong. 2010. “The Network Financial,” Mechanical Industry Publishing House.
 4 Mobile and third-party payments
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Paul Volcker, the former chairman of the Federal Reserve, once said that: “The only
useful thing banks have invented in 20 years is the ATM machine.” However, we
believe that the most significant innovation in the financial industry over the past
20 years arose in the field of payments.

SECTION 1: BASIC CONCEPT AND DEVELOPMENT

1 Mobile payments
Mobile payments involve the transaction of monetary value with mobile
communication equipment and wireless communication technology, thus clearing
and settling the positions of creditors and debtors.1 These are made possible by
the spread of mobile terminals and the development of mobile Internet. With
the popularization of mobile terminals, it is possible that mobile payments will
be widely accepted in the transaction of products and services as well as in the
liquidation of debt. It will then be able to replace currency and credit cards, thus
becoming a key form of electronic currency. The basic traits of mobile payments
are as follows:

1 They use mobile communication equipment, especially smartphones, as the

carrier.
2 They employ wireless communication technology.
3 Electronic currency is the basis for mobile payments’ existence, but they only
really function when they coexist.
4 Mobile payments change the form of money without touching upon its
essence.
5 The development of mobile payments relies on third-party payments.

In China, Mobile payments have been developing rapidly. According to data from
iResearch, mobile payments reached Renminbi (RMB) 296.51 billion in transac-
tions by the third quarter of 2013, an explosive growth of 185.3%. The number of
smart phones in China reached 580 million in 2013, a year-over-year (YoY) increase
 Mobile and third-party payments 41
of 60.3%, while mobile shopping made up 38.9% of the mobile Internet market.
2013 also witnessed the growth of mobile games, the market for which amounted
to RMB 11.24 billion, a YoY increase of 246.9%. The development of mobile
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Internet markets has also driven the development of mobile payments, which itself
is adding diverse formats such as SMS payments, near field communication (NFC)
payments, QR code payments, mobile banking payments, and even payments with
facial recognition, and so on.

2 Third-party payments
Third-party payments refer to helping clients quickly realize the function of
currency payments and fund settlements by making connection between clients,
third-party payment companies, and banks over the Internet, thus playing the role
of credit guarantor and technical support as well. According to data from iResearch,
the scale of market transactions of online third-party payments in China added
up to RMB 1.4 trillion in the third quarter of 2013, a link-relative growth rate of
26.7%. Alipay makes up 48.8% of this market, while Tenpay takes the second place
with 18.7%. At the same time, Easylink, quick money, remittance world, EPRO pay,
and IPS are also developing quickly (Figure 4.1).

3 Third-party mobile payments

Third-party mobile payments are executed by third-party payment companies
via mobile payment terminals. According to data from iResearch, the online third-
party payments market in China reached RMB 1.4 trillion in the third quarter of
2013, a link-relative growth rate of 26.7%.Third-party payments in China added up
to RMB 1.2 trillion, an increase of 707% YoY. Remote mobile Internet payments

YeePay
Others 2%
China Pay 3% IPS 3%
4%
Remittance
world 6% Alipay 49%

99Bill
7%

Easylink
7%

Tenpay 19%

Figure 4.1 The market structure of online third-party payments.

Data source: iResearch, by the third quarter of 2013
 42 Mobile and third-party payments
make up 93.1% of this total market, while near field mobile payments occupy only
0.8%. From this data, we conclude that the share of near field mobile payments is
still small.There thus remains vast space for development. Payment methods such as
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

voice payments, QR-code payments, as well as the integration of personal accounts

continue to drive change in this sector.
Alipay, Lakala, and Tenpay are the market leaders in this field, with respective
market shares of 58%, 21%, and 6% (Figure 4.2).

SECTION 2: BASIC PRINCIPLES AND THE ACCOUNT SYSTEM

1 Mobile payments: Forms and principles

While the forms of mobile payment in developed countries are similar to those
in China, that of developing countries is mobile banking, which usually does not
require third-party payments. Mobile payments executed by Chinese banks require
mobile banking. They may also need NFC-enabled mobile phones for near field
payments. If the payments go through the three major telecom operators,2 the
process uses microchips planted in SIM cards (such as the sticker cards on the cell-
phone, or the RFID-UIM3 card used for Bestpay payments). If the mobile payment
products are provided by pure third-party payment companies, mobile banking is
unnecessary and the payments can be completed directly. Examples of this include
“Pengpengshua” by Alipay and Wechat payment by Tencent. These methodologies
provide speed and convenience. However, the mobile payments from third-party
payment companies are generally less secure than mobile banks and tend to be
underwritten by insurance companies.
Mobile banking in African areas requires forward storing, while Chinese mobile
banking uses transaction size ceilings. Mobile payments provided by mobile

YeeBao Others
1% 3%
Lakala Payment
21%

99Bill
1%

UniPay
4%
AliPay
QianDaiBao 58%
5%
Tenpay
Union Mobile
6%
Pay 1%

Figure 4.2 The market shares for mobile third-party payment.

Data source: Enfodesk, by the second quarter of 2013
 Mobile and third-party payments 43
operators in China do, however, require forward storing. However, it is not neces-
sary for payments done by third-party companies such as Tencent and Alipay. As
forward storing makes mobile payments less convenient, those without the forward
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

storing will prove to be more popular in the future.

The relationship between mobile banking and mobile payment is reflected in
two aspects. First, mobile payments are more focused on banking. However, as
mobile banking has relatively high device and security requirements, one must
register in bank branches or on the online banks websites. This means less con-
venient payment operations with more complications, such as repeated entry of
account numbers and passwords. The second point is that mobile payments are
more focused on payments, which means a more streamlined process.

2 The principle of third-party payments

Third-party payments go around direct account settlement between clients and
banks. This is beneficial for a reason: third-party payments can play the role of
guarantor, numerous banks can be combined and payments completed without
online banking or mobile banking services, and it saves transaction costs. Mobile
third-party payments then reduce the costs even more.
Mobile payments ostensibly use mobile phones rather than computers as the
payment terminals. However, it is this kind of transformation that may lead to
revolutionary changes in the field of payments. As payments refer to the transaction
of currency between different accounts, payment itself indicates mobility, while
the most significant strength of terminals like mobile phones is also mobility. The
combination of mobile payments and the third-party payments then magnifies this
advantage.
Before the emergence of the third-party payments, the payment and clearing
system included the connection between commercial banks and clients, as well
as the connection between commercial banks and the central bank. As the com-
mercial bank’s payment and clearing counterparty, the central bank can conduct
liquidation by netting. In the already existing payment and clearing patterns, as the
clients are not able to establish a direct connection with the central bank, they have
to establish a connection with each commercial bank individually. This reduces the
efficiency of payment and clearing. After the emergence of third-party payments,
clients can build connections with the third-party payment companies, then the
third-party payment companies represent their clients when establishing the con-
nection with commercial banks. The third-party companies become payment and
clearing counterparties between clients and commercial banks, so they can net large
amounts of trading capital through intermediate accounts established at different
banks, while only small amounts of interbank payments are completed by the pay-
ment and clearing system of the central bank. The third-party payments complete
the clearing of large amounts of small transactions after the third-party payment
companies net them with secondary settlement. They thus undertake the functions
of the central bank in payment clearing, as well as offering credit guarantees. Before
 44 Mobile and third-party payments
the emergence of mobile payments, computers mainly undertake the connection
between clients and third-party companies. However, after the birth of mobile
payments, the connection will gradually migrate to mobile phone terminals.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

3 The function and categories of accounts

There is a close connection between payments and accounts. Accounts are the sine
qua non of payments, and they gain even more importance in the era of electronic
currency. The chairman of Guotai Junan Securities once said, “He who wins the
accounts wins power over the future of finance.”4 In the era of electronic currency,
it is essential to have a personal account to ensure that the currency has the pay-
ment functions and qualifies as a financial product, which is not necessary for cash
transactions. A financial product that can be used as means of payment generally has
low volatility and high mobility (see more details in Chapter 3).
There are many different formats for accounts. One includes personal accounts
at traditional financial institutions. Payments based on these accounts still dominate
the market. Inside a certain financial institution, this kind of account is forming
gradually (such as the Yiwangtong accounts of the China Merchants Bank and
Junhong accounts of Guotai Junan). This kind of account has not been totally inte-
grated between different financial institutions. However, after the emergence of
Internet banking, parts of these functions, such as balance inquiries, will be inte-
grated between banks. The securities company accounts may also implement the
integration. When clients open an account at securities companies, they actually
open an account at China Securities Depository and Clearing Co., Ltd. The clients
can thus use securities accounts at different companies by only applying for custody
transfer.
The second type includes accounts in third-party payment companies or finan-
cial services companies such as Alipay. Alipay began to open its account system to
third-party applications such as Dingding Discounts and iReaders in July 2013.
Clients can now log into the third-party applications directly with their Alipay
accounts, thus eliminating the cumbersome registration process. Clients can also
use the information in their Alipay accounts when they need to pay within these
applications, which is both convenient and safe.
The third type includes social networking platform accounts, such as QQ, which
can be used to buy various virtual goods with virtual currency. The QQ account
system is now gradually opening to third-party applications.
In conclusion, various kinds of accounts have only implemented the integra-
tion of partial functions within certain categories. They have not completed the
integration of different kinds of accounts. This greatly limits the development of
mobile and third-party payments. However, with the development of information
technology, personal accounts will take shape gradually and become consolidated
accounts, integrating all the personal business and all the assets and liabilities. It will
thus serve as the beginning of personal financial activities and even of daily life.The
account integration in turn promotes the development of mobile and third-party
 Mobile and third-party payments 45
payments, because people will be able to use the mobile payments anywhere once
the integration is complete. Without the integration, despite the existence of the
basis of mobile payments, people will abandon mobile payments because they either
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

lack a necessary account or tire of the cumbersome registration process.

With the rise of Internet finance, personal financial accounts are no longer con-
fined to the traditional financial institutions.They can now be provided by Internet
companies such as Alipay and Tencent. The account providers will also proliferate.
And with the gradual integration of personal accounts, even the central bank may
begin to provide them (see Chapter 2), because integrated personal accounts have
a nature similar to public goods.

SECTION 3: PROPERTIES OF FINANCIAL PRODUCTS AND

MONETARY CONTROL

1 Payments and electronic currency

The essence of both mobile payments and third-party payments is the fund flow
of electronic currency. Despite the different definitions of electronic currency, it
generally has the following features: currency value represented by virtual accounts,
stored in electronic devices, and possesses a general purpose. It thus becomes an
acceptable method of payment for issuers and entities beyond close business part-
ners. Electronic currency is considered a form of currency, so it is not only a means
of payment but also has the functions of means of exchange and store of value.
Electronic currency includes that based on bank cards (issued by commercial banks,
including savings and digital checks), or products that need transform between
bank deposits, cash, and electronic currency. Jin Chao and Leng Yanhua call this
type “electrified currency,” which includes phone cards, meal cards, and digital
cash.5 The second is based on virtual accounts, such as M-PESA and Internet cur-
rency (for more on Internet currency see Chapter 5).
Electronic currency can not only be transformed with the sovereign currency
issued by the central bank, but can also be separated from it, thus taking on an inde-
pendent existence as an Internet currency. Enterprises seeking profit maximization
and the central bank may also issue currency at the same time.
If this pattern of currency and payment methods (the mobile third-party pay-
ment) which is conventionally accepted is going to rise to become acknowledged
by law, it will require the gradual enlargement of the range of application (including
network scale effects) and its irreplaceability in the social economy. It can then in
turn force the law to stipulate it as a kind of currency pattern and payment method.
Mobile payment, including third-party payments, which is limited by net-
work and scale effects, still has a long way to go to be popularized. However, the
combination of mobile and the third-party payments will accelerate this process.
During the initial period, as the scale of mobile payment users is limited, suppliers
are unwilling to provide while consumers are reluctant to use. To break through
the chicken-egg problem, effective measures taken by each involved party are
 46  Mobile and third-party payments
needed, such as vigorous advocacy of the advantages of mobile terminals and broad
­prospects for the mobile payment, thus promoting the use of mobile payment. In
the ­beginning, the government can provide subsidies or promote the c­ onstruction
of related ­ infrastructure, while suppliers need to take preferential measures to
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

encourage ­consumers to participate in mobile payments.

The mobile payments have both network externalities and positive externali-
ties. The utility that consumers can get from consuming certain commodities such
as mobile payments relies on the consumption of other consumers. The core of
the network scale effects of mobile payments is the establishment of basic users.
According to rationality in economics, the premise of using mobile payments as a
payment method is that the benefits of using it are larger than its costs.To ­establish the
premise, the number of mobile payment users must reach a scale that is big enough
to make this the case. Consumer acceptance is important to make the number of
mobile payment users sufficient. Only if they except that large quantities of people
will use the mobile payments, can the number of users reach a certain scale.
We can use the analytical framework proposed by Nicholas Economides and
Charles Himmelberg to analyze mobile payments.6 The framework is mainly used
to analyze the network scale effects of the telecommunications industry. There are
many similarities between mobile payments and the telecommunications industry,
mostly in relation to network effects.
We suppose that the utility function of consumers is u( y, n e ) = y(a + b(n e ))y.
(a) y  ­represents the income of consumers. The function shows that the ­utility
­consumers gain from mobile payments are linked to their income; the higher
the income level, the more ability they have to participate in mobile payments.
(b) a represents the intrinsic value of mobile payments when there are no other
users.There is no necessity for the existence of mobile payments if there is no other
user, so a equals zero in this condition. (c) n represents the proportion of mobile
payment users among total users. 0 ≤ n ≤ 1. b(n e ) is used to measure the benefits
that consumers get from the network scale effects of mobile payments. b(n e ) is an
increasing function of n e , and b(0) = 0. n e is the proportion of mobile payment
users among the total users.
Given the price p7 and basis users n e , we have u( y, n e ) = y(a + b(n e )) > p . The
aggregate demand f­ unction for mobile payments can be represented as n = f (n e , p ).
For the given user n e, the aggregate demand of mobile payments moves inversely
with the price. However, if the basic users n e increase, the aggregate demand of
mobile payments moves with the price. The above-mentioned a­ ggregate demand
­function can also be represented as p = p(n,n e ) (Figure 4.3), there is n = n e when
the ­equilibrium is achieved, and then we call it the aggregate demand curve of
­perfect information, which is represented as p = p(n,n ).The curve p = p(n,n1e ) and
p = p(n,n2e ) ­represent consumers’ willingness to pay for mobile payments, in which
n2e > n1e . The more users the consumers accept, the higher their willingness to pay
and the more actual users of mobile payments.
The aggregate demand curve of perfect foresight is not monotone because
when the number of users reaches a certain value, the marginal cost of mobile
 Mobile and third-party payments 47

C‚P
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

p(n,n4e)

p(n,n3e)

p(n,n)
p(n,n2e) E2

E1
p(n,n1e)

0 n1e n2e n3e n4e n0 1 n

Figure 4.3 The demand curve for mobile payments.

payments become quite low, so the actual price that consumers pay becomes lower.
That is before reaching n0, the price of mobile payments moves the same way with
the number of users, while after n0, the two move inversely. When n is big enough,
lim n → 1 p(n, n ) = 0.
The key to understand the network scale effects is the consumers’ expectation
of the users of mobile payments.When people expect that the number of base users
is large, that is there will be a bright future for mobile payments, people are willing
to get involved. They are also willing to pay a higher price now because the price
will be low when the number of mobile payment users is large enough. Sometimes
the costs can approach zero.To make people expect that the number of mobile pay-
ment users will be high in the future, service providers and the government need
to cooperate. For instance, the government should lead the development of basic
infrastructure for mobile payments and set related rules and laws. This will send
a positive signal that leads to an improved outlook for mobile payments. On the
other hand, suppliers can provide consumers with subsidies and a more convenient
experience to make the consumers truly feel the advantages of mobile payments.
In conclusion, with the importance of network scale effects for mobile payments
and electronic currency, the range of application of electronic currency issued by
private enterprises will become wide spread. It can be used to purchase both virtual
commodities online and physical products in real life. Therefore, the importance
 48 Mobile and third-party payments
of network scale effects for mobile payments and electronic currency means that
electronic currency will either replace the sovereign currency issued by the central
bank or coexist with it, which will have a significant impact on the monetary policy.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2 Payments and financial product properties

In the era of electronic currency, with the rapid growth of information technology,
mobile and third-party payments have developed traits that resemble financial
products. This trait has not only made mobile and third-party payments more
attractive, but also has increased the difficulty of control over monetary conditions.
This trait is defined by having the potential to make profit for clients, while being
sensitive to the changes of currency. This resemblance to financial products has one
defining characteristic—either the currency in the payment method is a financial
product itself, or it can create value by switching between being a financial product
and being a payment method.
Currency can be used as a financial product when it is not making payments.
This is true for both precious metals and electronic currency.The main difference is
that during payment, the transaction costs of currency and financial products were
higher in the era of precious metals than in the era of electronic currency. Besides,
in the era of electronic currency, the mobility of mobile payments is consistent with
that of payments. This greatly reduces the transaction costs between currency and
financial products.
In the era of electronic currency, as long as one has an online account, such as
an Alipay account, currency can be linked to financial products with a few clicks
or the swipe of a finger. When no payment is being made, the balance on personal
accounts is a financial product. During a payment, the balance can be used as cur-
rency. This transformation can be made within a second, which was previously
unimaginable. However, the power of technology also brings new challenges to
monetary control.

3 Payment and monetary control

The rise of mobile and third-party payments has affected monetary control in
the following ways: First, the low transaction costs of mobile and third-party pay-
ments have reduced the need for cash. Second, the financial product resemblance
of mobile and third-party payments has obscured what counts as money, which
could have an influence on the monetary base and multiplier. At last, with the rapid
development of mobile and third-party payments, a private supply of electronic
currency is available, which could change the currency supply system of the central
bank and commercial banks.
With financial innovation, technological development, and the emerging third-
party and mobile payments, other assets with relative high yield can be used as
currency. This phenomenon is especially obvious in the era of electronic currency,
as low transaction costs help to reduce the demand for interest-free electronic
 Mobile and third-party payments 49
currency. For the first time, people can enjoy financial services anytime, anywhere,
and any way. This also includes converting high yield assets into electronic cur-
rency. People will put electronic currency into bank accounts or virtual accounts to
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

collect interest. Only when there is a purchase will withdrawals be made through
a simple operation on computers or mobile phones for the exact amount to com-
plete the purchase. The emergence of mobile and third-party payments will reduce
the demand for currency as well as accelerate the velocity of circulation.
The emergence of mobile and third-party payments obscures the bound-
ary of currency arrangements. Using mobile terminals to send instructions, one
can quickly transform non-liquid currency into liquid currency. It is now hard
to distinguish whether a currency is a demand deposit, a saving deposit, or cash.
Furthermore, some nonbanking institutions have begun issuing electronic cur-
rency. For example, mobile carriers can issue electronic currency through virtual
accounts. Many of these electronic currencies were not considered currency before,
which greatly increased the difficulty of regulation.
Mobile and third-party payments are accompanied by private issuance of
electronic currency. Some nonbanking institutions are able to issue electronic cur-
rency autonomously. Most of them are mobile carriers and Internet companies.
Since there is not enough regulation, the reserve ratio of electronic currency is
a company’s own decision. Of course, companies will put some money aside as
reserves because their reputations are at stake. However, this is only a soft con-
straint. As long as electronic currency issuance remains unregulated, companies
may minimize their reserve in order to maximize profit. This could be a shock to
the currency supply.
At the current stage, the private supply of electronic currency will not end the
fundamental system of central bank currency issuance. However, changes in the
format of currency demand, the unclear boundary of currency arrangements, and
the diversification of currency issuers will certainly pose a threat to central banks.

SECTION 4: ANALYSIS OF TENCENT’S WECHAT PAYMENT

1 A brief history
Since its start in 2013, Wechat Payment has rapidly spread across China. Businesses
that are currently supporting Wechat Payment include QQ top-up,Tencent voucher
center, Guangdong Unicom, WeLOMO, McDonald, Wechat Group Purchase, and
others. Banks’ personal financial services are also cooperating with Wechat, and
Tenpay has already started its cooperation with several banks, experimenting with
less risky fixed-rate products. Wechat Payment has also extended itself to payment
for everyday expenses. One example is the “Shenzhen Power Supply” service
provided by Shenzhen Power Supply Bureau, which provides services such as
online payment of electricity bills, electricity price checks, and balance inquiries.
Paying for everyday expenses may not be as profitable as credit cards and electronic
payments, but may increase client stickiness.
 50 Mobile and third-party payments
In order to make payments more convenient, Wechat is gradually partnering
with some companies to push voice payments. Clients can simply say the products
they desire on their mobile phones, and then both purchase and payments are
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

processed. Wechat also allows payments to be made through QR code scanning.

Wechat clients do not have to exit the Wechat application and while paying, they
simply need a bank card attached to a Wechat account to pay for products provided
by public accounts on Tenpay. The entire process takes less than a minute.
However, the progress of financial institutions’ adoption of Wechat Payment is
relatively slow. By December 2013, more than 40 securities companies have opened
accounts on Wechat. Most of them, however, do not involve payments. If there is a
need for financial products, these companies need to go on fund company websites
with their cell phones. The banking industry is the same. Furthermore, merchants
who joined Wechat Payment are predominately from the Chinese mainland. Since
it can only accept payment transactions from the Chinese mainland. The Wechat
platform cannot yet make overseas purchases.

2 Operating principles
The name “Wechat Payment” implies two things: It is a mobile payment application
where transactions are made through the Tenpay third-party payment platform. It
is also a payment method in which payments are made through mobile banking
by accessing banks’ Wechat public accounts. The Wechat Payment that we generally
refer to is the former. It integrates social network platforms, third-party payment
companies, as well as mobile banking, thus maximizing the client experience.
The bank must have a Wechat public account to handle payments. Banks can
interact with clients on Wechat, and then direct their clients to their mobile banks
to complete the payment. Of course, the clients must have a mobile bank account.
The core characteristic of Wechat Payment is its integration of a social networking
platform, third-party payments, and mobile banking. It combines their strengths to
provide a new service to clients (Figure 4.4).

3 Risk and risk management

The main risk facing Wechat is IT risk (see Chapter 10). This includes damage to
hardware, software failures, viruses, operating errors, potential mistakes in data trans-
mission and data processing, and various online frauds.

Third-party payments
(Tenpay)

Mobile
Transaction
Users social
completed
network

Bank’s Wechat Mobile bank

Figure 4.4 The process of Wechat payment.

 Mobile and third-party payments 51
The information technology risk of Wechat Payment cannot be managed
with just one method. A system of risk control must be in place, attacking the
problem with both technological and non-technological means. Some examples
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

of technological means include setting a separate password for Wechat Payment,

sending confirmation text messages, requiring voice authentication, verify-
ing client identity by using big data analysis, and so on. When a risk to client
accounts is confirmed, the system must notify the client immediately, and a series
of protective measure must be turned on to provide real-time protection for clients.
For example, Alipay developed a real-time risk monitoring system for its accounts
and transactions in 2005 to preempt and avoid losses. Non-technological means
include providing insurance, which can be negotiated with insurance companies.
In fact, Wechat Payment is now fully covered by Peoples Insurance Company of
China (PICC).

SECTION 5: A BRIEF ANALYSIS OF YU’E BAO

1 Creation and development

Yu’E Bao is a Chinese service developed by Alipay. It targets individual users to
provide interest on their savings. Since its creation in June of 2013, Yu’E Bao has
seen tremendous growth.According to Tian Hong Asset Management, in December
of 2013,Yu’E Bao reached over 43 million users, with total savings of RMB 18.53
Billion. The assets under management ranking of Tian Hong Asset Management
soared from 46th in the first quarter of 2013 to 2nd by the end of 2013.
Yu’E Bao has posed a threat to banks by challenging their deposit savings. It
has attracted numerous small investors and low value added money market funds.
Cumulating tiny savings was an exclusive right for the banks that is now threatened.
With the development of technology, Yu’E Bao may become a universal account
in the future, providing services in lending, saving, securities purchases, and asset
managing products, on top of its current services.

2 The fundamentals of Yu’E Bao

The success of Yu’E Bao lies in its ability to create value for all involved. The secu-
rities companies increase their sales through Alipay, and Alipay earns a commission
for bridging the gap between securities companies and individual buyers. Moreover,
the individuals can collect interest on savings with Yu’E Bao while keeping their
savings liquid. The innovation of Yu’E Bao is in purchasing and redemption.
In purchasing, because Yu’E Bao is not a qualified trustee, funds from the
customers must be quickly transferred to Tian Hong Asset Management’s escrow
account. Otherwise, there could be a suspicion of misconduct. The funds actu-
ally go through two steps: purchase and transfer. During purchase, the balance in
Alipay moves to Yu’E Bao. In transfer, the balance is automatically transferred to
Tian Hong’s escrow account.
 52 Mobile and third-party payments
The greatest innovation of Yu’E Bao with respect to redemption is real-time
redemption. If redemption is simply a reversal of purchasing, then there is no
misconduct. Through this process, however, redeeming T + 0 cannot be achieved.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

In order to achieve real-time redemption, there are two other methods. Both of
them, nevertheless, are on the edge of being illegal. One of them is to have Alipay
cover the cost, which Tian Hong later reimburses. The second method is for Tian
Hong to keep a reserve in Alipay’s payment account. When redemptions occur, the
reserve can be used to transfer money to the users (Figure 4.5).
Although Yu’E Bao has bypassed banking sales, it still cannot exist without banks.
Third-party payment companies needs banks to verify client information, transfer
funds and liquidate positions when necessary.

3 Risks
On top of IT risks,Yu’E Bao also has legal and market risks. Legally,Yu’E Bao lacks
risk disclosure to its users. “A Guide on Securities Investment Sales Applications”
states that institutions selling securities should investigate and evaluate the investor
risk tolerance at or before their first purchase. For investors who have already bought
securities, they should also investigate and evaluate the investor’s historical risk
tolerance. On Yu’E Bao redemptions, if Alipay covers the payments, it may violate
clause 23 of “Regulation on Excess Reserve for Payment Institutions” by complet-
ing payments requested by clients ahead of time. Tian Hong may also violate the
“Interim Measures for the Administration of Securities Investment Funds” with its
subscription and redemption activities.
There may also be market risks, since Yu’E Bao is not exactly savings. Users may
face interest rate declines or even losses due to mistakes in management. The initial
high return rate was possible due to a shortage of funds in the market, which is
likely to continue.Yu’E Bao may also be investing in non-monetary products with
high risk. If the economy slows down,Yu’E Bao could decide to take more risks in
pursuit of higher returns.

AliPay

Alipay Income Box

Transfer Yu’E Bao Support Money Market
Balance
Fund

Clients Tian Hong Asset

Management

Figure 4.58 The operational structure of Yu’E Bao.

 Mobile and third-party payments 53
4 Financial products as instruments of payment
The existence of Yu’E Bao gave rise to numerous financial products involving
third-party payment and funds. These products provide interest to users and can
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

be an instrument of payment when needed. This has blurred the line between
investment and payment, as well as financial products and currencies. With interest
rate marketization, these products are likely to grow in the next years. If their total
market cap reaches RMB 1 trillion, they could exert tremendous influence to the
financial system.
To be an instrument of payment, a financial product must have low volatility
and high liquidity. These two characteristics are closely correlated, but there is no
causality between them. Other than the composition of financial products, these
characteristics are also results of the market environment.
Statistically, low volatility requires a product to be stable in price over time,
which makes the product maintain its value. This is similar to the store of value
characteristic of currencies. Two approaches can be used to achieve low volatility.
Either the product is a diverse portfolio of securities, or the product has low risk
due to high quality issuers or counterparties, advantages in capital structure, short
maturities, or good contract terms.
Liquidity refers to a financial product’s ability to convert quickly to cash. On
some occasions, it may also refer to the product’s collateral quality—the higher the
quality, the lower the haircut.9 Market liquidity is plotted by the bid-ask spread in
the secondary market. If the spread is low, the liquidity is invariably high. In most
occasions, when the risk of securities is low, the secondary market is often well
equipped with high liquidity.
Most money market funds invest in treasury bonds, commercial paper, and
short-term financing bonds. They are generally low in volatility and highly liquid,
which makes them ideal for payments. In fact, M2 statistics include money market
funds and savings accounts with the ability to issue checks.
Nevertheless, money market funds are not risk free, and they are not protected
by deposit insurance as some savings accounts are. They are not always low in vola-
tility and highly liquid. In fact, money market funds experienced a major crisis in
September of 2008.Therefore, we can conclude that there are indeed risks to finan-
cial products combining third-party payment and funds. As a result, implementation
of appropriate regulations will be very important.

5 Main models
5.1 Self-built platforms to sell financial products
With the rise of Internet-based finance, major commercial banks have gradually
established their own e-commerce platforms. They provide both payment and
other one-stop financial services. For example, the “ShanRong” service of China
Construction bank is both a B2B and B2C platform. It is an in-depth incorporation
 54 Mobile and third-party payments
of e-commerce and financial services. Self-built platforms aim to develop custom-
ers. However, because of the closed nature of the platform, less-diversified product
portfolio, and non-existence of a price advantage, customers are not very active on
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

these platforms. This is evidenced by the shutdown of ICBC’s online store.

5.2 Utilizing third-party channels to sell financial products

There are various ways to run this model: to sell products through third-party
e-commerce business platforms, such as Taobao flagship stores; routing through
third-party financial products such as Yu’eBao (accomplished mainly through col-
laboration with mutual funds); and marketing through a supermarket of funds,
loans, or insurance. An example of each type of site is Shumi Funds (funds), Ron
360 (loans), and Huize (insurance).

5.3 Using social networks to sell financial products

This model refers to financial institutions that utilize social platforms to connect
financial institutions and users. This model takes full advantage of big data analysis,
data flows, cloud computing, and relationships on social network platforms to
capture “soft” information. Simultaneously, by creating a virtual online VIP room,
customers can enjoy the feeling of being served over the counter. Selling finan-
cial products through social networking platforms can fundamentally change the
relationship between financial institutions and customers. It achieves real-time
dialogue between financial institutions and customers. The more opportunities
there are for dialogue, the more information will be shared. Financial institutions
can therefore more accurately determine customer needs. Typical representative
include microblog banks, Wechat bank, and Baidu finance.

5.4 Integration of relationships and the Internet to launch a financial

supply chain
In this model, what is sold is more than just a financial product; it is a package of
financial services.The customers form a customer cluster in relationship with a core
enterprise. Through the “transfer” of target customers, the financial supply chain
can be an effective solution to information asymmetry. Because Internet supply
chain financing can achieve a high degree of integration of information flows,
capital flows, logistics, and online control, lending efficiency, and safety are vastly
improved. The financial supply chain of Jingdong is indicative of this trend. It is an
e-commerce company that provides guarantees for merchants to obtain loans from
financial institutions. Through guarantees, merchants more easily obtain loans.
Finally, not all financial products are suitable for online marketing, especially
products of high complexity, high customization and high risks, or products
requiring investors to make many sophisticated judgments.
 Mobile and third-party payments 55
Notes
1 Shuai, Qinghong. 2011. “Electronic Payments and Settlements,” Dongbei University of
Finance & Economics Press.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2 The services are actually provided by the subsidiaries of the telecom operators, including
Wing e-commerce(China Telecom), Unicom WO easypay(China Unicom), e-commerce
of CMCC(CMCC).
3 RFID-UIM card is a kind of mobile phone card with the function of radio frequency
identification.
4 Wan, Jianhua. 2003. “E-era of finance,” China Citic Press.
5 Jin, Chao, and Yanhua Leng. 2004, “Electronized Currency, Electronic Currency and
Money Supply”, Shanghai Finance.
6 Economides, Nicholas, and Charles Himmelberg. 1994? “Critical Mass and Network
Evolution in Telecommunications?” Working paper.
7 The p here can be understood as the transaction costs of mobile payments.
8 This diagram is built on “A Construction Bank’s Interpretation of Risks: How Will
Yu’E Bao Impact Banks”, especially its content involving Yu’E Bao’s structure.
9 Haircuts refer to reductions in the amount one can borrow based on collateral of a
certain present value.
 5 Internet currency
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

This chapter examines the form of money in the Internet world. The core of our
idea is that many quality online communities with payment functions will issue
their own “Internet currencies.” Internet currency will be widely used for economic
activity on the Internet, and society will revert to the state where currency issued by
the central bank and that issued privately coexist. It will create new challenges to the
foundations of current monetary, currency, and central banking theory.

SECTION 1: THE CONCEPT OF INTERNET CURRENCY

Virtual currencies have already emerged as a rudimentary form of Internet currency.

Classic examples include Bitcoin, Qcoin (created by the Tencent, one of the key
Chinese Internet companies), Facebook Credits, Amazon Coins, Gold (World
of Warcraft), and Linden Dollars. These virtual currencies can be used in social
networks, online games, or virtual worlds to purchase goods (hereafter referred to as
digital products) and services. This activity has already developed into an extremely
complex market mechanism.
Some of these virtual currencies are not exchangeable for legal money, such as
World of Warcraft’s G coin, a currency that may only be used in the online game’s
environment. Other virtual currencies such as Amazon Coins employ legal cur-
rency to allow holders of virtual currency to buy both online and physical products
and services, but they cannot be directly exchanged for legal currency. Bitcoins and
Linden Dollars are in a final category that can both be exchanged for legal cur-
rency and be used for purchases. According to research conducted by the European
Central Bank,1 virtual currency transactions in the United States amounted to $2
billion in 2011, a level that surpasses the gross domestic product (GDP) of a few
countries.Traditional payments companies are also jumping into the virtual currency
area. For example, in 2011 Visa spent $190 million to buy PlaySpan, a company that
manages electronic goods transactions for games, online media, and social networks.
American Express entered this market through its $30 million purchase of the vir-
tual currency platform Sometrics.Virtual currencies will prove even more beneficial
once the development of mobile payments improves their ability to transact.
 Internet currency 57
We identify the following characteristics that define virtual currencies:

1 Issued by an online community

Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2 Either not regulated at all or very lightly regulated, especially by central banks
3 Exists in digital form
4 Issuing online community has an internal payments system
5 Accepted and used by the members of the online community
6 Can be used to purchase the online community’s virtual goods or real goods
7 Serves as a unit of exchange for online products

Number five indicates that the virtual currency can be used as a unit of exchange.
A few online communities have user numbers that surpass the population of
many countries. For example, Facebook’s monthly active users (MAU) come
from many countries and have now surpassed 1 billion. The sixth indicates the
unit of exchange function, and the seventh means it can function as a price
discovery mechanism. If we consider the purchasing power of Internet cur-
rency and the relative prices of goods purchased, it also has the “store of value”
function. Internet currencies thus fulfill all of the standard requirements to be
considered a currency (unit of exchange, store of value, and price discovery).
Internet currencies also defy borders. They are both international and exceed-
ingly powerful.
Up to now, most Internet currencies are essentially fiat money with a central-
ized issuer. The value depends on the trust people have in the issuer. Bitcoin is thus
an exception. There is no central issuer, and its characteristics more approach that
of money backed by precious metals. Later sections will go into further detail on
this point.

SECTION 2: THE ECONOMICS OF INTERNET CURRENCY

1 Online communities and network economics

Internet currency’s benefits for online communities include the following:

1 Independent pricing for digital goods

2 Wealth effect due to online accounts
3 Facilitation of transactions between members
4 Increased “stickiness” due to more effective use rules than legal currency
5 New income sources such as seigniorage, exchange with legal currency, and
use of inactive members’ funds
6 Creation of economic activity on the online network
7 No counterfeit currency

The distinction between digital and physical goods is becoming increasingly unclear.
Digital goods such as software, e-books, music, films, and news are all digitized
 58 Internet currency
information.This means that consumer utility does not vary with their form. More
and more people recognize the value in these products. In the future, many prod-
ucts that do not require physical delivery will be produced, traded, and consumed
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

online (see Chapter 11).This type of goods will continually increase as a percentage
of total goods. For example, Tencent makes Renminbi (RMB) 22.8 billion from
online games, 51% of its total income. It is not necessary to use legal currency for
this type of activity.
It is also important to note that the relationship between the Internet economy
and the real economy is getting ever closer. Imagine the following scenario:
someone creates a software product online and sells it in exchange for Internet
currency. He then uses this currency to buy food at McDonalds, which in turn uses
this currency to buy digital products. The real and Internet economies combine
perfectly. Legal money is excluded and unnecessary.

2 The new form of currency: Facilitating network payments

Chapter 4 predicted the coming integration of forms of payment such as bank cards
and mobile payments. We will have the convenience of payments anytime, any-
where, and any way. This reality is already upon us if we consider virtual currency
accounts as a type of deposit. One can make quick payments over the Internet.
A possible future state may occur if everyone has both virtual and legal currency.
Internet currencies are both easily exchangeable across online communities and
into legal currency. Online currencies would be far more integrated with the real
economy, and we could even witness the emergence of Internet stocks, bonds,
deposits, and credit.
Up to the present, there have been three stages of currency development:

1 Barter economy, no currency exists

2 Hard goods as currency, such as gold, silver, or paper money redeemable for
precious metals. Money creation was mainly determined by discovery and
mining of precious metals
3 Fiat money without intrinsic value or guaranteed convertibility

Private organizations were the primary issuers at the beginning of stage three. Legal
currency only emerged when law specified the central bank’s currency as that
used in payment, settlement, and clearing, excluding others.The central bank, com-
mercial banks, savers, and lenders all participate in money creation. Base money
includes money in circulation and the reserves commercial banks store at the cen-
tral bank. Commercial bank credit and bond investments cause deposits to expand
to many times over. We separate money into three levels: M1, M2, or M3 based on
liquidity (from highest to lowest). Legal currency replaced private currency due to
the central bank’s creditworthiness and deep involvement in payment, clearing, and
settlement systems.
 Internet currency 59
However, we have not yet reached the end of money’s evolution. Hayek and
Friedman already expressed doubts about this system all the way back in the
1950s. Hayek argued that government currency issuance since the eighteenth
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

century was not natural, and he saw it as an important contributing factor of

both inflation and the business cycle. He recommended a return to multiple
issuers to reinstate competition for issuance that would lead to more stability
and societal benefits. Friedman wanted to do away with the central bank and
replace it with an automated mechanism that would increase the money supply
at a stable rate.
Although private money no longer circulates in large amounts, quasi-private
money is still common. In the twentieth century, China’s universities issued
cafeteria cards that could be traded or used to buy food and household prod-
ucts. Today, rewards programs, frequent flier miles, gift cards, and other forms of
quasi-private money are everywhere. Internet currencies combine fiat money
and private money. There are two main reasons why legal currency is unable
to replace Internet currencies: some areas of online activity cannot accept legal
currency, and Internet technology permits payment outside the central bank’s
systems. Since payments and money follow the same evolutionary trajectory,
currency’s fourth stage will witness the coexistence of legal money and Internet
currency.

3 The risks of Internet currency

One of the key risks inherent to Internet currency is that its issuers are unable to
match neither the creditworthiness nor the effective payment and clearing of a
central bank. For example, Bitcoin’s main risk is an insufficiently reliable clearing
process. It is impossible to avoid credit risk, liquidity risk, operations risk, and
payment problems. Since these currencies allow for significant anonymity, they
are difficult to regulate. They are thus more susceptible to illegal activities such as
money laundering, which give rise to both legal and reputational risk.
Internet currency may also affect price stability. Before the emergence virtual
deposits, the creation of virtual currencies did not involve the central bank or
commercial banking system. They did not create a multi-tiered monetary system.
However, excessive emission of Internet currency could create inflation in digital
goods. When digital goods enter the consumer price index (CPI) basket, they will
enter as relative prices between digital goods to create an average. Price risk is thus low.
Internet currencies can also affect physical goods prices. Once it integrates into
the real economy, it may even replace legal currency in some areas. This would then
affect legal money velocity through a contractionary effect. Monetary policy and the
quality of the central bank’s statistics would then be affected; since the central bank
does not necessarily have access to key information on Internet currency. As for
financial stability, fluctuation in the rate of exchange between the Internet currency
and legal currency is the main source of risk. This is especially clear for Bitcoin.
 60 Internet currency
SECTION 3: BITCOIN

Bitcoin is the first ever electronic currency based on P2P technology. It is both issued
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

and traded online. It was announced by Satoshi Nakamoto in 2008 and officially
entered circulation on January 3, 2009.2 By the end of December 2013, 12 million
Bitcoins had been issued.3 Assuming a value of $900 per Bitcoin,4 the total value
amounted to $10 billion, higher than the GDP of more than 60 countries.5
Cryptography and universal Internet serve together as Bitcoin’s foundation.
Its uniqueness comes principally due to its ability to process payments and issue
currency without either the central bank or third-party payments, and its modern
verification system allows for relatively high anonymity. Its initial supporters were
tech fanatics, anarchists, and criminals, who viewed it as an embodiment of the spirit
of democracy. It challenges the banking system and those in power by breaking the
state’s monopoly on issuing currency. It protects its holders from expropriation
through inflation. As it increased convertibility with physical currencies, it gained
the attention of the media, governments, scholars, and others. It also sparked
significant controversy. Competitors such as Litecoin, Peercoin, and Primecoin have
already emerged.6 These competitors retain Bitcoin’s main idea, but they differenti-
ate themselves through special characteristics.

1 Working mechanism
In early 1998, Dai Wei explained the thinking behind Internet currency in a cryp-
tography-focused e-mail newsletter. It would be “provided to and by untraceable
entities” and “the government is … unnecessary.”7 Bitcoin is the development and
extension of this thinking. The most important innovation is the decentralized
model it uses for its payment system (Figure 5.1)

Node
Person

Person
Node
Person

Person
Node

Figure 5.1 A decentralized payment system.

 Internet currency 61
In an entirely decentralized system, one can have any number of payment ter-
minals to confirm transactions and protect the accounts system. Payments are con-
firmed in two steps: it begins when a node successfully completes the initial step
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

through competition and finishes when this information is broadcast and finally
confirmed by the entire network.
Bitcoin’s new model requires it to deal with two important questions: how
to protect the anonymity of those transacting and how to ensure that the same
money is not used in two transactions. Bitcoin has dealt very creatively with these
problems. It uses public-key encryption to protect anonymity. There is a public key
that corresponds roughly to one’s address or account number at a bank, and there
is a private key held only by the Bitcoin holder. The public key acts as the address
for reception of Bitcoins, and the private key provides access to the account, its
Bitcoins, and payments. Thus the public key functions like an e-mail address, the
private its password.
Figure 5.2 shows how both keys are used for Bitcoin transactions. Assume that
in N transactions, person A wants to make a transfer to B, and in N + 1 transactions
B wants to further transfer the coins it received from A to a new person D. This
transaction requires four steps.
Person “A” first generates the information from the past N transactions, including
the connection from the last transaction, information from the current transaction
(including the amount to be transferred), and the public-key corresponding to
person “B”. “A” then uses his/her private key for the electronic verification process

N The connection from

the last transaction
Information from the
current transaction
Person A
1 The public-key
corresponding to person B Network

Person A’s Private-key 2

Node C

Person B
Node E
N+1 The connection from 4
the last transaction Node F
3
Information from the
current transaction

The public-key
Person D corresponding to person D

Person B’s Private-key

Figure 5.2 An illustration of Bitcoin transaction.

 62 Internet currency
and sends it to the network. In the second step, the network checks and confirms
A’s request, including whether it was sent from A, whether it possesses sufficient
Bitcoins, and whether these Coins have already been used. If this is confirmed, it
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

will be sent over the entire network and saved. In the third step, B undertakes the
same steps for transaction N + 1 as A did in transaction N. In the fourth, the net-
work then undergoes the same confirmation process as in the second step.
This is the complete process for Bitcoin transactions. It must prevent that the
same coins are used multiple times. In a traditional economy, the central bank per-
forms this function. It uses a centralized account system for the requisite checks,
while Bitcoin uses decentralization and real-time technology. Each computer in the
network verifies new transactions by referring to the entire “transaction chain” of
those previously completed. Only transactions checked and confirmed against this
chain become official. In fact, each new transaction is responsible for addition to
the end of the transaction chain, and there is only one valid transaction chain.Thus,
although it is on every computer (decentralized), it is in actuality a system based on
one central confirmation (Figure 5.3).
Bitcoin requires complex calculation for a node to add a transaction to the chain
due to failed transactions, fraud, and other risks. The calculation requires expendi-
ture of a great deal of computing power, which serves as proof of work performed.
The incentive to perform this work comes in the form of “Bitcoin mining”, the
way in which new Bitcoins are created and the payment platform continues to
function.
Bitcoin’s reliance on the Internet allows it to easily cross borders. Its unlimited
nodes allow it to easily withstand attacks on any individual or group of nodes.
Users are completely anonymous and can possess any number of accounts. Bitcoin
is also cheap for users, as the transfer fees and exchange charges are extremely low.

Node
Person

Bitcoin
transaction Person
(see Figure 5.2)
Node

Bitcoin
transaction
(see Figure 5.2)

Person

Figure 5.3 The full Bitcoin transaction chain.

 Internet currency 63
One rare characteristic is the irreversibility of transfers.There is no way to cancel or
reverse a transfer that has already been confirmed except through instituting a new
transfer in the opposite direction. It is both open and transparent, and all transaction
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

data are available on the Internet. Finally, it is divisible up to eight decimal places,
or 0.00000001 Bitcoin.
Bitcoin is, however, not perfect. Some believe the massive expenditure of com-
puting power to create new transactions is a waste. Microsoft’s Moshe Babaioff and
collaborators published research point out a potential “red balloon effect” in Bitcoin
mining in which the decentralized network do not contribute to the transmis-
sion of transaction information.8 For example, some Bitcoin miners could change
their distribution codes so that other nodes are unable to receive new transaction
information.

2 Issuance mechanism
The payment terminals that successfully confirm transactions receive newly minted
Bitcoins.The system therefore does not need a central bank. At the beginning, each
completed transaction link creates 50 Bitcoins, but this number will decline by 50%
every four years until the total Bitcoins in circulation maxes out (expected in 2040
at 210 million total Bitcoins, see Figure 5.4). This create scarcity, which in turn
could very well lead to a deflationary tendency that manifests itself in declining
goods prices in terms of Bitcoins.This increase in Bitcoin value could cause hoard-
ing that reduces liquidity and further increases deflationary pressure.This deflation-
ary tendency becomes a barrier to financial products based on Bitcoin prices due to
the continually increasing real debt repayment burden. All currencies, even Bitcoin,
need price stability. Although Bitcoin has dealt with the inflationary tendency of
other currencies through a fixed issuance schedule, its deflationary tendency will
encumber its future growth. However, it remains possible to change the code. For
example, the issuance could be changed to base itself on a certain level of inflation.
We recommend a report by the Paolo Alto Research Center9 to readers particularly
interested in this subject.
Another problem is the diminishing returns on Bitcoin mining. In the early days
of Bitcoin, individual “miners” with personal computers could receive Bitcoins.
Today, the computing power required to harvest the same number of Bitcoins is
extremely high, so many computers have to work together and split the winnings,
giving rise to a “mining pool” (see Figure 5.5).
The “mining pool” distribution implies a somewhat monopolistic pay-
ment mechanism. If there is a limited number of effective payment nodes, then
Bitcoin supporters’ belief in its democratic purity declines along with the distinc-
tion between Bitcoin and money issued from a central bank. The biggest ques-
tion is how to keep the mechanism afloat once Bitcoin miners cannot receive any
more Bitcoins for their efforts. Bitcoin’s founder has suggested a transaction fee to
maintain the system, and the Bitcoin Association has created a plan to implement
it. Transactors would pay the confirming nodes to confirm their transactions in
 64 Internet currency
21

18
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

15

12

Current volume of issuance

9

0
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
Figure 5.4 Bitcoin’s volume of issuance (by the end of January 2014, in millions)

this model. However, these are all in the planning stages. The current system does
not involve nodes that collect confirmation fees.

3 Bitcoin’s effect
Bitcoin has already run up an impressive transaction volume and is convertible
into many currencies at rates that fluctuate freely with demand. Mt. Gox is the
world’s largest Bitcoin exchange. It is responsible for over 80% of Bitcoin exchanges
and publishes daily exchange rates between Bitcoin and major currencies. Other
Internet sites such as Bitcoin.local allow accountholders to directly exchange with
each other.
Bitcoin’s price is famously unstable. Both user numbers and prices increased
dramatically from June to July 2011. The price started below $1, then quickly
rocketed up to $30. Prices have fluctuated up ever since. Figure 5.6 indicates the
staggering price volatility in 2013, reaching $1,200 and then receding. Although
Bitcoin has had a relatively small footprint, some service, software, and clothing
companies accept Bitcoins. Some questionable uses have also emerged out of its
anonymity. WikiLeaks began to accept donations in Bitcoins when it was cut
off from traditional payments. Some drug dealers on the “Silk Road” Online
Marketplace accept payment only in Bitcoins. Nicholas Christin of Carnegie
Mellon University estimated that this website generated monthly transactions of
 Internet currency 65

Person

Node Person
Person
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

The
“Mining pool”
node
Person Person
Node
Person
Person

Person
Node

Figure 5.5 The “mining pool” distribution mode.

Mt. Gox (USD)

Jan 18, 2014 - Daily
1300
Weighted Close: 905.4
1200
1100
1000
900
800
700
600
500
400
300
200
100
0
Feb 13 Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan 14
(CC) BY-SA
This chart is licensed under a Creative Common Attribution-ShareAlike 3.0 Unported License.

Figure 5.6 Bitcoin’s price volatility in 2013.

$1.2 million in 2012.10 In May 2013, the FBI closed the site and took possession
of its Bitcoins.
In a systematic analysis of Bitcoin, Ron and Shamir discovered that Bitcoins are
mostly untraded and highly concentrated.11 Most existing Bitcoins have not moved
from their initial accounts, an indication that they have been either lost12 or hoarded.
90% of accounts have made less than ten transactions and possess very few Bitcoins,
an indication that Bitcoin can hardly be considered a lively marketplace.The break-
down in Table 5.1 shows that 97% of accounts possess 10 Bitcoins or less.
Table 5.1 confirms Bitcoin’s scarcity and tendency to both deflation and appre-
ciation expectations. Bitcoin holders tend to remain just that, holders who do not
use them. This cannot be good for Bitcoin’s popularization.

4 Innovation and insufficiency

Bitcoin sparked a deeper discussion about the nature of money. We believe it is
a fitting reference point for questions about innovation and the future of money,
 66 Internet currency
Table 5.1 Balance on Bitcoin accounts13

Greater or equal to Lesser than Number of users Accounts

Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

0 0.01 2,097,245 3,399,539

0.01 0.1 192,931 152,890

0.1 10 95,396 101,186

10 100 67,579 68,907

100 1,000 6,746 6,778

1,000 10,000 841 848

10,000 50,000 71 65

50,000 100,000 5 3

100,000 200,000 1 1

200,000 400,000 1 1

400,000 0 0

including a possible world currency. Bitcoin’s main innovation is its independence

from the banking system, which allows even international payments to be both
direct and cheap. It could play a leading role in a future with many forms of money.
It has even inspired J.P. Morgan to apply to create its own digital currency.
The fixed issuance schedule has inspired new thinking on an issue that has
plagued monetary policy since the collapse of the gold standard and transition to
fiat money: inflation’s wealth destruction. Mainstream ideas focus on central bank
independence and monetary policy transparency. Bitcoin’s idea is something new.
It does not require a central bank and separates the payment and money issuance
functions in favor of a rules-based issuance.
However, a lack of stable purchasing power is holding Bitcoin bank from being
a completely developed currency. There is neither an intrinsic value (gold standard)
nor an issuing body (fiat money) to ensure this.This instability, combined with hold-
ers’ high concentration and hoarding behavior, mean Bitcoin badly needs a redesign.

5 Risk and regulation

In a fiat money system, base money includes bank reserve requirements and money
in circulation. Money and credit are inextricably linked through credit expansion,
which both creates money and causes deposits to proliferate. Bitcoin decouples
credit and money, it is not a fiat currency. A look at money creation under the gold
standard, with Bitcoin mining as a replacement for gold mining, gives us a better
 Internet currency 67
idea of why Bitcoin behaves as it does. The total amount of both gold and Bitcoins
is fixed, and the costs to extract them increase as the easiest accessible stocks are
exhausted.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Fiat money is priced based on faith in the issuer and total issuance. That of
metal-backed currency is supported by its intrinsic value (use in jewelry and other
products) and stabilized by the tendency for gold to exit circulation when prices
decline and enter when prices rise. Bitcoin’s price combines elements of both. It
could decline if faith in its mechanism declines due to worries of hacker attacks,
loopholes, Bitcoin seizures, or counterfeit coins. The cost of computing power that
miners expend could also influence the price. Just like gold, it is influenced by the
price of fiat currencies in which the prices are expressed.14
The February 2014 hacker attack on the Mt. Gox Bitcoin exchange demon-
strates the risks associated with Bitcoin. The Bitcoin losses were so staggering that
the exchange applied for bankruptcy, and the price of Bitcoins fell 50% in a single
day.The emergence of similar, competing currencies such as Litecoin, Peercoin, and
Primecoin all pose risks to Bitcoin’s value.
Bitcoin currently resides in a legal gray area. Should Bitcoins be classified and
protected as wealth? Does it violate current laws? How should policymakers and
regulators deal with Bitcoin? How can we adjust current laws to deal with Internet
currencies? Many users doubt that Bitcoin is ready for an expanded role due to
this uncertainty. Its anonymity facilitates money laundering along with drug and
weapon sales. This must be changed. Insufficient regulation has inhibited tax col-
lection on Bitcoins and focused attention on a Bitcoin transaction tax. For example,
the United Kingdom is considering classification of Bitcoin as a money replace-
ment, which will allow it to impose a 20% value added tax.15 Norway, Germany, and
Singapore have already provided for Bitcoin taxes.
How does Bitcoin fit into the existing US legal framework? In a comprehensive
review published in 2011, Ruben Grinberg, now a lawyer at Davis Polk, argued
that the Stamp Payments and federal counterfeiting statutes would need substantive
revision to apply to the issuance of Bitcoin.16 They also do not quite fit the many
definitions of securities under securities laws. He did, however, point out significant
risk for the Bitcoin community in relation to money laundering laws, as Bitcoin
does not fulfill the registration requirements for money services and it is, “gener-
ally known that Bitcoin is used to promote illegal activities.” Nikolei Kaplanov
argued strongly against the regulation of Bitcoin in a paper published by the Temple
University Law Review.17 He compared Bitcoin with existing community curren-
cies and questioned the legal basis for the federal government, including the Federal
Reserve, to regulate the use of Bitcoins for transactions between willing parties.
These admonitions appear not to have convinced the federal government.
Bitcoin’s greatest risk is uncertainty regarding its regulation. Regulators around
the world are paying ever-increasing attention to Bitcoin’s development. In August
2013, a federal judge officially ruled that Bitcoin is subject to regulation accord-
ing to American Securities Laws. European officials are issuing warnings about the
threat it and other virtual currencies pose to the central bank’s ability to safeguard
 68 Internet currency
fiscal and monetary stability. Russia, India, and Hong Kong have all issued warn-
ings and taken various steps such as investigation and temporary closure of Bitcoin
trading platforms (India) and bans on domestic transactions parallel with the Ruble
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

(Russia). However, these pale in comparison with the strict response from China.
In December 2013, the People’s Bank of China (PBoC) issued a directive banning
Bitcoin’s use as a liquid currency.18 The ban extended to financial institutions, pay-
ments companies, and retail businesses, which were forbidden from processing or
accepting payments in Bitcoins.
America’s attitude toward Bitcoin is still developing. Former Federal Reserve
Chairman Ben Bernanke expressed hope for Bitcoin’s ability to overcome legal
and supervisory questions to create a more effective international payments system
over the long term. On the state level, California’s House of Representatives passed
a bill permitting use of any alternative currency with monetary value in payments
and retail transactions in the state.19 It would put Bitcoin on the road to full legality
in California, but the bill still needs to pass through the state senate and receive the
governor’s approval to become law.
Based on this analysis, we have two recommendations. The first is that China
delays its clarification of Bitcoin’s legal role until the technical and market founda-
tions are fully developed. The second is to research ways to effectively regulate it
so as to protect consumers and prevent money laundering, speculation, and price
manipulation. The circulation of Internet currency indicates that it could one day
become an able competitor with central bank-issued legal money, especially if an
online currency emerges whose issuance is rules-based rather than predetermined
like Bitcoin. It could sensitively adjust issuance to take economic risk into account
rather than price and quantity, thus avoiding the deflationary pitfalls of Bitcoin and
the myriad issues with central bank-issued currency.There is a great deal of work to
do, but we see a bright future with many forms of currency on the horizon.

Notes
1 European Central Bank. 2012. “Virtual Currency Schemes.”
2 Nakamoto, Satoshi. 2008, “Bitcoin: A Peer-to-Peer Electronic Cash System.”
3 Blockchain.info. 2013. http://blockchain.info/charts/total-bitcoins.
4 Mt. Gox. 2012. www.mtgox.com/.
5 Gross Domestic Product. 2013. World Bank.
6 Sprankel, Simon. 2013. “Technical Basis of Digital Currencies,” Technische Universitaet
Darmstadt.
7 Source: http://weidai.com/bmoney.txt
8 Babaioff, Moshe, Shahar Dobzinski, Sigal Oren, and Aviv Zohar. 2012. “On Bitcoin and
Red Balloons.” “Red Balloons” refers to a 2009 DARPA experiment in which par-
ticipants were to locate ten red balloons placed randomly across the United States. The
winner received $40,000. The core of the Red Balloon Phenomena is cooperation and
competition.
9 Barber, Simon, Xavier Boyen, Elaisn Shi, and Ersin Uzun. 2012. “Bitter to Better—How
to Make Bitcoin a Better Currency.”
10 Christin, Nicolas. 2012. “Travelling the Silk Road: A Measurement Analysis of the Silk
Road Anonymous Marketplace.”
 Internet currency 69
11 Ron, Dorit and Adi Shamir. 2013. “Quantitative Analysis of the Full Bitcoin Transaction
Graph.”
12 Bitcoins can be linked to a specific piece of computer hardware. If that hardware breaks
or is lost, it may be impossible to access its linked Bitcoins.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

13 Up to May 3rd 2012.

14 If the US Dollar depreciates (appreciates), Bitcoin and Gold as expressed in dollars appre-
ciate (depreciate).
15 Chaturvedi, Neeblah. “U.K. Weighs How to Tax Dealings in Bitcoin.” 21 January 2014.
Wall Street Journal. http://online.wsj.com/news/articles/SB1000142405270230430270
4579334790053233278
16 Grinberg, Reuben. 2011. “Bitcoin: An Innovative Alternative Digital Currency.”
17 Kaplanov, Nikolei. 2012. “Nerdy Money: Bitcoin, the Private Digital Currency, and the
Case Against its Regulation.” Temple Law Review.
18 “Notification on containing the risks of Bitcoin.” 5 December 2013, People’s Bank of
China www.pbc.gov.cn/publish/goutongjiaoliu/524/2013/20131205153156832222251/
20131205153156832222251_.html.
19 California House of Representatives, AB129.
 6 Big data
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Big data is a new concept that also goes by Big Scale Data or Massive Data. So far,
there exists no universal definition for the concept. However, it is well understood
that the so-called Big data has four features, that is 4Vs: large Volume, low in Value
density, Variety in sources and features, and rapid Velocity.
The concept has emerged from society’s digitization process (see Chapter 1),
especially with the development of online social networks and sensory devices.
Updated cloud computing and search engines provide efficient ways to analyze big
data.Yet the core lies in how to quickly identify valuable information from the wide
range and massive data sources. Big data is playing an increasingly significant role
in social analysis, scientific discovery, and business decision making. Finance is only
one example of such an application.
Today Big data has become a hot topic. Some believe that it is an important or
even strategic component of a nation’s resources, just like oil and gas. We will not
discuss each idea in detail. Rather, we believe that most of the discussion is theo-
retical. Since it does not explain the mechanism of big data analysis technically, we
aim to fill this gap.
Based on the findings of data science and data mining literatures, we start by
introducing the main types of big data, and then discuss the goals for big data analy-
sis. In the end, we will compare it to econometrics. Chapter 7 will focus on the
big-data-based credit investigation and Internet lending. Chapter 12 will discuss the
application of big data in securities investment and actuarial science.

SECTION 1: BIG DATA: CONCEPTS AND MAIN TYPES

The basic unit of big data is a data set. First we will introduce two relevant concepts
for data sets: data objects and properties, and discuss the three different types of data
set: recorded data, graphic-based data, and ordered data.

1 Basic concept
A data set is a collection of data objects. Accordingly, the data objects are its
components. In statistics, data objects are equal to statistical units or sample points
 Big data 71
and are sometimes called records, points, vectors, models, events, cases, samples,
observations, or objects. The data objects are described via a set of properties.
The number of the data objects’ properties is called the “dimension.” A common
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

problem in the analysis of high dimensional data is the so-called curse of dimen-
sionality. The difference between the number of data objects and their dimensions
is the degrees of freedom. As the degrees of freedom decrease, some statistical meth-
ods will become inapplicable and the results will be less reliable. One way to deal
with this is dimensionality reduction, that is combining original dimensions into
a new one to reduce the number of dimensions. Usually, linear algebra methods
such as principal component analysis and singular value decomposition are used to
project high-dimensional data sets into lower ones.
In some data sets, object values in most properties are zero, and in most cases,
the nonzero terms account for less than 1%. This is known as sparseness. There are
certain ways to deal with the sparseness. Properties are equal to the variables in sta-
tistics, which stand for the data objects’ features or characters and vary from object
to object or from event to event. Properties are also called characters, fields, features,
or dimensions. The rules that connect numbers or symbols with the objects’ prop-
erties are the measurement scale. For example, GPA is one property of students, and
identifying a certain student’s score is actual measuring.
Properties can be divided into four categories based on their applicable
calculation types (i.e. comparison, order, addition and subtraction, multiplication
and division). The first category is a nominal property, of which the values are set
as different labels that provide sufficient information to identify each data objective.
For example, different IDs, genders, and so on. belong to this category. The second
category is the ordinal property, of which the values give information to order dif-
ferent data objects. House numbers belongs to this category.The third is the interval
property. The interval property has measurement units with meaningful differences
in properties. One example of this category is temperature. The fourth category
is the ratio property, of which the difference and ratio are both meaningful. For
instance, length and weight belong to this category. Of the four categories, nominal
and ordinal properties are qualitative properties while interval and ratio ones are
quantitative properties.
The possible number of values of each property can be divided into two catego-
ries. The first is a discrete property with limited or infinite numbers of values. The
second is the continuous property with values from the whole set of real numbers.
Usually, nominal and ordinal properties are discrete and interval and ratio properties
are continuous.

2 Recorded data
Recorded data is a collection of records (i.e. data objects) and each of them contains
fixed data fields (i.e. properties). For most basic forms of recorded data, the records
have no clear relevance to the fields. The recorded data is usually stored in flattened
files or relational databases. It is structural data and can be presented as a data matrix.
There are two conditions for this.
 72 Big data
Condition I: If all data objects have the same data property set, these data objects
can be regarded as points (or vectors) in multi-dimensional space. Each dimension
represents a different property. Cross-sectional data used in econometrics can there-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

fore be regarded as a data matrix, whose analysis method is represented by matrix

signals and computing.
Condition II: After it is transferred, some recorded data can be represented by
a data matrix, such as file data. If we ignore the order of words that appear in a
file, we can use word vectors to represent the file, of which each word is a com-
ponent of a vector (property). The value for each component corresponds to the
times of a word appears in the file. Data matrices based on this logic are called
document-term matrices. Transaction and market basket data are other examples
of this type. In a supermarket, the products bought by one customer at one time
can be regarded as “one transaction,” within which each commodity is one term.
Each customer buys different products, and each product type is of a different
amount. Still, the complete market basket can be represented by commodity vec-
tors, each of which is a component of the vector (property), and each vector’s
value corresponds to the market basket’s commodity type, commodity amount,
and expense. It is important to note that the data matrix under the second condi-
tion tends to have “sparseness.”

3 Data based on graphs

Data with relationships among data objects. The relationship among data
objects tends to carry important information. For example, website pages include
text and links that refer to other web pages.These links carry information about the
importance of this web page. Google’s Page Rank computing method is based on
link analysis, which we explore in detail below (see Link analysis in Section 2). We
can use graphs to represent this kind of data. Generally, data objects will produce
mappings on the graph’s nodes, and the relationship among data objects can be
represented by the direction and weight of links between different nodes.
Data with graph objects. If data objects have structure (i.e. objects include
sub-objects), they are usually represented by graphs. For example, the structure of a
compound can be represented by graphs, among which the atom is the nodes and
the chemical link is the link between different nodes.

4 Data in order
Sequential data is an expansion in recorded data in which each record includes
time-related information. For example, a supermarket cashier records the market
basket information for each customer. The recorded information is sequential data.
Also, all the time series data (such as macroeconomic series and financial price series
data) belong to sequential data.
Sequence data is the sequence of each entity. It is quite similar to time series,
except that sequence data do not have a timestamp. Typical sequence data includes
genome, word, or letter sequences.
 Big data 73
Spatial data can also be regarded as an expansion in recorded data, only that
each record includes the space of the area feature related to this data. Typical
spatial data includes meteorological data collected from different geographic posi-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

tions and different points in time, such as temperature, humidity and atmospheric
pressure.

SECTION 2: PRIMARY TASKS FOR BIG DATA ANALYSIS

Big data analysis is primarily used for prediction or description. For predic-
tion, it uses a set of independent variables to estimate or predict the value of
the dependent variable. Description summarizes potential correlations, trends,
clustering, and abnormal trajectories in the data. Descriptive tasks are usually
exploratory and often need post-processing to verify and interpret results. These
two types of tasks can be divided further into classification, regression, associa-
tion analysis, cluster analysis, recommender systems, anomaly detection, and link
analysis.

1 Classification
Classification aims to place data objects into a pre-defined target class. Typically,
these target classes are mutually exclusive. For example, credit rating agencies use
issuer data to place debt into categories based on creditworthiness from AAA (the
best rating) to D (default).
The input data is a collection of records. Each record is represented by the
tuple, (x, y), in which x is a collection of attributes and y is a special attribute that
indicates the class label record (also called categorical attributes or target attrib-
ute). The core of classification is to determine a target function by learning and
map each attribute set x to a pre-defined class number y. Sometimes, instead of
determining the set of attributes mapped to a certain category number y, x may be
mapped to all the target classes as long as it is subject only to a certain probability
distribution.
The general classification task has two steps.The first step requires a training test
consisting of records in which the label is known. It can develop into classifica-
tion models, including the Logit model, Probit model, decision tree classification,
rule-based classification, artificial neural networks, vector machines, and Bayesian
supported classification method (part of this model will be described in detail
in Chapter 7). Next, the classification model is applied to a test set consisting of
records in which the label is unknown. We then evaluate the classification model’s
performance by seeing which records were correctly classified.
Confusion matrices are often used in these problems. Table 6.1 is a typical con-
fusion matrix for a binary classification problem. There are positive and negative
types (a positive class usually represents a rare category while negative classes repre-
sents the majority); in each table entry, f ij indicates its actual class is i but is predicted
to be class j. The table contains four specific conditions:
 74 Big data
Table 6.1 Confusion matrix

Predicted Class
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Positive Negative

Actual Class Positive f + +(TP) f + −(FN)

Negative f − +(FP) f − −(TN)

• True positive (TP) or f + +: number of positive samples of correctly predicted

to be positive;
• False negative (FN) or f + −: number of positive samples erroneously predicted
to be negative;
• False positive (FP) or f − +: number of negative samples erroneously predicted
to be positive;
• True negative (TN) or f − −: number of negative samples correctly predicted to
be negative.

According to the confusion matrix entries, the total number of correctly

predicted samples by classification model equals to f + + + f − − and that of errone-
ously predicted ones equals to f + − + f − +. Accuracy and error rates are two common
indexes of the classification model’s performance. The accuracy rate is equal to the
f++ + f−−
proportion of correct predictions, , and the error rate is the
f++ + f+− + f−+ + f−−
f+− + f−+
proportion of wrong predictions, .
f++ + f+− + f−+ + f−−
Another common tool to measure the performance of classification models
is the receiver operating characteristic (ROC) curve based on the true and false
positive rates (FPRs). The true positive rate (TPR) is defined as the proportion of
TP
correctly identified positive samples, TPR = , and FPR is the proportion
TP + FN
FP
of negative samples predicted to be positive by the model, FPR = .
TN + FP
ROC curves can be applied to classification models that produce output values
continuously (such as Logit models, Probit models, artificial neural networks, and
Bayesian classifications). These output values can be used to sort orders into cat-
egories from the most likely to be positive to the least likely, specifically in the
following five steps:
Step 1 (assuming a continuous value output for the positive class) sorts samples
by their output value. The higher the output value, the more likely the sample is
to be positive.
Step 2 selects the sample with minimum output value and defines others as
positive. This makes centralized test samples positive. If all positive samples have
 Big data 75
been correctly classified, then all negative ones must be wrongly classified, so
TPR = FPR = 1.
Step 3 selects the next sample from the sorting list and defines it and those with
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

higher output values as positive. Samples with lower output values are negative.
Recalculate TPR and FPR.
Step 4 repeats the third step; the test selects the sample with the highest output
value.
Step 5 connects all ( FPR ,TPR ) in turn and get the ROC curve (Figure 6.1).
Two classification models’ ROC curves are rather special. One is perfect clas-
sification TPR = 1, FPR = 0. Its ROC curve is the horizontal line at the top of
Figure 6.1. The other is random classification, that is divide samples randomly into
positive category according to a fixed probability and get a diagonal ROC curve.
Other models’ ROC curves are between that of perfect classification and random
classification. Models with curves closer to the upper left corner curves perform
better. Thus, area under curve (AUC) also functions as a performance metric.

2 Regression
Regression is similar to classification. The key difference is that in regression, the
target attribute y is continuous. In classification, however, the target attribute y
is discrete. In other words, classification predicts whether something will happen,
while regression is to predict how much.
Regression is the most commonly used tool in econometric analysis, especially
the linear regression model:

y = α + β1x1 + β 2 x 2 + … + β p x p + ε (6.1)

1
0.9
0.8
0.7
True positive rate

0.6
0.5
Perfect classification
0.4
Classification models
0.3
0.2 Random classification

0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
False positive rate

Figure 6.1 ROC curve.

 76 Big data
Here, y is the dependent variable. x1 to x p are independent variables. β1 to β p are
parameters to be estimated. ε represents random disturbance.
Mean square error (MSE) is a prime measurement of a regression model’s
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

performance. Suppose there are n samples and the i sample’s dependent variable
value is yi which is predicted to be yˆi by the regression model. Then the MSE is:
1 n
∑ ( yi − yˆi )
2
MSE = (6.2)
n i =1
In the linear regression model, the least squares method (LSM) gets parameter
estimates by solving the MSE minimization problem.

3 Association analysis
Association analysis focuses on market basket data.The goal is to find hidden mean-
ingful relations in big data sets. One popular (and humorous) example is diapers and
beer. Retailers found that many customers buy diapers and beer at the same time, so
they placed diapers and beer together in stores to promote cross-selling. Association
analysis tries to find a link rather than an explanation. Here, we face two essential
issues: first, the calculation cost may be too high; second, some correlations may
occur by accident. The core of the association analysis algorithm is to deal with
these problems, expressed as follows:
I = {i1 , i2 ,… , id } stands for all the items in the market basket data, and
T = {t1 , t 2 ,… , t N } represents all matters. Every item contained by matter ti is subset
of I. For item set X , its support is:

σ ( X ) = # {ti X ⊆ ti , ti ∈ T } (6.3)

In it, # represents the number of elements in the collection.

Association rules are expressed as an implication expression like X → Y in
which X and Y are disjointed item sets X ∩ Y = ∅. Support and confidence show
the strength of association rules.
Support depicts the frequency of X and Y occurring together (i.e. X ∪ Y ) in T:
σ (X ∪ Y )
s (X → Y ) = (6.4)
#T
Occasional rules have low support. From business perspectives, low support rules
are mostly meaningless. Therefore, support is used to delete those pointless rules.
Confidence depicts the frequency of Y in matters that contain X and reflects
reasoning by the rule:
σ (X ∪ Y )
c (X → Y ) = (6.5)
σ (X )
The association rule’s mining problem is for a given set of matters T to identify
all rules whose supports are greater than or equal to minsup and confidences are
 Big data 77
greater than or equal to minconf, where minsup and minconf are the thresholds of
support and confidence. Here, we usually complete two steps:
First, generate frequent item sets to find all item sets that meet the minimum
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

support threshold;
Second, produce rules with high confidence from the previous step, which are
then called strong rules.
Typically, the amount of calculation to generate frequent item sets is much larger
than that of desired rules. Generating item sets frequently uses either the Apriori or
FP growth algorithm.

4 Cluster analysis
Cluster analysis divides data objects into groups based on descriptive objects and
their relationship information in the data. Objects are related within a group and
uncorrelated among different groups. The greater the homogeneity within the
group, the greater the gap between the two groups, which then improves the clus-
tering effect.
Cluster analysis and classification are similar. Clustering classes (clusters) can also
be considered as a data object classification, but this classification information can
only be derived from the data. In classification, however, the class labels are known
when the development of the object model is done. We then assign numbers to
new and unmarked objects. Thus, classification is also called supervised classifica-
tion, and clustering analysis is called unsupervised classification.
Common clustering algorithms include K-means, agglomerative hierarchical
clustering, and DBSCAN. They can be hierarchical/agglomerative or use point
assignment. The first type of algorithm regards each data object as a cluster at the
beginning and combines clusters according to their closeness, which can be defined
in many different ways. When the further combination leads to pre-defined unde-
sired results, the above combination process is ended. For example, combination can
stop when it reaches a pre-given number of clusters. The second category involves
point assignment process, which means considering each data object in a certain
order and assigning it to the most appropriate cluster. The process typically has a
short initial phase to estimate clusters. Some algorithms allow merging or splitting
for temporary clusters. They may also identify outliers when the data object may
not be assigned to any cluster.

5 Recommender system
The recommender system predicts user preferences. For example, an online news-
paper provides news reports based on the prediction of user interest, and online
retailers recommend products to customers that they might want to buy on the
basis of their shopping and/or goods search history.
There are two types of elements in recommender systems: users and items. User
preference for some items can be expressed as utility. Suppose there are N items
 78 Big data
and R (u, i ) represents user u′ s utility from item i. Since it is impossible to observe
R (u, i ), the recommender system must work this out and get prediction Rˆ (u, i ).
After obtaining N predicted utility values Rˆ (u, i1 ) , Rˆ (u, i2 ) ,… , Rˆ (u, iN ), the recom-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

mender system will first recommend K (usually much smaller than N) items on the
principle of utility maximization.
One type of recommender system is content-based. Such recommender sys-
tems use customers’ past preference to recommend similar ones. Similarity between
items is determined by calculating the similarity between attributes. Amazon uses
this to recommend books similar to the one currently viewed.
Another is collaborative filtering, which recommends items according to other
customers’ choice who have similar taste. The similarity between users depends on
their browsing and scoring records. For example, Dangdang’s “those bought this
item also bought” feature is a type of collaborative filtering. Collaborative filtering
is the most popular and also the most widely applied of all types.
Recommendations can also base themselves on demographic characteristics.
Such recommender systems recommend appropriate items based on users’ demo-
graphic characteristics (such as age, language, and country).
Knowledge-based systems make out yet another category. Such systems use
domain-specific knowledge to judge whether an item meets users’ needs, prefer-
ences, and practicability. The core is a similarity function used to measure users’
needs (i.e. description of the problem) and match degree of recommendations
(i.e. answers).
Community-based systems rely on the user’s friends. The basis for such recom-
mender systems is that people tend to accept the recommendation of friends. The
popularity of social networks has also promoted its development.
Hybrid recommender systems mix the above systems together. For example,
content-based recommender systems can make up for the new-item problems of
collaborative filtering.

6 Anomaly detection
Anomaly detection identifies characteristics that are significantly different from
other observed values.The abnormal observation value is called an anomaly or out-
lier. Anomaly detection tries to find real outliers and avoid false labeling. In statistics
parlance, a good anomaly detector must have a high detection rate and low false
alarm rate. Anomaly detection applications include fraud detection (such as credit
card fraud detection), network attacks, unusual disease patterns, and ecosystem
disturbance.
There are three main sources of outliers. First, data may come from different
classes. The statistician Douglas Hawkins defines outliers as observations with
differences so large that it seems impossible they come from the same mechanism.
For example, people who use credit cards fraudulently and legally are two different
types. The second source is natural variation. Many data sets can be modeled with
a statistical distribution (e.g. normal distribution) in which the probability of an
observation appearing decreases as it departs from the mean.The third source is data
 Big data 79
measurement and collection errors. Such anomalies reduce the quality of the data
analysis and should be thrown out.
There are a variety of ways to detect abnormalities. Many anomaly detection
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

techniques first create a model without points that are not perfectly fit. For
example, the data distribution model can be created by the parameter estima-
tion of probability distributions. If an object does not fit well with the model
(that is, not subject to the distribution), it is an outlier. If the model is a set of
clusters, then the outlier is not the object of any significant clusters. When using
the regression model, anomalies are relatively far from the predicted value of the
object.
Proximity-based techniques, on the other hand, measure the distance between
objects. Outliers are those objects furthest away from most of the other objects.
When the data can be displayed in a two-dimensional or three-dimensional scat-
ter chart, outliers can be detected visually by finding points separate from the
majority.
We also have techniques based on density. The relative density of an object can
be estimated directly, especially when there is already a proximity measure between
objects. Low-density areas relatively far from the nearest neighbor objects may be
seen as outliers. A more sophisticated approach is to consider that the data set may
have different density areas. Only if the local density of a point is significantly lower
than most of its neighbors can they be classified as outliers.

7 Link analysis
Google’s PageRank is representative of the link analysis algorithm. Before Google,
there were many search engines, most of which used a web crawler to obtain data
from the Internet, and then list the lexical items of each page with an inverted
index. When a user submits a search query, all pages containing these words are
extracted from its inverted index and listed in a way that that reflects inside items.
Therefore, words appearing on the page header items have a higher correlation
than lexical items that appear in text, and the more occurrences of lexical items, the
more relevant the web page. In this case, “term spam appeared” in large numbers
and some unrelated Web pages modified themselves (such as a large number of
duplicate keywords) to deceive search engines.To solve lexical item falsification, the
PageRank algorithm has developed two innovations.
First, it simulates the behavior of Internet surfers. These imaginary surfers start
from random web pages and pick a next destination link randomly selected from
current page. This multi-step process can be iterated. Ultimately, these surfers will
converge on a page. Pages with more visitors become more important than those
of few visits. PageRank measures this “importance” of web pages. Google lists pages
according to PageRank when answering queries.
Second, in determining the content of the page, both lexical items that appear
on the page and word entry point links to the page should be considered. The
implicit assumption is that web owners tend to link to good or useful web pages
instead of bad or useless ones. Even if spammers can easily add false lexical items on
 80 Big data
pages under their control, they can hardly do the same thing on pages referring to
theirs. Link spam has been developed to work against link analysis, which builds a
webpage collection that is called a “junk farm” to increase the PageRank of certain
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

pages. Correspondingly, some anti-spam methods appeared such as TrustRank and

“garbage quality.”
In this way, the PageRank algorithm actually uses the Internet as a directed
graph where pages are graph nodes: two can be connected with a direct edge if
there is a link between them. The PageRank algorithm simulates the behavior of
Internet surfers and gives transition probability meaning to edges. For example,
suppose a surfer is currently on page A, while page A has three links pointing to
page B, C, and D. It is believed that the surfers’ probability to visit B, C, or D is
equally 1/3, but staying on A has zero probability.Thus, a Markov process can char-
acterize surfers’ behavior on the Internet.
Assume webpages denoted by {1, 2,… , N } are in the state space of a Markov
process. X t represents surfers’ location at the moment of t and is a random variable
ranging from {1, 2,… , N }. The sequence of random variables { X t , t = 0,1, 2,…}
is a random process. In the PageRank algorithm, this random process satis-
fies Markov properties (visual description is given now, past and future are not
relevant):

∀t, n ( )
f X t + 1 X t , X t − 1, … , X t − n = f ( X t + 1 X t ) (6.6)

In Equation (6.6), f ( X t +1 ⋅) indicates X t +1’s conditional distribution.

Thus, {X t , t = 0,1,2,...} is a Markov process and its dynamic change can be
characterized by a transition matrix. Use N × N matrix P to represent the transi-
tional probability matrix, then pij in line i and column j means:

pij = Pr ( X t +1 = j X t = i ) (6.7)

Using a N × 1 matrix V t to describe surfers’ locational distribution on the

Internet at moment of t , the i component vit means:

vit = Pr ( X t = i ) (6.8)

So, at the time t + 1, surfers’ locational distribution V t +1 satisfies:

V t +1 = P ⋅ V t (6.9)

If the Internet is a strongly connected corresponding digraph, that is starting

from any node we can reach all other nodes and there is no end point (i.e. the
node does not exist in the chain), then regardless of the surfers’ initial locational
distribution on the Internet, after long enough, it will be close to a stationary
distribution π 1. The rigorous formulation is:

∀V 0 ,limV t = π (6.10)
t →∞
 Big data 81
The steady-state distribution π satisfies π = P ⋅ π (i.e. starting from the steady
state, the next moment is still a steady-state), so π is actually an eigenvector of
Matrix P with corresponding eigenvalue 1.
π i is the i component of π in the steady state that shows surfers’ probability to
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

stay on page i and PageRank value of page i.

In reality, the corresponding figure of the Internet generally does not have strong
connectivity features. For example, there may be no termination point for the chain
or there may be a set of pages that have a chain, but the chain does not point to
other web pages outside of this group. The PageRank algorithm solves these prob-
lems by modifying the transition probability matrix, such as “taxed” law, which will
not been discussed here.

SECTION 3: COMPARISON OF BIG DATA ANALYSIS

AND ECONOMETRICS

1 Introduction to econometrics2
Econometrics is the dominating quantitative empirical analysis used in the economic
and financial fields. Econometrics adds empirical content to economic theory,
which allows theories to be tested and used for solving real problems, especially
for forecasting and policy evaluation of government and enterprises. Econometrics
studies random economic relations by dealing with data from real observations,
usually without artificial control, with mathematic methods to build models and
reveal quantitative relations for an objective economic system.
Econometric analysis is generally based on four steps: variables and data selec-
tion, model specification, model fitting and testing, and applications. It should be
noted that model specification and testing are closely connected in real studies.

1.1 Variables and data selection

First, model variables are based on research purpose, theory, and empirical studies.
There are two kinds of variable: dependent and independent. These variables must
be observable and quantifiable.
Secondly, we select samples to assign data to each variable that has been chosen.
There are three main types of data: cross-sectional data that refers to observations
of many different individuals (subjects, objects) at a given time (each observation
belonging to a different individual), time series data that refers to a sequence of
observations which are ordered in time or space (each observation belonging to
the same individual), and panel data that refers to multi-dimensional data frequently
involving measurements over time. Panel data contains observations of multiple
phenomena obtained over multiple time periods for the same firms or individuals.
The data sample has a direct impact on an econometric model’s quality. General
samples must meet the following requirements: the sample should be representative
and randomly selected from the same population, the data should be comparable
to the same variable and have a consistent statistical standards, data should be as
 82 Big data
accurate as possible to reduce errors caused by measurement and merging, and data
should be as complete as possible.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

1.2 Model specification

When one has selected the variables and sample, one can set the mathematical form
of the econometric model. In principle, the model should be in line with economic
theory and reality.The models should also contain the important influential factors.
Econometric models can be abstracted into the following general form:

Y = g ( X,θ ) + ε (6.11)

In Equation (6.11), Y is the dependent variable, X is independent variable, ε is a

random disturbance term, and θ is the parameter to be estimated.
First, g (⋅)’s function (often a linear or log-linear model) and explanatory varia-
bles are the core problems that can generally be determined after repeated attempts,
inspection, and adjustment.
Second, the random disturbance ε is a random variable. ε is the sum of all factors
that influence the dependent variable but are not included in the model. It comes
mainly from omission of an explanatory variable, mathematical omission, observa-
tion error, accidental factors, and merging errors. ε is unobservable and generally
assumed to be white noise (expected value is zero; variance is constant; covariance
is zero; and sometimes assumed to be normally distributed).
Third, the role of the parameter θ is to describe the stable characteristics of
the economic system, generally having economic implications and used to analyze
the direction and strength of the relationship between explanatory and explained
variables. Parameter θ is objective but always unknown. The general range of θ ’s
values is provided by economic theory. Its estimated value is derived from sample
data based on econometric methods.

1.3 Model fitting and testing

Model fitting generally requires Stata, Eviews, SAS, SPSS, R, or other professional
software. Output includes the parameter estimates and various test statistics. There
are a variety of testing models. One uses the effect of model fitting tests, including
goodness of fit tests for the entire model or individual parameters, the residuals
(estimates of random disturbance) of the test, and whether the main test item meets
the same random noise variance, irrelevance, and normal distribution assumptions.
It also refers to improper specification, such as the presence of unrelated vari-
ables, the omission of important variables, and mathematical form error.The second
category tests economic rationality, including symbols, numerical size of the param-
eter estimates, and whether the relationships meet the predictions of economic
theory.
 Big data 83
1.4 Applications
First, one must predict. For example, one may use time series models to predict
future macroeconomic variables. Model predictions need certain conditions.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

History, reality, and future must have a consistent economic structure for the same
econometric model to have stable parameters. Explanatory variables for the next
period are known. The model must also have predictive ability and reach desired
accuracy. If economic development processes are not steady, lack standardized
behavior, or the model is out of date, then prediction will fail.
Second, we have the economic structure of the analysis, including fit with pre-
dictions of economic theory and quantitative relations between economic variables,
such as marginal analysis, elasticity analysis, and comparative static analysis.
The final step is policy evaluation. Econometric results can help us both evaluate
existing policies and select the appropriate solution for future policy.

2 Differences between big data analysis and econometrics

2.1 Different data types
Econometrics deals with structured data, including cross-sectional data, time series
data and panel data, which are generally presented in tabular form in Excel. They
have clear economic implications and consistent statistical standards for each rank.
These data can also be transformed into data matrices. Many of the basic concepts
of econometrics, analytical tools, and algorithms can be expressed as matrices or
matrix operations.
Big data analysis can handle unstructured data, including documents, video, and
images, which are generally difficult to present in tabular form. These unstructured
data need quantification before analysis, and quantification is generally accompa-
nied by loss of information. For example, in big data analysis, presentation of docu-
ments must be transformed into word vectors in order to do further processing.
Word vectors only reflect words’ location and frequency in the document, but
cannot abstract semantic information in words’ arrangement.
The curse of dimensionality exists both in econometrics and big data analysis.
However, in big data analysis, because of big data amount and lower signal noise,
the curse of dimensionality is more prominent, so the use of dimension reduction
techniques is more common.

2.2 Different focuses

Econometric analysis focuses on hypothesis testing. Its core concept is very close to
Popper’s falsificationism. In general econometric analysis, it is a must to first define
a number of hypotheses deduced from economic theory. Secondly, econometric
analysis must build null hypotheses that may be falsified. For example, correlation
 84 Big data
tests usually take “no association between the two” as the null hypothesis, while
independence tests take “a link between the two” as the null hypothesis. The null
hypothesis often constrains parameters to be estimated, such as whether an argument
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

is 0 or two arguments are equal. If the null hypothesis is valid, the sample data or
its constructed test statistics should follow a known probability distribution. If the
actual value of sample data or the test statistic only occurs with small probability
(i.e. the confidence level, for example 1%, 5%), then the sample data are considered
to correspond to the null hypothesis. One negates the null hypothesis under a cer-
tain confidence level. Otherwise, the null hypothesis cannot be rejected (but never
say “accept the null hypothesis”). Through hypothesis testing, econometrics falsifies
or supports (not confirms) economic theories.
In addition, parameter estimation is one of the important contents in econometric
analysis. However, hypothesis testing for parameters is more important than para-
meters’ specific values. Econometrics can forecast, but mainly in policy research areas.
In leading academic research, econometrics is mainly used to test economic theory.
In contrast, big data analysis is more pragmatic. Prediction occupies a large pro-
portion of the big data analysis. We use classification and regression to reveal rela-
tions between variables and further predict unknown variables. Association analysis,
cluster analysis, recommender systems, and anomaly detection detect potentially
relevant data, trends, clustering, trajectory, and abnormal patterns.These patterns are
expected to be useful on other occasions or in the future and have practical value.
Therefore, the assessment of predicted effect is an important part of data analysis
embodied in the confusion matrix, ROC curves, and other tools.

3 Inherent contact between big data analysis

and econometrics
3.1 Based on probability theory and mathematical statistics
Although big data analysis uses a number of specific terms, concepts and methods
from computer science, it has no essential difference from econometrics in terms of
random questions. Both are based on probability theory and statistics. Recognizing
this helps us to clarify certain misconceptions about big data.
Viktor Mayer-Schonberger and Kenneth Cukier believe that big data analysis is
for all the data, not random samples.3 This is debatable.While the data collection and
analysis methods are sophisticated enough for data collection and analysis, it is not
always necessary to do so. According to the central limit theorem, there is a square
root relation between the quality of statistical analysis and quantity of the sample.
For example, when the number of samples increase 100-fold, quality of analysis will
increase 10-fold. Yet the statistical analysis workload has a linear relationship with
the number of samples. For example, as the number of samples increases 100-fold,
increase in storage and computing capacity is generally 100-fold. Thus, the cost of
additional work to improve the quality will exceed the corresponding benefits gen-
erated as the sample size grows.Therefore, to obtain a representative sample through
sample surveys of scientific design, data analysis is still of great value.
 Big data 85
Secondly, they believe that big data analysis seeks correlation rather than
causation.This statement is a cliché in statistics. Statistical correlation between cause
and effect relationships can only be used for falsification instead of proving cau-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

sation. Theoretical analysis of big data is also based on theory of probability and
statistics.
Third, big data analysis is not a panacea. Estimation based on big data can
be abstractly described as: X is known information; Y is unknown information;
looking for a function h ( X ) of X to forecast Y . Prediction error is Y − h ( X ).
E [Y − h ( X )] (similar to the MSE) used to measure the predicted effect. It can be
2

proved4 for any h ( X ),

E [Y − h ( X )] = E ⎡⎣h ( X ) − E (Y X ) ⎤⎦ + E ⎡⎣Y − E (Y X ) ⎤⎦
2 2 2
(6.12)

and

E [Y − h ( X )] ≥ E ⎡⎣Y − E (Y X ) ⎦⎤
2 2
(6.13)

The equal sign is valid when h ( X ) = E (Y X ). So E (Y X ) is also known as the

best predictor.
Here we arrive at two conclusions. First, in big data analysis, optimal algo-
rithms’ core task is to make h ( X ) as close as possible to the theoretical prediction
E (Y X ). Second, even in the best prediction, prediction error represented by
E ⎡⎣Y − E (Y X ) ⎤⎦ still cannot be eliminated and is endogenous to information
2

structure. For instance, even if information technology is well developed, some

information cannot be digitized (and thus cannot be used in big data analysis). This
“mixed” message determines the efficient frontier of big data analysis.

3.2 Unification of our understanding

Both big data analysis and econometrics can use points, collection/space, distance,
and other concepts to unify understanding if data objects are regarded as points and
data set as a collection or pace.

3.2.1 Big data analysis

In classification, attributes can be recorded as points and divided into classes in dif-
ferent collections. In correlation analysis, the item sets can be viewed as points; their
support and confidence are like a kind of distance. In cluster analysis, an object is
a point, cluster is a collection, and proximity can be seen as distance. In a content-
based recommendation system, options and users in collaborative filtering can be
seen as points, while the same user’s utility differences between the different options
can be seen as a kind of distance. In anomaly detection, object is the point that
is characterized by abnormal deviation from the normal point of collection, and
proximity is a distance.
 86 Big data
Big data analysis has developed many dissimilarity or proximity metrics
to proxy for distance. For example, Euclidean distance, Manhattan dis-
tance, Hamming distance, Minkowski distance, correlation coefficient, Jaccard
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

distance, cosine similarity, Mahalanobis distance, Bergman divergence, and edit

distance are used in this way. Big data analysis can flexibly use various distances,
such as nearest neighbor classifier in classification, K-means method in cluster-
ing analysis, and outlier detection based on the proximity in anomaly detection.

3.2.2 Econometrics
3.2.1.1 LINEAR REGRESSION

Consider the following linear regression model

y = β1 x 1 + β 2 x 2 + … + β p x p + ε

Suppose there are n samples. In the i sample, the dependent variable is yi , and the
independent variable is ( xi 1 , xi 2 ,…, xip ). Introducing the following notation

⎛ y1 ⎞ ⎛ xi 1 ⎞ ⎛ X 1' ⎞ ⎛ β1 ⎞
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ y2 ⎟ ⎜ xi 2 ⎟
'
⎜ X2 ⎟ ⎜ β2 ⎟
Y = ⎜ X = X = ⎜ β =⎜
f ⎟, i ⎜ f ⎟, f ⎟, f ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ yn ⎟⎠ ⎜ xip ⎟ ⎜ X' ⎟ ⎜ β ⎟
⎝ ⎝ ⎠ ⎝ n ⎠ ⎝ n ⎠

According to the least square method, the coefficient is estimated

βˆ = ( X ' X ) X 'Y
−1
(6.14)

⎛ x 01 ⎞
⎜ ⎟
⎜ x 02 ⎟
Consider a new sample. The independent variable X 0 = ⎜ f ⎟ is known, so
the dependent variable is predicted to be ⎜ ⎟
⎜ x 0 p ⎟⎠
⎝

yˆ 0 = X 0' βˆ = X 0' ( X ' X ) X 'Y

−1

n
(6.15)
∑ ⎡⎣ X ( X X )
−1
= '
0
'
X i ⎤⎦ yi
i =1

The positive definite matrix ( X ' X ) can be decomposed as ( X ' X )

−1 −1
= Γ ' ΛΓ,
in which Λ is a diagonal matrix. Thus, ŷ0 can be expressed as
n
yˆ 0 = ∑
i =1
Λ 1/2 ΓX 0 , Λ 1/2 ΓX i yi (6.16)
 Big data 87
⋅ represents the vector product implying the cosine similarity concept:
x, y
cos ( x, y ) = (6.17)
x ⋅ y
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

⋅ represents the vector mode (length). When the angle between two vectors is
0, the cosine similarity is equal to 1. When two vectors are orthogonal, the cosine
similarity is equal to 0.When the angle between two vectors is 180 degrees (i.e. the
opposite direction), the cosine similarity equals to −1. Cosine similarity coefficient
corresponds to essentially random variables.
Therefore, ŷ0 can be further expressed as
n
yˆ 0 = ∑
i =1
Λ 1/2 ΓX 0 ⋅ Λ 1/2 ΓX i ⋅ cos ( Λ 1/2 ΓX 0 , Λ 1/2 ΓX i ) ⋅ yi (6.18)

Equation (6.18) shows that the predicted value ŷ0 is weighted by the
known sample’s dependent variable. Wherein the i sample’s weights is
Λ 1/2 ΓX 0 ⋅ Λ 1/2 ΓX i ⋅ cos ( Λ 1/2 ΓX 0 , Λ 1/2 ΓX i ). Given Λ 1/2 ΓX 0 and Λ 1/2 ΓX i ,
the higher cosine similarity of vectors Λ 1/2 ΓX 0 and Λ 1/2 ΓX i (i.e. more in the same
direction), the bigger i sample weight.

3.2.1.2 MAXIMUM LIKELIHOOD METHOD

The core logic of maximum likelihood is that the most reasonable estimate of the
parameters should allow the largest probability to pick samples in a group that is
randomly selected from the model population.
Assuming the overall distribution is f (Y θ ), f (⋅) represents the data generation
mechanism. θ is an unknown parameter vector to be estimated. Assuming
X 1 , X 2 ,…, X n is a set of observations from the overall sample distribution, and is
independently and identically distributed.The definition of the likelihood function
is the joint density function of the sample:
n
L (θ X ) = ∏ f (X θ )
i =1
i (6.19)

The maximum likelihood method obtains estimation parameters by solving the

maximum value problem:
n
θˆMLE = arg max L (θ X ) = arg max ∑ ln f ( X i θ ) (6.20)
θ θ i =1
n
LL (θ X ) = ∑ ln f ( X θ ) is known as the log-likelihood function.
i =1
i

The maximum likelihood method has tight logical link with anomaly detection
in big data analysis. An anomaly detection method based on density supposes an
object outlier score is the inverse of density. The common definition of density is
the number of objects around in a specified distance. Next, we use the density and
anomaly score to interpret the maximum likelihood method.
 88 Big data
Assume the value of the parameter vector θ is estimated within Θ. For every
θ ∈ Θ, we generate independent and identically distributed random variables
Y1 ,Y2 ,…,Ym following f (Y θ ). In the random variable space, the distribution
f (Y θ ) is represented by point set {Y1 ,Y2 ,…,Ym }. Thus, there is a series of point
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

sets corresponding to different values of θ .

Consider the distribution f (Y θ ) with a representative point set {Y1 ,Y2 ,…,Ym }
and an observation sample X i . With reference to the anomaly detection approach,
X i ’s relative density to {Y1 ,Y2 ,…,Ym } is:

density ( X i, d ) =
{
# Z Z ∈ {Y 1,Y 2,...,Y m } and Z − X i ≤d } (6.21)
m• ∫ dZ
Z − X i ≤d

# represents the number of elements in the set. Z − X i ≤ d means that the

distance between Z and X is no more than d. ∫
i
dz represents the volume of
Z − Xi ≤d

a sphere with center X i and radius d.

When {Y1 ,Y2 ,…,Ym } is sufficiently dense (i.e. m is sufficiently large), there
comes:

∫ f (Z θ ) dZ
lim denisty ( X i , d ) =
Z − Xi ≤d
m→∞
∫ Z − Xi ≤d
dZ

Further, when d is sufficiently small, we have:

lim denisty ( X i , d ) = f ( X i θ ) (6.22)

m → ∞ ,d → 0
1
denisty ( X i , d )
In abnormality detection, X i ’s abnormal point score is . Because in
the observed sample X 1 , X 2 ,…, X n are independent of each other, as a whole, their
n
1
relative outlier scores to {Y1 ,Y2 ,…,Ym } are ∏
i = 1 denisty ( X i , d )
. According to the

above analysis, when m is sufficiently large and d is sufficiently small, the abnormal-
point scores of observed sample are close to the inverse of the likelihood function:
n n
1 1 1
lim
m → ∞ ,d → 0
∏ denisty ( X , d ) = ∏ f ( X θ ) = L (θ X )
i =1 i i =1 i
(6.23)

Therefore, to maximize the likelihood function is equivalent to minimizing the

outlier score. Maximum likelihood and anomaly detection solve the same problem
from different points of view.
Finally, we would like to stress two points. First, unifying understanding of big
data analysis and econometrics by using point, collection/space, distance, and other
 Big data 89
mathematic concepts reflects the deep connections among the probability theory,
statistics, and functional analysis (or topology). Second, if data relation issues can be
flexibly expressed as points, collection/space, and distance, it may create many new
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

analytical methods.This shows that big data analysis still has much room for innova-
tion beyond the seven major tasks in this chapter.

Notes
1 Qian, Minping, and Gong, Guanglu. 1998. “The Application of Stochastic Processes,”
Peking University Press.
2 This chapter mainly refers to: Jin,Yunhui and Sainan Jin. 2007.“Advanced Econometrics,”
Peking University Press.
3 Mayer-Schonberger, Viktor, and Kenneth Cukier. 2013. “Big Data: A Revolution That
Will Transform How We Live, Work And Think,” Eamon Dolan/Houghton Mifflin
Harcourt.
4 Durrett, Richard. 1996. “Probability: Theory and Examples,” 2nd ed., Duxbury Press.
 7 Big data-based credit and
Internet lending
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

The core of loan provision is the management of credit risk. This begins with an
evaluation of the creditworthiness of borrowers, and then determines the interest
rate and term of the loan. Thus, creditworthiness is also the basis of Internet loan
provisions. In this chapter, we first introduce the process of credit provision based
on big data by contrasting the credit center of People’s Bank of China (PBoC), a
typical traditional means of credit, and that of Alibaba, one based on big data. We
then introduce lending based on big data by contrasting Kabbage (USA) and Ali
Small Loan (China).

SECTION 1: BIG DATA-BASED CREDIT

1 Basic concepts1
Credit is mainly about evaluating clients’ willingness and ability to repay loans based
on their financial status, behavior, occupation, and credit record. The result of this
evaluation is manifest in credit ratings and credit scores.
Credit rating involves the use of models and analytical methods to
comprehensively evaluate different economic entities’ credibility and ability to
repay principal and interest. This process is undertaken by professional credit rating
agencies. It is also an overall evaluation of repayment risk. Credit rating consists of
the evaluation of willingness and ability to pay, with ability as an objective param-
eter and willingness as a subjective one. The credit rating procedure combines
quantitative and qualitative methods. It emphasizes qualitative methods and uses
quantitative estimates for reference. The results of credit rating is presented in sim-
ple grades, such as AAA, AA, A, BBB, BB, B, CCC, CC, C, and D. Every grade
can be subdivided into three subcategories, positive such as AA+, the neutral AA,
and negative such as AA−. In addition, each grade suggests a certain probability of
default (PD). For example, the historical default rate of BBB bonds is approximately
3%, while that of AAA bonds is 0.003%.
A credit score is a score for individuals or small companies. Credit agencies
extrapolate them from mathematical models based on credit reports to evaluate
and estimate their creditworthiness. The higher the credit score, the better the
creditworthiness.
 Big data-based credit and Internet lending 91
Credit then becomes a problem of categorization (see Chapter 6). It categorizes
individuals and companies according to their default probabilities. Mathematically,
if we use X to represent the characteristics, features, and historical information
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

of individuals and companies as independent variables and use Y to represent the

credit rating, default probability and credit score as dependent variables, we actually
use g(X), a function of X, to predict Y based on an empirical analysis. The bank-
ing industry put a great deal of effort into setting and adjusting their predictive
functions2 for the Internal Rating-Based Approach in Basel II & III.
In contrast with traditional credit, big data–based credit mainly absorbs new
sources of data, but does not differ from traditional credit in approaches and
modeling techniques. The principal difference thus lies in new independent
variables available for model input. In the next two sections, we briefly introduce
the PBoC credit rating center (CRC) and the credit rating system at Alibaba. Here
we focus on the differences in their data sources. After that, we introduce the main
credit rating approaches.

2 Brief introduction of the credit rating center of People’s

Bank of China
The CRC of PBoC collects, matches, and processes data related to credit. It also
produces credit reports and credit service to lenders. The flow of information is
divided into data collection, product processing, and external service (Figure 7.1).

2.1 Data collection

The CRC collects credit trade information and material non-credit trade
information that has direct and specific influence on credit institutions, especially
credit information in the financial industry. It includes the following:

1 Lending information is provided by credit institutions certified by the China

Banking Regulatory Commission (CBRC). This is the main source of infor-
mation, as the CRC’s right to collect data from these agencies is guaranteed

Data collection
Data submission

Organization Product
/individual processing R&D

Credit rating External service

Figure 7.1 Business model of the credit rating center.

 92 Big data-based credit and Internet lending
by laws and regulations. These institutions are technically advanced, and the
quality of their data is generally reliable.
2 Uncertified credit institutions provide lending information. This is
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

a supplementary source of data, and the CRC’s right to collect data

from these agencies is also guaranteed. However, these organizations
are small in scale and may lack the requisite information systems to provide
reliable data.
3 Data is also collected from public sector entities that handle credit transactions,
such as telephone bill records. These data are collected by permission of
or through negotiation with the public sector. The quality of data is much
influenced by the information quality of these public entities.
4 Information is also produced through law enforcement. These data have an
essential impact on the credit report of the credit body. As the government
enlarges the scope of information disclosure, this information can be collected
from public channels.
5 Information from courts on the registration, litigation, judgment, and execution
of lawsuits. Barring extreme situations, judicial information will be disclosed
and can be collected through public channels.

2.2 Data processing

Data processing requires sorting, filtration, and processing of client-related informa-
tion. The framework of data processing at the CBC has six levels (Figure 7.2):

1 Data provision: extracts data in various ways from queues, documents and
databases
2 Data exchange: verifies data in format and logic and loads it into the basic
library
3 Basic data: stores verified data as post source layer and provides data for further
data processing
4 Data processing: differentiates and integrates credit institution information
5 Product processing: processes basic and value-added products
6 Data movement: includes dispatch, transfer, formatting, and other functions.

2.3 Main product

After examining, filtering, matching, storing, and managing the data, the CRC
processes data into products according to the product design and research results.
The CRC is designed to be a basic database in financial credit information, in
addition to its service for credit institutions and the government. According to
diversified scale and demands of clients, the CRC for now supplies five categories
of product: data products, tools, solution products, outsourced service products, and
credit institution service products (Figure 7.3).
 Big data-based credit and Internet lending 93
Based on Public information
accounts processing based on Based on
/transactions credit bodies products
Data provision level Data exchange level Basic data level Data processing level Product processing
level
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Data source1
Data source1

Ground data
Data source2

MDM Data source2

Data source3 (master data
management)
ODS Data source3
Data source4
Buffer (operational
data store)
Data source5 Data source4
Shared
Ground data

data store
Data source6 ……

Manual
recording
Data source n
Other
sources

Data migration Data transmission Data distribution

Unified management Log management
management management

Figure 7.2 Data processing framework of the credit rating center.

Including credit reports,

Information products digital tool development,
and authentication services
Including credit rating,
Electronic tools model building, regulatory analysis tools,
and asset management tools
Including loan application, fraud detection,
Problem solving client life cycle management, receivable collection,
products data management, model performance tracking and
model performance regulation

IT outsourcing Providing IT support and business process

products management to small financial firms and
other organizations
Providing the credit bodies with
Credit body services credit history, credit risk assessment,
and real time credit monitoring services

Figure 7.3 Products of the credit rating center.

3 Brief introduction of the Alibaba credit system3

The Alibaba credit system (ACS) emphasizes borrowers’ behavioral data
automatically recorded in the Alibaba internal system (Table 7.1).
Data analysis is the core of business decision making in Ali Finance, with the goal
to provide objective and reliable analysis, recommendations, and optimal solutions
to improve business procedures by inputting basic data on risk, sales, and policy into
analysis models. It serves business decision-making on micro loans, wealth man-
agement, insurance and retail, and procedures of marketing, loan approval, credit,
payments, supervision, and debt repayment.
 94 Big data-based credit and Internet lending
Table 7.1 Sources of Alibaba credit system

System Alibaba credit system PBoC credit system

Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Number of Corporate Over 6 million (TaoBao Over 10 million

businesses credit alone)
Individual 145 million (Taobao 600 million
credit alone)
Credit Corporate All information relating Identification, credit
information credit to business operation, information, social
including identification, security payments and
volume, activity level, housing acquisition
inventory, rating, cash fund, quality control
flow, and utility bills, etc. information, salary
arrears, and telecom
payments, etc.
Individual Buyer’s information and Personal credit
Credit corporate credit system, information,
buyer’s identification, identification,
online expenditures, social security,
utility expenses, and payments, and housing
social activeness, etc. acquisition fund, etc.
Source of data Automatic recording Commercial banks and
government

The core of data analysis is the PD model, mainly applied in loan provision,
automatic approval of loans, and post-approval risk monitoring.

4 Main credit evaluation approaches

Credit evaluation methods can be divided into qualitative and quantitative
categories. A representative qualitative method is the 5C assessment. It is based
on experts’ assessment of borrowers’ character, capacity, capital, condition, and
collateral. As statistical methods and technology improve, quantitative methods play
an increasingly important role in credit assessment.
Based on the current practical applications of quantitative methods, we focus on
the Merton model, credit default swap (CDS) model, Logit model, and Bayesian
Criterion. Among these methods, the Logit model and Bayesion Criterion are
applicable to credit assessment for both companies and individuals. They evaluate
creditworthiness according to borrower characteristics and historical credit
information.The Merton model and the CDS model can only be applied to compa-
nies with outstanding stock and CDS respectively.They extrapolate creditworthiness
from market information such as stock prices or CDS spreads, assuming that mar-
kets can fully reflect credit information. The Merton model is classified as a struc-
tural model because it analyzes credit risk through causality. The other three models
are based on correlation, and are thus called reduced-form models. The difference
between these two kinds of models will be discussed further in Chapter 11.
 Big data-based credit and Internet lending 95
4.1 Merton model
The Merton model was first developed by the renowned economist and Noble
laureate Robert Merton. The Merton model takes the stock of a company as a call
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

option with the asset of the company as underlying asset and the total liabilities of
the company as the strike price (Figure 7.4). First, it calculates the market value
of the company and volatility of stock based on historical stock prices, and then
calculates the probability that the company becomes insolvent. The Moody’s KMV
model is a commercialized form of the Merton model, and the progressive single
factor risk model applies the same logic.
The mathematics of the Merton model are illustrated here. We assume the
following: the total debt of a company is D and matures at time T; (to simplify, we
ignore interest payments), the market value (not the book value) of the company
assets is V0 at the present and will be VT at time T, the volatility of the stock price is
a constant σV (these three variables cannot be attained directly but can be calculated
from market data), the market value of equity is E0 at present, and will be ET at
time T, and the volatility of the equity value is a constant σ E (these three variables
also must be calculated).
Due to limited liability, we have the equation ET = max (VT − D,0 ). Thus,
E0 can be deemed the present value of a European call option with VT as the under-
lying asset and D as the strike price. According to the Black–Scholes formula,

E0 = V0 ⋅ Φ (d1 ) − D ⋅ e − rT ⋅ Φ (d 2 ) (7.1)

with

ln (V0 / D ) + ( r + σ V2 / 2 )T
d1 =
σV T

d 2 = d1 − σ V T

Asset price distribution

on maturity date
Asset price

Stock price Distance to default

Face value of debt Probability of default

The current Maturity Timeline

point date

Figure 7.44 Merton model.

 96 Big data-based credit and Internet lending
Φ (⋅) as the cumulative distribution function of a standardized normal
distribution.
According to the Ito formula
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

σ E ⋅ E0 = Φ (d1 ) ⋅ σ V ⋅ V0 (7.2)

After calculating E0 and σ E from market data, we solve the simultaneous

Equations (7.1) and (7.2), and then we can solve V0 and σ V . Thus, we have the PD

PD = Pr (VT < D ) = Φ ( −d 2 ) (7.3)

In practice, the capital structure of a company is much more complicated than

the model we have discussed, so we should set triggering conditions for default and
adjust the Merton model accordingly.

4.2 CDS model

A CDS is essentially insurance on the credit risk of one or more institutions,
called the “underlying.” Any CDS transaction requires a buyer and seller of the
protection. The buyer periodically pays a fixed amount, called the CDS spread, to
the seller. In return, if the underlying institution defaults or restructures its debt
before the CDS expires (also called a “credit event”), the seller has the obligation
to compensate the buyer for its loss. The CDS market is very active, and involves
significant disclosure.
Intuitively, to make CDS transactions fair, the CDS spread should equal the
expected loss of the underlying institutions’ default. Approximately5
CDS spread ≈ PD × (1 – reclaim rate of debt after default)
Given the recovery rate, (the recovery rate of a senior debt is usually assumed to
be 40%), the PD can be inferred from the CDS spread. The higher the CDS spread
is, the higher the PD.
Chapter 2, Section 2.1 discusses a simplified version of CDS, suggesting that the
CDS spread pools the credit information that CDS market participants have.This is
the economic basis of the CDS model.

4.3 Logit model6

The Logit model is simple in form and easy to regress.These advantages have made
it one of the most commonly adopted models in the banking industry for individual
and company credit evaluation. The Logit model is a nonlinear probability model
and is derived from the latent variable model.
We use the binary variable Y with the value of 0 or 1 to represent the credit
status of clients. Y = 1, means borrower default, while Y = 0 means no default.
Also, we use the scalar X to represent the features, characteristics, and historical
information of clients, including their demographic characteristics, financial status,
and behavioral information.
 Big data-based credit and Internet lending 97
Assume that Y and X are connected by the latent variable Y *. Though we cannot
observe the specific value of Y *, we assume that

Y * = X 'β + ε
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

(7.5)

With X ' β as a linear indicator function and constant β as indicator coefficient.

The cumulative distribution function of random disturbance ε is noted as F (⋅).
The value of Y is determined by Y *:

⎧1When Y * > 0
Y = ⎨ (7.6)
⎩0 When Y ≤ 0
*

Thus, the probability distribution of Y is

Pr (Y = 1 X ) = Pr ( ε > − X ' β ) = 1 − F ( − X ' β ) , Y = 1 (7.7)

or, the probability distribution of Y when Y = 0 can be similarly derived.

eε
If the cumulative distribution function of ε is the logistic function F ( ε ) = ,
1 + eε
Then Equation (7.7) can be transformed into

eX β
'

Pr (Y = 1 X ) = (7.8)
1 + eX β
'

Equation (7.8) is the core of the Logit model. By determining what independent
variables to include in X and examining the methodology according to empirical
data, such as the borrower characteristics, category, empirical information, and
default status, we can estimate the indicator coefficient β , Equation (7.8) that can
then be used to estimate PD.
The Logit model is very flexible in the selection of independent variables. For
example, more than 100 indicators in 7 categories can be adapted to non-retail
clients, such as: financial leverage indicators including asset-liability ratio, adjusted
asset-liability ratio, all-capitalization rate; solvency and liquidity indicators including
liquidity ratio, quick ratio, cash ratio, interest coverage ratio, operating net cash flow/
total debt ratio, operating profit/total loan ratio; profitability indicators including
operating profit/sales revenue ratio, EBIT/sales revenue ratio, and profit margin
volatility in the last three years; return indicators including ROA, ROE, adjusted
ROA and adjusted ROE; operating efficiency indicators, including asset turnover,
fixed asset turnover, inventory turnover, accounts receivable turnover and liquid
asset turnover; scale indicators including total assets, equity, sales revenue, operat-
ing profit, and the average of these indicators over the last three years; and growth
indicators including total asset growth rates, sales revenue growth rate, equity
growth rate, profit growth rate, and time since establishment. For online business,
Internet-based indicators such as online sales, numbers of hits, client comments,
shipment rates, logistics records, and promotion and activity on social networks.
 98 Big data-based credit and Internet lending
4.4 Bayesian Criterion7
We use Ω to represent the sample space. Assume that there are k overall
distributions, with probability density functions f 1 ( x ) , f 2 ( x ) ,…, f k ( x ). The
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

k
priory probability is sequentially q1 , q2 ,…, qk, while ∑q i = 1.
i =1
Assume that D1 , D2 ,…, Dk divide the sample space into k non-overlapping
k
comprehensive regions. Thus ∪D
i =1
i = Ω and Di ∩ D j = ∅ for any i ≠ j , ∅
referring to the empty set. We define the criterion as if a sample x ∈ Di , we take
that x is from overall distribution Gi . We use D = {D1 , D2 ,…, Dk } to represent the
criterion.
Under criterion D, the probability that a sample from the overall distribution Gi
is mistaken as from G j is:

Pr ( j i, D ) = ∫ f i ( x ) dx (7.9)
Dj

We use L (i, j ) to represent the loss caused by mistaking one sample from Gi for
being in G j . Thus, the total average loss under criterion D is:
k k
g (D ) = ∑∑q i ⋅ Pr ( j i, D ) ⋅ L (i, j ) (7.10)
i =1 j =1

The Bayesian Criterion is an improvement on Equation (7.10) to find a criterion

D = {D1 , D2 ,…, Dk } to minimize the total overall loss g ( D ) .
When k = 2 (in this case, overall distribution G1 represents clients with good
credit, while G2 represents clients with bad credit), it can be proved that the Bayesian
Criterion is:

D1 = {x : q1 ⋅ f 1 ( x ) ⋅ L (1,2 ) ≥ q2 ⋅ f 2 ( x ) ⋅ L ( 2,1)}
D2 = {x : q1 ⋅ f 1 ( x ) ⋅ L (1,2 ) < q2 ⋅ f 2 ( x ) ⋅ L ( 2,1)} (7.11)

If we further assume that the two overall distributions are the normal distributions
N ( μ1 , Σ ) and N ( μ2 , Σ ) , with μ1 , μ2 , Σ known and Σ referring to the covariance
matrix of the two overall distributions, it can be proven that the Bayesian Criterion
Equation (7.11) is equivalent to:

D1 = {x : w ( x ) ≥ d }
(7.12), with w ( x ) = ( x − μ ) Σ −1 ( μ1 − μ2 ),
'

D2 = {x : w ( x ) < d }

μ1 + μ2 q ⋅ L ( 2,1)
μ = and d = ln 2 .
2 q1 ⋅ L (1,2 )
 Big data-based credit and Internet lending 99
If that Criterion function w ( x ) is linear, there exists a weight vector A and
threshold B, deducting Equation (7.12) into:

D1 = {x : A' ⋅ x ≥ B}
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

D2 = {x : A' ⋅ x < B} (7.13)

We define the credit score of clients as

score = A' ⋅ x (7.14)

By Equations (7.13) and (7.14), our strategy is to decline clients whose credit
score is below threshold B and to accept those above it.

SECTION 2: BIG DATA-BASED INTERNET LENDING

1 Analysis on Kabbage8
Kabbage was founded in 2008 with the goal of supplying operating capital for
online businesses that do not qualify for loan requirements by commercial
banks. Numerous small- and medium-sized online businesses come together on
e-commerce platforms such as eBay,Yahoo, and Amazon. Their demands for capital
are characterized by a short duration and small amounts. However, because these
businesses’ FICO credit scores are often below 720 and their owners are reluctant
to put up their personal assets as collateral, it is very difficult for them to obtain
bank loans. Kabbage focuses on these businesses and provide them loans by analyz-
ing Internet-based statistics. Kabbage has now exceeded 100,000 customers, and
the annual total loan amount is now around 200 million. The average Kabbage
customer receives ten loans per year.
We introduce the business model of Kabbage in three aspects: data sources, loans
issuance, and post-lending management (Figure 7.5).

1.1 Data sources

Kabbage relies on data provided by third parties rather than gathering data itself.
The sharing and reading of these data connects authorized accounts, which require
them to be standardizable, transferrable time series. Kabbage’s main sources of data
are listed in Table 7.2.
Kabbage was the first financial service institution to introduce social network
analysis into credit assessment. Online businesses can gain access to a larger line
of credit from Kabbage by keeping close to potential clients on social networks.
Kabbage has developed its own credit scoring system, Kabbage Score, by analyzing
online businesses’ operation and interaction with clients. Kabbage Scores are actively
adjusted to include the latest information, and can better describe the operation of
online businesses than traditional FICO credit scores.
 100 Big data-based credit and Internet lending
Electronics Distribution Bookkeeping Social media
platforms platform software i.e. Facebook
i.e. eBay/Amazon i.e. UPS i.e. QuickBooks /Twitter
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Transaction Distribution Accounting

Authorize data data data Networking
data

Loan
Electronic application
Kabbage
businesses Credit
rating

Account Post loan

transfer monitoring

On-time
payment PayPal
Allocation
of funds

Figure 7.5 Business model of Kabbage.

Table 7.2 Data sources of Kabbage.

Data type Source Note

Information flow eBay/Amazon Amount of views, prices, ratings,

inventory, and turnovers

Cash flow PayPal Cash flow of online payment

accounts

Distribution flow UPS Data on distribution

Social network Facebook/Twitter Client maintenance and

socialized sales

Offline businesses QuickBooks Small bookkeeping software

The online business has the incentive to provide more useful information
in order to obtain credit, and increased information also means better terms
for loans from Kabbage. The Kabbage Score and other related reports can help
online business-owners supervise the operation of their businesses and may also be
provided to third parties. Business owners could find their operational problems
and devise solutions accordingly, thus improving both their businesses and their
Kabbage Score. They would then receive a larger line of credit, which all leads into
a virtuous cycle. The interests of Kabbage and the online businesses it serves are
ultimately very well aligned.
 Big data-based credit and Internet lending 101
1.2 Loan issuance
Kabbage’s slogan is, “Fund your business in 7 minutes.” After an online business
submits its registration, the Kabbage background system automatically checks
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

whether the business has online sales data over a sufficiently long period of time.
Online businesses can only submit their application for loans after they have passed
the vetting process, which is automatically carried out by Kabbage’s system. It
decides whether to issue lines of credit, in addition to the amounts, interest rates,
and maturities based on the credit assessment. The Kabbage algorithm guarantees
that the vetting results can be provided and money transferred to an appointed
account in a third-party online payment system within seven minutes. The loans
issuance can provide customized solutions tailored to every applicant’s need, such
as automatically adjusting the line of credit, maturity and interest rate according to
different purposes of loans and operational situations.
Kabbage has several patents including “an approach to provide liquidity loans in
online auctions and online exchanges.” Online businesses can apply for loans from
Kabbage by using inventories as collateral, and then repay the loans when the goods
are sold and payments are received. In this process, Kabbage receives interest or fees
from loans, and the online businesses receives cash flows in advance to maintain
working capital.
The lines of credit that Kabbage provide range from $500 to $40,000. Interest
rates are determined by maturities (up to six months) and the creditworthiness of
the businesses, from 2% to 7% in thirty days and 10% to 18% in six months.

1.3 Post-lending management

Kabbage loan repayment is simple. On a determined repayment day, Kabbage
deducts a fixed amount (including fees) from the business’ payment account.
Prepayment is optional and does not incur extra costs.
The core of Kabbage’s post-lending management is to use multiple sources of data
to cross-check. Cash flow data from the company’s accounts are especially important.
Kabbage has accordingly realized that the real-time supervision of sales and cash flow
among online businesses can provide an early warning if cash flows become tight. If
Kabbage has confirmed that a business has difficulty paying, it can directly withdraw a
portion of the loan from the business’ accounts and cut off future lines of credit.
Kabbage has also established a disciplinary mechanism. On payment day, if the
amount in the payment account is insufficient to cover the installment, Kabbage
generally charges US$35 as late fee and retains the right to refer the case to debt-
collecting agencies or take legal action. However, this is a relatively rare event at
Kabbage. Although bad debts average 5%–8% in US banks, Kabbage has managed
a rate of only 1%.

2 Ali (Alibaba) small loan analysis

Ali Small Loan was jointly founded by Alibaba with Fosun Group,Wanxiang Group
and Intime Group, in June, 2010. It is the first small loan company to focus on
 102 Big data-based credit and Internet lending

Businesses
Consumption
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Xyb100
(A risk
Ali loan TaoBao loan Factoring managing
platform)
TrustPass credit loan Taobao/Tmall Travel factoring 0-down payment loan
E-Business loan Purchase Order loan Travel credit loan Payment on digital
CBU purchase order loan Taobao/Tmall AE advance payment Entertainment
credit loan
Tmall start-up loan
Tmall supply
Chain finance
Taobao/Tmall
Juhuasuan margin loan

Figure 7.6 Main categories of loans by Ali Small Loan.

small and micro companies in e-business in China. Ali Small Loan issues loans
based on the mountains of data it accumulates on trade platforms such as Alibaba,
Taobao, and T-mall (see Figure 7.6 for main categories of loans). It does not require
collateral or guarantees, and it generally issues loans under RMB 1 million. The
whole lending process is completed online and issued through Alipay, so no off-line
vetting is required. The time it takes to issue a loan with this system is only three
minutes. Clients with excellent credit can apply for extra loans through a manual
review process to receive amounts up to RMB 10 million.
By 2013, Ali Small Loan had provided loans for more than 490,000 clients and
issued loans totaling more than RMB 12 billion. Since its foundation, it has issued
more than RMB 100 billion loans. On average, each client receives lines of credit
of around RMB 130,000 and loans for around RMB 40,000. The bad debt rate is
under 1%. As the source of capital, Ali Small Loan has around RMB 2 billion of
assets and has transferred RMB 8 billion assets in 2013.
We will introduce Ali Small Loan through four aspects: loan application, loan
approval, post-lending management, and IT systems (Figure 7.7).

2.1 Loan application

Clients first log onto the homepage of Ali Small Loan and submit application forms
online. The application information includes the line of credit, the name of the
company, the name of the legal representative, the mobile phone number of the
legal representative, the post box of the legal representative, the marital status of
the legal representative.
After receipt of the application, an investigation team accesses trade records,
credit records, comparison of peers, inventory records, financial data, non-financial
 Big data-based credit and Internet lending 103
Aliabab B2B
Submit Taobao C2C
application Tmall B2C
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Behavioral
Credit rating
data

Guarantee
Electronic Ali small Provide in transaction
business loan guarantee platform

Lending Post-loan monitoring

Pay on time
AliPay
Transfer funds

Figure 7.7 Business model of Ali Small Loan.

comments, credit reports, and bank statements of the applicant on the B2B platform
Alibaba, the C2C platform Taobao, the B2C platform T-mall, and visits to the
applicant’s facilities. During these visits, Ali Small Loan authorizes third-party inves-
tigators to visit companies that apply for loans and check on the operation of the
companies, then the client manager of Ali Small Loan communicates with the
client to confirm the materials that investigators have submitted for application.
It is worth mentioning that Ali Small Loan uses a hydrological transaction
forecast model (in Figure 7.8) to actively promote its business to clients. The
basic principle of the model is to use hydrological variables to forecast future
transaction amounts on Taobao, excluding the seasonal business fluctuation, and
to judge the amount of both client capital and ability to repay loans. It focuses
on marketing when the clients are at the peak of demand for capital. All the
marketing and feedback is recorded in the system and is further optimized in the
response model.

2.2 Approval and issuance of loans

Ali Small Pay uses the PD Model to score the credit of online businesses. The
model has three steps: first it gathers the client’s demographic information, credit
information, historical performance, transaction data, and operation status; second,
according to the gathered information, it filters out the variables that significantly
influence the credit status to establish the Model; third, it categorizes potential
clients according to their PD scores and distinguishes clients with good or bad
credit. In this model, credit records, transaction records and complaint records of
 104 Big data-based credit and Internet lending

Transaction
Peak
sales

#/$
Peak
payments
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Marketing
Financial point
requirement Sales
trend Weeks

Q-1 Q-2 Q-3 Q-4

Figure 7.8 The hydrological transaction forecast model of Ali Small Loan.

Current Customer PD Model PD Customer Categorization

G B G
G B G
BG GG G
G G
G B G G G B
B G B G G
G B G B
G B G B B B
B
B G B

Significant Variable 1
Personal Information Variable 1 Significant Variable 2
Variable 2 Significant Variable 3
Credit Information Variable 3
Variable Qualification Significant Variable 10
Customer History
Variable 10,000
Transactional Information
Variable N
Operating Status PD Rating
......
Performance Performance Forecast Window
Record Window

Observation Point

Figure 7.9 The PD model of Ali Small Loan.

the online businesses in the Ali ecological system is fully utilized and considered as
the grounds for loan approval, settling the information asymmetry and complicated
procedure in loans for individuals and small and middle enterprises in traditional
commercial banking (Figure 7.9).
 Big data-based credit and Internet lending 105
Ali Small Loan determines the lines of credit, interest rates and maturities of
loans according to the credit scores of the online businesses, financing, and guaran-
tees in transaction platforms. If the loan is approved, the client must sign a contract
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

with Ali Small Loan to connect the legal representative’s personal account to the
business’s Alipay account. Ali Small Loan then verifies the applicant’s real name and
the Alipay account to confirm the loan in the Alipay account.
Ali Small Loan perfectly merges the transaction and financing platforms. Lenders
have saved costs by providing required information as an independent third party
and discovering the opportunity to provide financing as one party in the transac-
tion in real time and at the frontier, improving the allocation of financial assets and
productivity.9

2.3 Post-lending management

Ali Small Loan monitors the usage of loans and the productivity of businesses by
monitoring the scoring and the debt-collection model. It does so according to
transaction data and financial reports of the clients to collect loan payments. Ali
Small Loan uses the installment method for repayment. Clients periodically transfer
the installment from their bank accounts to Alipay accounts or leave enough money
in Alipay accounts to be deducted as the installment. If the client chooses to prepay
the loan, Ali Small Loan charges a fee equal to 3% of the principal. If the loan is
overdue, the interest rate in the overdue period is 1.5 times the normal rate.

2.4 IT system
Ali Small Loan has designed a management system that covers the whole life cycle
of a loan, including: management before lending, management during and after
lending, anti-fraud, market analysis, and credit system.
For now, the following two types of models have been completed:

• Risk models including PD, operational risk models, monitoring and scoring
models, debt-collection scoring, and loss given default model
• Marketing models including client response, churn, client loyalty, life cycle,
cross-marketing, event-marketing, and client value.

These following three types of model are still being developed:

• Anti-fraud models which are essential for online financial transactions, includ-
ing false trading model, false identity model, and account hacking model.
• Models for client behavior including drip-type growth analysis model, custom-
ized differentiation pricing model, and hydrological transaction forecast model.
• Ali credit model including address standardization, individual identification,
natural and legal persons identification, credit scoring, credit evaluation of sell-
ers, credit evaluation of buyers, and performance ability model.
 106 Big data-based credit and Internet lending
Ali Small Loan’s decision-making system processes tens of millions of clients,
transactions and messages and over 10 terabytes data and outputs lines of credits for
RMB tens of billions.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Considering the 79.8 million registered users of Alibaba, with 10.3 million
online businesses and over one million paying members, if Ali Small Loan could
obtain a banking license, allowing it to absorb public savings and opening up its
ability to source capital, it would be very competitive.

Notes
1 Niu, Luchen. 2012. “Research on Loaning Reputation: Theory and Practice in a Credit
Center.”
2 As is pointed out in Chapter 6, the best estimate in theory is E(Y|X). Setting and
adjusting predictive function is to make the function g(X) as close to E(Y|X) as possible.
The Internal Rating-Based approach has four parameters to predict risk: (a) Probability
to Default (PD), the possibility that the borrower defaults in a certain period of time in
the future; (b) Loss Given Default (LGD), the percentage of economic loss in the total
risk exposure once a borrower defaults; (c) Exposure at Default (EAD), the estimate of
exposure on the bank once a borrower defaults; (d) Maturity (M). IRB measures two
dimensions of risk, the default risk of a borrower in Client Rating and the specific risk
in trade in Facility Rating. Client Rating, the rating on the default risk of the borrower,
suggests the credit status of the borrower itself, and adopts PD as the core variable.
Facility Rating, the rating of risk in a specific trade, suggests the specific risk in a trade,
such as collateral, priority, and the kind of trade, and adopts LGD as the core variable.
3 In this section, unless specified, all the materials on Ali Small Loan are from the speech,
“How We Operate Internet Finance.” by Lou Jianxun, manager of the Department of
Micro Loans in Ali Small and Mini Financial Services Group, in the Second Annual
Conference of the Zhejiang Finance Society in November 30, 2013.
4 Duffie, Darrell, and Kenneth Singleton. 2003. “Credit Risk: Pricing, Measurement, and
Management,” Princeton University Press.
5 The precise estimation of PD requires solving the CDS pricing problem. It is both very
complicated and technical. Interested readers may consult Hull, John, 2006. “Options,
Futures, and Other Derivatives,” 5th ed., Pearson Education Asia Limited.The net premi-
ums determined in life insurance that we discuss in Chapter 12 are in theory very similar
to CDS pricing. The Default Density Model we will discuss is the core of CDS pricing.
6 Our introduction on the Logit model mainly refers to Yunhui, Jin, and Sainan Jin. 2007.
“Advanced Econometrics,” Peking University Press.
7 Our introduction of Bayesian Criterion refers to Shidong, Liang. 2011. “The Theory and
Application of Measurement of Risk in Commercial Banking: The Core Techniques in
‘Basel Capital Accord’,” China Financial Publishing House.
8 Our analysis on Kabbage mainly refers to Liao, li. 2013. “Lectures on Internet Finance,”
PBC School of Finance, Tsinghua University.
9 This reveals the relationship between Internet exchange economy and Internet finance
that we discuss in Chapter 11.
 8 P2P network loans
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

The P2P network loan is a new model of person-to-person lending which has been
developed in recent years. The reality that formal finance has not efficiently solved
the financing problems of small and medium-sized enterprises as well as replacing
private finance became the background for the development of P2P network loans.
However, information technology, especially the Internet, has significantly reduced
asymmetrical information and transaction costs. This gives new impetus to person
to person lending which can be seen as the earliest financial model and make up for
the inefficiency of formal financial institutions. P2P network loans can benefit both
investors (lenders) and borrowers. In this way, borrowers can enjoy more conveni-
ent financing channels and lower borrowing costs, while investors can obtain better
returns than bank deposits pay.
Zopa, the world’s first P2P network loan platform, was established in the United
Kingdom in March 2005. At present, in the P2P network loan industry, Lending
Club and Prosper are in the spotlight. Their operation is standardized, regulatory
measures are comprehensive, and information disclosure is sufficient. Between
them, the development of Lending Club is the most advanced. As a result, we will
focus on the analysis of Lending Club and take it as an example to discuss the eco-
nomics of P2P network loans.

SECTION 1: ANALYSIS OF LENDING CLUB

Lending Club began operations in 2007 with its head office in Los Angeles, no sub-
sidiaries, and all business conducted by telephone or via the Internet. By the end of
October 2013, Lending Club had facilitated US$2.77 billion in credit transactions
and generated US$250 million interest income and now it is the world’s largest P2P
network loan platform. It is also still developing rapidly (Figure 8.1).

1 Operational framework of Lending Club

In order to conform to the laws and regulations of United States (especially
securities regulations), the operational framework of Lending Club is very specific.
There are four kinds of core players: Lending Club, investors, lenders, and WebBank
(Figure 8.2). WebBank is a commercial bank registered in Utah and is protected by
the Federal Deposit Insurance Corporation (FDIC).
 108 P2P network loans
2,900 M
2,610 M
2,320 M
Total loans issued ($)

2,030 M
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

1,740 M
1,450 M
1,160 M
870 M
580 M
290 M
0
2007

2008

2009

2010

2011

2012

2013
Figure 8.1 Credit transactions facilitated by Lending Club.

Borrowing Borrowing Borrowing

members members members
Cash
Apply for a loan, Charge management fee flow

Web Bankcash
monthly repay-provide
lending platform Raise money
Lending Club
Obtain note payment
Provide note
transactions Buy notes
Cash
flow
Lending Lending Lending
members members members

Figure 8.21 Operational framework of Lending Club.

Although P2P means “peer to peer,” with respect to the operational framework
of Lending Club, legally speaking there are no direct obligatory relationships
between investors and borrowers. In fact, they register with an account name, which
is anonymous and unknown to other users. They are also not allowed to know the
other party’s real name and address. Investors purchase notes issued by Lending
Club according to the securities laws of United States. Loans to borrowers are
initially provided by WebBank and then transferred to Lending Club. Each series
of notes correspond to a sum of loans, and one is similar to mirror image of the
other. Excluding the service fee charged to investors by Lending Club, the monthly
principal and interest collected from the borrower is paid by Lending Club to the
holder of notes. If a borrower defaults, a corresponding holder of notes will not
receive any payment from Lending Club (Lending Club provide no guarantees for
investors).This does not constitute a breach of contract because Lending Club does
not take on credit risks related to credit transactions. As for WebBank, as lending to
 P2P network loans 109
borrowers and transfers to Lending Club occur almost simultaneously, credit risks
related to credit transactions are not taken by WebBank. It is similar to the role of a
custodian bank. Credit risks are borne by investors.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

As a result, the key mechanism of Lending Club’s operational framework is notes

and loans with a mirror image relationship. Every pair of loans and notes has the same
principal, interest, period, and cash flow characteristics.These types of notes are called
payment dependent notes, which are similar to pass-through securities. Through
arrangement of loans and notes, although there are complicated relationships among
Lending Club, WebBank, and lending and borrowing parties; Lending Club and
WebBank are similar in that they do not take on credit risk. Therefore, the operation
of Lending Club involves the origination and transfer of loans as well as issuance and
transaction of notes. It thus straddles the two fields of banking and securities.
In the process of selling notes to the investors and using WebBank to issue loans,
Lending Club collects a service fee. For every payment investors receive, Lending
Club charges a 1% service fee. The borrower pays a one-off origination fee to
obtain funds. See Table 8.3 for a description of details.
2 Borrowers
Anyone who proposes to borrow money must register and then submit a loan
application. There are some limits for borrower qualifications, which include:
US citizenship or permanent residency, age above 18, have an e-mail address,
possess a US Social Security Number and financial institution account number,
have an individual credit profile, and have a FICO credit score over 660. Their
debt-to-income ratio must also be below 35% (excluding mortgage loans), and the
length of credit history must be over three years. One also must have borrowed
money through Lending Club less than six times over the past six months.
One who proposes to apply for a loan should provide information which can
reflect his/her credit status. Lending Club will review the application but not
necessarily examine the authenticity of the information that borrowers provide.
Loan applications are subjected to Lending Club’s strict examination. Until the
end of the year 2012, only 11% of the applicants received a loan. The borrowers
on Lending Club generally belong to US’s upper-middle class. For instance, until
the end of October 2013, the average FICO credit score was 703 and the average
length of credit history was 15 years. The average annual income was US$71,000,
which is in the top 10% of the American income distribution.
The borrower must state three core terms of a loan: amount, period, and purpose.
Permitted loan amounts are between US$1,000 and US$35,000 while the loan
maturity is according to the borrower’s preference. For loans which are between
US$1,000 and US$15,975, the loan period is three years unless otherwise specified.
The borrower lists the purpose of the loan without Lending Club’s supervision.
Up to the end of October 2013, the average loan amount on Lending Club was
US$13,500 mainly for refinancing and paying off credit card debt (Figure 8.3). It
thus basically belongs in the consumer credit category.
Risk pricing is one of the core technologies of Lending Club, consisting of two
parts: credit rating and loan interest pricing. Credit ratings are divided into seven grades
 110 P2P network loans

2.39%
4.79%
0.91% 1.85%
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

0.78%
Refinance
5.87%
Repayment of credit card
Home improvement
House purchase

20.67% Auto financing

Big-ticket purchase
Commercial use
Other
61.30%

Figure 8.3 Purpose of loans of Lending Club.

from high to low, and each grade has five segments (the total is thus thirty-five credit
rating levels).The ratings are generated in two steps. Step 1 goes according to the bor-
rower’s FICO credit score and other credit characteristics. Lending Club has a model
rank, and every model rank corresponds to a standard credit rating. Step 2 revises the
original rating according to loan amount and gives a final credit rating.The larger the
loan amount and the longer the loan period, the larger the downgrade from the stand-
ard credit rating (Table 8.1).

Table 8.1 Credit rating method of Lending Club

Model rank Rank

1 2 3 4 5

Credit rating A 1 2 3 4 5

B 6 7 8 9 10

C 11 12 13 14 15

D 16 17 18 19 20

E 21 22 23 24 25
(The relationship between standard credit rating and model rank)

Grade adjustment Standard credit rating

A B C–E

Loan amount <$5,000 –1 –1 –1

$5,000–<$10,000 0 0 0
 P2P network loans 111

Grade adjustment Standard credit rating

A B C–E
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

$10,000–<$15,000 0 0 0

$15,000–<$20,000 0 0 –1

$20,000–<$25,000 0 –1 –2

$25,000–<$30,000 –1 –2 –3

$30,000–<$35,000 –2 –3 –4

$35,000 –4 –5 –6
(Loan amount and credit rating adjustment)

Loan period Credit rating Downgrade

3 years A–G 0

5 years A–G –8 to –4
(Loan period and credit rating adjustment)

On Lending Club, loan interest is market-oriented and fixed. Overall, the

interest rate is connected with credit rating and is equal to the sum of risk and
volatility. The purpose of the risk and volatility adjustment is to cover the expected
loan losses. The lower the credit rating, the higher the loan interest is (Table 8.2).
Service fees that Lending Club charges the borrower are around 1.1%–5.0% of
the loan amount, directly deducted from the principal of a loan. The service fee

Table 8.2 Loan pricing mechanism of Lending Club

Interest Grade

1 2 3 4 5

Credit rating A 6.03% 6.62% 7.62% 7.90% 8.90%

B 9.67% 10.99% 11.99% 12.99% 13.67%

C 14.30% 15.10% 15.61% 16.20% 17.10%

D 17.76% 18.55% 19.20% 19.52% 20.20%

E 21.00% 21.70% 22.40% 23.10% 23.40%

F 23.70% 24.08% 24.50% 24.99% 25.57%

G 25.80% 25.83% 25.89% 25.99% 26.06%

 112 P2P network loans
rate is related to credit rating and loan period, and the lower the credit rating or the
longer the loan period, the higher the service fee (Table 8.3).
Lending Club has the exclusive power to receive monthly repayment from bor-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

rowers through normal electronic fund transfers. Lending Club also has recourse to
any defaulted loans and is empowered to decide whether or when to transfer the
loans to a third-party collection agency.

3 Investors
For investors, Lending Club has set qualifications such as minimum income and
wealth (measured by net assets), and investment on Lending Club can be no more
than 10% of his/her total wealth, but a credit check is not necessary. Moreover,
Lending Club has established an investment consulting company called LC
Advisors. LC Advisors resembles a fund manager, raising external capital to invest
in notes issued by Lending Club.
Investors can choose notes that they wish to purchase on Lending Club’s
website. Due to the great amount of notes, it provides search and filter tools in
addition to a portfolio building tool. The minimum investment in each note is
US$25. For example, Lending Club will recommend a notes portfolio after the
investor specifies risk and return parameters (Figure 8.4).
For investors, it is a very effective method of risk diversification. For instance,
statistics indicate that if an investor purchases 100 notes, the possibility of incurring
loss is 1%. If an investor purchases 400 notes, the possibility of incurring loss is 0.2%.
If an investor purchases 1,000 notes, incurring loss is nearly impossible (Figure 8.5).
Later we will explain this phenomenon.

Table 8.3 Loan service fee rate

Service fee rate Loan period

Credit rating and grade 3-year period 5-year period

A 1 1.11% 3.00%

2–3 2.00% 3.00%

4–5 3.00% 3.00%

B 1–5 4.00% 5.00%

C 1–5 5.00% 5.00%

D 1–5 5.00% 5.00%

E 1–5 5.00% 5.00%

F 1–5 5.00% 5.00%

G 1–5 5.00% 5.00%

 P2P network loans 113
Build a portfolio from 836 Notes
A
7.51%
Grade
6.03% 24.4%
B «Less options
11.62% 14.89%
Grade
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

C
14.37%
Risk / Reward

Grade Portfolio composition Summary

Term | Grade
D Average interest rate: 14.89%
17.03% A $1,750 (40)
Grade
Expected default rate: 4.29%
B $2,750 (49)
E
19.18% C $2,525 (41) Service charge: 0.64%
Grade
D $1,450 (24)
F E $800 (13) Projected return: 9.97%
21.48%
Grade
F $550 (10)
G G $175 (3)
22.31% Continue >
Grade Total: $10,000 (180)

Figure 8.4 Portfolio building tool of Lending Club.

To be clear, when investors subscribe to notes, the notes are actually not
issued and loans are not made. When the subscription is full, notes are issued to
investors and Lending Club receives the subscription payment (historically, 99%
of notes are fully subscribed). Meanwhile, WebBank makes corresponding loans
and transfers those loans to Lending Club. The notes that Lending Club issued
will not be listed and traded on any exchange, but Lending Club has established
a notes transaction platform called FOLIOfn to provide notes transfer services
with investors, equivalent to setting up a secondary market for notes to provide
liquidity.

4 Supervision of Lending Club in the U.S.

Supervision of Lending Club in the United States reflects a concept of func-
tional supervision (in contrast with that of institutional supervision), which means
that it is supervised by the business in which it engages and the risk it generates
(Figure 8.6).
The US Securities and Exchange Commission (SEC) is the main regulator of
Lending Club, because it issues notes to investors that are regarded as securities
issuance. The priority of the SEC’s supervision is to ensure that Lending Club fol-
lows its disclosure guidelines rather than to check or to monitor the operation of
Lending Club or audit the notes themselves.
Lending Club has adopted shelf registration, which means registration of
prospectuses at the SEC.The prospectus should disclose the operational mechanism
and corporate governance structure of Lending Club, along with basic clauses of
notes and detailed disclosure of possible risk to investors. At the time of issuance,
Lending Club will make a representation to SEC about the notes (called sales
reports), including anonymous information like terms of corresponding loans,
purpose of loans, borrower’s job, income, and so on.The information Lending Club
discloses can be found on the SEC’s EDGAR system and the Lending Club website.
 114 P2P network loans

45%
Percent of Lending Club investors
40%
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

35% 33.61%
30% 28.74%

25%
20%
15% 14.64%
11.12%
10% 8.93%

5%
1.00% 1.96%
0%
< 0% 0% – 3% 3% – 6% 6% – 9% 9% – 12% 12% – 15% > 15%
Net annualized return
100 loans
45%
Percent of Lending Club investors

40%
36.66%
35% 32.40%
30%
25%
20%
15% 14.96%

10% 7.86%
7.33%
5%
0.20% 0.59%
0%
< 0% 0% – 3% 3% – 6% 6% – 9% 9% – 12% 12% – 15% > 15%
Net annualized return
400 loans
45%
Percent of Lending Club investors

40% 38.33%

35% 33.33%
30%
25%
20%
15.32%
15%
10%
6.42% 6.27%
5%
0.00% 0.34%
0%
< 0% 0% – 3% 3% – 6% 6% – 9% 9% – 12% 12% – 15% > 15%
Net annualized return
800 loans

Figure 8.5 Risk diversification effect of loan investment.

 P2P network loans 115

SEC+
FDIC UT DFI state
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

regulators

Loan Loan Notes

Self-directed
Borrower WebBank Lending Club
investors
$ $ $

Federal+ Servicing Loan $

state
regulators
Independent LC Trust 1
trustee (SPV)

Certificate $ SEC

LC Advisors
Funds &
SEC(IA) A Lending Club
SMAs
Company Management

Figure 8.6 Supervision on Lending Club in the United States.2

SECTION 2: ECONOMICS OF P2P NETWORK LOANS

There are many perspectives we may use to analyze P2P network loans. The
first is the legal perspective, which analyzes legal contracts and legal risks of P2P
network loans. There is much debate as to whether it involves illegal fund-raising
and illegal pooling of deposits. The second perspective analyzes the capital flows
of P2P network loans. The third perspective is risk, which analyzes major risk
types, risk-taking behavior, and risk transfer. We will follow the third perspective
because we consider it the best way to examine the economic impact of P2P
network loans.

1 Comparison with private finance

If P2P network loans do not provide guarantees for the investors, they can be seen
as direct lending from peer to peer. This lending form is old but is ubiquitous in
modern society as “private finance.” A bidding society is the form of private finance
most similar to P2P network loans.
A bidding union is one type of ROSCA, a mutually assisting credit union.
Generally, an initiator (the head) invites several relatives and friends (the feet) to
participate in and make an agreement to hold a meeting where they pay a mem-
bership fee to a certain member in turns for mutual help. The head of the society
collects the first membership fee payment, and then they determine the collection
sequence. Every member generally has an opportunity to collect the member fee,
but the head of the society normally collects the first round. The head must gather
and organize the bidding, collect and give out the member fees, pay other members,
 116 P2P network loans
and deal with member fees when someone cannot pay for his/her member fee
on time. Bidding societies can be either discount bid or premium bid societies.
Discount means the collector receives an agreed amount of member fees and the
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

member fee paid by other feet of society in every round should deduct the interest
portion that the collector is willing to pay. In premium bid, after the member who
collects member fee receives the current member fee, he/she pays other members
an agreed amount of interest.3
The head grants a loan and repays it in amortization.The last round foot of soci-
ety essentially make an installment saving. Middle feet of society participate in an
installment saving and then obtain a loan with amortized repayment.To better ana-
lyze the debt and liability relationship of bidding society, we describe a four-person
bidding society. Relevant conclusions are true even in more complicated situations.

Assume the member fee is m dollars, the bidding price of second and third
round collectors are b2 and b3, then the cash flow is

⎛ 3m −m −m −m ⎞
⎜ −m 3m −m −m ⎟
⎜ ⎟ (8.1)
⎜ − m − m − b2 3m + b2 −m ⎟
⎜ − m − m − b2 − m − b3 3m + b2 + b3 ⎟⎠
⎝

Each row indicates the cash flow of every participant in sequence.The first row
is the cash flow of the head of society and the last row is the last collector’s cash flow,
and so on. Every column indicates the cash flow of participants in t = 1, 2, 3, 4, ….n
round in sequence. A plus sign signifies capital inflow while minus sign is a capital
outflow. We can decompose the cash flow of this bidding society:

⎛ 3m −m −m −m ⎞
⎜ −m 3m −m −m ⎟
⎜ ⎟
⎜ −m −m − b2 3m + b 2 −m
⎟ (8.2)
⎜⎝ −m −m − b2 −m − b3 3m + b 2 + b 3 ⎟⎠

⎛ m −m 0 0 ⎞ ⎛ m 0 −m 0 ⎞ ⎛ m 0 0 −m ⎞
⎜ −m m 0 0 ⎟ ⎜ 0 0 0 0 ⎟ ⎜ 0 0 0 0 ⎟
= ⎜ ⎟ + ⎜ ⎟ + ⎜ ⎟
⎜ 0 0 0 0
⎟ ⎜ −m 0 m 0
⎟ ⎜ 0 0 0 0
⎟
⎝ 0 0 0 0 ⎠ ⎝ 0 0 0 0 ⎠ ⎝ −m 0 0 m ⎠

⎛ 0 0 0 0 ⎞ ⎛ 0 0 0 0 ⎞ ⎛ 0 0 0 0 ⎞
⎜ 0 m −m 0 ⎟ ⎜ 0 m 0 −m ⎟ ⎜ 0 0 0 0 ⎟
+⎜ ⎟ +⎜ ⎟ +⎜ ⎟
⎜ 0 −m − b2 m + b2 0
⎟ ⎜ 0 0 0 0
⎟ ⎜ 0 0 m −m
⎟
⎜⎝ 0 0 0 0
⎟⎠ ⎜⎝ 0 −m − b2 0 m + b2 ⎟⎠ ⎜⎝ 0 0 −m − b3 m + b3 ⎟⎠
 P2P network loans 117
Equation (8.2) decomposes the bidding society to a series of debt and credit
⎛ 0 0 0 0 ⎞
⎜ 0 m −m 0 ⎟
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

between two sides. Such as: ⎜ ⎟ means that in t = 2 round,

⎜ 0 −m − b2 m + b2 0 ⎟
⎜⎝ 0 0 0 0 ⎟⎠

the current collector borrows m Dollars and repays m + b2 Dollars in round t = 3. b2 is

the interest amount. Other debt between two sides can be explained in the same way.
Hence we can see that bidding societies are equivalent to a series of peer to peer
credit agreements in which every member is obliged to lend money to other mem-
bers who precedes him/her and also has the right to borrow money from the other
members who are after him/her. Among them, the head of society is a borrower.
The last collector is a lender. Debt and credit relationships between member A and
member B (assume A collects the member fee before B) occur when A becomes the
collector and ends when B becomes the collector. Every member is a lending party
before he/she becomes a collector, then turns to a borrowing party afterwards. In
sum, the preceding member for collection is a borrowing party while next member
for collection is a lending party.
From the above analysis, we can see that there are three common points between
bidding societies and P2P network loan. First, they are essentially peer to peer
lending. Second, the lending relies on credit completely without any collateral or
guarantee. Third, the interest rate is market-oriented. In P2P network loans, the
interest rate is determined by risk pricing mechanism. In bidding societies, interest
rates fluctuate in line with market conditions, which can include the credit risk
premium of the participants.
Bidding societies have an elaborate contract form and risk control mechanism,
especially using social capital which is formed in a long-term cooperative game,
including non-statuary ethics, custom constraints, mutual trust of acquaintances, and
social sanctions as a performance bond in order to reduce information asymmetry
and transaction costs. However, In comparison with P2P network loans, bidding
societies have two shortcomings.
First, bidding societies are personalized transactions essentially based on social
networks. Bidding society participants are largely relatives and friends, which restricts
the potential scope. P2P network loans use a third party to enable impersonal
transactions. Moreover, although in a bidding society the interest rate can include
credit risk premium, it is basically determined by experience. Thus possibly irra-
tional factors may influence these risk premia. P2P network loans have a more
scientific risk pricing mechanism.
Second, private finance has internal instability. As private finance has to be inside
a certain social network, it acts as a series of separate local markets. These local
markets have different participants and risk control mechanisms without identical
interest rates. Low connectivity means risk raised in a local market does not have
an overall impact. However, someone may make use of differences of interest rates
among private finance markets for arbitrage by borrowing in a low interest rate
 118 P2P network loans
area and lending in high interest rate area. When arbitrage is common enough,
every local market for private finance is connected, generating risk transmission
channels. In this case, the activities of private finance tend to be dynamic. Private
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

credit expands, risk control mechanisms become ineffective, and risk accumulates.
Private credit will then rack up bad debts and harm social networks’ trust relation-
ship. Incremental private credit will then sharply decrease, and a credit squeeze
will result. This credit squeeze will directly affect the local real economy, whose
weakness will weigh on the private finance market.The squeeze can only be termi-
nated after bad debts have been written off, balance sheets have been repaired, and
private social trust relations are rebuilt. In recent years, this case actually happened
in Wenzhou and Ordos in China. P2P network loans are different. They involve
an impersonal transaction that does not considerably rely on social networks, so
investors are sufficiently diversified. On the other hand, P2P network loans may be
subject to a credit cycle.

2 Comparison with bank deposit

P2P network loans are similar to direct financing. They actually resemble a bond
market as well if we consider notes as bonds. Investors buy the bonds and bear the
borrowers’ credit risk. As there is no maturity transformation, P2P network loans
bear no liquidity risk. P2P network loan platforms themselves bear neither credit
nor liquidity risk.Their profits come from services offered to investors and borrow-
ers, such as facilitation of credit transactions, risk pricing, loan collection, and notes
services. It is essentially an intermediary business.
Bank deposits represent another type of financing. Banks take on the maturity
mismatch problem between suppliers and demanders of capital. Demanders gener-
ally need long-term capital for investment. However, suppliers generally are only
willing to provide short-term capital to insure against liquidity shocks. Due to the
Law of Large Numbers, a bank only needs to store a fraction of this capital in the
form of high liquidity assets. Depositors’ ordinary cash requirements still can be
met. Banks can use other capital to grant long-term loans. Second, banks provide
an agency function because they represent depositors, supervise borrowers’ use of
money, and control the borrowers’ credit risk. By contrast, P2P network loan plat-
forms do not confirm or supervise the real purpose of loans. Banks also bear credit
and liquidity risk.They earn profit through compensation for bearing risk, which is
mainly represented as the margin between lending and borrowing rates. Banks are
also subject to capital adequacy ratios, liquidity risk regulations, and deposit-reserve
ratios because of bearing the above risks.4

3 Core technology
3.1 Risk pricing
The core technologies of P2P network loans for borrowers are mainly internal credit
rating and lending rate determination. Internal credit rating is essentially the classifica-
tion problem discussed in Chapter 6 (the main models are in Chapter 7), which means
dividing borrowers into different grades by the probability of default. If lower credit
 P2P network loans 119
ratings mean the credit profile is worse, the credit rating system is effective. One of the
measurement tools is Receiver operating characteristic (ROC) curves. The larger the
area below the ROC curve, the more effective the credit rating. For instance, the ROC
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

curve of Lending Club indicates that it is more effective than FICO08 (see Figure 8.7).
Theoretically, interest rate pricing of P2P network loans resembles bond pricing.
The lending rate equals the risk-free interest rate plus a risk premium. The lower
the credit rating, the higher the risk premium must be in order to reach a balance
(Table 8.4).

Gains chart: Pricing model vs. FICO08

100%

90%

80%

70%

60%

50%

40%
Pricing model KS: 27.1%
30%
FICO08 KS: 14.5%
20%
Random
10%

0%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Figure 8.7 ROC curve of Lending Club.

Table 8.4 Characteristics of risk and return in Lending Club by rating

Credit rating Bad loan ratio Average lending rate Net annual rate
of return

A 1.38% 7.56% 5.48%

B 2.03% 11.74% 8.80%

C 2.41% 15.16% 10.61%

D 4.33% 17.98% 11.80%

E 4.86% 20.44% 13.23%

F 5.77% 22.76% 13.41%

G 8.81% 23.51% 11.40%

All ratings 2.91% 17.02% n.a.

 120 P2P network loans
3.2 Portfolio tools
There are many notes for investors to choose on P2P network loan platforms. For
asset allocation convenience, platforms provide investors with portfolio building
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

tools in order to recommend a notes portfolio according to investors’ risk and

return preferences. As notes and loans have mirror image relationship, the theoreti-
cal basis is Portfolio theory (related to the Markowitz Mean Variance Model dis-
cussed is Chapter 11). We explain it in a simple model as the following:
Assume there is n sum of loans to invest. Assume in the loan portfolio, loan i has
n
weight w i , which satisfies ∑ w i = 1. For P2P network loans, the core of an investor’s
i =1
asset allocation problem is to solve w i , i = 1, 2,…, n which can make the loan port-
folio satisfies the investor’s preference for risk and return. Assume the interest rate of
loan i is ri , unconditional default probability is Pi , and loss given default is λ i .We use
random variable X i to indicate loan i’s default rate, E ( X i ) = Pi , therefore the return
rate of loan i is L i = (1-X i ) ri − X i λi .The same assumptions are true for other loans.
n n
Then the rate of return for the loan portfolio is L = ∑ wiLi = ∑ w ⎡⎣(1-X ) r
i i i −
i =1 i =1

X i λi ⎤⎦ . The return characteristics of the loan portfolio can be described by expected

value:
n
E (L ) = ∑ w ⎡⎣(1-P ) r
i i i − Pi λi ⎤⎦ (8.3)
i =1

The risk characteristics of the loan portfolio can be depicted by Credit Value at
Risk (CVaR). Assume confidence level as θ , let CVaR ( L ,θ ) indicate the credit risk
value under a certain confidence level, and the definition is:

Pr ( L > −CVaR ( L ,θ )) = θ (8.4)

Credit risk value actually also describes the distribution of the portfo-
lio’s rate of return. For example, assuming the cumulative distribution func-
tion of L is G (l ) = Pr ( L ≤ l ), we have an implicit function relationship:
G ( −CVaR ( L ,θ )) = 1 − θ .
In the Asymptotic Single Risk Factor (ASRF)5 model, the value at risk of the
loan portfolio equals the sum of every sum of loan value at risk
n
CVaR ( L ,θ ) = ∑ w CVaR ((1 − X ) r
i i i − X i λi ,θ ) (8.5)
i =1

According to Vasicek’s research6 and Basel committee documents,7

 P2P network loans 121

⎛ Φ −1 ( Pi ) ρi ⎞
CVaR ((1 − X i ) ri − X i λi ,θ ) = λi Φ ⎜ + Φ −1 (θ )⎟ (8.6)
⎝ 1 − ρi 1 − ρi ⎠
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

⎛ 1 − e −50 P ⎞ i

Of which, ρi = 0.12 ⎜ 2 −
⎝ ⎟ , Φ (⋅) indicates a cumulative distribution
1 − e −50 ⎠
function with standard normal distribution. Equations (8.4) through (8.6) describe
the risk characteristics of the loan portfolio. By similar means we can explain the
risk diversification effect of P2P network loan investors. For brevity, assume in the
above model that investments are even in every sum of loans and the risk charac-
1
teristics of loans are the same. That is, w i ≡ , ri ≡ r , Pi ≡ P , λi ≡ λ , i = 1, 2,… , n .
n
1 n
Then the rate of return for a loan portfolio is L = ∑ L i .We introduce an indicator
n i =1
⎧⎪ 1 Default 1 n
that indicates if the loan i defaults: L*i = ⎨ and define L* = ∑ L*i
0 No Default n i =1
⎩⎪
to indicate the ratio of default in the loan portfolio. As L i = r − ( r + λ ) L i , the rate
*

of return for the portfolio is L = r − ( r + λ ) L*.

According to the ASRF model, when the risk is fully diversified or big enough,
the cumulative probability distribution is:

⎛ 1 − ρ Φ −1 ( x ) − Φ −1 ( P ) ⎞
F ( x ) = Pr ( L* ≤ x ) = Φ ⎜ ⎟ (8.7)
⎝ ρ ⎠

⎛ 1 − e −50 P ⎞
Of which ρ is 0.12 ⎜ 2 − ⎟ , the probability of facing a loss for the
⎝ 1 − e −50 ⎠
portfolio is:

Pr ( L < 0 ) = Pr ⎛ L* >
r ⎞
= 1− F⎛
r ⎞
(8.8)
⎝ r + λ⎠ ⎝r + λ⎠

For comparison, assume the loan portfolio is constituted by a certain loan. We

might assume it is loan i.The rate of return for the loan portfolio is L i , and the prob-
ability of facing a loss for the portfolio is Pr ( L i < 0 ) = P .
If parameters as r , λ , P have ordinary values, it is not difficult to prove that, com-
pared with centralized loans portfolio, a diversified loan portfolio is more like to
avoid losses (that is the phenomenon that Equation [8.5] describes).
Last, we want to point out three possible development trends for P2P net-
work loans. The first one is the extension of “p.” P2P network loans are now
mainly among natural persons, but it could be extended to person to organization,
 122 P2P network loans
organization-to-organization, and so on.Theoretically, it has an unlimited possibility
frontier. Second, along with the development of big data credit evaluation, pricing
efficiency of P2P network loans will increase. Third, P2P network loan platforms
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

can offer a credit insurance function. After investors pay a certain insurance pre-
mium (the rate is determined by big data analysis, as in Chapter 12), part or all of
the borrowers’ credit risk can be transferred out. This will not only provide inves-
tors with new risk control tools but also extend the transaction probability frontier
of P2P network loans.

Notes
1 Refer to “Internet Finance Handouts,”Wudaokou school of finance,Tsinghua University.
2 Abbreviations in this graph mean: (1) UT DFI, Utah Department of Financial Institutions;
(2) FDIC, Federal Deposit Insurance Corporation; (3) SEC, Securities and Exchange
Commission.
3 Description of bidding society in this period in cited from: Xiang Zhang, Chuanwei
Zou: “Generating Mechanism of Bidding Society,” “Financial Research,” 2007–2011.
4 Xie, Ping, and Chuanwei Zou. 2013. “Fundamental Theoretical Research on Bank’s
Macroprudential Regulation.”
5 Gordy, M. 2003. “A Risk-Factor Model Foundation for Ratings-Based Bank Capital
Rules.” Journal of Financial Intermediation, no. 12, 199–232.
6 Vasicek, O. 2002. “Loan Portfolio Value.” Risk, no. 15, 160–162.
7 Mainly two documents: BCBS, 2004, “International Convergence of Capital
Measurement and Capital Standards: A Revised Framework” and BCBS, 2005, “An
Explanatory Note on the Basel II IRB Risk Weights Function.”
 9 Crowdfunding
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Crowdfunding is equity financing raised on the Internet. Producers can obtain

funding from potential consumers before incurring large costs, giving them the
opportunity to carry out self-made production. Also, if the crowdfunding receives
a cold welcome, producers should think twice before starting production. This saves
investment cost compared with serious reflection after failure. Crowdfunding is not
only a source of funds, but also a platform to evaluate product design and market
prospects. In addition, it helps producers get rid of constraints frequently imposed by
large funders. For consumers, they can contact the producer prior to final production
to obtain the latest products.They can choose the share of their involvement accord-
ing to their income level and judgment of the product’s value. Compared with a fixed
investment share, multi-level choices of investment scale improve consumer utility.
AngelCrunch in China and Kickstarter in the US are two typical crowdfunding
platforms.This chapter starts with an analysis of Kickstarter and then introduces the
operating mechanism and development of crowdfunding based on a 2013 paper1 by
Agrawal, Catalini, and Goldfarb. In the end, it will carry out an economic analysis
along the lines of Hardy’s 2013 paper2 on crowdfunding.

SECTION 1: ABOUT KICKSTARTER

1 The establishment of Kickstarter

Kickstarter launched on April 28, 2009. It was founded by Perry Chen, Yancey
Strickler, and Charles Adler in New York. It is an online platform to publically
raise funding for creative projects,3 such as movies, music, theater, comic books,
video games, and food-related items.4 However, instead of offering cash rewards
to funders, these projects can only return a unique experience or material objects,
such as a written note of thanks, custom T-shirts, dinner with the writer, or the
initial experience of a new product.5

2 The operating model of Kickstarter

The fundraisers of projects on Kickstarter need to explain deadlines and minimum
funding targets. If the target is not achieved by the deadline, then Kickstarter will
 124 Crowdfunding
apply its refund guarantee.6 Funds are transferred through Amazon from funder to
project manager. Kickstarter is open to funders around the world as well as project
managers from the United States and United Kingdom.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

3 The pricing structure of Kickstarter

In crowdfunding cases, Kickstarter typically charges 5% of raised funds as
commission. Amazon generally charges an additional 3%–5% of the cost.
The platform holds neither ownership of projects nor of work produced on it.
Project launches on Kickstarter are permanently archived and open to the pub-
lic. After the crowdfunding round is completed, both the project and uploaded
media materials cannot be edited or deleted from Kickstarter. However, there
is no guarantee that all the funds will be used exclusively for realization of the
project, nor to meet the expectations of supporters. Funders also have no way
to confirm the status of the project other than to consult the fundraiser directly.
Kickstarter proposal funders decide for themselves whether to support a pro-
ject or not. It also warns project managers of possible damage compensation for
funders if commitments are not met.

4 Projects on the Kickstarter platform

Projects released on the Kickstarter platform are divided into thirteen categories
and thirty-six subcategories. The thirteen categories are: art, comics, dance, design,
fashion, film, food, music, games, photography, publishing, technology, and comedy.
Television and music is the biggest category, accounting for more than 50% of
Kickstarter projects and attracting most of the funding.

5 Kickstarter guidelines7
In order to keep its focus on innovative project financing, Kickstarter asks all
fundraisers to obey following three criteria:

• The fundraiser must have innovative projects

• The project must fit in one or more of the thirteen major categories of
Kickstarter
• Fundraisers shall not engage in prohibited conduct defined by Kickstarter.

Kickstarter has additional requirements for hardware and product design projects
including: prohibiting the use of photo-realistic renderings and simulations to dem-
onstrate product, limiting the amount of individual projects or “set of ideas” project
donations, requiring physical prototypes, and a requirement to lay out plans. These
guidelines are intended to support Kickstarter's policy of supporting funding to
complete projects rather than seek product orders. Kickstarter has also stressed the
notion that the creation of a project depends on a collaboration of fundraisers
and funders. All types of projects should describe the risks and challenges faced
 Crowdfunding 125
in the creative process. To educate the public and encourage their contribution to
society is also one of Kickstarter’s goals.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

SECTION 2: OPERATIONAL PRINCIPLE AND DEVELOPMENT

1 Relation between crowdfunding and crowdsourcing

Crowdfunding derives from crowdsourcing. The purpose of crowdsourcing is to
effectively make use of knowledge, wisdom, and skills from potential participants
in a new product and build a large pool of funds. The rapid development of the
Internet provides a novel network foundation for crowdsourcing. Many specialized
crowdsourcing platforms have been used as crowdfunding for enterprise incubation.
There are two kinds of crowdfunding users. One is producers or project managers.
They introduce new items to the platform and execute the project in accordance
with funding plan once they have raised enough money.The other is crowdfunders.
Although most crowdfunding platforms serve as “pay-what-you-want,” they
generally provide other immaterial rewards for crowdfunders such as an e-mail of
thanks, Music CD, film studio visits, playing a small role in the movie, or souvenirs
of cultural value. The product is the main return of crowdfunding. But for the par-
ticipants who contribute a large amount, there are other incentive programs, such
as a higher rate of return on equity.
Among many crowdfunding platforms, Kickstarter has two reasons to be a leader.
First, it has a wide range of projects and attracted many participants and followers.
Second, of the early crowdfunding models, Kickstarter’s robust mechanism has been
imitated or copied by latecomers.

2 Operational principles and project classification

Corporate funding is generally divided into accepting donations, accepting spon-
sorship, front-end (or pre-scheduled) sales, loans, equity sales, and so on. The com-
plexities of these areas vary substantially, as one can see in Figure 9.1. Financing
by donation is the least complex. Sponsors may add to the funding complexity by
attaching other requirements. Crowdfunding usually takes a front-end sales approach.
Front-end sales must have samples of innovative products, corresponding share con-
figuration rules, and plans for prospective returns. Thus it is more complicated than
donation and sponsorship. When enterprises are stable and reach a certain size, they
often use more elaborate financing tools like borrowing, including bank loans and
bond issuance. The highest degree of complexity is the public offering of stock.
However, crowdfunding also requires certain conditions such as sufficient crowd
funders and management of a large number of micro-payment transactions. Many
companies’ startup projects lack either experience or interest in conducting the
crowdfunding process. More often than not, they delegate the task to so called
“intermediary,” crowdfunding platforms. These service intermediaries often started
as Internet or software companies. They had wide ranges of activities and activity
 126 Crowdfunding
Investment
equity
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Loan

Front-end sales
Donation Sponsor

Figure 9.1 Complexity variation of corporate finance tools.

intensity, served as an information release platform at the beginning, and acted as

a neutral intermediary between the project fundraisers and public funders. With
the development of participants’ demands and crowdfunding mechanisms, they
added functions to confirm the financing and monitor project execution capabili-
ties, and so on. Some crowdfunding platforms also provide financing suggestions,
build broader social networks for project managers, and help members find com-
mon funders.
The rise and rapid development of these platforms can be ascribed to their
accordance with financing logic as well as key functions dealing with information
and cost. First, as the capital demander, the innovative projects manager (producer)
can communicate with the crowdfunding platform, take its advice, and promote
according to its rules on detailed project information, development plans, con-
figuration options, reward, and other information. Second, after innovative pro-
jects are released on the platform, fund providers (crowdfunders) can search for
qualified projects according to their preferences, negotiate directly with project
managers; or entrust banks to invest in certain or related projects, which is similar
to trust loans. Of course, the majority of this process is conducted by payers of
small amounts. Then, after the transfer of funds, project managers start production
and give rewards to funders. Its operational relationship is described in Figure 9.2.
Notably, crowdfunding platforms have certain nationality requirements for project
managers that do not exist for funders.
Projects on crowdfunding platforms can be classified into non-profit, for-
profit, and neutral categories. Non-profit projects generally have significant social
purposes, such as public health, public infrastructure, foreign aid, general charity, and
public research projects. For-profit projects generally have clear business profit tar-
gets, for example, to set up a company or provide funding for commercial projects
within an existing company, internal research and development projects, commer-
cial movies, and music albums. Neutral projects temporarily lack clear long-term
 Crowdfunding 127
business prospects. Examples include Skype, Facebook, and YouTube, which mainly
provided new services or social networks on the Internet at first, and only later
developed into commercial services.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Project organizations can also be divided into independent, embedded, and

startups. An independent project generally does not have any organizational back-
ground and is initiated by an individual. An embedded project is originally initiated
by private or public organizations, such as companies, non-governmental organi-
zations, consortiums, or multinational corporations. It then works as part of the
organization. Startup projects may start independently, but transform into a corpo-
ration, association, club, or other forms after it achieves success. A few examples in
each category are laid out in Table 9.1.

Reward

Application

Information Intermediary Information

Crowd Crowd
crowdfunding
fundraise funder
platform

Information
Bank

Micro-payment Financial transaction

provider and payment

Figure 9.2 The basic process of crowdfunding.

Table 9.1 Classification matrix of crowdfunding projects

Organization Business background

Non-profit Neutral For-profit

Independent I am Verity, Lynch Three Million Dollar

SmallcanBeBig, Project, Homepage
Solarimpulse, Love Like hers, Exthanded,
Friendly Fire Iron Sky, lunatik.com
The Age of Stupid,
The Cosmonaut,
Artemis Eternal
Embedded Blender, Racing Shares, Hotel Chocolat,
Reduce the Cost of Project Franchise, Media No Mad,
Energy in Africa Justin Wilson plc Trampoline Systems,
Cintep
Startup Buy this Satellite, The Independent Outvesting
4th Revolution, Collective
Energy Autonomy MyFootballClub
 128 Crowdfunding
3 Development overview
Crowdfunding is booming in developed countries. Constrained by limited
material, we indirectly invoke relevant data,8 to select ten crowdfunding plat-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

forms in the United States, Britain, France, the Netherlands, Belgium, and Finland
to sketch the developments (see Table 9.2). Our ten crowdfunding samples
include Kickstarter (US), IndieGoGo (US), RocketHub (US), SliceThePie
(UK), Sonicangel (UK), Ulule (France), MyMajor (France), PledgeMusic
(US and UK), SellaBand (Netherlands and Belgium), and Grow VC (Finland
and UK). From this examination, we outline a few salient characteristics of
crowdfunding.
First, crowdfunding platforms can usually be classified as emerging enterprises.
The longest amount of time from the launch date to the time of the literature
references9 given (as of January, 2011) is fifty-three months. The shortest is only six
months. Second, crowdfunding projects depend on platform visibility and sound
rules. Kickstarter issued a total number of 12,000 projects in this time period,
an average of 571 per month. In total, the ten platforms we examined released
51,477 items and averaged 258 projects per month.
Second, only a small fraction of released projects will be chosen. Among
Kickstarter’s 12,000 items, more than 5,000 projects were funded.This selection rate
(the number of projects funded by the selected/number of items released) is slightly
greater than 42%. The ten platforms together selected a total of 11,414 items, with
a selection rate of around 25%.
Third, the number of successful crowdfunding projects is even lower. Under
the crowdfunding provision point mechanism (explained in detail below), the pro-
ject cannot carry on once it failed to get enough funding during a defined period
and needs to return previously raised funds to investors. Pre-selected items may
not be able to reach a predetermined threshold. 3,500–4,000 projects succeeded
during the period, with a success rate (number of successfully funded projects/
number of items chosen by funders) of 70%–80%. The success rate over the ten
platforms was 64%.
Fourth, a large number of funders participate in crowdfunding. Kickstarter
funders total 400,000, an average of 19,000 per month of operation; seven
crowdfunding platforms indicated that the average number of contributors is
84,200, which means each platform has an average of 3,900 funders per month
of operation.
Fifth, the minimum financial contribution is low. Kickstarter raised €24.6 million
for an average per project over €4,920, and the total amount over ten platforms
is €45 million for an average of €3,942 per project. Each funder contributed an
average of €62.9.
 Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Table 9.2 Ten selected crowdfunding platforms’ performance data10

(2010.12-2011.2d)

Platform (country) Field Start date Projects Selected Successful Number Funding Paid Average Platform
(moment till amount projects(selecting amount (suc- of funders amount € amount funding fee
2011.1) (monthly) rate) cessful rate) (monthly) (each pro- (€) (€)
ject) (€)

Kickstarter (US) For-profit 2009.4 (21) 12,000 >5,000 3,500–4,000 >400,000 >€24.6 m ? 50 5%
(571) (>42%) (70%–80%) (>19,000) (>4,920)

IndieGoGo (US) Any 2008.1 (37) >15,000 >4,000 Thousands Million 56 4%

(405) (>27%) dollars

SellaBand Music 2006.8 (53) ? 54 38 CDs >70,000 >€ 2.7 m 2.7 m 41 15%
(NL/DE) (70%) (>1,320) (>50,000)

RocketHub (US) Any 2010.2 (12) 350 (29) 75 (21%) ? ? ? 300,000 ? 8%

Ulule (F) Any 2010.10 (4) 169 (42) 53 (31%) 42 (80%) 4,818 €100,000 70,000 32 0%
(1,204) (1,887)

SliceThePie (UK) Music 2007.6 (43) ? 31 26 (84%) ? ? 750,000 ? ?

PledgeMusic Music 2009.7 (19) >2,700 2,079 (77%) 132 (6%) 74,000 ? ? 65 15%
(UK/US) (>115) (3,895)

Sonicangel (B) Music 2010.4 (11) 1,500 (142) 13 (0.8%) 12 (92%) 3,500 (318) ? ? 46 0%
(dividend)
(continued)
Crowdfunding
129
 Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Table 9.2 Ten selected crowdfunding platforms’ performance data10 (continued )

130

Platform (country) Field Start date Projects Selected Successful Number Funding Paid Average Platform
(moment till amount projects(selecting amount (suc- of funders amount € amount funding fee
2011.1) (monthly) rate) cessful rate) (monthly) (each pro- (€) (€)
ject) (€)
Crowdfunding

MyMajor (F) Music 2007.10 18,000 (473) 36 (0.2%) 15 (42%) 3,0000 (789) €5 m 360,000 150 0%
(38) (13,8889) (dividend)

Grow VC (FIN Startup 2010.8 (6) 1,758 (293) 73 (4.1%) ? 7,229 €11.6 m ? ? 25% of
UK internat.) (1,205) (148,904) ROI

Total or average 51,477 (258) 11,414 (25%) 64% 84,200 >€45m 62.9
(51.7) (3,942)
 Crowdfunding 131
SECTION 3: THE ECONOMICS OF CROWDFUNDING11

1 Common characteristics
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Funds have no geographic restrictions.The advent of the Internet enables transactions

between funders and fundraisers less constrained by spatial distance. For example,
SellaBand is a music crowdfunding platform allowing musicians and fans to interact.
Musicians can showcase their work quickly and efficiently, attract support and spon-
sorship of enthusiasts, and raise funds to release the album. Then through websites’
exposure and support from hardcore fans, musicians can attract more supporters,
expand their influence and stimulate sales of their albums. On SellaBand, over 86%
of the funders are over 60 miles away from the musicians they are supporting. For a
normal crowdfunding platform, the average distance is about 3,000 miles.12
Funds are also highly skewed.According to statistics for the general crowdfunding
platforms between 2006 and 2009, 61% of the producers have not raised any money
while 0.7% raised more than 73% of the total funds. One percent of the projects
raised 36% of the funds on Kickstarter while 10% of the projects raised 63% of
the funds.
In the early stages of crowdfunding, capital from friends and family play a key role.
Friends’ and family’s early investments produce a signal for later funders through the
promotion of capital accumulation. Crowdfunding may replace traditional sources
of funding, such as home equity loans.

2 Incentives
Project managers, funders, and platforms are the three major players in the
crowdfunding process. Here we discuss their incentives.

2.1 Project manager (producer)

Two main advantages attract project managers to use crowdfunding: lower funding
costs and the ability to obtain more information.

2.1.1 Lower cost

Early project funding comes from personal savings, home equity loans, personal
credit cards, friends and family investments, angel investors, and venture capitalists. In
some cases, project managers can obtain funds at a lower cost through crowdfunding
due to following reasons:

1 Better matching: Project managers can match investors with the highest
willingness to fund. Funders of these projects are no longer limited to specific
areas (e.g., in close geographic proximity).
2 Bundling: During the crowdfunding process, funders can get products in
advance and confirm innovative value under certain conditions. However, to
 132 Crowdfunding
some extent, crowdfunding also helps project managers bundle share sales to
obtain funding and reduce the cost of capital by pre-selling the product.
3 Information: To some extent, crowdfunding can produce more information
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

than traditional capital sourcing, which may increase funders’ willingness to

pay and reduce cost of capital. For example, Pebble Technology's founder Eric
Migicovsky published the Pebble Smart Watch project on a crowdfunding
platform after receipt of donations and funds from sponsors and friends.
He disclosed relevant information to prove that some funders recognized
the product’s innovative value, thus attracting more funders to purchase the
product and increasing the scale of funding. This improved the overall value of
the product and reduced financing costs.

2.1.2 Information about potential demand

Project managers evaluate the innovation and practicability of products according
to funders’ reaction and choice, and further amend preliminary ideas and plans.
Project managers can analyze the expectations for market demand, provide products
to meet this demand, and enhance the probability of success for later products.13
In addition, signals about market demand for products can assist project managers
in understanding potential users’ and funders’ improved product recommendations
that will help project managers quickly develop better products. For example, Pebble
Smart Watch took users’ advice and developed software applications compatible
with iPhone and Android phones, expanding its functions to checking iMessage
SMS, displaying caller information, Internet browsing, real-time alert messages,
SMS, microblogging, and social networking information, all of which exampled
new potential users.

2.2 Funders
One of the key benefits to funders is the ability to seek investment opportunities.
Early corporate funders are traditionally located near businesses, but crowdfunding
offers investment global opportunities.14 Another benefit is early access to new
products. Crowdfunding projects turn product fans into early shareholders, and
their participation can enhance the value of the company. Crowdfunding also func-
tions as a built-in social network. For investors, crowdfunding is essentially a social
activity through which they can achieve improved, low cost communication with
project managers. Investors can also support a product, service, or idea. Charity
plays an important role on crowdfunding platforms. Some funders neither receive
tangible material rewards nor participate in related online communities. Their goal
is simply to find new products and new businesses with high potential.

2.3 Crowdfunding platforms

Crowdfunding platforms’ main purpose is profit. They generally charge 4%–5%
of the funding as transaction fee. Therefore, their goal is to maximize the number
 Crowdfunding 133
and size of successful projects. This requires soundly designed market operating
rules to attract high-quality projects, reduce fraud and match creativity and capital
efficiently. Successful crowdfunding projects have advertising effects.They can draw
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

media attention and coverage, which further expands existing funders, enhances
projects’ probability of success and benefits the platforms.

3 Crowdfunding risks
3.1 Project manager
Information disclosure is the number one problem for crowd fundraisers. In most
cases, a project manager must reveal some private information to early funders
before selling his/her goods. However, the recipients are limited, which poses
far fewer risks than posting similar information on the Internet for all, including
potential competitors and imitators, to see. It could also be detrimental to bargain-
ing power with potential suppliers.
Another problem stems from the fact that crowdfunders tend not to be profes-
sional investors. Angel investors and venture capitalists tend to bring additional value
to the company through expertise and networks that can help a company further
develop its capacity. Crowdfunders on the other hand provide limited investment,
bring little experience or value and are not likely to make much post-investment
effort.
Additionally, the cost of managing funders may rise significantly as funders
multiply. Unlike angel investors or other venture capital, crowdfunding typically
requires small amounts of money from a large audience in order to obtain the
expected capital scale for project operational needs. This dispersion of ownership
can add costs to dividend payment, voting, and eventual reorganization of a com-
pany or its equity.

3.2 Crowd funders

Producers may raise only a part of the funding they need and lack the ability to
raise enough money for planned delivery commitments. This is a problem com-
mon to funding rounds, whether with crowdfunding or venture capital/PE, but is
especially difficult due to the small amounts supplied by users. If funders are overly
optimistic, they may not only invest in bad projects but also face blatant fraud. For
example, project managers may use false information to make very real-looking,
but fake, fundraising page. Despite all the attempts crowdfunding platforms make
to filter and control, career criminals are still able to game the system. In addition,
the small investment share reduces investor incentives to conduct due diligence that
could mitigate this risk.
In any case, there is a high risk of project failure for early-stage projects. Even
if funders can integrate risk into investment decisions, information asymmetry
(i.e. creators have more information than capital providers) makes a proper evalua-
tion difficult. Good projects may become bad due to rapid market changes.
 134 Crowdfunding
4 Market design
Market design affects market performance. Currently, there are four main designs
crowdfunding platforms use to reduce investment risk caused by information
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

asymmetry: reputation signaling, rules and regulation, group due diligence, and
provision point mechanisms.The first three reduce information asymmetry between
producers and funders, while the fourth helps to reduce the free-rider problem.

4.1 Reputation signaling

The early stage of traditional public financing largely depends on face-to-face due
diligence and personal relations. On crowdfunding platforms, producers send a sig-
nal through disclosure of as much information as possible on their projects and
reward programs. Market design helps market funders evaluate producers and cre-
ates market reputations. Reputation and trust are particularly important on crowd-
funding platforms. Cabral sees reputation as a mechanism to reduce the risk of
fraud for online transactions, “Although there are a variety of mechanisms to deal
with fraud, reputation is one of the best choices for enterprises.”15 Internet mar-
ket designers established many trust-building tools through reputation mechanisms.
Broadly speaking, these tools can be divided into three types: quality signals, feed-
back systems, and trustworthy intermediaries.

4.1.1 Quality signals

Waldfogel and Chen explained the importance of quality signals in online trading
markets. They believe the importance of brands decreases with the ability of con-
sumers to obtain information.16 Lewis further tested the role of access to informa-
tion and found that personal information disclosure can increase secondhand car
prices on eBay.17 Even if the product information cannot be convincingly con-
veyed, there are other methods to display quality. For example, Roberts showed
that the guarantor can provide a reliable quality signal.18 Elfenbein, Fisman, and
McManus concluded that charitable donations send a signal of higher quality signal
for Internet transactions.19 In the early financing stages when information asymme-
try is high, patents can also be used as a quality signal. Similarly, funders often regard
producers’ past successful experience as a quality signal, for instance, entrepreneurs
or founding team members with doctoral degrees.20 Finally, education level is posi-
tively correlated with crowdfunding success.

4.1.2 Feedback system

Many crowdfunding platforms provide a system for buyers and sellers to build
reputations through feedback. The most basic version simply reports sales
information.Tucker and Zhang demonstrated that sales information had a significant
impact on financing decisions.21 As part of a crowdfunding signal mechanism, online
transaction materials can help discover project managers’ behavioral characteristics
 Crowdfunding 135
and provide reference material for the investment. More complex mechanisms rely
on information on credibility from rating systems, which generally allow evaluation
after transactions. For example, buyers can evaluate sellers on eBay. High-quality
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

service results in positive evaluations, which then attract new customers. The
literature confirms the importance of seller and customer evaluations on Internet
platforms.22 However, even a good producer is not likely to raise money repeatedly
on crowdfunding platforms over a short time period. A potential solution is to
divide a sizeable project into a series of smaller ones.

4.1.3 Trustworthy intermediaries

Quality signals provided by third-party intermediaries can promote trust among
market participants. For example, Jin and Kato demonstrated the significance of
third-party certification for the prosperity and collectibles markets.23 Credit rat-
ing agencies should be able to provide an accurate and verifiable quality standard.
On crowdfunding platforms, funders tend to use information posted on Facebook,
Twitter, LinkedIn, and other large social networks to avoid moral hazard.

4.2 Rules and regulation

4.2.1 Platform rules
In order to maximize transaction scale, crowdfunding platforms modify their rules
constantly based on user behavior. For example, Kickstarter added human and sys-
tem resources to monitor the risk of fraud in 2013. Kickstarter managers believe
that this reduces investment risk, protects funders, and encourages participation
despite increasing cost. However, Kickstarter ultimately still relies on funders’ own
due diligence and must find a balance between many competing objectives.

4.2.2 Industry regulation

The US JOBS Act requires the SEC to establish industry rules for crowdfunding
(see Chapter 10 for details). These rules are mainly for investor protection, but they
will have a profound impact on the industry.
One of the regulatory areas includes risk exposure restrictions. According to
crowdfunding law, such investments must not exceed 10% of an investor’s annual
income, net assets, or $100,000, whichever is smallest. In addition, if their incomes
or net assets are less than $100,000, then the investment may not exceed the lesser
of 5% of income/assets or $2,000.
Crowdfunding platforms need to register with the SEC, conduct investment risk
education for investors, take measures to reduce fraud risk (e.g., check the history
of directors and shareholders holding more than 20% of shares), and prove that the
funders’ investment on every platform is within the aforementioned limitations. It
will also set a threshold to prevent producers from continually seeking financing if
they fail to raise enough money within a certain time.
 136 Crowdfunding
4.3 Group due diligence
Crowdfunders are more vulnerable than traditional funders. They usually hold a
very small stake and thus lack the incentive to conduct due diligence.This may lead
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

them to “free ride” on due diligence done by others. However, a large number of
funders means an examination of the project’s prospects from many different angles,
which together could lead to a unique sort of due diligence.
However, problems such as herding may reduce the effectiveness of collective
due diligence. Research shows that crowdfunders regard accumulated capital as
an important signal of quality. Project financing generally increases in accordance
with the degree of difficulty (see Figure 9.1). In the initial phase, crowdfunders
invest in projects sponsored by the project owner’s relatives and friends, which stand
in for quality and professional acceptance. Funders thus “herd” by choosing pro-
jects that are already funded. Besides, project managers may use prophase-financing
data to attract funders. In extreme conditions, producers may take advantage of
this by devoting much capital in early stages to then attract new funders, then
withdraw the capital they invested earlier. Platform rules can work to prevent this
from happening.

4.4 Provision point mechanisms

Provision point mechanisms can help to solve the “the free rider” problem. For
example, early funders can produce valuable signals for latecomers through due
diligence and capital accumulation, thus creating an incentive for investors to
wait. As for “free rider” problem, almost all crowdfunding platforms adopt some
form of provision point mechanism. The mechanism requires producers to reach
certain funding thresholds to continue financing. Projects that fall short must
return all funds accumulated from investors and declare failure. In the provision of
public goods and “free rider” problem, this particular type of contract is a classic
coordination solution. Through the provision point mechanism, crowdfunding
platforms can lower funders’ risk and help to ensure that project managers do not
engage in indefinite funding rounds.

5 Several open questions

Crowdfunding gets policy support because of its potential for societal benefit.
Crowdfunding may improve the speed and direction of technological innovation.
It may also help along the application of new technology to practical production.
Although it is likely to improve overall social welfare, increase private returns, and
generate positive externalities, crowdfunding may lead to greater financial risks.
The way this is handled will have a strong influence on the future development of
crowdfunding.
 Crowdfunding 137
5.1 Social welfare
Crowdfunding creates two types of social welfare. Platforms create a win–win sce-
nario by raising transaction revenue, while producers and funders obtain operating
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

funds and equity transactions to meet their needs.This activity is also beneficial due
to a positive spillover effect. In particular, crowdfunding focuses on enterprises in
early stages, many of which may produce innovative products. The production this
finances meets societal needs that may otherwise go unsatisfied.
Crowdfunding is, however, not perfect. It will definitely result in some social
loss. For example, new forms of fraudulent activity and new securities sales methods
could lead inexperienced or reckless individuals to make imprudent investment
decisions. Regulation should be designed to minimize these effects while maximiz-
ing the positive aspects.

5.2 Technical innovation

Crowdfunding can affect the speed and direction of technological innovation. First,
successful crowdfunding projects have generally already survived a public test. They
are more innovative and in sync with societal needs. Projects with no practical value
cannot succeed. Crowdfunding is essentially a mechanism for public screening of
innovative projects. Traditional government support and bank loans cannot react in
the differentiated manner required to effectively meet the needs of society through
innovation. Second, creative productions and innovative business solutions cannot
easily obtain operating funds through traditional financing channels. Crowdfunding
helps them obtain funds to enter production as soon as possible, thus accelerating
the pace of technological innovation.

5.3 Geographic distribution

Compared with conventional financing, crowdfunding makes it much easier to
allocate capital globally. Traditional geographic restrictions will play a less impor-
tant role in early stage investments as trading increasingly occurs on the Internet.
For example, SellaBand’s financing activities are almost completely unrestricted
by geographical factors. Therefore, crowdfunding’s capital has a completely differ-
ent spatial distribution from that of the early traditional finance. Expect stepped
up competition between traditional financing channels and crowdfunding in the
future.
Finally, we believe that crowdfunding may develop into a “financing toolbox.”24
With sufficiently transparent information and low transaction costs, some companies
(especially better qualified enterprises) will be able to raise money directly on
crowdfunding platforms and integrate financing tools instead of operating only
through the stock or bond markets. Companies will be able to issue stocks, bonds,
or hybrid capital instruments dynamically according to their needs.
 138 Crowdfunding
Notes
1 Agrawal, Ajay K, Christian Catalini, and Avi Goldfarb. 2013. “Goldfarb Catalinimics of
Crowdfunding,” NBER Working Paper 19133. http://www.nber.org/papers/w19133.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2 Hardy, Wojciech. 2013. “How to Perfectly Discriminate in a Crowd? A Theoretical

Model of Crowdfunding,” Working paper. No. 16.
3 Wauters, Robin. 2009. “ Kickstarter Launches Another Social Fundraising Platform.”
4 Levy, Shawn. 2010. “Kickstarter Raises Money Online for Artistic Endeavors, Tapping
into Portland Ethos.”
5 Walker, Rob. 2011. “The Trivialities and Transcendence of Kickstarter,” The New York
Times Magazine, August 5.
6 Musgrove, Mike. 2010. “At Play: Kickstarter is a Web Site for the Starving Artist,” The
Washington Post, March 7.
7 Blattberg, Eric. 2012. “Kickstarter Bans Project Renderings, Adds ‘Risks and Challenges’
Section,” Crowdsourcing.org.
8 Andrews, Robert. 2011. “Crowdfunding: How Does The Scene Stack Up?.” http://
paidcontent.org/table/crowdfunding. 1US$= 0.7057€.
9 Hemer, Joachim. 2011. “A Snapshot of Crowdfunding,” Working Papers Firms and
Region, No. R2/2011.
10 Andrews, Robert. 2011. “Crowdfunding: How Does The Scene Stack Up?” http://
paidcontent.org/table/crowdfunding. 1US$= 0.7057€.
11 This section mainly refers to Hardy,Wojciech. 2013. “How to Perfectly Discriminate in a
Crowd? A Theoretical Model of Crowdfunding,” Working paper. No. 16.
12 Agrawal, Ajay K, Christian Catalini, and Avi Goldfarb. 2013. “Goldfarb Catalinimics of
Crowdfunding,” NBER Working Paper 19133. http://www.nber.org/papers/w19133.
13 Lauga, D, and E. Ofek. 2009. “Market Research and Innovation Strategy in Duopoly,”
Marketing Science, 28(2), 373–396.
14 Gubler, Z. J. 2013.“Inventive Funding Deserves Creative Regulation,”Wall Street Journal,
January 31, http://online.wsj.com/article/SB1000142412788732346860457825191386
8617572.
15 Cabral, L. 2012. “Reputation on the Internet,” in Martin Peitz and Joel Waldfogel, eds.
Ch. 13, The Oxford Handbook of the Digital Economy, pp. 343–354.
16 Waldfogel, J, and L. Chen. 2006. “Does Information Undermine Brand? Information
Intermediary Use and Preference for Branded Web Retailers,” Journal of Industrial
Economics, 54(4), 425–449.
17 Lewis, G. 2011. “Asymmetric Information, Adverse Selection and Seller Disclosure: The
Case of eBay Motors,” American Economic Review, 101(4), 1535–1546.
18 Roberts, J. 2011. “Can Warranties Substitute for Reputations?” American Economic
Journal, 3(3), 69–85.
19 Elfenbein, D. W, R. Fisman, B. McManus. 2012. “Charity as a Substitute for Reputation:
Evidence from an Online Marketplace,” Review of Economic Studies, 79(4), 1441–1468.
20 Hsu, D. 2007. “Experienced Entrepreneurial Founders, Organizational Capital, and
Venture Capital Funding,” Research Policy, 36(5), 722–741.
21 Tucker, C, and J. Zhang. 2011. “How Does Popularity Information Affect Choices?
A Field Experiment,” Management Science, 57(5), 828–842.
22 Cabral, L. 2012. “Reputation on the Internet,” in Martin Peitz and Joel Waldfogel, eds.
Ch. 13, The Oxford Handbook of the Digital Economy, pp. 343–354.
23 Jin, G. Z, and A. Kat. 2007. “Dividing Online and Online: A Case Study,” Review of
Economic Studies, 74(3), 981–1004.
24 This concept is similar to Internet Financing Trading Platform described in Chapter 3.
 10 The regulation of Internet finance
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Regulation of Internet finance remains an open question and mature regulations

are not yet in place. Currently, governments are exploring possible regulation.
This chapter discusses three issues. Whether it is necessary to supervise Internet
finance, functional supervision of Internet finance, and organizational supervision
of Internet finance.

SECTION 1: SHOULD INTERNET FINANCE BE REGULATED?

After the most recent financial crisis, we believe that governments should regulate
Internet finance. However, it is also necessary to take into account the special char-
acteristics of Internet finance.

1 Necessities of regulation
When the market is efficient (the ideal case of Internet finance, see Chapter 1), mar-
ket participants are rational and their self-interested behaviors make it possible to
automatically achieve market equilibrium through the “invisible hand.” The market
both balances prices fully and accurately reflects all available information.When this
point is reached, Internet finance regulation should follow the concept of “laissez-
faire,” the key objectives of which are to remove factors causing market inefficien-
cies and to allow the market mechanism to play a role with little or no regulation.
We justify this assertion through three principal assumptions. First, the market price
signals will be correct, so we can rely on market discipline to take effective control
of harmful risk-taking behavior. Second, we make it possible for financial insti-
tutions to fail, thus achieving the survival of the fittest in a competitive market.
Finally, it is not necessary to supervise financially innovative products. Unnecessary
or non-value-creating innovations will be eliminated under market competition
and discipline.Well-regulated financial institutions will not develop high-risk prod-
ucts, and consumers with full information will only choose products that meet their
own needs. Additionally, in terms of determining whether the financial innovation
is valuable or not, regulatory authorities may not take the best position. Instead,
regulation may inhibit beneficial financial innovation. However, before reaching the
 140 The regulation of Internet finance
ideal situation, non-effective factors like asymmetric information and transaction
costs still exist, making the concept of laissez-faire regulation not applicable.
First of all, individuals may be irrational in Internet finance. For example, in
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

P2P network loans, investors lend personal credit loans to borrowers. Even if the
P2P platform accurately reveals the borrower’s credit risk and the investment is
diversified enough, it remains a high risk investment. Investors may not be able to
fully understand the impact on individuals when the investment fails. Therefore,
for P2P network loans, it is necessary to introduce regulation to protect investors.
Second, individual rationality does not necessarily imply that collective ration-
ality is achieved. For example, in the “third-party-payments plus money market
fund,” exemplified by Yu’E Bao, investors buy fund shares in the money market (see
Chapter 3). Investors can redeem their fund at any time, but money market fund
positions generally have a longer maturity. This means one may be forced to sell at
a discount in the secondary market, initiating problems like maturity mismatch or
insufficient liquidity. If money market volatility occurs, investors may redeem their
funds in order to control risks, which is absolutely rational from the individual’s
perspective. It is problematic that money market funds will encounter a run if there
are large-scale redemptions, which is irrational from the collective point of view. In
September 2008, one of the oldest US Money market funds, the Reserve Primary
Fund, suffered this after the bankruptcy of Lehman Brothers. The Reserve Primary
Fund was exposed to Lehman Brothers, and the writedowns on the Lehman com-
mercial paper they held led them to “break the buck,” or fall below a net asset value
(NAV) of $1. Therefore, institutional investors scrambled to redeem their invest-
ments although the net loss was no more than 5%. Thus, the fund had to be liqui-
dated.The whole money market fund industry then suffered the hit of redemptions
overnight. The liquidity crunch had also spread to the entire financial system, and
the central banks of related countries had to team up to launch massive liquidity
support measures. Such collectively irrational behavior exhibited by institutional
investors is entirely possible for individual investors as well.
Third, market discipline may not able to control harmful risk-taking behaviors.
In China, there exist various implicit and explicit guarantees against investment risks
(e.g. implicit deposit insurances, implicit promises from banks for financial products
sold at their branches), and investors have become accustomed to “fixed pay-outs.”
This implies that risk-based pricing mechanisms have failed. In this environment,
some Internet financial institutions have launched high-risk and high-yield products.
They try to attract investors and achieve a large scale through expected high-returns.
However, they may not truthfully reveal risks. There is a huge moral hazard.
Fourth, if Internet financial institutions involve in large number of users and
reach a certain scale of funds, it is difficult to solve the problem by clearing the
market in a crisis situation. If the institution provides payment, clearing, and other
basic financial services, its bankruptcy may also damage the infrastructure of the
financial system, posing a systemic risk. For instance, the scale of people involved
and business funds are so large in Alipay and Yu’E Bao that they have reached
systemic importance.
 The regulation of Internet finance 141
Fifth, Internet financial innovation may have major defects. For example, P2P
network loans in China appear to be a mixed bag. In some P2P platforms, customer
funds and platform funds are not effectively separated. There are thus situations in
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

which platform managers escape with customer’s money. Some P2P platforms
have aggressive marketing strategies, selling high-risk products to people who do
not have the capacity for risk identification and risk bearing (e.g. retired people).
Some P2P platforms even breached the regulatory red line of illegal fund-raising
and illegal deposits. Take Bitcoin as another example: because of the characteristic
of anonymity, it is used in money laundering, drug trafficking and other illegal
activities.
Sixth, there may be fraud and irrational behavior in the consumption of Internet
financial products, finance institutions may develop or sell overly risky products,
and consumers may buy products they do not understand. For example, when
sold online, some products generally disclose their expected rate of return without
explaining how they achieve it or the potential risk factors. Some products even
use subsidies, guarantees, “loss leaders,” and other ways to magnify gains, which are
not pure market competitive behaviors. On the other hand, some consumers are
even unclear on the differences between P2P network loans, deposits and bank
financial products because of their limited financial knowledge and, expectation of
“fixed-payouts.”
Meanwhile, behavioral finance also supports the need for regulation of Internet
finance. It studies the irrational behavior of individuals and problems with the
market. On the one hand, psychological research on cognition and preferences
are introduced, implying that individual behavior does not necessarily meet the
description of the rational expectations hypothesis; on the other hand, it studies the
limits of arbitrage, which can hinder the achievement of market equilibrium. Thus,
it can prove that the efficient market hypothesis is not necessarily true. The revela-
tion behavioral finance gives us when considering questions in Internet finance
are as follows: first, we must curb excessive speculation. For instance, the deflation-
ary effect of Bitcoin comes with serious problems of speculation (see Chapter 5).
Second we should restrict market access. Internet financial institutions and investors
are not completely rational, so certain markets or products should only open to
those who satisfy certain conditions. Third, we should strengthen the supervision
of Internet financial innovations and promptly correct any problems. Fourth, we
must strengthen protection for financial consumers. Fifth, we need regulation that
fits investor needs.
Therefore, for Internet finance, we cannot adopt the concept of laissez-faire
because of its immaturity, instead we should promote it with regulation. We need
to encourage limited, well-regulated innovation in Internet finance.

2 Specialties of regulation
Compared with traditional finance, Internet finance has two unique characteristics.
These present several risks to which we should pay attention.
 142 The regulation of Internet finance
2.1 Information technology risks
Because of increased connectivity (see Chapter 1), IT risk becomes a problem in
the Internet finance, such as computer viruses, computer hacking, insecure pay-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

ments, online financial fraud, financial phishing sites, customer data leaks, illegal
identity theft, or tampering. The Vice Chairman of the China Banking Regulatory
Commission, Yan Qingmin, explained that information technology risk can be
understood from three perspectives: exposure source, objects impacted, and the
influence on responsible units.1
Sources of information technology risk are divided into four categories:

1 Risks based on natural factors

2 Systemic risks caused by related hardware and software defects in information
systems. They include aging infrastructure and hardware devices, applications
and systems software quality defects, etc.
3 Risks caused by management defects, mainly reflected by the lack necessary
management systems, imperfect organizational structure, or inadequate man-
agement processes
4 Operational risks caused by intentionally or unintentionally illegal operations
of personnel.

Objects of information technology can also be divided into three categories:

Data risks, risks in information technology platforms, and risks posed by the physi-
cal environment. In information technology areas, financial services involve a great
deal of data processing. Once mismanagement rears its head, data errors like cus-
tomer information leaks and mistakenly allocated capital will easily appear. Data
processing in financial services need a robust operation system. Inherent defects
or mistaken management of hardware devices, networks, operating systems, data-
bases, middleware, and operations systems will affect the quality of the information
systems, giving rise to potential risks. As for risks in the physical environment, the
security of information system operation platforms relies on a suitable physical
environment. Earthquakes, thunderstorms, and equipment failures in engine rooms
will affect the supply of electrical power, along with temperature and humidity of
engine rooms, potentially causing equipment malfunctions.
The influence of informational risks to organizations are divided into four
categories:

1 Security risks, such as information that has been tampered with, stolen or used
by unauthorized organizations
2 Availability risks, which means information or applications are unavailable due
to system failures or natural disasters
3 Performance risks, meaning that the poor performance of systems, applica-
tions, or personnel lead to low efficiency of transaction and operation, as well
as value destruction
4 Compliance risks, including the handling and processing of information
that does not meet the requirements of laws, regulations or policies made by
 The regulation of Internet finance 143
IT or financial institutions, which may damage the reputation of financial
institutions
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Regulatory approaches for information technology risks include2 off-site

operation (using regulatory compliance), on-site inspections, risk assessment and
supervisory ratings, forward-looking risk control measures, and mathematical models
for measuring IT risks (e.g. measurement methods based on the loss distribution).

2.2 “Long Tail” risks

As Internet finance expands the boundaries of transactions and serves a large part
of the population that is not under the coverage of traditional finance (the so-called
“long-tail” feature), Internet finance exhibits quite different risk characteristics than
those of traditional finance. First, target groups for Internet finance know little about
finance and have a poor ability to identify risk and risk-taking. They are especially
vulnerable to misleading, deceptive, and unfair treatment. Second, their investments
are small and scattered. For individuals, the costs of financial institution regulation
far outweigh their benefits, thus the “free rider” problem is more prominent,3 and
Internet financial market discipline fails easily. Third, situations of individual and
collective irrationality are more likely to appear. Fourth, once Internet risks break
out, a large set of negative externalities will be unleashed on society (the specific
meaning of externalities can be seen below) due to the population involved (while
the funds involved may not). In general, because of the “long tail” risks of Internet
finance, mandatory, professional, knowledge-based, and lasting financial regulation
is indispensable, while financial consumer protection is particularly important.

SECTION 2: FUNCTIONAL REGULATION OF INTERNET FINANCE

The core of functional regulation is to set regulation based on the business and risks
of Internet finance. Internet finance should accept regulation consistent with that
of traditional finance if its function is similar. Different Internet financial institu-
tions should be subjected to the same regulation if they are engaged in the same
business or are subject to the same risks, otherwise it is likely to cause regulatory
arbitrage. This is not only harmful to fair competition in the market, but also gives
rise to regulatory blind spots. Institutional regulation also corresponds with func-
tional regulation. Although institutional regulation is clearer in regulatory issues,
functional regulation involves more basic theories and methodology. It is necessary
to discuss this before discussing institutional regulation.
Functional supervision mainly refers to supervision of risk. It is based on
risk identification, measurement, prevention, early warning, and disposal. Like
traditional finance, the risks of Internet finance mean the possibility of future loss.
Conceptual and analytical frameworks of market risks,4 credit risks,5 liquidity risks,6
operational risks,7 reputational risks,8 and legal compliance risks9 are adaptable to
Internet finance.10 There also exist problems of misleading consumers, exaggerated
marketing, fraud, etc. Therefore, there are no significant differences between
 144 The regulation of Internet finance
conventional finance and Internet finance in the functional regulation area. They
can be divided into three categories: prudential regulation, behavioral regulation,
and protection for consumers of financial products (although specific measures may
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

be different from traditional finance).

1 Prudential regulation
The goal of prudential regulation is to control the externalities of Internet finance and
protect the public interest.11 According to microeconomic theory, externalities refer
to a situation in which the behavior of economic actors will directly affect the welfare
of other consumers and the production capacity of other vendors. If additional restric-
tions are not imposed on the market, the equilibrium yield is lower than the socially
optimal level when the externality is positive, while higher when the externality is
negative (in general, the financial sector belongs to the latter case). The basic meth-
odology of prudential supervision is to introduce a series of risk management tools
based on risk identification (generally using methods of regulatory limits) to control
risk-taking behaviors of Internet financial institutions and the negative externalities
they produce for society. Therefore, the socially optimal level may be reached.
At the moment, the externalities of Internet finance revolve primarily around
externalities of credit and liquidity risks. To deal with this, we can design practices
of banking regulation according to the principle of “substance is over form,” and
design regulatory measures for Internet finance.12

1.1 Regulation of credit risk externalities

Some Internet financial institutions are engaged in credit intermediation. For
example, in P2P network loans, some platforms are directly involved in the bor-
rowing chain or provide guarantees for lending activities. Overall, they are taking
credit and will bring credit risk externalities. If they go bankrupt, their creditors
and counterparties are not the only one whose interests are going to be damaged,
because creditors and counterparties of similar Internet financial institutions would
also doubt the solvency of these institutions. This is a key component of financial
contagion. According to the “notice on strengthening regulation of shadow banks
and related issues” (i.e. “the 107th text”) by the State Council, 2013, Internet finan-
cial institutions engaging in intermediation activities without financial licenses and
unregulated are shadow banks, and the Chinese regulatory authorities should thus
formulate regulatory measures for them.
We can refer to the practices of banks for the supervision of external credit
risks. Under Basel II and III, banks need to provision for loss reserves and capital
assets to remain as solvent “going concerns,” even under the impact of credit risks.
Among them, losses are divided into expected and unexpected categories. Expected
losses are not real risks, rather they represent the mean of future possible losses and
can be covered by asset reserves. Unexpected losses refer to fluctuations from the
value of expected losses and are certainly real risks. They must therefore be covered
 The regulation of Internet finance 145
by capital. Concrete manifestations include loan loss provision coverage, capital
adequacy ratios, and other regulatory indicators.
For example, P2P platforms generally allocate part of their revenue to risk
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

reserve pools (e.g. 2% of total loans) to offer capital protection for investors. Risk
reserve pools are functionally consistent with bank reserves for asset losses. How do
we determine the appropriate size of the risk reserve pool? The capital adequacy
ratio for banks is 8%, calculated as the bank’s capital divided by risk-weighted assets.
The average risk weight is about 50%, thus bank capital assets must account for 4%
of total assets. Correspondingly, the risk reserve pool of P2P platforms should be
4% of total loans. Of course this is just a simple calculation to illustrate the relevant
regulatory logic. More specific standards for risk reserve pools should be deter-
mined according to risk measurement.

1.2 Regulation of liquidity risk externalities

Some Internet financial institutions undertake credit intermediation through
maturity transformation and/or liquidity provision. This is exemplified in Yu’E
Bao’s “third party payment and money market fund” product, in which investors
can always redeem their fund shares, but the duration of the fund may be longer.
Such an Internet financial institution will bring about liquidity risk externalities. If
they suffer from a liquidity crisis, creditors, and counterparties of similar Internet
financial institutions will suspect the liquidity situations of their own, thus causing
contagion. Moreover, when financial institutions suffer from a liquidity crisis, they
usually recover their cash through the sale of assets to meet liquidity requirements.
Large-scale asset sales in a short time (a “fire sale”) will cause a decline in assets
prices. Under a fair value accounting system, other financial institutions holding
similar assets will also be harmed. In extreme cases, a vicious cycle of “asset prices
fall, sales are triggered, asset prices fall further” may even appear.
We can also refer to bank practices for the supervision of external credit risks.
Basel III introduces two regulatory liquidity indicators: the Liquidity Coverage
Ratio (LCR) and Net Stable Funding Ratio (NSFR). Among them, the LCR has
been implemented, requiring banks to set aside high-quality liquid asset reserves to
deal with net cash outflows within thirty days in amounts estimated with liquidity
pressure tests. The regulatory logic for the NSFR is similar.
For Yu’E Bao, we should estimate the volume of investor redemptions by taking
stress tests in situations of large shopping seasons and money market volatility. We
should then accordingly limit the position of money market funds to guarantee a
sufficient proportion of high liquidity assets (this would of course sacrifice some
profitability.)

2 Behavioral regulation
Behavioral regulation, including regulation of Internet financial infrastructure,
financial institutions, and related behavior of participants, is aimed to make Internet
 146 The regulation of Internet finance
finance transactions more secure, fair and effective. In a certain sense, behavioral
regulation optimizes Internet finance operations.
We first emphasize regulation of the shareholders and managers of Internet
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

financial institutions.The certification process should exclude those who are impru-
dent, lack requisite capacity, are dishonest, or have a poor record as shareholders
and managers. On the other hand, once operations have started, regulators should
strictly control transactions among shareholders, managers, and Internet financial
institutions to prevent them from damaging the legal interests of Internet financial
institutions and customers through means of asset expropriation and fraud, etc.
Second, behavioral regulation focuses on deposits, trusts, the trading and clearing
system of funds, and securities related to Internet finance. On the one hand, we
must improve the efficiency of Internet financial operations and control operation
risks; on the other hand, platform-based Internet financial institutions must prevent
misappropriation of client funds by clearly separating platform and customer’s
money, and taking measures against other potential malfeasance.
Third, Internet financial institutions are required to have a sound organizational
structure, internal control system and risk management measures, business premise,
IT infrastructure, and security arrangements.

3 Protection for consumers of financial products

Protection for consumers of financial products refers to protection of their inter-
ests in Internet finance transactions. Protection for consumers is closely related to
behavioral regulation, but we list it separately because consumer protection is aimed
to serve the “long tail” groups in financial services, while behavioral regulation is
aimed at Internet financial institutions.
The background of consumer protection is the theory of consumer sovereignty
and damages to interests of consumers from financial institutions under asymmet-
ric information.13 We must realize that interests of Internet financial institutions
and consumers of financial institutions are not exactly consistent, so the sound
development of Internet financial institutions cannot fully protect the interests of
consumers.
In reality, due to the limitations of expertise, financial consumers cannot fully
understand the costs, risks, and benefits of financial products as Internet financial
institutions do.They are at a disadvantage and cannot afford the monitoring and due
diligence required. The consequence is that Internet banking companies have the
appearance of financial products and pricing information dominance, and they will
consciously take advantage of the relatively weak information possessed by con-
sumers. In economics, we call this the implicit fraud tendency (or quasi-fraudulent),
but it is not necessarily equivalent to the legal definition of fraud. Moreover, there
is a “lock-in effect” between Internet financial institutions and consumers of finan-
cial products, meaning that the fraud activities are generally subtle and cannot be
eliminated by competition in the market (in another words, even if consumers have
discovered that there is a fraud, they may not be able to choose another institution).
 The regulation of Internet finance 147
To protect consumers, we can use methods of self-regulation. However, if there
are no appropriate, low-cost activist channels, or if financial institutions are too
strong while the self-regulatory bodies lack effective measures, the aforementioned
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

quasi-fraud activities are generally difficult to stop and punish. Many cannot even be
disclosed. In this case, self-regulation will fail, and the government will take manda-
tory regulatory powers. The primary measures which must be taken are as follows:
First, we must require financial institutions to strengthen information disclosure.
Terms and conditions of Internet financial products should be simple and clear, and
the information must be transparent. Then, consumers of financial products can
understand the relationship between risks and returns.
Second, we should give consumers their own advocacy channels. The first is
a compensation mechanism. Just as those who buy electrical equipment are able
to ask for compensation when encountering fake products, consumers of finan-
cial products should also be able to make a claim when they encounter situations
of misleading or exaggerated marketing and fraud. The second requirement is an
action mechanism. The original provisions of US law stipulated that class actions
lawsuits are only available for stock investment. However, after this round of finan-
cial crisis, consumers are allowed to sue banks, insurance companies and securities
companies if there is fraud, and sales agencies may also be forced to take joint
responsibility. This mechanism can be used in consumer protection.
Third, we must be able to detect regulatory loopholes promptly by receiving
complaints from consumers and taking swift action. At the moment, some quasi-
fraudulent products are difficult to be discovered by regulators, but this must be
improved.
Fourth, the Internet should be used as a platform to allow consumer complaints
to be broadcast widely.When a consumer of financial products finds problems with
products and publishes them online, other consumers can become “free riders.”
Protection thus expands to all consumers of the financial product.This is equivalent
to the use of the principles of social networking and big data.

SECTION 3: ORGANIZATIONAL REGULATION OF

INTERNET FINANCE

Among the six major types of Internet finance listed in this book (see Chapter 1),
P2P network loans and public financing loans most urgently require regulation.
Other forms more or less already have an established regulatory framework. This
section will focus on the regulation of P2P network and public financing loans.14
Organizational regulation operates on the premise that similar institutions engage
in similar businesses and produce similar risks. They should therefore be subject to
similar regulation. However, there have been mixed results for this approach so far
in Internet finance. In this case, we need to formulate regulatory measures from
an organizational perspective according to specific businesses and risks of financial
institutions. We must also coordinate supervision more effectively.
 148 The regulation of Internet finance
1 Current regulatory framework
Online banking, mobile banking, online securities companies, online insurance
companies and online financial trading platforms (see Chapter 1) act as substitutes for
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

traditional financial intermediaries such as banking, securities companies, insurance

companies, and exchange institutions. The application of big data to credit assess-
ment, network loans, securities investment, and actuarial services (see Chapters 7
and 12) mainly improves information processing chains in related financial activities.
Although these Internet financial institutions are more transparent, have lower
transaction costs, and allocate resources more efficiently, their financial capabilities
and risk characteristics do not differ much from traditional financial intermediaries.
Therefore, the regulatory frameworks for traditional financial intermediaries and
markets are still applicable; we need only add regulation for information technology
risks.15 Relevant laws and regulatory frameworks for China are on the respective
of the People’s Bank of China, China Banking Regulatory Commission, China
Securities Regulatory Commission, and China Insurance Regulatory Commission.
Second, China has established a comprehensive regulatory framework for mobile
and third-party payments. The specific laws are too numerous to list here, but can
be found on the website of the China Association for Payment and Clearing.
Third, protecting consumers` rights is the regulatory focus for online sales of
financial products, which needs strict control over problems such as misleading buyers,
exaggerated advertising and fraud. According to the CBRC’s “Securities Investment
Fund Sales Management” regulation, Article 35: “The fund publicity and promo-
tion materials must be true, accurate, and consistent with its contract and prospectus.
They may not contain false records, misleading statements, or major omissions; may
not forecast the performance of the fund’s securities investments; may not make a
commitment of gain or loss; ….must avoid exaggerated or one-sided marketing; can-
not use terms which imply that the investment is risk-free such as ‘safe,’ ‘guarantee,’
‘promise,’‘hedge,’‘secured,’‘high yield,’ and ‘no risk’ or emphasize a time limit for col-
lecting funds; and may not publish endorsements or recommendations from entities
or individuals.” The CBRC expressly forbids financial and trust products to promise
yields or forecast returns. Marketers must repeatedly warn investors that investors are
responsible for investment risk. In January 2014, the Zhejiang branch of the China
Securities Regulatory Commission issued the first violation for Internet wealth man-
agement products. Shumi Fund (Fund123) was found to be using improper wordings
like “maximum yield of 8.8%” in its materials and was required to correct them.
For cooperative products of “third party payment and money market funds”
represented by Yu’E Bao, in view of the possible liquidity risks, we should refer to
the US’s regulatory measures for money market fund after the international finan-
cial crisis:
First, cooperative products of “third party payment and money market funds”
are required to reveal risks to investors so as to avoid the incorrect assumption that
there can be no loss in money market funds. Article 43 of “Measures for the Sale of
Securities Investment Funds,” stipulates that: the marketing and recommendation
 The regulation of Internet finance 149
materials of a money market fund shall remind investors that their investment in the
fund is not the same as a bank deposit, and the fund manager does not guarantee
the profitability or minimum yield of the fund.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Second, cooperative products of “third party payments and money market fund”
are required to fully disclose information of fund distribution (including security
types, issuers, counterparties, amounts, duration, rating, and other dimensions) and
redemption.16
Third, cooperative products of “third party payment and money market funds”
should meet conditions like average duration, rating, and investment concentra-
tion to ensure that there are sufficient liquidity reserves to cope with large investor
withdrawal under stress scenarios.

2 Regulation of P2P network loans

Theoretically, if the loans operate purely in a platform mode (neither bear credit
risks associated with loans, nor conduct liquidity or maturity transformation), and
investor risks are adequately diversified, a P2P platform itself does not require pru-
dential supervision.
P2P network loans represented by Lending Club and Prosper have the following
characteristics (see Chapter 8 for more details):

1 There is no direct credit and debt between investors and borrowers. Investors
purchase bills (i.e. usufruct certificate) registered and issued by P2P platforms
according to US Securities Law, while loans to borrowers are first offered by
third-party banks, and then transferred to the P2P platform.
2 The relationship between bills and loans is like a mirror. Borrowers repay the
loan principle and interest, and P2P platforms pay holders of corresponding
bills in the same quantity.
3 If the borrower defaults on the loan, the holder of the corresponding notes
will not receive payment from P2P platforms (i.e., P2P platforms do not offer
guarantees for investors), but this does not constitute a breach of contract for
the P2P platform itself.
4 Personal credit is highly developed (such as FICO credit score). P2P platforms
do not have to carry out a much due diligence.

Under these circumstances, the regulation of the US Securities and Exchange

Commission (SEC) focuses on information disclosure rather than the operation of
P2P platforms. P2P platforms must constantly update information about each bill
in issue instructions, including terms of corresponding loans and anonymous bor-
rower information. We should pay particular attention to the flexibility of capital
instruments provided by the US Securities Act; usufruct certificates are not only the
core of P2P network operating framework, but also serve as the “starting point” for
regulation of P2P network loans.
 150 The regulation of Internet finance
P2P network loans in China have some unique characteristics, including the
following:
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

1 Online information is insufficient to meet credit assessment needs, therefore

underlying due diligence investigation are generally carried out.
2 Investors are used to fixed payment schemes, so it is difficult to attract inves-
tors without this structure. P2P lending platforms generally allocate part of the
revenue to risk reserve pools for the protection of investor principal.
3 The “professional lenders and debt transaction” model is aimed to better
connect the financial needs of borrowers and demands for investor funds. It
actively conducts business in large amounts rather than passively waiting for
their match.
4 A large number of underlying promotional activities are carried out, so it is
important to strengthen protection for consumers of financial products.

Overall, Chinese P2P network loans are more similar to private loans over the
Internet. Currently, Chinese P2P network loans exceed other countries in terms of
both the number of loan institutions and total loans issued. The “Chinese” process of
P2P network loans has produced many unique business models, operating systems
and potential risks. Some regulators have already expressed their concern with this
growing business. The deputy governor of the People’s Bank of China Liu Shiyu
(also the group leader of “Development and Regulation of Internet Finance” in the
State Council) stated on December 4, 2013 that central banks and financial regula-
tory authorities will cooperate with security organs and all levels of government to
deal a “heavy blow” to violators in order to promote the healthy development of
Internet finance. Illegal fund-raising and illegal deposits from the public are two red
lines which cannot be crossed. Capital pools are thus unacceptable, especially for P2P
platforms.
We believe that the regulation of P2P network loans should introduce
the following measures, whose core idea is “open access, tracing activity, and
post-issuance accountability.”

2.1 Market access regulation

1 Establish basic entry criteria. Directors, supervisors, and senior executives
should have certain financial knowledge and experience, which must be
reviewed. P2P platforms should be subject to minimum operating conditions.
For example, the IT infrastructure should safely store and manage customer
information and transaction records, in addition to running a solid risk man-
agement system.
2 Establish the principle that “people who set and approve an institution are
responsible for its supervision and risk evaluation.” (this is also the spirit of
the “Notice on Issues of Strengthening the Regulation of Shadow Banking” by the
Chinese State Council, released in 2013.)
 The regulation of Internet finance 151
2.2 Operational regulation
P2P Platforms:
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

1 Should be limited to the financial information services business and estab-

lish directly corresponding relationships between investors and borrowers.
However, P2P platforms cannot directly involve themselves in lending activi-
ties and are not allowed to go beyond a certain scope of operation even when
technical means improve.
2 Must accept regulatory standards for non-performing assets and capital if
they bear credit risks (such as through risk reserve pools). These are normally
applied to banks to ensure that the risk reserve pool has adequate risk absorp-
tion capacity. The core objective of this requirement is to adapt the business
scale of P2P platforms to their risk appetite, thus protecting the continued
viability of the business.
3 Should strictly separate own funds and client funds. Client accounts should be
managed by a third party fund (e.g. third-party payment agencies approved by
the People’s Bank of China). P2P platforms shall not appropriate client funds
in any way.
4 Should know their customers and take effective means to identify and authen-
ticate identities in order to prevent criminals from trading, fraud, finance scams,
money laundering, and other illegal activities.
5 Should establish qualified investor systems to ensure that investors have suffi-
cient financial knowledge, risk management capabilities, and financial capacity
to invest in P2P network loans (e.g. investors must meet certain income and
asset thresholds).
6 May not use false advertising.

2.3 Information regulation

P2P platforms should:

1 Save credit rating information (including application and credit assessment

data), matching customer information from the lending and borrowing sides,
repayment, and other transaction information.
2 Not tamper with lending information. If shareholders or staff of a P2P platform
raise finance on their own platform, they need to truthfully disclose this to
prevent conflicts of interest and related party transactions.
3 Fully fulfill the obligations to disclose risks to ensure that investors and
borrowers clearly know their rights and obligations (including the loan amount,
term, interest rates, service rates, repayment, etc.), and protect the customer’s
right to know and choose.
4 Truthfully disclose operating information, including corporate governance,
platform operational model (such as assessment method, matching mechanism
between lenders and borrowers, customer funds management system, and
 152 The regulation of Internet finance
guarantees), business data (such as transactions, the cumulative number of users,
the average single loan amount, investor earnings, NPL indicators, etc.) for
customer reference.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

5 Protect the security of customers’ information. They should not use this
information for purposes not related to the lending or borrowing activity
requested by the customer, and any such use should require customers’ express
authorization.

3 Public finance regulation

Because of current restrictions on the number of investors in the Securities Act
(not more than 200), public finance is mainly confined to “charitable donation” of
creative art projects, and does not extend to SMEs. They thus cannot set up plat-
forms to link SMEs and capital markets. Doing so will not produce insurmountable
financial risks. In the future, if China allows public finance to give investors returns
in the form of equity financing, the securities regulatory framework will require
some changes. In this regard, the US. JOBS Act is worth careful study. The bill is
based on the US’s developed capital markets and securities regulatory mechanisms.
It allocates the rights and obligations between all participants in public finance in
order to achieve a balance between investor protection and investment and financ-
ing promotion.
Finally, regulatory coordination for Internet finance is also an important issue.
At present, China has adopted the framework of “separate operation, separate
supervision” between banking, securities and insurance, while the regulatory
authority is highly concentrated in the central government. This approach may not
fit the Internet financial industry. For example, Internet sales of financial products,
bank financial products, securities investment products, funds, insurance products,
and trust products are sold at the same network platform.Yu’E Bao combines third-
party payments and money market funds, and is also involved in broad money
creation. In addition, large numbers of Internet financial institutions have emerged
with small, scattered scale and a variety of business models that may defy the
unified central financial regulation. Therefore, the allocation of responsibility for
issuing licenses, daily supervision, and risk disposal of Internet financial institutions
between different central government departments and between central and local
governments is a very complicated question.
In August 2013, the State Council of China established the “financial
supervisory coordination leading group” led by the People’s Bank of China, whose
responsibilities include the “coordination of cross-financial products and cross-
marketing financial innovation.” The Third Plenum document states a priority
for “improving regulatory coordination mechanisms” and “defining the duties of
financial regulation and responsibilities of risk disposal between central and local
government.” The Institutional framework for Internet financial supervision and
coordination has thus already taken shape.
 The regulation of Internet finance 153
Notes
1 Yan, Qingmin. 2013. “Research on Supervision over Information Technology Risks of
Banking Financial Institutions,” China Financial Publishing House.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2 For specific regulatory measures, interested readers can refer to the book written by Yan
Qingmin mentioned on the previous page.
3 More discussion about “free-rider” issues can be found in Chapter 9.
4 Risks of losses caused by adverse changes of market prices.
5 Risks of losses caused by failure of debtors to fulfill debt obligations.
6 Risks of being unable to obtain sufficient funds timely or at reasonable cost to compensate
for with asset growth or paying debts due.
7 Risks of losses caused by inadequate or failed internal processes, personnel and
information technology systems, or external events.
8 Risks of negative evaluation for financial institutions from stakeholders caused by
inappropriate management, administration, and other activities or external events.
9 Risks caused by financial institutions failing to comply with laws, regulations, regulatory
requirements, rules made by self-regulatory organizations, or codes of conduct applicable
to financial institutions in their business activities, and may be subject to legal sanctions
or regulatory sanctions, material financial loss or loss of reputation.
10 This was discussed in more details in the section on IT risks.
11 Prudential regulation can be divided into two categories: microprudential supervision
and macroprudential supervision. Microprudential supervision refers to regulation for
safety and soundness of single Internet financial institutions; macroprudential supervision
is aimed to the regulatory impact of the Internet for safety and soundness of the financial
system and the real economy.
12 For more information about the banking regulatory measures involved, interested
readers can refer to: Xie Ping, Zou Chuanwei. 2013. “Basic Theory of Macroprudential
Supercision of Banks,” China Financial Publishing House.
13 Xie Ping. 2010. “Consumer-Oriented Financial Regulation,” New Century.
14 Chapter 5 discussed the currency regulation of Internet (especially Bitcoin).
15 China always has reservations about various of financial trading platforms hold by
non-central-government institutions. In 2011, the State Council has made policies of
“Decision on Straighten Out All Kinds of Trading Venues and Effectively Guard Against
Financial Risks” (i.e. “No. 38”).
16 Not necessarily the details of each position.
 11 Internet exchange economy
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

The exchange economy is the economic concept that describes exchange after
products are produced and find themselves in the hands of owners. The problem
then becomes the allocation of these goods between different people.The exchange
economy abstracts away the process of production and consumption from eco-
nomic activities, and its foundation is the different resource endowments or divi-
sion of work from person to person. It is essentially everywhere, from goods trade
markets to art auctions. The Internet-based exchange economy is a concept we
propose to highlight the influence that the Internet has on patterns of exchange,
which represents both e-commerce and the sharing economy.
In this book on Internet finance, we discuss the Internet-based exchange econ-
omy in a single chapter because there is a close connection between two con-
cepts. First, Internet finance can be regarded as a special case of the Internet-based
exchange economy, and some perspectives in this chapter are actually complemen-
tary to Chapter 2. Second, Internet finance can quickly apply its innovations to
the Internet-based economy, which also helps drive the development of Internet
finance itself.We study the Internet-based exchange economy mainly by discussing
some basic economic concepts in the context of the Internet: such as preference,
utility, market, exchange, and resource allocation, which help us understand the
mechanisms of Internet finance.

SECTION 1: ANALYSIS OF THE SHARING ECONOMY

1 Definition of sharing economy

In this section, we focus on the sharing economy, another representation of the Internet-
based exchange economy, since there has been a flood of literature on e-commerce.
The sharing economy refers to an economy and social system where commodities,
services, data, and intelligence can all be shared. Even though there are various forms of
the sharing economy, one thing they have in common is the use of information tech-
nology. In the sharing economy, it is possible to share and reuse surplus or idle goods
and services as long as market participants have the requisite information.
 Internet exchange economy 155
Rachel Botsman and Roo Rogers fully introduce the sharing economy in a
2010 monograph in which they classify the sharing economy into three types. The
first is product service systems that focus on utilization of idle resources, which
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

therefore pay more attention to resource accessibility than to ownership. Examples

include Airbnb (an online platform for house rental) and ZipCar (an online plat-
form for car rentals). The second is redistribution markets, which focus on the
reuse or recycling of old commodities with environmental goals and a reduction
in overconsumption. Examples include Freecycle, an online exchange platform of
used commodities, and thredUP, an online exchange platform for used clothes.
The third type is collaborative lifestyles, which focuses on trades between friends
or neighbors with the abandonment of impersonal trade and the return to nature.
Examples include TaskRabbit and Etsy, online distribution platforms for small tasks
and self-produced goods. Additionally, R Botsman and R Rogers also consider P2P
network loans and crowdfunding as a type of sharing economy.

2 Case analyses
We will primarily introduce four cases in this section, Airbnb, Zipcar, TaskRabbit,
and BarterCard. From the logistics demands of exchanging, Airbnb, ZipCar,
TaskRabbit just represent three different types. On Airbnb, houses (i.e. the
exchangeable goods) cannot move. On TaskRabbit, little tasks can be delivered over
the Internet. However, BarterCard represents bartering in the Internet era.

2.1 Airbnb
Airbnb, founded in 2008 and headquartered in San Francisco, focuses on home
exchange. It offers an online service platform for homeowners to rent out unused liv-
ing space (including the whole house, single rooms, beds, boats, and even tree rooms)
over a short period to tourists who travel to cities in which the homeowners live.
The business model of Airbnb is much more like the hotel industry than an ordinary
housing rental intermediary. Moreover, landlords can benefit from otherwise idle res-
idences through Airbnb while tenants can get a cheaper and more personalized hous-
ing experience than they can with hotels. By September 2013, users of Airbnb have
come from 192 countries and 33,000 cities, with more than 500,000 total rentals.
Landlords and tenants need to register and create a profile on Airbnb.com. It
encourages landlords and tenants to validate their identities and evaluate each other,
so as to establish their reputations for the site. The identification process may be
completed either by scanning ID cards or associating with websites such as Google,
Facebook or LinkedIn. Additionally, landlords must supply basic information and
photos of the housing. They can make the price flexible through negotiations with
potential renters or reduce rates for longer leases.
Potential tenants search Airbnb by city, travel dates, and specific housing
requirements (such as types of housing, prices, sizes, surroundings, facilities, and
acceptance of pets.), Airbnb then returns the best match to potential tenants.
 156 Internet exchange economy
Airbnb also set a response rate index system for each supplier to measure their
responsiveness to tenants’ questions and booking applications.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2.2 ZipCar
ZipCar is an online car-sharing company founded in 2000 with business in cities and
college campuses in the United States, Canada, Spain, Australia, and other countries.
ZipCar both manages and owns cars. People who want to rent cars must first
become members of ZipCar, and then obtain membership cards with antenna chips
embedded inside. As of July 2013, ZipCar has 10,000 cars and 810,000 mem-
bers. Each ZipCar is equipped with Radio frequency identification (RDIF) tran-
sponders to connect membership cards. ZipCar’s central system tracks car positions,
members’ rental periods and distances.
ZipCars are parked at special parking lots near residential areas. Members can
reserve cars online or over the phone at any time in any participating city, either
to use right now or up to one year later. After receiving the reservation, ZipCar
offers car profiles and prices on the e-map for members to choose according to the
distance between cars and members. Members obtain the cars from special parking
lots. At the designated reservation time, membership cards will activate. This allows
the member to enter and start the car. In order to protect members’ privacy, ZipCar
does not track cars during the rental period, but the car security systems remain
activated. Members must return cars to the original parking places within the res-
ervation time and lock them with their membership cards.
ZipCar’s completely self-service rental model not only reduces labor costs, but
also gives consumers more choice. The fee structure includes an application fee, an
annual fee and a reservation fee, for each use. Cars are paid by the hour, including
gas and insurance. Electronic bills are conveniently sent by ZipCar over the Internet
and paid automatically.

2.3 TaskRabbit
TaskRabbit, founded in 2008, is an online post and claim tasks community with
the concept of “Do More, Live More, Be More.” The platform allows some users
called TaskPosters to distribute small tasks to others called TaskRabbits.TaskRabbits
are then paid after completing tasks from the TaskPosters. On the website of
TaskRabbit, TaskPosters can solve many problems at low prices. On the other
hand, TaskRabbits get opportunities to show their abilities. TaskRabbit’s business
is mainly located in big cities around the east and west coast of the US including
Boston, Chicago, New York, San Francisco, Los Angeles. TaskPosters can post just
about any task on TaskRabbit, including everything from installing furniture to dog
walking. They just describe the tasks and set price caps based on the prices paid for
similar ones.
The other side of the equation is the TaskRabbits, who offer help. They are
mainly retirees or full-time parents who have the time and ability to fulfill tasks.
 Internet exchange economy 157
If one wants to become a TaskRabbit, he or she must apply on the website,
including a video interview and criminal background check. Those who pass will
be distributed to different communities depending on their location and skills.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

TaskPosters select the most suitable TaskRabbit, depending on applicant skills,

fee, and so on. After the tasks, TaskPosters evaluate the work of the TaskRabbit.
To motivate TaskRabbits, the website designs an incentive “point” mechanism.
TaskRabbits are ranked by their points according to the accomplishment of tasks.
The website displays a ranking list to show those TaskRabbits with the most points
and best evaluations.
The website of TaskRabbit also provides a “Deliver Now” service. For example,
if a TaskPoster needs someone to help buy some food or get clothes from the
laundry immediately, the website will provide tools for the TaskPoster to help track
the location of the TaskRabbit after a TaskRabbit claims the task.The website earns
a 13%–30% referral commission for each completed task.

2.4 Bartercard
Bartercard mainly operates in Australia, New Zealand, the United Kingdom, the
United States, and Thailand. It currently has 55,000 cardholders throughout the
world, who together trade in excess of US$60 million per month. Members earn
Bartercard Trade Dollars for the goods and services they sell. This value is recorded
electronically in the member’s account database or goes toward repaying credit that
the member may have used. Bartercard provides services like billing, matching, and
providing monthly statements. It makes its money through service fees on transac-
tions between its members.
There are three main reasons why barter exchanges like Bartercard can still thrive
in modern society. First, the development of Internet technology expands the scope
of barter and promotes the “double coincidence of wants” (see Section2 1.2 below).
Second, Bartercard uses credit as a medium of exchange. In fact, the use of credit in
barter has a very long history. For example, reciprocity among relatives and friends
can be regarded as a form of barter. If Alice gives Bob a gift or does Bob a favor, Bob
will not immediately pay Alice back with money. Instead, Bob will think of himself
as owning Alice a favor. He will then give a gift to Alice or return Alice a favor
afterward. Here the “favor” can be regarded as a kind of credit. Third, this Internet-
based barter has a background in environmental protection and consumer culture.

SECTION 2: PRINCIPLES OF THE INTERNET-BASED

EXCHANGE ECONOMY

1 Basic framework
Combining e-commerce and the sharing economy, we consider three key elements
of the Internet-based exchange economy: the exchanged commodity, the medium
of exchange, and those who undertake the exchange.
 158 Internet exchange economy
1.1 Exchanged commodity
The exchanged commodity is the commodity which is being exchanged. It can be
one single item or multiple items.What is most important is the commodity’s form
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

of existence, its divisibility, the form of property transfer, and whether it is a public
good or a private good.

1.1.1 Form
Exchanged commodities can be classified into four forms:The first is substantial and
electronic goods, such as houses, cars, books, electronics, clothing, household goods,
music, and videos. The second includes services such as health care, education, and
tasks on TaskRabbit. The third is information, such as news and knowledge. The
fourth includes rights, such as ownership, use rights, and operation rights. Because
stocks, bonds, loans, and other financial instruments are in essence claims on an
individual or an organization (see Chapter 1), they also count as rights.

1.1.2 Divisibility
According to the divisibility concept, exchanged commodities fall into two
categories. The first is indivisible commodities, which must be exchanged as a
whole. When merchandise is being exchanged, it is usually indivisible. The other
type is composed of divisible commodities. When financial resources are being
exchanged, they can be divided into a number of small and homogeneous shares.
For example, the sale of financial products on the Internet (Chapter 3), notes of P2P
loans (Chapter 8), and equities in crowdfunding (Chapter 9).

1.1.2.1 FORM OF PROPERTY TRANSFER

The two main forms of property transfer are exchanges which are accompanied
by the transfer of ownership, and exchanges in which rights to use or to operate
are transferred but the ownership remains the same. In the sharing economy, most
exchanged commodities belong to this category. For example, landlords on Airbnb
own the houses and ZipCar owns the cars.Tenants on Airbnb and ZipCar members
have only temporary access to the houses or the cars.

1.1.2.2 PUBLIC OR PRIVATE GOODS

Public and private goods are two opposing concepts in economics. In simple terms,
public goods can be used by a group of people, while private goods can only be
used by one person at any time. A person can only consume items. To distinguish
these two more rigorously, we need to introduce two concepts: exclusivity and
rivalry. A good is called exclusive if it is possible to prevent other people from
having access to it. A good is rivalrous if consumption by one consumer prevents
simultaneous consumption by other consumers.
 Internet exchange economy 159
Thus, exchanged commodities include public goods that are both non-exclu-
sive and non-rivalrous, such as national defense, television and radio broadcast-
ing, and clean air. Public goods rarely become exchanged commodities. They also
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

include private goods that are both exclusive and rivalrous. Most exchange com-
modities are private goods, such as commodities, financial resources, and health care.
Additionally, many exchanged commodities fall between public goods and private
goods. For example, some music, videos, information, educational resources, and so
forth on the Internet are exclusive but non-rivalrous.These are called “club goods.”

1.2 Medium of exchange

Most Internet-based economies need to use currency as the medium of exchange.
Both fiat money issued by central banks and Internet currency may serve this func-
tion (see Chapter 5). This is, however, not always true. Some Internet-based econo-
mies reject this and use barter, thus taking up one of the earliest, pre-fiat money
human means of exchange. This can be problematic and even break down unless
both parties possess something the other desires, the so-called “double coincidence
of wants.”
Some online companies have used innovative approaches to overcome these
issues. For example, websites such as YouTube offer free digital goods but insert
advertisements into them. Each party has something the other wants, in this case
digital goods and the attention of the viewer. These transactions flow effortlessly,
with no exchange of money, by involving a third-party advertiser that provides funds.

1.3 Parties to the exchange

An Exchange requires a supplier and a consumer and can be classified into three
types based on the quantity of each:

1.3.1 One-to-one
One-to-one means there is only one supplier and one demander. This type rarely
occurs in the Internet exchange economy.

1.3.2 One-to-many
In this type, there can be either one supplier and many demanders or vice versa.
This is a more common form of Internet-based exchange than one-to-one, but is
less common than many-to-many.

1.3.3 Many-to-many
Most Internet-based economies thrive in this form. There are many sup-
pliers and demanders interacting together in a market that can shift into
 160 Internet exchange economy
one-to-many transactions. A P2P lending platform clearly illustrates this
phenomenon. Multiple investors face multiple potential investment projects
and are thus clearly many-to-many. From the perspective of a single project, it is
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

one-to-many, with one project and many funders. From that of an individual inves-
tor, it is a many-to-one with one funder and many projects.
Different types of exchanges have very different modes of resource allocation.
One-to-one functions like negotiation; one-to-many is similar to an auction; and
for many-to-many, there are many different modes, such as the Edgeworth Box and
stable matching. These forms involve very complex economic problems that we
discussed in detail in a later section.

1.4 Three pillars of the Internet-based exchange economy

The Internet-based exchange economy is incredibly diverse. There are many dif-
ferent combinations of the exchanged commodity, the medium of exchange, and
actors who undertake the exchange. Just like Internet finance, we believe this
unique economy can be understood as standing on three pillars.
The first pillar is logistics and payment. This is the physical process of complet-
ing the exchange. The second pillar is information processing, including identi-
fication and selection of exchanged commodities (types and amounts) and the
other participants. The third pillar is the allocation of resources. The core issue is
to design a mechanism that is able to match supply and demand efficiently based
on a comprehensive consideration of the participants’ endowments, preferences,
and utility.

2 Logistics and payment1

Logistics is to the Internet-based exchange economy as payment is to the Internet
finance. Exchanged commodities in different forms have different logistical
requirements. Some Internet-based exchange economies have little or non-existent
logistic requirements. Electronic products, information, and financial products are
transferred, paid for, and consumed online. Traditional logistics are unnecessary.
When services and rights are being exchanged, though the final consumption
is physical, the logistics are mobile and are usually embedded in the exchange pro-
cess. For example, the houses on Airbnb cannot be moved, but the arrangements
can be done from anywhere with an Internet connection. These core processes of
exchange are all online, so logistics costs are minimal.
However, in cases in which exchanged commodities are physical goods, logistics
play a key role in delivery and cost. Chinese E–commerce is a perfect example.
Alibaba has developed a symbiotic relationship with the logistics industry due to its
massive employment of logistics resources to deliver products ordered on its sites.
In 2013, Alibaba co-founded “CaiNiao,” or Rookie Logistics, to create a massive
open social logistics platform in five to eight years that can reach any place in China
within 24 hours. In order to ensure its delivery speed and consumer experience,
Jingdong Mall (JD.com) chose to build a logistics network itself. Another example
 Internet exchange economy 161
is Amazon. Amazon started testing delivery with drones in late 2013. These drones
can deliver packages up to 2.3 kilograms (86% of Amazon’s packages would qualify)
in half an hour door-to-door. These examples show how closely the e-commerce
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

and logistics industry are connected. Logistics give rise to relatively large transac-
tion costs for e-commerce, but new technology has the potential to significantly
reduce these costs.
One of these new technologies is intelligent logistics, which uses sensor tech-
nologies like RDIF to track the location of every piece of merchandise in real time,
feeding back information that includes aspects of the goods’ location and status.
Intelligent logistics reduce transaction costs by optimizing the geographical distri-
bution of warehouses so as to better match customer needs. It can also save costs by
optimizing warehouse management, minimizing inventory, using less capital, and
increasing inbound, storage, and outbound efficiency. Optimization of the trans-
portation routes with these methods can also reduce transportation time and cost.
The second technology is 3D printing, which allows for a new type of “additive”
rather than “subtractive” manufacturing. It prints objects layer by layer using digital
files as the model and powdered metal or plastic as the material. 3D printing digi-
talizes physical commodities, which allows the creation of more personalized and
customized products. E-commerce of the future will allow consumers to purchase
the plans for products, and then print them out nearby or even at home. 3D print-
ing will make exchange of physical goods become like the exchange of electronic
goods today. It can be carried out on the Internet, significantly reducing or even
eliminating the demand for logistics. We can see that Intelligent logistics and 3D
printing together will both lower costs and bring the exchange economy of physi-
cal commodities increasingly online.

3 Information processing
Just as we saw in Internet finance, big data is the core information processing
technology in the Internet-based exchange economy. Since data usage makes no
fundamental distinction between finance and real economy, both can use the same
data analysis tools.
The Internet-based exchange economy generates large amounts of informa-
tion. One key type of information involves preference and utility, two fundamental
economic concepts. Preference shows how much consumers enjoy goods, mainly
reflecting consumer priority when selecting among several goods. Utility is a quan-
tification of preference that measures how consumption satisfies consumer needs
and wants. We will illustrate these two concepts with some examples.
Investors allocate their funds, subject to a budget constraint, among various
financial products, such as crowdfunding, P2P network loans, or traditional financial
products. Their preferences are, however, not uniform For example, young people
tend to select stocks due to their high expected return while the old prefer bonds
because of the increased stability of payment and lower volatility. We can express
their preferences with utility functions that describe the risk-return characteristics
of financial products. The most important parameter is investor risk aversion.
 162 Internet exchange economy
In e-commerce, every consumer can choose between many goods. For instance,
if one wants to buy a book on Internet finance online, he/she will make the pur-
chasing decision based on the contents, authors, price, reviews, and many other
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

factors. Websites like Amazon and Dangdang use browsing and purchasing histories
to predict consumer preferences, according to which they recommend books (see
Chapter 6).
We can similarly understand decision making in sharing economy. For example,
Airbnb helps tenants and landlords select each other according to certain prefer-
ences. ZipCar lets members select cars according to distance, price, and model. The
ultimate goal of information processing is to discover preference and increase the
utility of exchanges.The essential problem is information asymmetry.2 For instance,
the realized return for financial products is not knowable ex-ante. Book customers
are not usually able to read the whole book themselves before purchase. Property
owners on Airbnb cannot be sure how well prospective tenants will take care of
their property. Such examples abound in all exchanges, online and off, meaning
that virtually all decisions are made in the presence of asymmetric information.
Mitigating asymmetric information can thus reduce non-beneficial exchanges and
provide more efficient resource allocations.

4 Resource allocation
The core of resource allocation is to increase utility by matching demand and supply.
First, it is important to design a matching mechanism based on the set of transactors.
Generally, more transactors lead to a more efficient allocation of resources through
a diversification of supply and demand. Following the logic outlined in Chapter
2, we can prove that the set of possible exchanges will increase as transaction cost
and/or information asymmetry decline. For instance, strangers are able to lend to
each other through P2P network loans because of the innovative credit evaluation
system. It is possible now to stay in strangers’ homes because of the trust we place in
the evaluation system provided by Airbnb. These examples show that the Internet
exchange economy expands the range of possible transactions by making it possible
to accept recommendations from people other than friends and family. Below we
introduce four representative supply and demand matching mechanisms.

4.1 Auction3
The auction is one of the oldest market mechanisms. Auctions differ based on two
main parameters: open- vs. sealed-bid and ascending vs. descending price. In a
sealed bid auction, bidders submit their prices simultaneously without knowledge
of others’ bid amounts. In an open auction, bidders bid directly against each other
until the final sale price is reached with the highest bid. In ascending-price auc-
tions, the auctioneer continues raising the price until the highest bid emerges and
the sale concludes. In a descending-price auction, the auctioneer begins with a high
asking price. This price is then lowered until a participant is willing to accept the
auctioneer’s price.
 Internet exchange economy 163
There are four primary forms of auction used today. English auctions, the most
common form of auction, are open ascending-price auctions. Companies such as
eBay use this form. Dutch auctions are open, descending-price auctions.TaskRabbit
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

takes this form, as TaskPosters select the lowest price among TaskRabbits with the
same skills. In a first-price sealed auction, all bidders simultaneously submit sealed
bids. The highest bidder wins and pays the price they submitted. A second-price
sealed or “Vickery auction” is identical to the first-price sealed auction except that
the winning bidder needs to pay the second highest bid.
Bidders offer prices based on their valuation of the good, which in turn is a
function of the utility they expect to gain from its ownership. If a bidder knows
his/her expected value, the auction is termed “private value.” In this case, a bidder
is not able to accurately know or influence others’ valuations. Most Internet-based
exchange auctions are in the private value category. Another kind of auction is
common value auction, which means that the item’s value is the same to all the bid-
ders, but the bidders are unsure of the value. For example, in an offshore oil-license
auction, bidders have different valuations based on their exploration, but no matter
who the winner is, the oil license has the same market value because oil reserves
are certain.

4.2 Edgeworth box4

The Edgeworth Box applies to multiple, separable, many-to-many exchanges, such
as Bartercard. Each transactor has an endowment of certain consumer goods. Each
exchanger has individual preference for the consumption group, and he only cares
about his own utility. Every exchanger can choose to consume by himself or vol-
unteer to exchange with others. This kind of trade is the only way to reallocate
the initial resource endowment. The Edgeworth box results correspond to a Pareto
efficient equilibrium resource allocation.5
We generally develop an Edgeworth box with two exchangers and two types
of exchange. Here we use “A” and “B” to denote exchangers and “x” and “y” to
denote two items for exchange. At first, A’s initial endowment is (ω Aχ, ω Aχ ),6 and B’s
initial endowment is (ω Bx ,ω By ).Thus, their total endowment is (ω Ax + ω Bx ,ω Ay + ω By ),
which is marked with a blue star in Figure 11.1.
In Figure 11.1, the horizontal axis represents the quantity of x with length of
ω Aχ + ω Bχ , while the vertical axis represents y with length of ω Ay + ω By. Exchanger
A’s original state is on the lower left of the box, from which the distance upward
shows his endowment or consumption of “x.” Exchanger B’s original state is on the
top right corner, from which the distance downward represents his endowment or
consumption of y. The beauty of the Edgeworth box is that no matter how they
exchange, the total consumption of two exchanged items equals their total endow-
ment (ω Aχ + ω Bχ , ω Ay + ω By ), therefore the exchange between the two people can
always be reflected by points in Edgeworth box.
In an Edgeworth box, individual utility is depicted with indifference curves.
Under the classical assumption of utility in economics, indifference curves are
 164 Internet exchange economy
ωB x Barbara

Y
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

ωB y

Endowment
ωAy

Aaron ωAx X

Figure 11.1 Edgeworth box.7

convex to the origin.The utility declines when indifference curves get further away
from the origin. Our two exchangers correspond to two sets of indifference curves.
The equilibrium resource allocation is marked with a red star in the Figure 11.1.
Three lines are tangent at this point: the line from the red-star point to the ini-
tial endowment point (blue line in Figure 11.1), A’s indifference curve and B’s
indifference curve. It is Pareto efficient because A and B’s indifference curves are
tangential at this location. At any place other than this point, possible exchanges
lower either A’s utility or B’s utility or both. Therefore, each exchanger achieves
maximum utility. The slope of the blue line also shows the equilibrium price for
the two items.

4.3 Stable matching

Stable matching applies to the Internet-based exchange economy in the context
of two-way choices. For example, at Airbnb.com, landlords and tenants both make
choices. Now we explain the concept and mechanism of stable matching through
the lens of Airbnb.
We assume that the quantity of landlords and tenants are the same. Every land-
lord can sort tenants according to his preference and vice versa. To simplify the
problem, let us suppose that landlords believe it is better to rent out their houses
rather than leave them idle, no matter to whom they rent. Also, tenants believe it is
better for them to have a room to rent than no place to live, no matter who provides
the room.
There is then a one-to-one correspondence between landlords and tenants.This
kind of correspondence is called matching. Stable matching means that both the
landlords and the tenants believe that the combinations between them are the best
under the present conditions, and are thus Pareto efficient.
 Internet exchange economy 165
Stable matching exists all the time as long as certain assumptions are made. The
most famous matching mechanism is the deferred acceptance algorithm, as shown
in the following steps:
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

1 Each tenant applies to his favorite landlord for a reservation.

2 Each landlord examines the reservations he has received. Some may receive
more than one application while some may receive none.
3 Those landlords who have received reservations select their favorite tenants and
reject others. These landlords then no longer accept reservation applications.
4 The rejected tenants apply to their second-choice landlords for reservations
according to their own preferences.
5 Steps 2–4 repeat until all landlords and tenants find a match.

In the deferred acceptance algorithm, the power of the initial choice first influences
both the final matching results and the welfare effects.

4.4 Markowitz mean-variance model8

Investors build portfolios by allocating funds between financial products to maximize
their expected utility. The most important asset allocation model is the Markowitz
mean-variance model, in which fund allocation depends on the expected rate of
return and volatility. It assumes investors’ utility functions are quadratic, or the rates
of return for financial products are normally distributed.
Suppose there are n types of financial products A1, A2, …,An (excluding risk-free
assets). Within the given investment horizon, the rate of return of Ai has random
variance r i. The Markowitz mean-variance model solves the following problem: if
the expected rate of return of portfolio μ is given, find the weight of each financial
product w = ( w1 , w 2 ,..., w n ) to minimum the variance of the rate of return σ w2 .
'

Assume that the rate of return and the covariance of financial products are
μ = ( μ1 , μ2 ,..., μn ) , Σ = (σ ij ) , in which μi = E ( ri ), σ ij = cov ( ri , r j ). The
'

i , j =1,2,...,n
optimization problem becomes:
n
min
w
σ w2 = w ' Σw = ∑σ ij wi w j
i, j =1

n n (11.1)
s.t. w 'e = ∑w i = 1, w ' μ = ∑w μ i i = μ
i =1 i =1

in which e = (1,1,...,1)'.
We can prove that if the expected rates of return for n kinds of financial products
μi are all different, then the covariance matrix Σ is positive definite. If the expected
rate of return of portfolio μ is given, then Equation (11.1) has a unique solution:

⎛ μ ⎞
w = Σ −1 ( μ , e) A −1 ⎜ ⎟ (11.2)
⎝ 1 ⎠
 166 Internet exchange economy
in which A = ( μ, e ) Σ −1 ( μ, e ).
'

⎛ a b ⎞ ⎛ μ ' Σ −1 μ e ' Σ −1 μ ⎞
Denote A = ⎜ ⎟ = ⎜ ⎟ , then the expected rate of
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

⎝ b c ⎠ ⎜⎝ e ' Σ −1μ e ' Σ −1e ⎟⎠

2
return of the portfolio μ and the minimum variance σ satisfy:
2
2 a − 2b μ + c μ
σ = (11.3)
ac − b 2
With risk on the horizontal axis and the rate of return on the vertical axis,
we draw the graph of σ and μ (Figure 11.2) to obtain the portfolio frontier, here
expressed by the right branch of the hyperbola. The hyperbola’s apex corresponds
Σ −1e
to the minimum variance portfolio wG = T −1 , whose expected rate of return
e Σ e
eT Σ −1 μ 1
is μG = T −1 , and whose standard deviation is σ G = . The upper half
e Σ e eT Σ −1e
of the hyperbola is called the efficient frontier, and each point on it constitutes a
portfolio, which is called an efficient portfolio. The bottom half of the hyperbola is
the inefficient frontier.

0.1

0.09

0.08

0.07

Efficient frontier
Rate of return

0.06

0.05

0.04 The minimum

variance portfolio
0.03

0.02
Inefficient frontier
0.01

0
0.05 0.1 0.15
Deviation

Figure 11.2 Markowitz mean-variance model.9

 Internet exchange economy 167
SECTION 3: THE RELATIONSHIP BETWEEN THE
INTERNET-BASED EXCHANGE ECONOMY AND
INTERNET FINANCE
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

1 Internet finance: a special case

Internet finance is in essence the exchange of financial resources between indi-
viduals or organizations. It is within the realm of the exchange economy because
it does not refer to specific production and consumption processes. The Internet-
based exchange economy and Internet finance reflect the impact of the Internet on
real economic and financial activities, particularly in three areas: physical exchange
(logistics and payment), the processing of exchange-relevant information, and
resource allocation mechanisms. Thus, we can apply the same analytical framework
to both the Internet-based exchange economy and Internet finance.
Internet finance has unique characteristics. One of them is that the financial
resources are claims on an individual or an organization and thus do not require
any attached physical objects or logistics. Second, it can be broken down into small,
homogeneous shares. Third, the transfer of ownership usually accompanies the
exchange. Last, financial resources are private goods.
It is not an exaggeration to say that the impact of the Internet is all encom-
passing. It has already disrupted telecommunications, journalism, book publishing,
broadcasting, music, and retail businesses. In the future, we may see its disruptive
impact on more industries such as cinema, education, and advertising. These com-
plex phenomena share the logic inherent to the Internet-based exchange economy.

2 Internet finance derives from the Internet-based

exchange economy
E-commerce exemplifies the derivation of Internet finance. For example, in order to
promote online shopping and improve the consumer experience, Alibaba developed
Alipay to solve the payments problem (see Chapter 4), and then opened up a microfi-
nance business with the data it accumulated online (see Chapter 7). It finally developed
Yu’E Bao (see Chapter 3) to make use of Alipay account balances and meet consum-
ers’ financial needs. Yu’E Bao is successful because it connects financial products and
payment instruments, improving capital efficiency without affecting payment effi-
ciency. Alibaba’s financial innovation shows that the real economy is the foundation of
Internet finance.Without the real economy, Internet finance is like a fish out of water.
Internet finance has even more potential as the Internet-based exchange
economy develops. Internet finance can then promote the Internet-based exchange
economy and form a virtuous circle through deep integration.

3 The Internet-based exchange economy serves

Internet finance
Credit collection and Internet lending based on big data (see Chapter 7) use the
records of the Internet-based exchange economy (such as e-commerce) to assess
creditworthiness. The core here is the application of behavioral data.
 168 Internet exchange economy
Traditional credit quality assessment is based on financial statement analysis,
mainly focusing on the gearing ratio, debt-to-income ratio, ratio of interest income,
and a few other indicators. This assumes that a default is triggered by insolvency,
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

income insufficient to repay the debt, and other situations.The basis of this approach
is that credit risks have a clearly identified causal chain.
However, in many real cases, the causal chain of credit risks is not so clear (especially
ex-ante). Also, in a rapidly changing market, causal analysis is not conducive to
capturing changes in credit risks. More importantly, financial statements only reflect
a small portion of an individual or an organization’s information, while a large
amount of their behavioral information on the Internet remains unused. Sometimes
it is hard to see if there is a causal relationship between behavioral information
and credit risks. The key of behavioral information application is to predict, and
these predictions are based on correlation, not causation (see Chapter 6).Therefore,
as long as the behavioral information on credit risks has predictive abilities, the
behavioral information is certainly valuable to the pragmatic evaluator.
In credit risk management, analysis methods based on causality are called
structural approaches, represented by the Merton model. Analysis methods based
on behavioral information are called simple approaches, such as the credit default
swap (CDS) model, the Logit model, the Bayesian Criterion (see Chapter 7), and
the default density model (introduced in Chapter 12). Simple approaches turn out
to be superior to structured approaches when applied in practice, and are thus
becoming more mainstream.10

Notes
1 Since Chapter 4 is devoted to payments, so we will not repeat the discussions here.
2 Information asymmetry typically arises when the buyer is unable to obtain all relevant
information about the product he is buying, which can lead to situations in which the
seller takes advantage of this informational advantage.
3 Data sources: (a) Varian, Hal R. 2009. “Intermediate Microeconomics: A Modern
Approach,” 8th ed., W. W. Norton & Company, Inc.; (b) Wolfstetter, Elmar. 1999. “Topics
in Microeconomics: Industrial Organization, Auctions and Incentives,” Cambridge
University Press.
4 Data sources: (a) Ping, Xinqiao. 2001. “Eighteen Lessons of Microeconomics,” Peking
University Press. (b) Varian, Hal R. 2009. “Intermediate Microeconomics: A Modern
Approach,” 8th ed., W. W. Norton & Company, Inc.
5 Pareto optimality implies a state in which no party can be made better off without
making another worse off.
6 A has x of ω Aχ unit and y of ω Ay unit.
7 The graph is selected from a public lecture of Hal R.Varian.
8 Huang, Chi-fu, and Robert H. Litzenberger. 1988. “Foundations for Financial
Economics,” Elsevier Science Publishing Co., Inc.
9 Figures in the graph are for demonstration purposes only.
10 See Duffie, Darrell, and Kenneth Singleton. 2003. “Credit Risk: Pricing, Measurement,
and Management,” Princeton University Press.
 12 Issues requiring further research
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

In this book, we have already outlined the principles and basic types of Internet
finance. However, much of Internet finance is still open. There are many unde-
veloped areas and unanswered questions that deserve thorough discussion. In this
chapter, we focus on big data’s utilization in two areas: securities investment and
actuarial science. We both share our understanding on these two questions and
welcome interested readers to undertake further research that could lead to future
breakthroughs.

SECTION 1: UTILIZING BIG DATA FOR SECURITIES

INVESTMENT

Utilizing big data in securities investment has always been a sensitive topic. If the
securities market is fully functional, then the price should fully reflect all informa-
tion available to investors. In this case, choosing any security would only yield
a gain equal to the investment risk. In reality, however, investors seeking excess
returns (alpha), tend to perform active portfolio management and diverge from the
benchmark index. This indicates that investors do not believe the securities market
is fully efficient. Investors create value by conducting market studies to form dis-
tinctive judgments and decisions. Insider information, however, is not part of big
data. Utilizing big data in securities investment is very different from insider trading.
We will start by introducing two famous models in active portfolio management—
the Black–Litterman model and the fundamental law of active management to
show how information affects decision-making. Afterwards, we will discuss how big
data plays an important role in securities investment.

1 The Black–Litterman model

The Black–Litterman model was first proposed in 1992 by Fisher Black and
Robert Litterman. Fisher Black was renowned for his Black—Scholes formula,
while Robert Litterman was working at Goldman Sachs Asset Management.
The Black–Litterman model improved on several shortcomings of the
Markowitz mean-variance model (See Chapter 11).The Markowitz mean-variance
 170 Issues requiring further research
model has three main weaknesses: the portfolio is often non-intuitive or highly
concentrated, the model is very sensitive to the change of variable inputs, and the
parameter estimation error is often magnified (estimation error maximization).The
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Black–Litterman model uses a Bayesian method to combine the market’s equi-

librium return on assets (ROA) and the investors’ subject view of the ROA. The
resulting estimation of the ROA is used to construct the optimal portfolio. The
Black–Litterman model is now increasingly accepted on Wall Street.

1.1 Reverse optimization

In the Black–Litterman model, the optimal portfolio is the result of the following
optimization problem1.

max w ′μ − λ w ′Σw/2 (12.1)

w

With w being the weight of the assets, μ being the expected excess return on
assets over the risk-free rate, Σ being the covariance matrix of the excess return, λ
being the coefficient of absolute risk aversion.
Without any constraints, when solving for w , we obtain:
1 −1
w = Σ μ (12.2)
λ
The Black–Litterman model considers the opposite scenario: given the weight
of each asset in the portfolio, what would be the expected excess return on assets?
The solution to this reverse optimization problem is:

μ = λΣw (12.3)

The Black–Litterman model inherited Fischer Black’s view on universal hedg-

ing. w mkt marks the weight of the portfolio where each asset is weighted according
to its market capitalization (The Market Benchmark Index). The Black–Litterman
model states that the reverse optimization problem has a special connotation, that
the resulting expectation of the excess return on assets can clear the market. This is
called the implied excess equilibrium return.

∏ = λΣw mkt (12.4)

The implied excess equilibrium return ∏ is the starting point of the Black–
Litterman model.

1.2 Investor views

Investors often form distinctive views on the amount of a particular asset’s excess
return.These views are different from the implied excess equilibrium return. In the
 Issues requiring further research 171
Black–Litterman model, these views are categorized into the absolute view and
relative view. Confidence levels are used to measure them: a 100% confidence level
indicates that the investor is absolutely certain, while a 0% confidence level indi-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

cates that the investor rejects the view completely. The Black–Litterman model
assumes that the views are independent of each other.
The absolute view involves judgment on the ROA of a single asset, for example:
– View 1: The excess return on the housing market is 0.5% (25% confidence
level). The relative view, on the other hand, involves comparing the ROA of
multiple assets, for example:
– View 2: The ROA of the global stock market will be 2% higher than that of
the global bond market (65% confidence level).
– View 3: The ROA of the Chinese stock market will be 3.5% higher than that
of the other emerging markets (50% confidence level).
A key question to the Black–Litterman model is: how do you quantify the
investors’ views? We will use the three views mentioned above as examples.
Assuming there are nine types of assets, as shown in Table 12.12:
Set the excess return of the nine assets as r = ( rC , rUS , rDE , rEE , rGFI , rHY , rPE , rRE , rHF )′.
View 1 only involves Real Estate (RE) and can be expressed as rRE = 0.5% + ε 1,
where ε 1 follows a normal distribution N ( 0,ω 1 ), which reflects the uncertainty
of the view. A greater value of ω 1 translates to a higher level of uncertainty (lower
confidence level). Next, we will discuss the method of setting the value of ω 1
relative to the confidence level.
View 2 involves the global stock market, which includes Chinese stocks (C),
US stocks (US), stocks of developed (Non-US) economies (DE), and stocks of

Table 12.1 Asset types and symbols

Asset type Symbol

Chinese stocks C

US stocks US

Stocks of developed (Non-US) economies DE

Stocks of emerging (Non-Chinese) economies EE

Global fixed income GFI

High yield HY

Private equity PE

Real estates RE

Hedge fund HF
 172 Issues requiring further research
emerging (Non-Chinese) economies (EE). It also involves the global bond market,
which includes global fixed income (GFI) and high yield (HY).
As shown by View 2, relative views always have two elements: a set of
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

outperforming assets and a set of underperforming assets. The relative view implies
the existence of a portfolio with both a long and short side—going long outper-
forming assets and shorting underperforming assets.There is no guidebook when it
comes to the asset allocation, so we can allocate them evenly or according to their
market cap. We will choose the former in this example.
View 2 can be expressed by the following equation:
0.25rC + 0.25rUS + 0.25rDE + 0.25rEE − 0.5rGFI − 0.5rHY = 2% + ε 2, where ε 2
follows a normal distribution N ( 0,ω 2 ). Similarly, view 3 can be expressed as
rC − rEE = 3.5% + ε 3, where ε 3 follows the normal distribution N ( 0,ω 3 ). The
implication of ω 2 and ω 3 are the same as before.
We introduce the following symbols:

p1 = ( 0,0,0,0,0,0,0,1,0 ) , q1 = 0.5%
p2 = ( 0.25,0.25,0.25,0.25, −0.5, −0.5,0,0,0 ) , q2 = 2%
p3 = (1,0,0, −1,0,0,0,0,0 ) , q3 = 3.5%
⎛ p1 ⎞ ⎛ q1 ⎞ ⎛ ε1 ⎞ ⎛ ω1 0 0 ⎞
⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
P = ⎜ p2 ⎟ , Q = ⎜ q2 ⎟ , ε = ⎜ ε 2 ⎟ , Ω = ⎜ 0 ω2 0 ⎟
⎜ p3 ⎟ ⎜ q3 ⎟ ⎜ ε3 ⎟ ⎜ 0 0 ω3 ⎟
⎝ ⎠ ⎝ ⎠ ⎝ ⎠ ⎝ ⎠

Then, views 1–3 can be expressed as the following:

P ⋅ r = Q + ε , where ε follows the normal distribution N ( 0, Ω ). (12.5)
Equation (12.5) has a general connotation. Assuming there are n types of assets
and m mutually independent views, then P is the matrix created by m × n , Q is the
magnitude of m × 1 and Ω is the diagonal matrix created by m × m . We mark row
k of P as pk , element number k of Q as qk , element number k on the diagonal Ω as
ω k , as a result, view k can be expressed as:
pk ⋅ r = qk + ε k , ε k follows the normal distribution N ( 0,ω k )
Where pk ⋅ r can be viewed as a portfolio, qk + ε k is the investor’s view on the
excess return of the portfolio, and ω k reflects the uncertainty of the view.

1.3 The Bayesian method

Under the equilibrium approach of the Black–Litterman model, the excess return r
follows the normal distribution N ( ∏,τΣ ), where ∏, the implied excess equilibrium
return, is provided by Equation (12.4). τ is a scaling factor.The reason we introduce
τ is that the covariance matrix of the excess return Σ is usually estimated with
historical data. Any covariance matrix with a value higher than the excess return
under market equilibrium must be adjusted using τ . The distribution N ( ∏,τΣ ) is
equivalent to the prior distribution.
 Issues requiring further research 173
The investor views essentially create m portfolios P ⋅ r , while these portfolios
follow a normal distribution N (Q, Ω ), and are equivalent to new information.
Using the Bayesian method, we can calculate the posterior distribution of the
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

excess return.

( −1
r ~ N ⎡⎣(τΣ )−1 + P ′T Ω−1P ⎤⎦ ⎡⎣(τΣ )−1 ∏ +P ′Ω−1Q ⎤⎦ , ⎡⎣(τΣ )−1 + P ′Ω −1P ⎤⎦
−1
) (12.6)

1.4 Asset allocation

According to Equations (12.1) and (12.6), the asset allocation given by the Black–
Litterman model is:
−1
w BL = ( λΣ ) ⎡⎣(τΣ )−1 + P ′Ω −1P ⎤⎦ ⎡⎣(τΣ )−1 μ + P ′Ω −1Q ⎤⎦
−1
(12.7)

We have already explained how to set parameters Σ, ∏, P ,Q ,τ in Equation (12.7)

and will now introduce the method of setting parameters λ , Ω. The method of set-
ting Ω is especially important.
To estimate the coefficient of risk aversion λ , we use the following formula:

E ( rmkt − rrf )
λ = (12.8)
σ2

Where E ( rmkt ) is the expected return of the market portfolio, rrf is the risk-free
rate, σ 2 = w mkt
′ Σw mkt is the variance of the excess return on the market portfolio.
Thomas Idzorek3 has proposed a calculation for Ω. The logic behind his cal-
culation is that for every view, if the confidence level of that view is 100%, then a
strategy for asset allocation can be determined. If the confidence level of that view
is 0%, then the assets should be allocated according to their market capitalization,
which is w mkt . If the confidence level of the view is between 0% and 100%, then the
allocation strategy should be based on w mkt (the asset allocation strategy for a 0%
confidence level), and tilt toward the asset allocation strategy for a 100% confidence
level.
We use view k , pk ⋅ r = qk + ε k as an example. If the confidence level of this
view is c k (c k is between 0% and 100%), we can introduce a indicative vector
indexk : if a certain element of pk is not 0, then the respective element of index k
equals 1. On the other hand, if an element of pk is equal to 0, then the respective
element of index k equals 0. In short, index k shows the non-zero elements in pk ,
which is the asset mentioned in view k . Idzorek’s calculation follows the following
six steps.
Step 1: if the confidence level is 100%, the expected posterior distribution of the
excess return can be calculated using the following formula:

E ( rk ,100% ) = ∏ + τΣpk′ ( pkτΣpk′ ) (qk − pk ∏ )

−1
(12.9)
 174 Issues requiring further research
Step 2: calculating the asset allocation strategy at a 100% confidence level:

w k ,100% = ( λΣ ) E ( rk ,100% )
−1
(12.10)
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Step 3: calculating the difference between the asset allocation strategy at a 100%
confidence level and w mkt :

Dk ,100% = w k ,100% − w mkt (12.11)

Step 4: calculating the tilt:

Tilt k = Dk ,100% . ∗ index k ∗ c k (12.12)

Where . ∗ represents the pair-wise multiplication, the above formula shows that
the calculation of tilt involves only the assets mentioned in view k , as the confi-
dence level c k rises, the tilt also steepens.
Step 5: calculating the asset allocation strategy resulted from the tilt method:

w k ,c = w mkt + Tilt k
k
(12.13)

Step 6: solving the optimization problem involving ω k :

max
ωk >0
(w k ,c k − w k )′ ( w k ,c − w k )
k
(12.14)
−1
w k = ( λΣ ) ⎣⎡(τΣ )−1 + pk′ pk / ω k ⎤⎦ ⎡⎣(τΣ )−1 ∏ + pk′ qk / ω k ⎤⎦
−1
s.t.

For every view, we use the six steps listed above to calculate ω k , k = 1,2,..., m, the
⎛ ω 1 0 ... 0 ⎞
⎜ ⎟
⎜ 0 ω 2 ... 0 ⎟
result is the following: Ω = ⎜ .
f f _ f ⎟
⎜ ⎟
⎜⎝ 0 0 ... ω m ⎟⎠

2 The fundamental law of active management

Richard Grinold and Ronald Kahn first introduced the fundamental law of active
management.4 It explained the source of excess return alpha and the composition of
information ratio.We will present the relevant hypothesis as we discuss this law further.

2.1 Information model

Hypothesis 1: Suppose there are N types of securities available for investment.
We use the N × 1 random variable r to demonstrate their expected returns (dis-
counted by the risk free rate, similarly hereafter). The random variable rB repre-
sents the expected return of the market benchmark index. If the allocation ratio of
 Issues requiring further research 175
the securities in the market benchmark index is N × 1 vector hB, then rB = hB′ ⋅ r .
N × 1 vector β demonstrates the beta coefficient of the N securities relative to the
market benchmark index.
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Then the following must be true:

r = β ⋅ rB + θ (12.15)

Where N × 1 vector θ demonstrates the residual return of the N securities rela-

tive to the market benchmark index, and hB′ ⋅ β = 1, hB′ ⋅ θ = 0.
Hypothesis 2: The statistical characteristic of the residual return θ is
E (θ ) = 0,var (θ ) = Σθ .
Typically, the components of θ are correlated, for example, Σθ is not a diagonal
matrix. We can prove that there exists a N × 1 random variable x and a N × N
matrix A where:

1 θ = A⋅x
2 The components of x are not correlated; their expected values are 0, and their
standard deviations are 1
3 Σθ = A ⋅ A ′

Hypothesis 3: We use BR × 1 vector z to demonstrate BR units of signals

(earnings forecasts, price trends, analyst opinions, etc.). Suppose the statistical char-
acteristics of z are E ( z ) = 0,var ( z ) = Σ z .
Typically, these BR units of signals are correlated, for example Σ z is not a diago-
nal matrix. Like θ , there exists a BR × 1 random vector y and a BR × BR matrix
B where:

1 z = B⋅y
2 The components of y are uncorrelated; their expected values are 0, and their
standard deviations are 1
3 Σ z = B ⋅ B ′.

We use N × BR matrix Q to demonstrate the covariance matrix of residual

return θ and signal z, Q = cov (θ , z ).
We use N × BR matrix P to demonstrate the correlation coefficient matrix of θ
and z, P = corr (θ , z ) = cov ( x, y ) = ( ρn , b )1≤ n ≤ N ,1≤ b ≤ BR, where ρn ,b = corr (θ n , zb ).
The correlation between Q and P is: Q = A ⋅ P ⋅ B ′ .
After acquiring signal z, the investors renew their expectations of the residual
return θ , which is demonstrated in the form of the conditional expectation and
conditional variance of θ (We introduce symbol D = B −1, where I represents the
N × N identity matrix):

a ( z ) = E (θ | z ) = A ⋅ P ⋅ D ⋅ z (12.16)

C = var (θ | z ) = A ⋅ ( I − P ⋅ P ′ ) A ′ (12.17)
 176 Issues requiring further research
2.2 Investment target
We use N × 1 vector hp to demonstrate the allocation ratio of the securities, N × 1
vector h = hP − hB, which demonstrates the active positions relative to the market
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

benchmark index, where the projected rate of return for investors is:

rP = hP′ ⋅ r = hP′ ⋅ β ⋅ rB + h ′ ⋅ θ (12.18)

Where hP′ ⋅ β is the beta coefficient of the investors relative to the market
benchmark index, and h ′ ⋅ θ is the active return of the investors relative to the
market benchmark index. The volatility of h ′ ⋅ θ is called the tracking error, which
demonstrates the active risk of investors who deviate from the market benchmark
index.
Given active position h and signal z, the expected active return of the inves-
tors can be written as E ( h ′ ⋅ θ | z ) = h ′ ⋅ α ( z ), and the active risk squared is
var ( h ′ ⋅ θ | z ) = h ′ ⋅ C ⋅ h .
Hypothesis 4: Investor utility is expressed as a quadratic function with avoid-
ance level of active risk λ .
Thus, with signal z already provided, the utility function of the investors is
h′ ⋅ α ( z ) − λ ⋅ h′ ⋅ C ⋅ h (12.19)

2.3 The optimal active position

Using h ( z ) to demonstrate the optimal active position given signal z. h ( z ) can be
determined by the following optimization equations:
h ( z ) = arg max h ′ ⋅ α ( z ) − λ ⋅ h ′ ⋅ C ⋅ h (12.20)
h

The first order condition is: α ( z ) = 2λ ⋅ C ⋅ h ( z ). Plugging in the expression of

α ( z ) and C (E = ( I − P ⋅ P ′ ) ), then the first order condition can be simplified as:
−1

1
A′ ⋅ h ( z ) = ⋅E ⋅P ⋅D ⋅Z (12.21)
2λ

2.4 Information ratio

The information ratio is equal to the annual active return divided by active risk,
which reflects the risk-adjusted rate of active management.
On the optimal active position h ( z ), the conditional informa-
tion ratio of the investors IR ( z ) satisfies the following equation:
IR 2 ( z ) = Z ′ ⋅ D ′ ⋅ P ′ ⋅ E ′ ⋅ P ⋅ D ⋅ z = y ′ ⋅ P ′ ⋅ E ′ ⋅ P ⋅ y , and the information
ratio squared is:

IR 2 = E [ IR 2 ( z )] = Trace ( P ′ ⋅ E ′ ⋅ P ) (12.22)

Where Trace (⋅) indicates the trace of the matrix (sum of the elements on the
diagonal line).
Through a series of approximation, the following can be proved5
 Issues requiring further research 177
N BR
IR 2 ≈ Trace ( P ′ ⋅ P ) = ∑∑ ρ 2
n ,b (12.23)
n =1 b =1
N
We use ξb2 = ∑ρ
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

2
n ,b to demonstrate the sum of squares of each correla-
n =1
tion coefficient within signal zb and residual return θ ; this reflects the value of

signal zb . Therefore,
BR
IR 2 = ∑ξ 2
b (12.24)
b =1

Hypothesis 5: All signals have the same connotation, for every b, the following
is true ζ b2 = IC2.
Under hypothesis 5, the fundamental law of active management can express the
information ratio:

IR = IC ⋅ BR (12.25)

BR is called the breadth. It is equivalent to the number of independent signals

acquired by investors. Each independent signal represents a forecast by the investor
and results in an active position or active investment strategy. This process is called
an active bet. Consequently, the breadth BRreflects the number of independent bets
placed by the investor.
IC is called the information coefficient; it demonstrates the level of correlation
between the forecast (signal z) and the actual result (the residual return θ ).
The fundamental law of active management denotes the following:

1 Active management is essentially a forecast.

2 The consensus forecast will lead to passive investment based on the market
benchmark index.
3 To outperform the market benchmark index, an investor must make different
forecasts and create an active position based on this forecast.
4 To achieve a high information ratio, either the breadth BR must be large
(making a large number of forecasts) or the information coefficient IC must be
high (making accurate forecasts).

3 How is big data implemented?

The Black–Litterman model is different from the fundamental law of active
management, yet they are closely connected. The Black–Litterman model is more
practical, and the fundamental law of active management is useful in discussing
general trends. Both of the models, however, argue that:

1 The market benchmark index is the starting point of investment activities and
the default portfolio.This means that to some extent, the models imply a belief
in market efficiency.
 178 Issues requiring further research
2 Investors can only deviate from the market benchmark index if they decide
to make forecasts that differ from the average market forecast. Of course, they
would have to evaluate the validity of their forecasts. In the models, this is
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

reflected by the confidence level parameter in the Black–Litterman model

and the information coefficient in the fundamental law of active management.
3 Integrating the two models, we can conclude that active management can be
generalized to the following five steps: gather valuable signals from data, make
distinct forecasts, form strategies based on forecasts, evaluate the strategies by
observing results or performing back tests, and finally evaluate the validity of
the forecasts based on their results and make corrections (Figure 12.1).

Big data analysis is mostly used in step 1 and step 2.These two steps are typically
taken together. Investors can extract signals by using economic principles or data
mining.The basic principle is that the signals must be useful to improve the forecasts,
otherwise they are just noise. Richard Grinold and Ronald Kahn have established
the following formula for the basic forecast:

φ = E ( r | g ) − E ( r ) = cov ( r , g ) ⋅ var −1 ( g ) ⋅ ( g − E ( g ) ) (12.26)

Where r is the return of the security, g is the signal, E ( r ) represents the aver-
age market forecast, and E ( r | g ) represents the forecast based on signals. Equation
(12.26) is the general form of Equation (12.16).
If r and g are both random variables and not random vectors, then Equation
(12.26) can be simplified as:
g − E ( g)
φ = corr ( r , g ) ⋅ std ( r ) ⋅ (12.27)
std ( g )

Data

Signal

Projection
Feedback
and
correction
Strategy

Evaluation

Figure 12.1 The general steps of active management.

 Issues requiring further research 179
Where std (⋅) is the standard deviation of the random variable, corr ( r , g ) is the
g − E ( g)
information coefficient IC, and demonstrates the number of standard
std ( g )
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

deviations the signal deviates from the average, we call it the z-score. The connota-
tion of Equation (12.27) is:

Forecast = Information coefficient ∗ volatility ∗ Z − score of signal (12.28)

We will use the article written by Hristidis et al. to exemplify how big data is
used by securities investors. This article focused on correlation between informa-
tion on Twitter and stock prices. They started by analyzing all the data on Twitter
about one stock, and designed several quantitative indexes to measure its activity on
Twitter. The result showed that some indices were significant correlated with the
trade volume of the stock in the next 1–3 days, as well as with the rate of return.
The above demonstrated the first step—extracting signals from data.
Next, they created a Twitter-augmented regression on the rate of return using
the aforementioned Twitter activity indexes. This model used vector auto-regres-
sion (VAR) with the rate of return of the stocks as the dependent variable, while
the independent variables included the historical rate of return of the stocks as well
as the Twitter activeness indexes. This is step 2—forming forecasts based on signals.
Step 3 was the formulation of strategy based on forecasts. This was done by
investing in the stocks with the highest projected rate of return based on the
Twitter-augmented regression.
At last, they used back tests to compare the result of their strategy with other
strategies. The following were used in this comparison: (a) a portfolio that pas-
sively invested only in the market benchmark index; (b) a randomly selected
stock; (c) a portfolio whose stocks were selected based on market cap, size, total
liability, and so on; (d) a portfolio that invested in several stocks with the highest
projected rate of return provided by the auto regression. Essentially this was very
similar to the Twitter strategy, except that it did not include the Twitter activity
indices.
In the end, they found out that the Twitter-based strategy had the best result.
This proved that information on Twitter can actually be used in security investment.
The above summarized steps 4 and 5—evaluation, feedback, and correction.
We would like to suggest that in Internet finance, the securities market may pos-
sess characteristics described in both behavioral finance and the efficient markets
hypothesis. Investors affect each other in decision-making thanks to the prevalence
of social networks, which means that their behavior will be similar to that described
in behavioral finance. For example, they may exhibit a greater level of conformity,
which could affect a particular security or even the entire securities market. On the
other hand, big data is closing the gap in obtaining information. Securities pricing,
which used to require complex calculations, is now made simple by apps. Pricing
efficiency is becoming extremely high. As a consequence, the behavior of securities
markets may be similar to the efficient markets hypothesis.
 180 Issues requiring further research
SECTION 2: BIG DATA IN ACTUARIAL SCIENCE

The essence of insurance is to provide financial compensation to the insured.

Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Risk events that require compensation can be either personal or property. The
two major types of insurance, personal insurance and property insurance, are the
result. Actuarial science analyzes future risk events and calculates their potential
consequences. These calculations help to set the premium and margin. It is also
divided into two categories, life contingencies and non-life contingencies. Though
the fundamentals of the two categories are the same, the methods they implement
diverge greatly.
We will provide a brief summary of life contingencies and non-life contingen-
cies with a special focus on contingency rate setting. Afterwards, we will discuss
how big data can be used in actuarial science.

1 An introduction to life contingency6

1.1 Basic concepts
Life contingency focuses on mortality risks. If X is the life expectancy for a 0-year-
old infant, then X is a random variable between (0, ω ], where ω is called the limiting
age, usually 105 or 110.We plug in the following functions to track the distribution
of variable X .
The accumulated distribution function of longevity is:

F ( x ) = Pr ( X ≤ x ) , x ≥ 0 (12.29)

F ( x ) is the probability of the person dying before age x. F ( x ) increases mono-

tonically in interval (0, ω ], and satisfies F ( 0 ) = 0, F (ω ) = 1.
The survival function of longevity:

S ( x ) = Pr ( X > x ) , x ≥ 0 (12.30)

S ( x ) is the probability of the person surviving beyond age x. S ( x ) and F ( x )

satisfy the following: S ( x ) = 1 − F ( x ).
Mark a person x years old as ( x ). The years left for ( x ) until death is called ( x )’s
time until death, or T ( x ). The accumulated distribution function of ( x )’s time until
death is marked t qx , which is the probability of a person x years old dying in the
next t years is:
S (x ) − S (x + t )
q = Pr (T ( x ) ≤ t ) = Pr ( X ≤ x + t | X > x ) = (12.31)
S (x )
t x

The survival function of ( x )’s time until death is marked t px , which is the prob-
ability of a person at the age of x living beyond x + t years old:
S (x + t )
px = Pr (T ( x ) > t ) = Pr ( x > x + t | X > x ) = (12.32)
S (x )
t
 Issues requiring further research 181
The force of mortality at the age of x is μx , which means that a person can live
to x years old, and dies right after that moment:
Pr ( x ≤ X ≤ x + Δx | X > x ) S′ (x )
μx = lim = −
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

(12.33)
Δx → 0 Δx S (x )
The correlation between the force of mortality and the survival function is
demonstrated in the follow equation:

( x
)
S ( x ) = exp − ∫ μ s ds , t px =
0
S (x + t )
S (x ) ( x +t
= exp − ∫ μ s ds
x ) (12.34)

d
The density function of the remaining life duration T ( x ) is fT ( t ) = t q x, thus
d d dt
S (x + t ) S (x + t )
fT ( t ) = − dt = μx + t ⋅ t px fT ( t ) = − dt = μx + t ⋅ t px (12.35)
S (x ) S (x )
Generally speaking, the curve of the force of mortality is bowl shaped. Infants
have high forces of mortality, while the forces of mortality are lowest during youth.
The forces of mortality then gradually increase as people grow older. Some of
the most famous models for the force of mortality are: the De Moivre model, the
Gompertz model, the Makeham model, and the Weibull model.

1.2 Setting net premiums

To demonstrate the core skills of life contingency, we will set the net premium of a
single payment and build its actuarial models.The net premium covers only the cost
of risk, not operating costs and other profits. A single premium is a payment plan in
which all the charges are paid with one payment.
To set a net premium, we must follow the net equilibrium theory. This implies
that the net premium is equal to the sum of future insurance payments. The net
equilibrium theory is an underlying principle of the insurance industry, in that it is
a guideline every type of insurance follows in setting net premiums.
There are two assumptions in setting net premiums. First, the residual life data
sets of a group of policyholders who share the same age, gender, and starting date
are independently, identically distributed. Second, the insurance company can fore-
cast the investment rate. With these two assumptions, we can convert individual risk
events to a sum of homogeneous risk events. On an individual basis, one policyholder
cannot predict the time he will encounter a risk event, nor the amount that he will be
compensated. However, in total, the pattern for residual life is very clear statistically.
As mentioned before, T ( x ) is the residual life variable of a group of policy-
holders who share the same age, gender, and policy start date. If a mortality event
occurs after t units of time, the compensation is payable at the moment of death,
the amount of compensation is bt , and the discount function at this point is vt . The
present value function of compensation is then:

zt = bt ⋅ vt (12.36)
 182 Issues requiring further research
The density function at the moment of payment is f T (t ), and the expected
compensation function is:

E (Z t ) = ∫z f (t ) dt
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

t T (12.37)

The individuals do not necessarily receive compensation equal to their pay-

ments. Rather, we view the entire body of policyholders with the same risks as one
entity, and balance achieved on a larger scale. Equation (12.37) is very universal.
The structure of the model remains the same despite different insurance types. The
differences lay only in the function format of bt , vt, f T (t ).
Take whole life insurance as an example. Whole life insurance is insurance that
covers any applicable mortality event. If ( x ) applies a whole life insurance policy with
the coverage of $1, where bt = 1, t ∈ (0, ω ], and the annual compound rate is v, then
vt = v t , t ∈ (0, ω ]. The present compensation value function is zt = v t , t ∈ (0, ω ].
The net insurance amount is:

E ( zt ) =
w w
∫
0
zt ⋅ f T (t ) dt = ∫
0
v t ⋅ t px ⋅ μx +t dt (12.38)

1.3 The life table

Though the force of mortality models demonstrated the longevity variable by using
extremely simple models and equations, the actual distribution of life expectancy is
complex. To avoid errors, life contingency often implements life tables to approach
the distribution of longevity.
The life table describes the mortality patterns of a closed population from birth
to death, supposing that there is no mobility or reproduction, and that the only
factor able to reach the population is death. Assuming that the population is con-
sisted of n individuals, observation shows that kn people within the group die at the
k k
age of x, then n is the mortality probability at age x . n is an empirical estimation
n n k
of F ( x ). The theories of big data dictate that, if n is large enough, n converges
n
to F ( x ).
Official life tables often use a number of 100,000 or 1,000,000 as the surviving
population at age 0(l0). Life tables are divided into different age groups (from x to
x + t ). Except for infants, whose data is often measured in days or weeks, the other
age groups usually use years as inspectional units. For every age group from x to
x + t , the life table provides the following statistics:

1 Initial surviving number l x = l0 ⋅ S ( x )

2 Mortality rate t qx
3 Number of mortality t d x = l x ⋅ t qx
4 Numbert of years where l0 units of newborns survive in their age groups
t L x = ∫ l x + s ds
0
 Issues requiring further research 183
5 The total amount of residual life of all individuals who will live to age x.
ω −x
Tx = ∫ l x +t dt
0
T
6 The average residual life of individuals who will live to age x. e x0 = x
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

lx
The life table provides the life distribution in whole numbers. In practice, the
survival rate of the integer years is often used to estimate the probability of survival
and mortality between the integer years. This is basically an interpolation problem.
Provided ( x, l x ) and ( x + 1, l x +1 ), estimate one point between the integer years
( x + t,lx +t ), 0 < t < 1. Three approaches are used to solve this problem. The first
uses linear interpolation in between the integer years. The second uses geometrical
interpolation, and the last is harmonic interpolation.

2 An introduction to non-life contingency7

2.1 The compound risk model
Non-life contingency analyzes risk on property and obligation.The compound risk
model is a key tool in non-life contingency.
Suppose non-negative random variable sequences {X i , i ≥ 1} are mutually inde-
pendent. N is a non-negative whole number, and {X i , i ≥ 1} and N are mutually
0
independent given ∑ X i = 0.
i =1
N
S = ∑X i (12.39)
i =1

In reality, the compound risk model is used to plot the total loss of one policy
or policy set over a period of time:

1 A model for one policy: assuming that one policy will encounter N losses
within a certain period, and X i represents the amount of loss of i, the total
N
policy loss of is ∑ X i .
i =1
2 A model of a set of policies: assuming that N policies within the policy set will
encounter losses within a certain period, and X i is the amount of loss incurred
N
by policy i, the total loss of the set is ∑ X i .
i =1
N
In the mixed risk model S = ∑X , X i t is called the individual loss, and its dis-
i =1
tribution is called the individual loss distribution or the loss distribution. Some
common examples include: normal distribution, logarithmic normal distribution, Γ
distribution, B distribution, and Pareto distribution.
N is the number of claims, and its expectation E ( N ) is called the claim fre-
quency. Common distributions of N include: the Poisson, binomial, and negative
binomial distributions.
 184  Issues requiring further research
In the compound risk model, the following is true:

E (S ) = E ( N ) ⋅ E ( X i ) (12.40)
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Therefore, the expected loss of the model is equal to the product of claim fre-
quency and individual loss.This is the theoretical basis of non-life contingency setting.

2.2 Setting risk premiums

The amount and period of coverage often differ for the same type of insurance.
Thus, the risk or exposure unit is introduced as the basic unit for plotting the
amount of risk. The risk unit must be able to quantify risk, easy for insurance com-
panies to measure, easy to understand, hard to manipulate, and easy to manage. The
sum of risk units in a policy is called the risk amount. For example, the risk unit for
vehicle insurance is usually one vehicle-year. If two vehicles both have a coverage
period of half year, then the total risk amount is 1.
The risk premium is the main component of non-life contingency. The process
to set risk premiums has three steps.The first step is to estimate the claim frequency
N and the expected individual payment Xi per risk unit. In a statistical sense, this
step solves for the parameter values of distribution N and Xi. Methods that are
commonly used in this step include the moment estimation method and maximum
likelihood method. (See Chapter 6)
We then implement net equilibrium theory (Equation 12.40) for the next step.
The resulting formula for the risk unit policy is the following:

The risk premium of the risk unit policy = The claim frequency of the risk unit policy
* The average coverage of each payment
 (12.41)

The third step is to find the formula for calculating the risk premium:

The risk premium of the policy = The risk premium of the risk unit policy
* The risk amount of the policy  (12.42)

2.3 The experience rating

The Experience Rating Method is a quantification method designed to eliminate
the heterogeneity of risks. This method enables insurance companies to modify
premiums based on historical data. The basic logic is that if one policy is outper-
forming its original projection, then the policyholder may ask for lower payments.
From the perspective of the insurance company, the reliability of its own data must
be considered carefully. Moreover, the company needs to judge the risk based on
the performance of one policy versus other policies in the same category.The expe-
rience rating method includes complete credibility, partial credibility, and greatest
accuracy credibility. For clarification, we will use the Bühlmann model from the
greatest accuracy credibility theory as an example.
 Issues requiring further research 185
The Bühlmann model focuses on the following problem: for a risk unit policy, if
the compensations of the past n years are X 1 , X 2 ,…, X n, how do we determine the
risk premium X m +1 for the next year?
Suppose that X 1 , X 2 ,…, X n share the same type of distribution and the dis-
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

tribution parameter Θ is a random variable. If X 1 , X 2 ,…, X n are independ-

ent under the conditions provided by distribution parameter Θ, and they
have the same conditional expectation and conditional variance so that
μ ( Θ ) = E [ X i | Θ ], v ( Θ ) = var [ X i | Θ ]. μ = E [ μ ( Θ )] is the rate in the premium
chart or the estimated value of compensation based on analyses of similar policies.
The fundamental idea of the Bühlmann model is to estimate X n +1 using the
linear combination of historical compensations:
2
⎡ n
⎤
min E ⎢ X n + 1 − α 0 − ∑ α i xi ⎥ (12.43)
α ,α , … ,α
0 1 n
⎣ i =1 ⎦
The solution of optimization problem (Equation 12.43) is the credibility
premium:

Z ⋅ X + (1 − z ) ⋅ μ (12.44)
n

∑X i
n E [ v ( Θ )]
i =1
Where X = , credibility factor z = (k = ).
n n+k var ( μ ( Θ ))
The credibility factor z illustrates the reliability of future risk premiums on
historical data. When z = 1, the future risk premium relies completely on its own
historical data X . This is called complete credibility. When 0 < z < 1, the future
risk premium relies on both X and μ, the weight of μ is less than 1. This is called
partial credibility.

3 How is big data implemented?

3.1 The core variable of actuarial insurance
Force of Mortality μx is the core variable of actuarial science’s application to life
insurance. If μx is known, then we can calculate the cumulative distribution function
of life span F ( x ), survival function S ( x ), and the survival function of residual life t qx .
Since the life variable X represents the time point of mortality events, X is the stop-
Pr ( x ≤ X ≤ x + Δx | X > x )
ping time in the probability theory, and μx = lim
Δx → 0 Δx
marks the intensity of mortality events.
The claim frequency E ( N ) in non-insurance actuarial science is one of the
two core variables. The claim frequency is essentially a counting process. Assuming
we have a counter that starts from 0, and we raise the number every time the
insurance policy suffers a loss (if we assume the duration is T), then the last read-
ing is N (T ). More precisely, Tk represents the time interval between loss number
 186 Issues requiring further research
k and loss number k + 1 ( k = 0,1, 2,… ). The time point of loss number k is
Sk = T0 + T1 + … + Tk − 1, and the claim frequency N (T ) can be expressed as:
∞
N (T ) = ∑ 1{
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Sk ≤T } (12.45)
k =0

Where 1{⋅} represents the indicative function, the value is 1 if the logic in the
parenthesis is true, 0 in all other cases.
The time point of loss also involves the stopping time concept. If we look for-
ward from Sk , the occurrence of loss number k , we can describe the duration of
time before loss k + 1 by using the intensity of loss event:
Pr (t ≤ Tk ≤ t + Δt | Tk > t )
λ (t ) = lim (12.46)
Δx → 0 Δt
λ (t ) fully describes the distribution of claim frequency N (T ).
Consequently, the force of mortality μx and the occurrence of loss have the same
statistical connotation in reaching intensity λ (t ).

3.2 Responsiveness of actuarial science to individual heterogeneity

Technically, the force of mortality is applicable to everyone, and the intensity of
loss events are applicable to every vehicle, apartment, or other non-life objects.
Therefore, these two variables can be used to create models and perform projec-
tions on an individual level. Based on this, actuarial science can reflect differences
between individuals, which may lead to the creation of a personalized premium
rate. To some extent, the current trend of actuarial science reflects this logic.
In life contingency, the life table is categorized into male, female, and general life
tables. Other categories include smoking and non-smoking life tables, national life
tables targeting national citizens, as well as experience life tables that target only the
policyholders of the insurance company.
Many types of insurance require the applicants to pass certain health require-
ments, and only those who pass may enter the plan.The health exam is like a door-
step of selection, only admitting people in a superior health condition.This is called
selection bias. In order to assign risks fairly, the mortality rate of these new appli-
cants is very low. However, this effect does not last forever. Within 2–3 years, their
health condition would be indistinguishable from the rest. The select-and-ultimate
tables are designed specifically for this. Within the selection period, life tables with
smaller mortality rates are used. After the selection period, the ultimate table, or life
table with normal mortality rates, is used. These adjustments to the life tables are
essentially setting differentiated force of mortality based on individual cases.
In non-life contingency, the experience rating method adjusts future payments
based on historical experience such as historical compensation. If each compensa-
tion is consistent with the rest, the experience rating method would essentially be
forecasting the future value based on the historical values of the intensity of loss
events.
 Issues requiring further research 187
Overall, actuarial science does not reflect individual characteristics adequately.
There are countless factors that could affect one’s risk, such as his genes, inheritance,
habits, occupation, climate, and pollution. The life table is still lacking in inclusion
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

of these factors for analysis. In non-life contingency, aside from historical

compensation, cross sections of the policyholder and macroeconomic variables can
also be included in actuarial science.

3.3 Thoughts on using big data in actuarial science

In actuarial science, we believe that big data analysis, instead of rules of thumb,
can be used to build a connection between force of mortality and intensity of loss
events. A related field—credit risk management has been explored with big data.
There is much that actuaries can learn from this field.
We will focus on the simplicity method for risk management (See Chapter 7).
The simplicity method does not discuss why a default may happen. Rather, it views
default as a random event that could happen at any time. Hazard ratios and default
intensity are used to plot the default rate. If τ is the time point for the occurrence
of a credit event, then risk rate means that:
Pr (τ ≤ t + Δt | τ > t )
h (t ) = lim (12.47)
Δx → 0 Δt
Comparing Equations (12.33), (12.46), and (12.47), we can tell that the risk rate
has the same connotation as in force of mortality and in intensity of loss events.
In credit risk management, there are many interesting methods to build models.
For more information, you can refer to the book by Darrell Duffie and Kenneth
Singleton.8 These models are designed to compare the force of mortality and the
intensity of loss events. Here we will use the corporate credit risk model by Sudheer
Chava and Robert Jarrow.9
Suppose that there are n policyholders. For each policyholder, a one-time risk
event may occur—personal or property. For policyholder i, we use τ i to represent
the points in time when his risk events happen. τ i also demonstrates the stopping
time concept.
For a set of scattered time points t = 0,1, 2,…, T , we know that X it is the basic
information of the policyholders at time point t regarding policyholder i. We use
N it to determine that there are no risk events at time point t . N it is a variable that
ranges from 0 to 1. We assign the values in the following manner:
⎧⎪ 0 t < τ i
N it = ⎨ (12.48)
⎪⎩ 1 t ≥ τ i
For each i, N i 0 = 0.
Pr (τ i ≤ t + Δt | τ i > t )
Risk event λit = lim represents the intensity of risk
Δx → 0 Δt
event at time point t for policyholder i. Since the data points are scattered, a more
convenient analysis is the conditional probability of default (equivalent to λit ).
 188 Issues requiring further research
If no risk event has happened at time point t , then the probability of a risk event
occurring before time point t + 1 is:

pit = Pr (τ i ≤ t + 1 | τ i > t, X it )
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

(12.49)

Assuming that there is a functional relationship between p and X it : it

exp (α + β ′ X it )
pit = (12.50)
1 + exp (α + β ′ X it )
Next, we will estimate α and β in Equation (12.50) based on the observation
{( X it , N it ) : i
= 1, 2,…, n, t = 1, 2,…,T }.
N iT = 0 means that until reaching time point T, the beneficiary i does not
experience a risk event, where τ i > T . N iT = 1 means that the beneficiary i expe-
rienced risk events before reaching T. In this case, data after τ i has no significance.
Consequently, we define Di = min (τ i ,T ).
For the following equations
⎧ Pr (τ = D ) N = 1
⎪ i i iD
L ⎡⎣ N iD | X it ,1 ≤ t ≤ Di ⎤⎦ = ⎨
i

⎪⎩ Pr (τ i > Di ) N iD = 0
i
(12.51)
i

= Pr (τ i = Di ) Pr (τ i > Di )
N iDi 1− N iDi

Di − 2 Di −1
Where Pr (τ i = Di ) = piD −1 ∏ (1 − pit ), Pr(τ i > Di ) =
i ∏(1 − p ). it
t =0 t =0
We define the likelihood function as:
n
L ⎡⎣α , β | ( X it , N it ) : i = 1, 2,…, n, t = 1, 2,…,T ⎤⎦ = ∏ ⎡⎣ N
i =1
iD
i
| X it ,1 ≤ t ≤ Di ⎤⎦

The log-likelihood function is equal to:

n n Di −1
piD −1
LL (α , β ) = ∑N iDi log i

1 − piD −1
+ ∑ ∑ log (1 − p ) it (12.52)
i =1 i i =1 t = 0

Optimizing Equation (12.52), we obtain estimates on parameters α and β .

Consequently Equation (12.50) can be used to assess the risk of the policyholders.
Looking at the current trend, there has been some big data utilization in non-life
contingency. Usage-based insurance could be a good example. This type of insur-
ance sets its rate based on vehicle usage. The insurance requires the installation of a
GPS, a velometer, and a radio device. All data collected by the insurance company
can help to set the rate. Another benefit is that claims are easier to settle once the
company obtains information about the vehicle and the driver.
At last, it is important to point out that the insurance component of Internet
finance is an extremely interesting topic. We have already stated that the essence of
insurance is financial compensation, which means that using theories of big data,
the insurance companies would provide financial compensation for unexpected
losses. In this process, the insurance payments of the policyholders who do not
 Issues requiring further research 189
experience unexpected losses would pay for those who incur losses. In a perfectly
competitive environment, the total amount of all insurance payments should be
equal to the total amount of the unexpected losses (the net equilibrium theory),
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

while the insurance company oversees the insurance payment transfer. Of course,
it is impossible to foretell which policyholders will experience losses and which
will not, thus it is impossible to predict the direction in which payments will flow.
Therefore, we can conclude that the risking sharing with insurance is different
from the risk shifting of usual derivatives (options, futures, swaps, etc.). The former
is a group contract and the latter is a bilateral contract. In Internet finance, even
if the level of information asymmetry and the transaction costs are low, the core
characteristic of insurance being a group contract will not change.10 While the busi-
ness model of insurance remains the same, the specific organizational structures are
subject to change. For example, a group of individuals with similar risk level can
sign a contract through the Internet stating that if one of them suffers unexpected
losses, the rest will be obligated to provide compensation. Once this group reaches
a certain size, it could replace insurance companies.

Notes
1 This is different from the mean-variance optimization problem (Equation 11.1). However,
it can be proven that if a utility function is quadratic, or if the excess return is normally
distributed, then we can conclude that this problem is equivalent to the mean-variance
optimization.
2 This is designed solely for demonstrational purposes and does not necessarily reflect
reality.
3 Idzorek, Thomas. 2004. “A Step-by-Step Guide to the Black-Litterman Model:
Incorporating User-Specified Confidence Levels.”
4 Grinold, Richard, and Ronald Kahn. 1999. “Active Portfolio Management:
A Quantitative Approach for Providing Superior Returns and Controlling Risk,” 2nd
edition, McGraw-Hill.
5 We can decompose E = ( I − P ⋅ P ′ ) to E = I + P ⋅ P ′ + P ⋅ P ′ ⋅ P ⋅ P ′ + ….
−1

Therefore P ′ ⋅ E ′ ⋅ P can be decomposed to P ′ ⋅ E ′ ⋅ P = P ′ ⋅ P + P ′ ⋅ P ⋅ P ′ ⋅

P + … . Since the elements within P are correlated coefficients, we can neglect the
higher-order terms to get P ′ ⋅ E ′ ⋅ P ≈ P ′ ⋅ P .
6 Wang,Yan. 2008. “Life Contingency,” China Renmin University Press.
7 Yang, Jingping. 2008. “Non-Life Contingency,” Peking University Press.
8 Duffie, Darrell, and Kenneth Singleton. 2003. “Credit Risk: Pricing, Measurement, and
Management,” Princeton University Press.
9 A fundamental model for risk management company Kamakura Corporation.
10 As we pointed out in chapter 1, the terms of financial contracts remain the same in
Internet finance.
 Appendix 1
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Overview of SFI
Founded on July 14, 2011, the Shanghai Finance Institute (SFI) is a leading
non-governmental, non-profit institute dedicated to professional academic financial
research. SFI is operated by the China Finance 40 Forum (CF40) and has a strategic
cooperation with the Shanghai Huangpu District government. The Shanghai
Financial Services Office is SFI’s regulator, and SFI is registered with the Shanghai
Administration of Social Organizations.

The mission of SFI is to explore new trends in the global financial market, pursue
solutions to novel problems associated with China’s financial development, and
support the development of Shanghai as an international financial center.

Known for its professionalism and openness, SFI is an independent think tank
dedicated to promote academic exchange. It conducts high-level research activities
to provide first-class research products.

SFI hosts events such as closed-door seminars, the Shanghai Annual Conference of
New Finance, and the Bund Forum of Internet Finance. It also conducts research
projects and is responsible for the publication of the New Finance Review, the
New Finance Book Series, various academic journals and articles, and media
columns.

The China Finance 40 Forum (CF40) is a non-government, non-profit, and

independent think tank dedicated to policy research on economics and finance.
CF40 was founded on April 12, 2008, and operates as a “40 × 40 club” that consists
of forty influential experts around 40 years old. CF40 aims to enhance the academic
foundation of China’s finance, provide high-quality research on emerging financial
issues, and promote financial reform and development.
 Appendix 2
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

Organizational structure
Members of SFI Advisory Committee:
FANG Xinghai Director, International Economic Bureau, Office of the
Central Leading Group for Financial and Economic Affairs
HU Huaibang Chairman, China Development Bank
JIANG Yang Vice Chairman, China Securities Regulatory Commission
TU Guangshao Executive Vice Mayor of Shanghai
WANG Jiang Professor of Finance, MIT Sloan School of Management
Weng Zuliang Secretary of CPC Huangpu District Committee, Shanghai
WU Xiaoling Vice Chairman, the Financial and Economic Affairs Committee,
the National People’s Congress of the People’s Republic of
China
YAN Qingming Vice Mayor, Tianjin
YI Gang Deputy Governor, the People’s Bank of China; and Chief
Administrator, State Administration of Foreign Exchange
YUAN Li Vice President, China Development Bank

The Chairman of SFI Executive Council:

WAN Jianhua Chairman, E-Capital Transfer Co., Ltd

The Vice-Chairmen of SFI Executive Council:

ZHENG Yang Director, Shanghai Financial Services Office
TANG Zhiping Governor of Huangpu District, Shanghai
WANG Haiming Secretary-General, China Finance 40 Forum

Members of SFI Executive Council:

CHEN Jiwu Principal Partner and Chairman. Shanghai VStone Asset
Management Co., Ltd
GAO Keqin President, ABC Financial Leasing Co., Ltd
Gregory D Gibb Chairman, Lufax
KWAN Tatcheong Executive Director & Chief Executive, The Bank of East Asia
(China)
 192 Appendix 2
HOU Fu’ning President, Shanghai Rural Commercial Bank
LAN Rong Chairman, Industrial Securities Co., Ltd
LI Jianguo Vice President, Shanghai Bank
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

LI Lin Head of Strategic Development Department, Shanghai Pudong

Development Bank
LING Tao Chairman, Shanghai Huarui Bank
LI Xunlei Vice President & Chief Economist, Haitong Securities
LIAN Ping Chief Economist, Bank of Communications
PAN Weidong Chairman, Shanghai International Trust Co., Ltd
PAN Xinjun Chairman, Orient Securities Co., Ltd
PENG Lei CEO, Zhejiang Ant Small & Micro Financial Services Group
QIU Guogen Chairman, Shanghai Chongyang Investment Management
Co., Ltd.
Wang Song President, Guotai Junan Securities
XU Luode Chairman, Shanghai Gold Exchange
XU Zhen Chairman, Shanghai Clearing House
YANG Huahui Chairman, China Industrial International Trust Limited
YAO Wenping Chairman, Tebon Securities Co., Ltd
ZHAO Linghuan Chief Executive Officer, Hony Capital
ZHOU Xiong President, Zhongtai Trust Co., Ltd.
ZHOU Ye President, China PnR

Member of SFI Council:

GUO Yuhang Co-CEO & Founding Partner of DianRong
LI Guohong Chairman & President, Shangdong Gold Financial Holding
Capital Management Co., Ltd.
SHENG Jia CEO, Net Credit Finance Group
TANG Ning CEO & Founder, CreditEase
Yang Yifu Chairman, Renrendai.com
YAO Naisheng Vice CEO, JD Finance
ZHANG Jialin Chairman, Beijing Zipeiyi Investment Management Co., Ltd.

Members:
CHINA RAPID FINANCE
HUCHEN Investment
HUAHUI Fortune
NI WO DAI
YIBank
TAIRAN Internet Finance
100credit.com

Chairman of SFI Academic Committee:

QIAN Yingyi Dean, School of Economics and Management, Tsinghua
University
 Appendix 2 193
SFI Academic Committee Members:
LI Xunlei Vice President & Chief Economist, Haitong Securities
LIAN Ping Chief Economist, Bank of Communications
Downloaded by [Run Run Shaw Library, City University of Hong Kong] at 03:25 05 September 2017

LIAO Min Director, Shanghai Office, China Banking Regulatory

Commission
MIAO Jianmin President, China Life Insurance (Group) Company
WANG Qing President, Shanghai Chongyang Investment Management
Co., Ltd
ZHANG Chun Executive Dean, Shanghai Advanced Institute of Finance,
Shanghai Jiaotong University
ZHENG Yang Director, Shanghai Financial Services Office
ZHONG Wei Director, Finance Research Center, Beijing Normal University

The Chairman of SFI Supervisory Committee:

XU Zhen Chairman, Shanghai Clearing House

SFI Supervisors:
GUAN Tao Senior Fellow, China Finance 40 Forum
WU Cheng Deputy Governor, Huangpu District, Shanghai

The Director of SFI:

QIAN Yingyi Dean, School of Economics and Management, Tsinghua
University

The Deputy Director of SFI:

ZHONG Wei Director, Finance Research Center, Beijing Normal University

The Executive Deputy Director of SFI:

WANG Haiming Secretary-General, China Finance 40 Forum