Determinants of financial development

Huang, Yongfu

2010

Palgrave Macmillan

"As the world has witnessed the worst financial crisis and climate crisis of our age, during the period of 2007-2009, the issues surrounding the emergence and development of financial markets and carbon markets is becoming an increasingly significant area of research and debate worldwide. By engaging with recently developed methods of research and new areas of practice, this book investigates the political, economic, policy and geographic determinants of the development of financial markets. The volume examines the causality between financial development and aggregate private investment from an economic perspective. It also explores the consequences of political liberalization, focusing on the impact of institutional improvement on financial development. It studies what stimulates governments to initiate reforms aimed at boosting financial development, and analyses the determinants of carbon markets in developing countries from a geographic point of view. This book is essential reading for all interested in economic and financial development, climate change, environmental economics, and applied econometrics. "

Business and Management

Environmental Sciences

Economic

English

978–0–230–27367–2 (print)

Determinants of Financial Development

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE i — #1

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE ii — #2

Determinants of
Financial Development

Yongfu Huang
University of Cambridge

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE iii — #3

© Yongfu Huang 2010
All rights reserved. No reproduction, copy or transmission of this
publication may be made without written permission.
No portion of this publication may be reproduced, copied or transmitted
save with written permission or in accordance with the provisions of the
Copyright, Designs and Patents Act 1988, or under the terms of any licence
permitting limited copying issued by the Copyright Licensing Agency,
Saffron House, 6-10 Kirby Street, London EC1N 8TS.
Any person who does any unauthorised act in relation to this publication
may be liable to criminal prosecution and civil claims for damages.
The author has asserted his right to be identified as the author of this work
in accordance with the Copyright, Designs and Patents Act 1988.
First published in 2010 by
PALGRAVE MACMILLAN
Palgrave Macmillan in the UK is an imprint of Macmillan Publishers Limited,
registered in England, company number 785998, of Houndmills, Basingstoke,
Hampshire RG21 6XS.
Palgrave Macmillan in the US is a division of St Martin’s Press LLC,
175 Fifth Avenue, New York, NY 10010.
Palgrave Macmillan is the global academic imprint of the above companies
and has companies and representatives throughout the world.
Palgrave® and Macmillan® are registered trademarks in the United States,
the United Kingdom, Europe and other countries.
ISBN: 978–0–230–27367–2 hardback
This book is printed on paper suitable for recycling and made from fully
managed and sustained forest sources. Logging, pulping and manufacturing
processes are expected to conform to the environmental regulations of the
country of origin.
A catalogue record for this book is available from the British Library.
A catalog record for this book is available from the Library of Congress.
10 9 8 7 6 5 4 3 2 1
19 18 17 16 15 14 13 12 11 10
Printed and bound in Great Britain by
CPI Antony Rowe, Chippenham and Eastbourne

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE iv — #4

To Benrun and Benpei

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE v — #5

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE vi — #6

Contents
List of Figures

x

List of Tables

xi

List of Abbreviations

xiii

Preface

xvii

1 Introduction
1.1
1.2

1.3

Background
Origins of financial development: A review
1.2.1 Institutions
1.2.2 Policy
1.2.3 Geography
1.2.4 Other variables
Structure of the book

2 General Determinants of Financial Development
2.1
2.2

Introduction
The data
2.2.1 Samples
2.2.2 Measures of financial development
2.2.3 The potential determinants
2.3 Empirical strategy
2.3.1 Bayesian Model Averaging
2.3.2 General-to-specific approach
2.4 Empirical results (I): Overall financial development
2.4.1 Some stylized facts
2.4.2 What are the main determinants of FD?
2.5 Empirical results (II): Specific financial developments
2.6 Conclusions
Appendix text
Appendix tables

3 Private Investment and Financial Development
3.1
3.2
3.3

Introduction
The data
Analysis on data for five-year averages
vii

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE vii — #7

1
1
3
3
5
6
7
7
10
10
13
14
14
16
20
21
22
24
24
27
36
46
48
49
64
64
67
69

viii

Contents

3.3.1 Methodology: System GMM
3.3.2 Empirical results
3.4 Analysis on annual data
3.4.1 Methodology: Common factor approach
3.4.2 Panel unit root tests
3.4.3 Panel cointegration tests
3.4.4 Estimation on annual data
3.5 Conclusion
Appendix tables
Appendix figures

69
73
77
78
81
84
85
92
94
99

4 Political Institutions and Financial Development

101
101
102
104
104
105

4.1
4.2
4.3

Introduction
Institutions, democratization and finance
The measures and data
4.3.1 The sample
4.3.2 The measure and data for financial development
4.3.3 The measure and data for institutional
improvement
4.4 Methodology
4.5 Evidence
4.5.1 Preliminary evidence
4.5.2 Regression results
4.6 Conclusion
Appendix tables

5 Financial Reforms for Financial Development
5.1
5.2

Introduction
Methodology
5.2.1 Model specifications
5.2.2 Econometric methods
5.3 Empirical evidence
5.3.1 Analysis on the original dataset
5.3.2 Analysis on a larger dataset
5.4 Discussions
5.5 Conclusion
Appendix tables

6 Geographic Determinants of Carbon
Markets (CDM)
6.1
6.2

Introduction
Data and stylized facts

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE viii — #8

106
106
109
109
114
121
123
125
125
127
127
131
133
134
143
147
149
151
161
161
164

Contents

6.3 Econometric method: Spatial econometric approach
6.4 Empirical evidence
6.5 Concluding remarks
Appendix table

ix

168
171
178
180

Conclusion

181

Notes

183

Bibliography

194

Index

203

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE ix — #9

Figures
2.1
2.2
2.3
2.4
4.1
4.2
6.1
6.2
6.3

Scatter plots of institutions and financial development
Scatter plots of policy and financial development
Scatter plots of geography and financial development
Median Liquid Liability by different country group over
1960–2003
Financial development ten years before and after
democratization
Volatility of financial development ten years
pre/post-democratization
Scatter plots of CDM and geography
CDM and resource endowments
CDM and distance to biggest and smallest host countries

x

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE x — #10

25
26
27
28
113
113
166
167
172

Tables
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
3.1
3.2
3.3
3.4
3.5
3.6
4.1
4.2
4.3
4.4
4.5
5.1
5.2
5.3
5.4

Determinants of FD by using BMA
Determinants of FD
Top ten models and their posterior probabilities for FD
Geography, policy, institutions and FD
Determinants of FDBANK
Determinants of FDSTOCK
Determinants of FDEFF
Determinants of FDSIZE
Does private investment cause financial development?
1970–98 (five-year-average data)
Does financial development cause private investment?
1970–98 (five-year-average data)
Unit root tests in heterogeneous panels
Panel cointegration tests between FD and PI
Does private investment cause financial development?
1970–98 (Annual data)
Does financial development cause private investment?
1970–98 (Annual data)
Change in FD standardized before and after
democratization
Institutional improvement and financial development
(whole sample), 1960–99
Institutional improvement and financial development
(lower-income countries), 1960–99
Institutional improvement and financial development
(ethnically diverse countries), 1960–99
Institutional improvement and financial development
(French legal origin countries), 1960–99
Within estimates: Benchmark specification (Equation 4)
(A, B)
Within estimates: Alternative specification (Equations 5
and 6)
Within estimates: Alternative specification (Equation 8)
Error dependence across countries and over time
considered separately (A, B, C)

30
32
34
35
38
40
42
44
74
75
83
85
90
91
111
115
117
119
120
135
137
138
140

xi

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xi — #11

xii Tables

5.5
6.1
6.2
6.3

Augmented dataset with Chinn-Ito measure (2006)
(A, B, C)
Moran’s I and Geary’s C for CDM
Geography and CDM (by inverse-distance weights)
Geography and CDM (by binary weights)

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xii — #12

145
173
175
177

Abbreviations
2SLS
ADF
AM
AREA
AR(1)
ARDL
ASIA
BACE
BMA
BMP
BTOT
CCE
CCEMG
CCEP
CDM
CER
CIVLEG
COMLEG
CRIGHT
CTRADE
DGP
DURABLE
EBA
ELEV
ETHNIC
ETHPOL
EURFRAC
EURO1900
EXPMANU
EXPPRIM
EXPSERV

two-stage least squares estimator
augmented Dickey-Fuller test
Abiad and Mody (2005)
land area of a country in square km
first-order autoregression
autoregressive distributed lag
dummy variable for Asian countries
Bayesian averaging of classical estimates
Bayesian model averaging
black market premium (%)
index of commercial/central bank
common correlated effect approach
common correlated effect mean group estimator
common correlated effect pooled estimator
clean development mechanism
certified emission reductions
dummy variable for civil law legal origin
dummy variable for common-law legal origin
index of creditors’ rights
natural log of the Frankel-Romer measure of
predisposition to external trade
data-generating process
index of political stability
extreme bounds analysis
elevation in metres above sea level
index of ethnic fractionalization
index of ethnic polarization
index of European first language
percentage of population in 1900 European or of
European descent
dummy variable for manufactured goods exporting
countries
dummy variable for fuel and non-fuel primary goods
exporting countries
dummy variable for service exporting countries

xiii

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xiii — #13

xiv

Abbreviations

FD
FDBANK
FDBOND
FDEFF
FDSIZE
FDSTOCK
FL
FREE
GDN
GDP
GDP03
GDP90
Gets
GMM
GS2SLS
GUM
HINFL
INCLOW
INCMID
IC
KKM
LAC
LANDLOCK
LANGUAGE
LATITUDE
LEG_FR
LEG_GE
LEG_SC
LEG_UK
LLY
LR
LSDV
LSDVC
MC3
MCAP
MEDSHARE
MG
MINDIST
NIM
OLS

index of overall financial development
index of extent of bank-based intermediation
index of bond market development
index of financial efficiency
index of size of financial system / financial depth
index of measure of stock market development
index of financial liberalization
averaged indices of civil liberties and political rights
World Bank Global Development Network Database
gross domestic product
initial GDP per capita in 2003
initial GDP per capita /initial income in 1990
General-to-specific approach
generalized method of moments estimator
generalized two-stage least squares estimator
general unrestricted model
dummy variable for periods of high inflation
dummy variable for low-income countries
dummy variable for middle-income countries
information criterion
index of governance
dummy variable for Latin American countries
dummy variable for landlocked countries
index of language fractionalization
absolute latitude of a country from the Equator
dummy variable for French legal origin countries
dummy variable for German legal origin countries
dummy variable for Scandinavian legal origin countries
dummy variable for British legal origin countries
index of liquid liabilities
long-run
Least Squares Dummy Variable estimator
corrected LSDV estimator
Markov Chain Monte Carlo technique
index of stock market capitalization
index of market share of state-owned media
mean group estimator
minimum distance from USA, Japan and Belgium
index of net interest margin
ordinary least squares estimator

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xiv — #14

Abbreviations

OPENC
OVC
PC
PcGets
PCI
PIPs
PMG
PMP
POLITY2
POP03
POP90
POP100CR
PMP
PRIVO
R
REGEAP
REGLAC
REGEMENA
REGSA
REGSSA
REGWENA
RELIGION
RESCOFF
RESDIFF
RESPOINT
RMSE
RSS
SDBMP
SDGR
SDPI
SDTP

xv

trade openness (at current prices) or the sum of exports
and imports over GDP
index of overhead costs
principal components
Gets computer algorithm
index of political constraints
posterior probabilities of inclusion
pooled mean groups estimator
posterior model probabilities
index of democracy from Polity IV Database
initial population in 2003
initial population in 1990
share of population in 1994 within 100 km of coast or
ocean-navigable river
posterior model probability
index of private credit
a free software environment for statistical computing
and graphics.
dummy variable for East Asian and Pacific countries
dummy variable for Latin American countries
dummy variable for Middle Eastern and North African
countries
dummy variable for South Asian countries
dummy variable for Sub-Saharan African countries
dummy variable for Western European and North
American countries
index of religious fractionalization
dummy variable for coffee/cocoa natural resources
exporting countries
dummy variable for diffuse natural resources exporting
countries
dummy variable for point source natural resource
exporting countries
root mean square error
residual sum-of-squares
std. dev. (or volatility) for the black market premium
std. dev. for annual growth rate real, chain-weighted
GDP 1960–89
std. dev. for annual inflation 1960–89
std. dev. for volatility of GDP per capita growth of
trading partners

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xv — #15

xvi

Abbreviations

SDTT
SRIGHT
SSA
SYS-GMM
TOPEN
TOR
TVT
USINT
WG
YRSOFFC

std. dev. for volatility of the terms of trade index for
goods and services
index of shareholders’ rights
dummy variable for Sub-Saharan countries
System generalized method of moments estimator
index of trade openness policy
index of turnover ratio
index of total value traded
index of US Treasury Bill rate
within groups estimator
dummy variable for the first year in office

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xvi — #16

Preface
While it is clear that financial depth has a positive effect on economic
growth, the questions of what determines financial development and
how to develop financial markets remain imperfectly understood. More
specifically, economists still have an insufficient understanding of the
following key issues. What brings about the emergence and development of financial markets? What are the reasons why different financial
structures, bank-based or market-based, exist in countries where similar
levels of economic development have been reached? What accounts for
the differences in the levels of financial development in countries like
the OECD member countries which have similar income levels, and geographic conditions? The world witnessed the worst financial crisis and
climate crisis of our age during the period 2007–09. This highlights the
significance of the research into what is essential to the development of
financial markets and what is key to develop carbon markets for tackling
climate change.
Against this background, my book seeks to investigate the fundamental determinants of the development of financial markets and carbon
markets. It starts with a general examination of the determinants of
financial development in Chapter 2 and moves on to specific studies
in the following chapters. Chapters 3 and 4 examine two specific determinants of financial development in the context of globalization. To be
more specific, Chapter 3 provides an exhaustive analysis of the causality
between aggregate private investment and financial development from
the economic point of view while Chapter 4 explores the determinants of
financial development from a political perspective, namely, the impact
of institutional improvement on financial development. Chapter 5 looks
at what induces governments to undertake reforms aimed at boosting
financial development. Chapter 6 is concerned with the development of
carbon markets, which is a newly developed/recently emerging area for
both research and practice. It examines what could explain the uneven
development of carbon markets in developing countries from a geographic point of view, with an aim of encouraging further research into
other determinants of carbon market development.
This book constitutes a unique addition to the expanding literature
in this field, and its contribution is highlighted by its title. It could be

xvii

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xvii — #17

xviii

Preface

the first comprehensive book of this kind to explore this subject systematically by using various recently developed econometric methods.
It provides a very general but comprehensive overview of modern financial development theory and incorporates cutting-edge research in this
field, along with a huge number of relevant literature citations. This book
also presents the latest thinking on how to develop financial and carbon
markets. The findings of this book have rich implications for the conduct of macroeconomic policies in developing countries in an integrated
global economy.
This book is suitable for the students of financial development and
climate change at the advanced undergraduate or graduate level, for
economists and applied econometricians who are interested in economic
and financial development, financial liberalization and climate change
and for policy-makers and government agencies. This research topic will
continue to be of great interest to academics and practitioners across
the globe, which is underlined by the number of recent international
conferences and symposia devoted to the financial and climate crises.
I would like to avail myself of this opportunity to extend my sincere
thanks to all those who have made my research into these issues and
my writing of this book a truly fulfilling and unforgettable experience. It
goes without saying, or it should, that there are various people without
whom this book would never have been possible.
A great debt of gratitude is owed to my PhD supervisor, Professor
Jonathan Temple, for giving me his time, insights, enthusiasm, incredible help and constant support. His remarkable insights into various
development issues, his erudition in economics and his willingness to
discuss and blue-sky with me, have enriched both my academic life and
this book. Also, high tribute should be paid to my Centre Director at
Cambridge University, Dr Terry Barker, who has kindly advised me on
various climate change issues, for example, the last chapter of this book.
His generous support and assistance have been of inestimable worth to
the conduct of my research and the accomplishment of this book. If this
book looks good, it is only because of their insightful suggestions and
invaluable help.
A number of academic members were no less critical to my research
during the years of the preparation of this book. It would be impossible to give a comprehensive list, but I would like to thank Professor
Stephen Bond and Professor Frank Windmeijer for their expert comments and advice. Dr Sonia Bhalotra, Dr Edmund Cannon, Dr David
Demory and Dr Andy Picking (in alphabetical order) kindly provided
thought-provoking input during my research at Bristol. I am also deeply

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xviii — #18

Preface xix

indebted to Professor Philip Arestis, Professor Hashem Pesaran, Dr Mark
Roberts and other colleagues at Cambridge from whom I have greatly
benefited in terms of valuable suggestions.
I owe a special debt of gratitude to Professor Yuguang Yang at Fudan
University, who has played an important role in the course of my career
development. His professional conscientiousness (or rigorousness), positive attitude and strong thirst for knowledge have inspired me to go
forward over the years. My appreciation also goes to my close friends
from Fudan University, Youqiang Li, Zhiqun Lin, Zhuwu Xu, Muqing
Zheng and Xiaoxin Zhou (in alphabetical order) among others for their
heartfelt sincerity, encouragement and help in various circumstances.
I would like especially to acknowledge Taiba Batool and Gemma Papageorgiou at Palgrave Macmillan and Cathy Lowne, who have been
remarkably patient and helpful and whose expert jobs have helped to
make this book a reality. I also highly appreciate the contribution to the
book made in various ways by other people at Palgrave Macmillan. The
stamp of their illuminating advice and careful checking appears on every
page of my book.
On a personal note I wish as always to thank my beloved family for
their constant encouragement, unwavering support and love. From the
earliest time I can remember, my parents have instilled in me a love
of learning that has only grown over time. The incredible help and
love of my sister and brother enabled me to go through frustration and
depression. Their patience is legendary. To all these and more, I shall be
eternally grateful.
Yongfu Huang
Cambridge, March 2010

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xix — #19

HUANG: “FM” — 2010/9/30 — 18:48 — PAGE xx — #20

1
Introduction

1.1

Background

Among the profound evolutions in development economics in recent
decades has been the renewed interest in, and growing contributions
on, the role of financial systems in economic development. While it is
clear that a positive effect exists between financial depth and economic
growth, the questions of what determines financial development and
how to develop financial markets remain imperfectly understood.
Research on the role of financial development in growth can be
traced back at least to Bagehot (1873) who claims that large and
well-organized capital markets in England enhanced resource allocation towards more productive investment. Other historical antecedents
before 1970 include, among others, Schumpeter (1911), Hicks (1969) and
Goldsmith (1969). Schumpeter (1911) emphasizes the critical role of a
country’s banking system for economic development in mobilizing savings and encouraging productive investment. Hicks (1969) highlights the
importance of financial markets in the process of industrial revolution
with an observation that the development of financial systems facilitates
the applications of new technologies and innovations. Goldsmith (1969)
finds evidence of a positive link between financial development and economic growth from a comparative study with data for 35 countries over
the period 1860–1963.
Over the past three decades, the financial repression and financial development framework proposed by McKinnon (1973) and Shaw
(1973) has been the main intellectual basis of financial market analysis and policy advice. Before the 1970s most developing countries had
been financially repressed in the sense that their financial systems had
imposed upon them discriminatory taxation in the form of low interest
1

HUANG: “CHAP01” — 2010/9/29 — 20:05 — PAGE 1 — #1

2

Determinants of Financial Development

rate policies, high reserve requirements and high inflation rates. Keynes
(1936) and Tobin (1956) are among the various justifications for maintaining these policies. The McKinnon-Shaw model of financial repression
formulates the phenomenon of financial repression and points out that
financial repression reduces both the quantity and quality of aggregate
investment in the economy in the sense that a lower deposit rate of
interest discourages households from holding deposits that would be
used to finance productive investment. The policy implication of the
McKinnon-Shaw model is that government’s repressive policies towards
financial systems (such as interest rate ceilings, high reserve requirements and credit control) retard financial development, and therefore
economic growth. On the contrary, financial liberalization and financial
development can stimulate investment and its productivity, and ultimately foster economic growth. Since 1973, the McKinnon-Shaw model
has influenced financial sector policies in many developing countries
considerably.
Motivated by the McKinnon-Shaw model, a number of studies in this
area have been undertaken, such as Kapur (1976) and Mathieson (1980)
among others. However, these works in general treat financial intermediation and financial institutions as exogenous. The last two decades have
witnessed a resurgence of interest in the relationship between financial
development and economic growth which incorporates the insights of
endogenous growth models. These works include Townsend (1979), Diamond (1984), Gale and Hellwig (1985), Williamson (1986, 1987), Bencivenga and Smith (1991), Greenwood and Jovanovic (1990), Saint-Paul
(1992), King and Levine (1993) and Bernanke et al. (1999) among others.
Apart from a standard Arrow-Debreu framework, these studies make
use of the assumption of information asymmetry between lenders and
borrowers, producing significant findings. Due to the presence of information asymmetries, the problem of adverse selection and moral hazard
might arise, since the borrowers (typically entrepreneurs) have incentives
to hide their actual (or expected) return on their investment, calling for
costly state verification. The financial contract and financial intermediation are therefore endogenously determined. Not only do these models
demonstrate how financial intermediaries emerge, they also analyse how
financial intermediation promotes economic growth. The inherent functions of financial systems, including mobilizing savings to their highest
valued use, acquiring information, evaluating and monitoring investment projects and enabling individuals to diversify away idiosyncratic
risk, have been widely believed to encourage productive investment and
therefore total factor productivity.1

HUANG: “CHAP01” — 2010/9/29 — 20:05 — PAGE 2 — #2

Introduction

3

Given the broad consensus on the substantial role of financial development in economic growth, it is of great practical importance to
understand the origins of financial development. Economists still have
an insufficient understanding of what brings about the emergence and
development of financial markets, what are the reasons why different
financial structures, bank-based or market-based, exist in countries where
similar levels of economic development have been reached and what
accounts for the differences in the level of financial development in
countries like the OECD member countries which have similar income
levels and geographic conditions.
This research seeks to investigate the political, economic, policy and
geographic determinants of the development of financial markets. In
addition, it attempts to examine the causality between financial development and another important aspect of economic activities, namely
aggregate private investment. It also aims to explore the consequences
of political liberalization in terms of institutional improvement for financial development and whether we should expect any changes in the
political system, from autocracy to democracy for example, to exert
any influence on the speed of financial development. It then studies what stimulates governments to initiate reforms aimed at financial
development. This research ends up in the last chapter by studying
the determinants of carbon markets in developing countries from a
geographic perspective.
The following section provides a brief review on the determinants of
financial development. Section 1.3 describes the structure of the book.

1.2

Origins of financial development: A review

Recent years have witnessed burgeoning research into the potential
determinants of financial development. This section briefly outlines the
main possible determinants of financial development, including institutional factors, macroeconomic factors, geographic factors and others
which have been studied in the literature.
1.2.1

Institutions

Research on the role of institutions in financial development has been
considerable, especially research on the effects of the legal and regulatory environment on the functioning of financial markets. A legal
and regulatory system involving protection of property rights, contract
enforcement and good accounting practices has been identified as essential for financial development. Most prominently, La Porta et al. (1997,

HUANG: “CHAP01” — 2010/9/29 — 20:05 — PAGE 3 — #3

4

Determinants of Financial Development

1998) have argued that the origins of the legal code substantially influence the treatment of creditors and shareholders, and the efficiency of
contract enforcement. They document that countries with a legal code
like Common Law tend to protect private property owners, while countries with a legal code like French Civil Law tend to care more about
the rights of the state and less about the rights of the masses. Countries
with French Civil Law are said to have comparatively inefficient contract
enforcement and higher corruption, and less well-developed financial
systems, while countries with a British legal origin achieve higher levels of financial development. Among others, Mayer and Sussman (2001)
emphasize that regulations concerning information disclosure, accounting standards, permissible banking practice and deposit insurance do
appear to have material effects on financial development.
Beck et al. (2003)’s application of the settler mortality hypothesis
of Acemoglu et al. (2001) to financial development is another significant work in this context. They argue that colonizers, often named as
extractive colonizers, in an inhospitable environment aimed to establish institutions which privileged small elite groups rather than private
investors, while colonizers, often named as settler colonizers, in more
favourable environments were more likely to create institutions which
supported private property rights and balanced the power of the state,
therefore favouring financial development. Both the legal origin theory of La Porta et al. (1997, 1998) and Beck et al. (2003)’s application
are related to colonization, but the former is more concerned with how
colonization determines the national approaches to property rights and
financial development, whereas the latter is more about the channel via
which colonization influences financial development.
The recently developed “new political economy” approach regards
“regulation and its enforcement as a result of the balance of power
between social and economic constituencies” (Pagano and Volpin, 2001).
It centres on self-interested policy-makers who can intervene in financial
markets by either overall regulation or individual cases for purposes such
as career concerns and group interests. Rajan and Zingales (2003) emphasize the role of interest groups, and especially the incumbent industrial
firms and the domestic financial sector, in the process of financial development. They argue that, in the absence of openness, incumbents have
strong incentives to block the development of a more transparent and
competitive financial sector which undermines the incumbents’ vested
interests and relationships. When both trade openness and financial
openness are encouraged, the incumbents have incentives to support
financial development from which more funds can be sought to meet

HUANG: “CHAP01” — 2010/9/29 — 20:05 — PAGE 4 — #4

Introduction

5

foreign competition and new rents can be generated to compensate
partially for their loss of incumbency.
Generally speaking, institutions might have a profound impact on the
supply side of financial development. The level of institutional development in a country to some extent determines the sophistication of the
financial system.
1.2.2

Policy

The policy view highlights the importance of some macroeconomic policies, openness of goods markets and financial liberalization in promoting
financial development. The significant effect of policy on financial development could be working through either its demand side or its supply
side.
Some major national macroeconomic policies such as maintaining
lower inflation and higher investment have been documented as being
conducive to financial development. Huybens and Smith (1999) theoretically and Boyd et al. (2001) empirically investigate the effects of
inflation on financial development and conclude that economies with
higher inflation rates are likely to have smaller, less active and less efficient banks and equity markets. Some recent work has supported the
view that policies which encourage openness to external trade tend to
boost financial development (Do and Levchenko, 2004).
In addition, research has been carried out to study the effects of financial liberalization on financial development over the past three decades,
following the McKinnon-Shaw model (McKinnon, 1973; Shaw, 1973),
which concludes that while financial repression reduces the quantity
and quality of aggregate investment, financial liberalization can foster
economic growth by increasing investment and its productivity. The
positive link between domestic financial liberalization and financial
development is supported by evidence (World Bank, 1989), although
domestic financial liberalization is not without risks (Demirgüç-Kunt
and Detragiache, 1998). Research on the positive correlation between
external financial liberalization, especially capital account openness, and
financial development is discussed in the panel data studies of Bailliu
(2000) and Chinn and Ito (2006), although potential destabilizing effects
may also exist. Claessens et al. (1998) present evidence that opening
banking markets improves the functioning of national banking systems
and the quality of financial services, with positive implications for banking customers and lower profitability for domestic banks. Laeven (2000)
examines whether the liberalization of the banking sector may help to
reduce financial restrictions and the external cost of the capital premium,

HUANG: “CHAP01” — 2010/9/29 — 20:05 — PAGE 5 — #5

6

Determinants of Financial Development

stimulating investment and financial development. Bekaert et al. (2002)
provide evidence that opening up the stock market to foreign investors
renders stock returns more volatile and more highly correlated with the
world market return.
1.2.3

Geography

There is less work directly addressing the potential correlation between
geography and financial development in comparison to that for policy and institutions. However, much research attention has been paid
to the importance of geography for general economic development,
emphasizing three aspects in particular.
The first group is concerned with the correlation between latitude and
economic development. Countries closer to the equator typically have
a more tropical climate. On the one hand, research by Kamarck (1976),
Diamond (1997), Gallup et al. (1999) and Sachs (2003a, 2003b) suggests
that tropical location may lead directly to poor crop yields and production due to adverse ecological conditions such as fragile tropical soils,
unstable water supply and prevalence of crop pests. On the other hand,
tropical location can be characterized as an inhospitable disease environment, which is believed to be a primary cause for “extractive” institutions
(Acemoglu et al., 2001).
A second strand of research relates to countries being landlocked, distant from large markets or having only limited access to coasts and rivers
navigable to the ocean (Sachs and Warner, 1995a, 1995b, 1997; Easterly and Levine, 2003; Malik and Temple, 2009). As natural barriers to
external trade and knowledge dissemination, geographic isolation and
remoteness to some extent determine the scale and structure of external
trade in which countries engage. The potential to enter a large economic
market and exploit economies of scale may be limited by particular
geographic circumstances. The ability to develop a competitive manufacturing sector may be constrained when some intermediate inputs
for the production of manufactured goods need to be imported from
distant markets. As the main feature of external trade for these countries, the limited range of primary commodities exported determines the
vulnerability of these countries to external shocks.
The last strand of research focuses on the link between resource
endowment and economic development. Diamond (1997) suggests that
countries with a richer endowment of grain species have more potential
for high-yielding food crops and technological development. Isham et al.
(2005) argue that a developing country’s natural resource endowment
affects its economic development through an unique channel in which

HUANG: “CHAP01” — 2010/9/29 — 20:05 — PAGE 6 — #6

Introduction

7

natural resource endowment is linked to different export structures,
different export structures determine institutional capacities towards
coping with external shocks and finally institutional quality is reflected
in the level of GDP per capita. Easterly and Levine (2003) argue that
the natural endowment of tropics, germs and crops indirectly influences
income through the impacts of these on institutions.
In general, geography is likely to work mainly through the demand
side of financial development, although it may affect its supply side
by influencing the quality of institutions. For instance, the production
of particular agricultural products or primary goods and exploitation of
some natural resources could reduce the demand for external finance,
relative to other countries at a similar level of GDP per capita.
1.2.4

Other variables

Other variables considered as determinants of financial development
are economic growth, the income level, population level and religious,
language and ethnic characteristics, etc. Greenwood and Jovanovic
(1990) and Saint-Paul (1992) document that as the economy grows, the
costs of financial intermediation decrease due to intensive competition,
inducing a larger scale of funds available for productive investment.
The importance of income levels for financial development has been
addressed in Levine (1997, 2003, 2005). In considering banking sector
development in 23 transition economies, Jaffee and Levonian (2001)
demonstrate that the level of GDP per capita and the saving rate have
positive effects on the banking system structure as measured by bank
assets, numbers, branches and employees.
Stulz and Williamson (2003) stress the impact of differences in culture, proxied by differences in religion and language, on the process
of financial development. They provide evidence that culture predicts
cross-country variation in protection and enforcement of investor rights,
especially of creditor rights. The evidence also shows that the influence
of culture on creditor rights protection is mitigated by the introduction
of trade openness. Djankov et al. (2003) shed light on the role of state
ownership of the media in the extent of financial development.

1.3

Structure of the book

This research starts from a general examination of fundamental determinants of financial market development, and moves on to specific
studies as to the effects of aggregate private investment and institutional
improvement on financial development. It ends up with a study on the

HUANG: “CHAP01” — 2010/9/29 — 20:05 — PAGE 7 — #7

8

Determinants of Financial Development

geographic determinants of carbon market development in developing
countries, mainly the Clean Development Mechanism (CDM) markets.
The structure of this book is outlined as follows:
Chapter 2 is concerned with the main determinants of cross-country
differences in financial development. Two prominent tools for addressing model uncertainty, Bayesian Model Averaging and General-tospecific approaches, are jointly applied to investigate the financial
development effects of a wide range of variables taken from various
sources. The analysis suggests that the level of financial development in a
country is mainly influenced by the latter’s overall level of development,
the origins of its legal system and the quality of its institutions.
Chapter 3 provides an exhaustive analysis of the causality between
financial development and another important aspect of economic activities, namely aggregate private investment. It uses recently developed
panel data techniques on data for 43 developing countries over the
period 1970–98. GMM estimation on averaged data, and a common factor approach on annual data allowing for global interdependence and
heterogeneity across countries, suggest positive causal effects going in
both directions. This finding has rich implications for the development
of financial markets and the conduct of macroeconomic policies in developing countries in an integrated global economy. GMM results based on
averaged data appear in the Journal of Statistics: Advanced in Theory and
Applications, 2009, 2(2), whilst GMM results based on annual data appear
in an Empirical Economics Special Issue on “New Perspectives on Finance
and Development”, 2010.
Chapter 4 studies the effect of institutional improvement on financial
development in two steps. It examines whether political liberalization
in terms of institutional improvement promotes financial development,
using a panel dataset of 90 developed and developing countries over the
period 1960–99, revealing a positive effect on financial development at
least in the short run, particularly for lower-income countries, ethnically
divided countries and French legal origin countries. The results of this
chapter appear in World Development, 2010 38(12).
Chapter 5 studies what induces governments to undertake reforms
aimed at financial development. Its starting point is Abiad and Mody
(2005). Rather than their ordered logit technique, it uses a within groups
approach allowing for error dependence across countries and over time.
This chapter finds that policy change in a country is negatively rather
than positively associated with its liberalization level, while the regional
liberalization gap appears less relevant. On the effects of shocks and
crises, it suggests that some of the Abiad and Mody (2005) findings are

HUANG: “CHAP01” — 2010/9/29 — 20:05 — PAGE 8 — #8

Introduction

9

robust, but others are fragile. Furthermore, it claims that the extent of
democracy is important for this analysis, and identifies a negative effect
of the extent of democracy on policy reform. Some results of this chapter
appear in the Journal of Applied Econometrics, 2009, 24(7).
Chapter 6 examines whether certain geographic endowments matter
for the CDM market development. It suggests that CDM credit flows in
a country are positively affected by those in its neighbouring countries.
Countries with higher absolute latitudes and elevations tend to initiate
more CDM projects, whereas countries having richer natural resources
do not seem to undertake more CDM projects. This finding sheds light
on the geographic determinants of uneven CDM development across
countries, and has implications for developing countries in terms of
international cooperation and national capacity building for effective
access to the CDM.

HUANG: “CHAP01” — 2010/9/29 — 20:05 — PAGE 9 — #9

2
General Determinants of
Financial Development

2.1

Introduction

This chapter attempts to examine systematically the factors that might
account for cross-country differences in financial development. It
employs two modern quantitative methods, Bayesian Model Averaging
(BMA) and General-to-specific (Gets) approaches, to gauge the robustness of a selection of possible determinants of financial development.
Special emphasis has been placed on the contributions that institutions,
policy and geography may have in developing financial markets.
First, we take a look at some simple contrasts in the financial development experience. The United Kingdom and France have similar levels of
GDP per capita, democratic institutions and geographic characteristics
in terms of latitude, access to the sea and distance from large markets.
Nevertheless, they follow different legal traditions, reflected in different
legal practices towards the protection of private property rights. In the
1990s, stock market capitalization to GDP ratio in the UK was more than
three times higher than that in France, while the ratio of private credit
to GDP in the UK (112%) was noticeably higher than the same ratio in
France (89%). How much of the difference in financial depth between
the UK and France is due to the difference in their legal traditions and
practices?
The financial development experience in Latin American countries
provides an enlightening example of the possible role of macroeconomic
policies in financial development given the similarities of geographic
conditions, institutional development and cultural characteristics. After
implementing market-oriented policies in the 1970s and establishing
prudential regulations in the 1980s, Chile achieved remarkable growth
in financial intermediary development and stock market capitalization,
10

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 10 — #1

General Determinants of Financial Development 11

and has been regarded as the financial leader in Latin America since the
mid-1980s. In the 1990s both the ratio of liquid liabilities to GDP and the
ratio of private credit to GDP in Chile were 50 percentage points higher
than those of Brazil, the second best country in the region. Stock market
capitalization as a fraction of GDP in Chile in the 1990s was 78%, at least
three times larger than that in any other Latin American country. How
much of the success of Chilean financial development is due to better
macroeconomic policies?
In the 1990s the ratio of credit issued to the private sector to GDP
in Canada was 94%, more than four times higher than that in Mexico
of 23%. Stock market capitalization as a fraction of GDP in Canada in
1990s was 65%, more than twice as high as in Mexico (31%). Canada and
Mexico share a number of similarities in terms of geographic endowments and institutional development. More specifically, both of them
have access to the sea, have a long border with the biggest developed
country, have a large land area and a democratic political system, etc.
However, among other factors, Canada and Mexico apparently differ in
income level and latitude, which is associated with historical dominance
of tropical cash crops in Mexico and grain in Canada. How much of the
difference in financial depth between Canada and Mexico is due to the
difference in income level and how much is due to their geographic
endowment, and its long-run effects on institutions?
Exploring what determines financial development has become an
increasingly significant research topic in recent years. Examples are La
Porta et al. (1997, 1998), Beck et al. (2003), Rajan and Zingales (2003)
and Stulz and Williamson (2003) to mention a few. La Porta et al. (1997,
1998) have made a significant contribution to this topic with regard to
the legal determinants of financial development. By applying the settler
mortality hypothesis of Acemoglu et al. (2001) to financial development,
Beck et al. (2003) address how institutions matter for financial development. The Rajan and Zingales (2003) interest groups theory argues that
politics matter for financial development. Stulz and Williamson (2003)
illustrate that culture matters, although it may be tempered by openness. As to the role of policy, among others, Baltagi et al. (2009) study
the importance of trade openness, whilst Chinn and Ito (2006) focus on
the effect of financial openness.
Besides this, there is a large body of research aiming to identify the
determinants of financial development, ranging from some emphasizing
macroeconomic factors such as inflation, the income level (in terms of
GDP per capita) and the saving rate to others stressing institutional and
geographic factors. Since the relevant economic theories provide limited

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 11 — #2

12 Determinants of Financial Development

guidance on the specification of a cross-country regression for financial development, it is not clear which of these factors, acting relatively
independently, plays the primary role in determining financial development when they are all taken into consideration. Formally speaking,
there is a model uncertainty problem concerning which variables should
be included in the model to capture the underlying data-generating
process.
When facing a situation where a vast literature suggests a variety
of economic policy, political and institutional factors as determinants
of long-run average growth rates, Levine and Renelt (1992) raised a
concern over the robustness of existing conclusions in cross-section
growth regressions. They found that only a few variables can be regarded
as robust determinants of growth and almost all results are “fragile”.
They suggested applying a version of “extreme bounds analysis” to the
problem of model uncertainty. Motivated by this influential work, Salai-Martin (1997a, 1997b), Fernandez et al. (2001) and Sala-i-Martin et al.
(2004) are significant works among others that have investigated the contributions of various factors to cross-country growth. These works have
emphasized the Bayesian method as a potential technique for addressing
model uncertainty.
Empirical research on the determinants of financial development
encounters a similar model uncertainty problem to that on economic
growth. This chapter is the first attempt to study extensively the structural determinants of financial development using a large array of
variables, by jointly applying BMA and the so-called LSE Gets approach,
which is another modern method aiming to recover the true datagenerating process. The Gets method has been recently developed and
advocated by David Hendry and other practitioners (Hoover and Perez,
1999; Krolzig and Hendry, 2001 and Hendry and Krolzig, 2005 for example). To date, BMA and Gets have become more and more popular for the
purpose of model selection, although the theory of econometric model
selection is still underdeveloped.
Not only will this chapter look at each individual factor, but it also
pays special attention to the roles of institutions, policy and geography
in the process of financial development.2 There has been substantial
research on the role of institutions, policies and geography in the process of economic development in which much work regards institutions
as the fundamental factor in long-run growth (Acemoglu et al., 2001;
Dollar and Kraay, 2003; Easterly and Levine, 2003 and Rodrik et al.,
2004). In particular, research by Easterly and Levine (2003) and Rodrik
et al. (2004) highlights the dominant role of institutions over those

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 12 — #3

General Determinants of Financial Development 13

of geography and policy. They argue that geography and policy affect
economic development through institutions by influencing institutional
quality, and the direct effect of geography and policy on development
becomes weaker once institutions are controlled for. Is this also the case
for financial development?
In three aspects, this chapter exhibits distinct innovations and
strengths. First, it considers a wider assortment of economic, political
and geographic variables than any previous study. The second aspect
is its joint application of the BMA and Gets procedures, which combines the strengths of each method. By jointly applying two modern
methods using a wide range of variables, more reliable conclusions can
be expected. Third, since, as pointed out by Levine (2005), there is
no uniformly accepted proxy for financial development currently available, this paper constructs a composite index of financial development
using principal component analysis, which enables us to look at different
dimensions of financial development including overall financial development, financial intermediary development, stock market development, financial efficiency development and financial size development
(usually called “financial depth”).
The analyses based on the BMA and Gets procedures lead to the following findings. Institutions, macroeconomic policies and geography, when
taken as groups, together with cultural characteristics and the income
level of a country, are significantly associated with the level of financial
development. Of 39 variables taken individually, legal origins, a government quality index, a trade policy index, land area, initial GDP, initial
population and the population fraction of speakers of the main Western
languages are found to be important determinants of financial development. In particular, this research highlights the dominant roles played
by initial GDP, legal origin and institutional quality in the process of
financial development.
The following section includes a description of the data. Section 2.3
discusses the empirical strategy and is followed by the empirical results
of both BMA and Gets in Section 2.4. Section 2.5 summarizes the
conclusions.

2.2

The data

This section describes the sample of countries on which this study is
undertaken, and the measures of financial development and potential
determinants. Appendix Table A2.1 contains the description and sources
of these variables and Appendix Table A2.2 presents summary statistics.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 13 — #4

14 Determinants of Financial Development

2.2.1

Samples

This study mainly investigates key determinants of five specific indices
of financial development discussed in more depth below. For each financial index, there are three samples on which the investigation is based:
the whole sample, a developing country sample and a smaller sample for
which the La Porta et al. (1998) data are available. The whole sample is the
main focus of the analysis. The developing countries in the settler mortality dataset of Acemoglu et al. (2001) form the main part of the developing
country sample here. Looking at the smaller La Porta et al. (1998) sample makes it possible to examine whether differences in legal tradition,
reflected in the protection of shareholders’ and creditors’ rights, determine cross-country differences in financial development. The countries
included are listed in Appendix Table A2.3.
Note that the transition economies and small economies with a population of less than 500,000 in 1990 are excluded from the sample. The
information on the transition economies and population size is from
the World Bank Global Development Network Database (GDN) and the
Penn World Table 6.2 from Heston et al. (2006), respectively.
2.2.2

Measures of financial development

Since there is no single aggregate index for financial development in the
literature, we use principal component analysis based on widely used
indicators of financial development to produce new aggregate indices.
Essentially the principal components analysis takes N specific indicators and produces new indices (the principal components) X1 , X2 ,...XN
that are mutually uncorrelated. Each principal component, as a linear
combination of the N indicators, captures a different dimension of the
data. Typically the variances of several of the principal components are
low enough to be negligible, and hence the majority of the variation
in the data will then be captured by a small number of indices. This
chapter uses the first principal component, which accounts for the greatest amount of the variation in the original set of indicators, in the sense
that the linear combination corresponding to the first principal component has the highest sample variance, subject to the constraint that the
sum-of-squares of the weights placed on the (standardized) indicators is
equal to one.
The conventional measures of financial development on which the
principal component analysis is based are as follows.3
The first measure, Liquid Liabilities (LLY), is one of the major indicators used to measure the size, relative to the economy, of financial

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 14 — #5

General Determinants of Financial Development 15

intermediaries, including three types of financial institutions: the central
bank, deposit money banks and other financial institutions. It is calculated as the liquid liabilities of banks and non-bank financial intermediaries (currency plus demand and interest-bearing liabilities) over GDP.
The second indicator, Private Credit (PRIVO), is defined as the credit
issued to the private sector by banks and other financial intermediaries
divided by GDP, excluding credit issued to government, government
agencies and public enterprises, as well as the credit issued by the monetary authority and development banks. It measures general financial
intermediary activities provided to the private sector.
The third, Commercial-Central Bank (BTOT ), is the ratio of commercial bank assets to the sum of commercial bank and central bank assets. It
proxies the advantage of financial intermediaries in channelling savings
to investment, monitoring firms, influencing corporate governance and
undertaking risk management relative to the central bank.
Next are two efficiency measures for the banking sector. Overhead
Costs (OVC) is the ratio of overhead costs to total bank assets. The
Net Interest Margin (NIM) equals the difference between bank interest income and interest expenses, divided by total assets. A lower value
of overhead costs and net interest margin is frequently interpreted as
indicating greater competition and efficiency.
The last are three indices for stock market development.4 Stock Market
Capitalization (MCAP), the size index, is the ratio of the value of listed
domestic shares to GDP. Total Value Traded (TVT ), as an indicator to
measure market activity, is the ratio of the value of domestic shares traded
on domestic exchanges to GDP, and can be used to gauge market liquidity
on an economy-wide basis. Turnover Ratio (TOR) is the ratio of the value
of domestic share transactions on domestic exchanges to the total value
of listed domestic shares. A high value of the turnover ratio will indicate
a more liquid (and potentially more efficient) equity market.
The data are obtained from the World Bank’s Financial Structure and
Economic Development Database (2008) and averaged over 1990–2001.
Any country for which fewer than three years of data are available is
omitted from the sample.
Appendix Table A2.4 presents the eigenvalues, proportion explained
and the eigenvector of each first principal component from which the
new indices of financial development are defined. It reports the sample variance of each first principal component (linear combination), the
proportion of the variance in the raw data the first principal component
accounts for and the coefficient (weight) of each existing standardized
measure in the linear combination.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 15 — #6

16 Determinants of Financial Development

(1) The first is a measure of overall financial development, denoted by
FD. This is based on eight components, namely Liquid Liabilities, Private
Credit, Commercial-central Bank, Overhead Cost, Net Interest Margin, Stock Market Capitalization, Value Traded and Turnover. The first
principal component accounts for 49% of the variation in these seven
indicators. In Appendix Table A2.4 the coefficients of each financial
indicator for FD indicate the negative correlations between the Overhead Cost and Net Interest Margin and FD, and the positive correlations
between the rest and FD.
(2) A second measure, FDBANK, captures the extent of bank-based
intermediation. It uses five indicators, Liquid Liabilities, Private Credit,
Commercial-central Bank, Overhead Costs and Net Interest Margin.
FDBANK accounts for 61% of the variation in these five indicators.
(3) A third measure, FDSTOCK is a measure of stock market development, based on Stock Market Capitalization, Value Traded and Turnover.
FDSTOCK accounts for 66% of the variations in these financial indices.
(4) A fourth measure, FDEFF, captures financial efficiency. The four
indicators of financial efficiency used are Overhead Cost, Net Interest
Margin, Value Traded and Turnover. FDEFF accounts for 54% of the total
variation in these indicators. Lower values of this index indicate a higher
level of financial efficiency.
(5) A fifth measure, FDSIZE, based solely on Liquid Liabilities and Stock
Market Capitalization, captures the size of financial system (also called
“financial depth”). The first principal component of these two measures
accounts for 81% of the variation.

2.2.3

The potential determinants

Potential determinants of financial development considered in this analysis are widely selected from various sources. To discover the structural
determinants of financial development, they are either those “predetermined” like fixed factors, or those “evolving slowly over time” like some
institutional factors which are averaged over 1960–89. All variables that
could potentially cause serious endogeneity problems are excluded.5 The
candidate determinants are grouped into four categories as showed in
Appendix Table A2.1. The problem of missing data has been addressed
by using a set of fixed factors as independent variables to impute the
missing data. The fixed factors used include some regional dummies,
dummies for income levels and geographic factors for which we have
a complete set of data. The imputation procedure is summarized in
Appendix Table A2.5.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 16 — #7

General Determinants of Financial Development 17

2.2.3.1

Institutional variables

This analysis firstly considers legal origin dummies from the GDN dataset
in the work by La Porta et al. (1997, 1998) on the legal determinants
of financial development. The relevant variables are the common law
legal origin dummy (COMLEG) for countries with British legal origin
and a civil law legal origin dummy (CIVLEG) for countries with French,
Germany and Scandinavian legal origins. Two variables closely related
to the financial system itself are also considered.6 Taken from the dataset
of La Porta et al. (1998), SRIGHT is the aggregate index for shareholders’
rights ranging from 0 to 6, while CRIGHT is the aggregate index for
creditors’ rights ranging from 0 to 4. These variables measure directly
the extent to which the government protects the rights of shareholders
and creditors.
In addition, this research makes use of some general institutional indicators. POLITY2 and DURABLE are taken from the PolityIV Database
(Marshall and Jaggers, 2009), and averaged over 1960–89. POLITY2 is
an index of democracy, seeking to reflect government type and institutional quality based on freedom of suffrage, operational constraints and
balances on executives and respect for other basic political rights and civil
liberties. It is called the “combined polity score”, equal to the democracy
score minus the autocracy score. The democracy and autocracy scores
are derived from six authority characteristics (regulation, competitiveness and openness of executive recruitment, operational independence
of chief executive or executive constraints and regulation and competition of participation). Based on these criteria, each country is assigned a
democracy score and autocracy score ranging from 0 to 10. Accordingly,
POLITY2 ranges from -10 to 10 with higher values representing more
democratic regimes. DURABLE is an index of political stability, using the
number of years since the last transition in the type of regime or independence. The next variable is FREE, the average of the indices of civil
liberties and political rights from the Freedom House Country Survey
(2008) over 1972–89. Higher ratings indicate better civil liberties and
political rights such as freedom to develop views, institutions and personal autonomy from government. I also employ KKM and PCI. The
KKM measure from Kaufmann et al. (2008) is a widely used indicator of
the quality of government in a broader sense, derived by averaging six
measures of government quality: voice and accountability, political stability and absence of violence, government effectiveness, light regulatory
burden, rule of law and freedom from graft. The variable PCI, measuring narrowly the constraints on the executive, is derived by Henisz

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 17 — #8

18 Determinants of Financial Development

(2000). The last institutional variable I use is EURO1900, the percentage
of population that was European or of European descent in 1900, taken
from Acemoglu et al. (2001).
Although missing values for EURO1900, SRIGHT , CRIGHT and the
market share of state-owned media (discussed below) are imputed, the
variable EURO1900 appears only in the developing country sample while
the others appear only in the La Porta sample.
2.2.3.2

Policy variables

To examine whether macroeconomic policy variables explain crosscountry variation in financial development, this research makes extensive use of five economic volatility indicators and three trade openness
indicators. It uses output volatility and inflation volatility to capture
macroeconomic mismanagement and fluctuations. The output volatility measure (SDGR) is defined as the standard deviation of the annual
growth rate of real, chain-weighted GDP per capita over 1960–89 from
the Penn World Table 6.2. Inflation volatility (SDPI) is defined as the
standard deviation of the annual inflation rate over 1960–89 from the
World Development Indicators (2008). Taken from the GDN, the volatility of the black market premium (SDBMP), volatility of the terms of trade
(SDTT ) and trading partners’ output volatility (SDTP) are used to reflect
the extent of external shocks. SDBMP is defined as the standard deviation of the annual black market premium (BMP) over 1960–89. SDTT is
defined as the standard deviation of the first log-differences of a terms
of trade index for goods and services. SDTP is the standard deviation
of trading partners’ GDP per capita growth (weighted average by trade
share).
To assess the role of trade factors, this research uses dummies for fuel
and non-fuel primary goods exporting countries (EXPPRIM) and manufactured goods exporting countries (EXPMANU ) from the GDN. A trade
openness policy index, TOPEN, available from the database of Harvard University’s Center for International Development (Gallup et al.,
1999), is utilized to measure the extent of openness to external trade
in the presence of government intervention over 1965–90, while the
trade share proposed by Frankel and Romer (1999), denoted by CTRADE,
is employed to capture natural openness to external trade. CTRADE is
derived by Frankel and Romer (1999) by summing up all bilateral trade
with all potential trading partners from a bilateral trade equation that
controls for population and land area of the home country and trading
partners, the distance between any two trading partners and whether or
not the home country is landlocked.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 18 — #9

General Determinants of Financial Development 19

2.2.3.3

Geographic variables

To examine the role of geography, this study takes six regional dummies from the GDN for East Asian and Pacific countries (REGEAP),
Middle Eastern and North African countries (REGMENA), Western European and North American countries (REGWENA), South Asian countries
(REGSA), Sub-Saharan African countries (REGSSA) and Latin American
and Caribbean countries (REGLAC), respectively. It also uses the following two geographic variables from the GDN. The landlocked variable
(LANDLOCK) is a dummy variable that takes the value of 1 if the country has no coastal access to the ocean, and 0 otherwise. There are 17
landlocked countries in the whole sample. Absolute latitude (LATITUDE)
equals the absolute distance of a country from the Equator. The closer to
the equator the countries are, the more tropical climate they have.7 Latitude potentially has an institutional interpretation since smaller absolute
latitudes are associated with more unfavourable environments, which
are associated with weaker institutions according to the settler mortality hypothesis of Acemoglu et al. (2001). The land area (AREA) in
square kilometres for each country, taken from Hall and Jones (1999), is
in logs.
This study also makes use of three additional geographic variables.
One is POP100CR from the database of Harvard University’s Center for
International Development. It is the 1994 share of population within
100 km of a coast or navigable river for a country. Another is MINDIST ,
based on data from Jon Haveman’s International Trade website. This
captures the minimum distance from the three capital-goods-supplying
centres in the world (USA, Japan and the EU, the centre of the latter
represented by Belgium). The study uses the logarithm of the minimum
distance from the three capital-goods-supplying centres plus one. These
variables might be highly correlated with external trade and manufacturing, since lack of access to coasts or rivers navigable to the ocean
and geographic remoteness constitute natural disadvantages to external trade. A further variable for geographic endowment is a dummy
for the point source natural resource exporting countries (RESPOINT )
from Isham et al. (2005), who find that, in comparison to manufacturing exporters and exporters of “diffuse” natural resources (e.g. wheat, rice
and animals) and coffee/cocoa natural resources, the exporting countries
of “point source” natural resources (e.g. oil, diamonds and plantation
crops) are more likely to have severe social and economic divisions,
and less likely to develop socially cohesive mechanisms and effective
institutional capacities for managing shocks.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 19 — #10

20 Determinants of Financial Development

2.2.3.4

Other variables

Other variables included in this analysis are initial income (GDP90), initial population (POP90), an ethnic fractionalization index (ETHNIC),
an ethnic polarization index (ETHPOL), a religious fractionalization
index (RELIGION), a language fractionalisation index (LANGUAGE),
a European first language index (EURFRAC) and the market share of
state-owned media, either television or newspapers (MEDSHARE).
The inclusion of the level of GDP per capita in 1990 (GDP90) is stimulated by work such as Greenwood and Smith (1997) on the feedback
from growth in the economy to the development of financial markets.
Population size is also closely related to indices of financial development
since small countries tend to have higher ratios of liquid liabilities and
private credit, having the potential to affect the overall results substantially. GDP90 and POP90, the level of the population in 1990, are from
the GDN and used in logs.
The variables ETHNIC, RELIGION and LANGUAGE, taken from Alesina
et al. (2003), characterize social divisions and cultural differences, as does
the variable ETHPOL, which is taken from Reynal-Querol and Montalvo
(2005) to capture the extent to which a large ethnic minority faces an
ethnic majority in a society. The EURFRAC measure, taken from Hall and
Jones (1999), is the fraction of population speaking one of the major
languages of Western Europe (English, French, German, Portuguese or
Spanish) as a mother tongue. To some extent, this variable reflects not
only the culture of the country, but also the history of colonization. It is
therefore closely linked to some other variables like EURO1900, CIVLEG
and COMLEG.
The market share of stated-owned media (MEDSHARE) is from Djankov
et al. (2003), which shows that greater state ownership of the media is
associated with less political and economic freedom, inferior governance,
less developed capital markets and poor health outcomes. Djankov et al.
(2003) consider two kinds of media state ownership. One is press state
ownership, the market share of state-owned newspapers out of the aggregate market share of the five largest daily newspapers (by circulation), and
the other is television state ownership, the market share of state-owned
television stations out of the aggregate market share of the five largest
television stations (by viewer). The index used here is the average of the
two media state ownerships.

2.3

Empirical strategy

This section discusses the empirical strategies for dealing with model
uncertainty faced by research on the determinants of financial development, with the central focus placed on BMA and Gets approaches.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 20 — #11

General Determinants of Financial Development 21

As summarized in the Introduction, substantial research has been
done to explore the origins of financial development, leading to a large
number of candidate determinants. Essentially the associated theories,
developed under specific settings, are not mutually exclusive, raising
concern over the robustness of these candidate determinants in any
cross-section regression used to explain financial development.
Usually, the uncertainty about the composition of a regression model
is called “model uncertainty”. To handle the model uncertainty issue,
a number of methodologies have been proposed and widely debated.
Among others, the Extreme Bounds Analysis (EBA), BMA and Gets are
the most famous methods.
To handle the model uncertainty issue, a number of methodologies
have been proposed and widely debated. Among others, the EBA,8 BMA9
and Gets10 are the most widely used methods. Although the BMA and
Gets procedures have respective advantages in handling model uncertainty, neither of them is without limits or exempt from criticism.11 This
research chooses to apply the BMA and Gets procedures jointly to handle
model uncertainty in this context. The combination of Gets and BMA
analyses has the advantage of incorporating their merits while circumventing some of their limitations. In what follows, I set out the methods
of BMA and Gets in more detail.
2.3.1

Bayesian Model Averaging

This section begins with a brief review of the development of BMA
approach.
Following the seminal work by Levine and Renelt (1992), Sala-i-Martin
(1997a,b)12 , Fernandez et al. (2001)13 and Sala-i-Martin et al. (2004)
are among the significant works using BMA to study the robustness
of cross-country growth regressions. Based on work by Raftery (1995),
Sala-i-Martin et al. (2004) propose a version of BMA called Bayesian Averaging of Classical Estimates (BACE), in which diffuse priors are assumed
for the parameters and only one other prior, relating to the expected
model size, is required. This approach has generated evidence in favour
of Sala-i-Martin (1997a,b)’s original findings as well.
Essentially, BMA treats parameters and models as random variables
and attempts to summarize the uncertainty about the model in terms
of a probability distribution over the space of possible models. More
specifically, it is used to average the posterior distribution for the parameters under all possible models, where the weights are the posterior
model probabilities (PMPs). To evaluate these, the BMA uses the Bayesian
Information Criterion (BIC) to approximate the Bayes factors which are
needed to compute the posterior model probability, whose derivation is
described in the Appendix Text.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 21 — #12

22 Determinants of Financial Development

Typically, the number of possible models, 2p given p candidate variables, is large. Most applications of BMA to larger datasets do not average
over all possible models, but use a search algorithm to identify the subset of models with greatest relevance. The Occam’s Window and Markov
Chain Monte Carlo techniques can be adopted for this purpose.14 The
approach developed by Hoeting et al. (1996) has the advantage of selecting variables and identifying outliers simultaneously, but requires a larger
sample size relative to the regressor set, and so this method will be applied
only in Table 2.1 below. The simpler version of BMA used elsewhere in
this study follows Raftery et al. (1997) which focuses only on the subset
defined by the Occam’s Window technique and treats all the worstfitting models outside the subset as having zero posterior probability.
Embodying the principle of parsimony,15 the use of the Occam’s Window technique considerably reduces the number of possible models, and
in the meantime encompasses the inherent model uncertainty present.
Once the Occam’s Window technique excludes the relatively unlikely
models, the posterior model probabilities for the well-fitting models are
then calculated.
Once we have posterior model probabilities, we are ready to implement a systematic form of inference for different quantities of interest.
For example, when the interest is one of the regression parameters being
present, whether positive or negative, what we need to do is to sum up
the posterior model probabilities for all models in which the parameter
is non-zero, be it positive or negative. In Sections 2.4 and 2.5 below, on
the empirical results, the output of the BMA analysis includes the posterior inclusion probabilities for variables and a sign certainty index. The
posterior inclusion probability (PIP) for any particular variable is the sum
of the posterior model probabilities for all of the models including that
variable. The higher the posterior probability for a particular variable,
the more robust that determinant for financial development appears to
be. For PIPs greater than 0.20, a sign certainty index rather than sign
certainty probability is presented, indicating whether the relationship
appear to be either positive or negative.16
2.3.2

General-to-specific approach

The Gets modelling strategy starts from the most general unrestricted
model (GUM), which is assumed to characterize the essential datagenerating process (DGP), applies standard testing procedures to eliminate statistically insignificant variables and ends up with a “congruent”
final model, which should be free of significant mis-specification. Hoover
and Perez (1999) make important advances in practical modelling, like

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 22 — #13

General Determinants of Financial Development 23

the multiple-path approach to Gets model selection. Based on these, the
PcGets algorithm has been developed to embody the principles of the
underlying theory of Gets reductions extensively discussed in Hendry
(1995).
The selection of models by PcGets roughly includes three stages.17
The first stage concerns the estimation and testing of the GUM. The
GUM should be formulated carefully based on previous empirical and
theoretical findings, institutional knowledge and data characteristics.
The specification of the GUM should be sufficiently general with a relatively orthogonal parameterization for the N candidate regressors. The
next step is to conduct a mis-specification test for “congruence” of the
initial GUM. The congruence of the initial GUM is maintained through
the selection process to ensure a congruent final model. Once the congruence of the GUM is established, pre-search reduction tests are conducted
at a loose significance level. The statistically insignificant variables are
eliminated both in blocks and individually, and the GUM reformulated
as the baseline for the next stage.
The second stage is the search process. Many possible reduction paths
are investigated to avoid path-dependent selection. The terminal model
emerges from each path when all reduction diagnostic tests are valid and
all remaining variables are significant. At the end of the path searches,
all distinct terminal models are collected and tested against their union
to find an un-dominated encompassing contender. If a unique model
results, it is selected; otherwise, the “surviving” terminal models form
a union as a new starting point for reduction. The search process continues until a unique model occurs, or the union coincides with either
the original GUM or a previous union. If a union made up of mutually
encompassing and un-dominated models results, PcGets employs the
BIC to select the unique final model.
The third stage is the post-search evaluation. At this stage PcGets uses
post-selection reliability checks to evaluate the significance of variables
in the final model selected in two overlapping subsamples.
Obviously, the choice of critical values for pre-selection, selection
encompassing tests and subsample post-selection is important for the
success of the PcGets algorithm. It provides two basic strategies, liberal
and conservative, for the levels of significance, degree of pretesting and
so on. The liberal strategy tries to equate the probability of deleting relevant and retaining irrelevant variables, whilst the conservative strategy
tries to reduce the chance of retaining irrelevant variables. The choice of
different strategies hence affects the chance of either retaining irrelevant
variables or dropping relevant variables. Throughout the chapter, PcGets

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 23 — #14

24 Determinants of Financial Development

is conducted with a more liberal strategy than the default setting of the
“liberal strategy” as presented in Appendix Table A2.6,18 aiming to keep
all promising variables in the final model. The final conclusions are then
based on the intersection of the BMA and Gets results.

2.4

Empirical results (I): Overall financial development

This section begins studying the determinants of various indices of financial development. The BMA and Gets methods are applied and compared
in three different samples (the whole sample, the developing country
sample and the La Porta sample) for each index. This section, the central
contribution of this analysis, studies the determinants of overall financial development (FD). Section 2.5 is concerned with the determinants
for four specific indexes of financial development, followed by a study
of the determinants of bond market development.
2.4.1

Some stylized facts

As a starting point, it might be useful to look at some stylized facts on
the links between some important institutional, policy and geographic
variables and FD. These figures are based on the whole sample.
Figure 2.1 presents two scatter plots for the links between institutions and financial development. Better institutional quality, captured
by KKM, and a more democratic regime, captured by POLITY2, are associated with higher values for FD. The trade policy index denoted by TOPEN
and Frankel-Romer trade share denoted by CTRADE are positively related
to FD in Figure 2.2. The upper chart of Figure 2.3 indicates that countries closer to the main world market centres achieve a higher level of FD,
while the lower chart shows that financial markets in countries further
from the equator are relatively more advanced.
Figure 2.4 portrays the evolution of averaged liquid liability (LLY) over
1960–2003 by different country groups. Note from the upper-left chart
that countries in all income groups experienced an increase in LLY,
although higher-income countries remain at a higher level of financial
development than lower-income countries throughout. The upper-right
chart shows considerable differences in averaged LLY between manufactured goods exporting countries and primary goods exporters in which
the latter remain at lower levels or at least partially financially repressed.
The lower-left chart shows that the level of LLY in West European and
North American countries was much higher and more stable than that in
other country groups. The development process of LLY in East Asian and
Pacific countries was much more pronounced relative to that in any other
country group. In the lower-right chart, the development performance

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 24 — #15

General Determinants of Financial Development 25

Institutional quality and new index FD
6
CHE

4

MYS
JPN

GBR

NLD

USA
CHN

2

KOR

THA KWT

FD

JOR
ZAF ISR
MUS
EGY PHL
ITA
PAN
TUN
IND
MAR
OMN
TTO
IDN
BGD
PAK
LKA
BOL
NPL
MEX
CRI
ARG
KEN
URY
CIV
SLV BRA
GTM
TUR
COL
PER
JAM
ECU
PRY

0
−2

ZWE

NGA

SWE

CYP MLT

CANNZL
IRL
AUS
DEU
FIN

BHR

BRB
CHL

NOR
ISL
DNK

GHA

−4

VEN
ZMB

−2

−1
0
1
Institutional quality denoted by KKM

2

Democracy and new index FD
6
4
JPN
USA

2

CHN

KOR

CYP
CAN

FD

AUS
JOR
NOR

0

EGY

TUN

ITA

PAN
CHL
PAK

FIN
ISR
IND

BOL
MEX
KEN

−2

TUR

ARG

BRA

PER
ECU

COL

SLV
JAM

GHA

−4
−10

Figure 2.1

VEN

−5

0
5
Democracy index POLITY2

10

Scatter plots of institutions and financial development

Note: Variables and data sources are described in Appendix Table A2.1. These
figures show scatter plots of the institutional quality denoted by KKM, and the
democracy index POLITY2, against the new index FD.

of LLY in common law countries was in general much more gradual, with
the whole process stretching over four decades compared to that in civil
law countries, which experienced surges in the 1970s and late 1990s, but
a decline in the late 1980s.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 25 — #16

26 Determinants of Financial Development
Trade policy and new index FD
6
CHE

4

MYS

2

CHN

SWE
THA
IRL CAN
AUS
DEU

0

EGY
IND
TTO
BGD
PAK

FD

NLD
JPN
GBR
USA
KOR
NZL

−2

FIN
JOR
NOR
MUS
DNK
ITA

ISR

ZAF
TUN PHL

MAR

CHL

IDN

LKA

BOL

NPL
MEX
ARG
KEN CRI
URY
CIV
COL
BRA SLV TURGTM
PER
ZWE
PRY
NGA

JAM

ECU

GHA

−4

VEN
ZMB

0

.2

.4
.6
.8
Trade policy denoted by TOPEN

1

Frankel−Romer trade share and new index FD
6
CHE

4

MYS
JPN

NLD

GBR

USA

FD

2

CHN

THA
CAN NZL
AUS

KOR SWE

KWT
FIN

ZAF
PHLEGYITA
IND
MAR
CHL
IDN PAK
BGD
LKA
BOL
NPL
MEX
ARG
KEN URY
CIV
TUR
COL
BRA PER
ZWE
ECU
PRY

0

−2

NGA

−4

CYP

IRL

DEU
NOR
PAN
TUN

ISR
MUS
DNK
TTO OMN

JOR

BRB

CRI
GTM
JAM

SLV

GHA

VEN
ZMB

0

Figure 2.2

20
40
60
Frankel−Romer trade share denoted by CTRADE

80

Scatter plots of policy and financial development

Note: Variables and data sources are described in Appendix Table A2.1. These
figures show scatter plots of the trade policy index from Gallup et al. (1999),
and the trade share constructed by Frankel and Romer (1999), against the new
index FD.

The figures above have shown some interesting facts on the determinants of FD. However, a clear conclusion on the robustness of any
variable presented cannot readily be drawn. The task of the subsequent
Section 2.4.2 is to examine these links more systematically.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 26 — #17

General Determinants of Financial Development 27
Minimum distance and new index FD
6
CHE

4

MYS
NLD

GBR
SWE
KOR

2
FD

DEU

CHN CYP
MLT

DNK

NZL
AUS

BHR
FIN

0

KWT
THA

CAN
IRL
NOR
ITA

JOR
ISR

ZAF
MUS

EGY
PAN
TUN ISL PHL
MAR
TTO

IND
OMN CHL
IDN
BGD
PAK
LKA
BOL
NPL
MEX
CRI
KEN ARG
URY
SLVCOL CIV BRA
TURGTM
PER ZWE
JAM
ECU
PRY

−2

NGA
GHA

−4

VEN
ZMB

5
6
7
8
9
Minimum distance denoted by MINDIST
Absolute latitude and new index FD
6
CHE

4

MYS
NLD
GBR

JPN
USA

2

CHN
KWT

THA

FD

BHR

0
−2

AUS

SWE

KOR
CYP
MLT
NZL

ISR
ZAF JOR
MUS
PHL
EGY
PAN BRB
TUN
IND
MAR
OMN
CHL
TTO
IDN
BGD
PAK
LKA
BOL
NPL
MEX
CRI
ARG
KEN
URY
CIV
SLV
GTM
COL
BRA
PER
JAM
ZWE
ECU
PRY

CAN

DEU

IRL
FIN
NOR

ITA

ISL

DNK

TUR

NGA
GHA

−4

VEN
ZMB

0

Figure 2.3

20
40
Absolute latitude denoted by LATITUDE

60

Scatter plots of geography and financial development

Note: Variables and data sources are described in Appendix Table A2.1. These
figures show scatter plots of the logarithm of minimum distance, and the absolute
latitude, against the new index FD.

2.4.2

What are the main determinants of FD?

As mentioned earlier, much research regards institutions as the fundamental factor in long-run growth, and some even argue that the only
effect of geography on development is via institutions (Acemoglu et al.,

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 27 — #18

28 Determinants of Financial Development

0.75

High-income
Mid-income
Low-income

0.50

0.25

1960

1970

1980

1990

2000

0.5

0.4
Common law
Civil law
0.3

1960

1970

1980

1990

2000

0.7
0.6
0.5
Manufactured exporters
Primary goods exporters

0.4
0.3
0.2
1960

1970

1980

1990

2000

0.75
REGEAP
REGSA
REGWENA
REGSSA
REGLAC
REGMENA

0.50

0.25

1960

Figure 2.4

1970

1980

1990

2000

Median Liquid Liability by different country group over 1960–2003

Note: Variable descriptions are from Appendix Table A2.1. These figures plot the
median liquid liabilities by different income groups in the upper-left chart, countries with different law traditions in the upper-right chart, different exporting
countries in the lower-left chart and different regions in the lower-right chart
over 1960–2003.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 28 — #19

General Determinants of Financial Development 29

2001; Dollar and Kraay, 2003; Easterly and Levine, 2003 and Rodrik
et al., 2004). Before proceeding to study the main determinants of overall FD, this section starts by testing the hypothesis of whether any of
three determinants (institutions, policy and geography), considered as a
whole, dominates the other two.
Table 2.1 reports the BMA results for determinants of FD, which is measured over 1990–99, for 64 countries in the whole sample. All possible
explanatory variables are grouped into four blocks in the order of “other”
variables, geographic variables, policy variables and institutional variables. In addition to including the “other” variables, models 1–3 include
any two of the three blocks (geographic variables, policy variables and
institutional variables) to examine the combined effects of any two types
of determinants on FD.19
The BMA analysis yields posterior inclusion probabilities (either “PIPs”
or “MC3 ”),20 the total posterior model probabilities for the set of models which include a given variable of interest and the sign certainty
index (“Sign”) of a relationship discussed above. The PIPs are the posterior inclusion probabilities calculated by using the method from Raftery
(1995). A sign certainty index is provided where the PIPs are above
0.2. The MC3 denotes the posterior inclusion probabilities computed
by using the Markov Chain Monte Carlo techniques due to Hoeting
et al. (1996), which conduct variable selection and outlier identification
simultaneously. Any MC3 greater than 0.2 is shown in bold.
Looking at the first block of “other” variables across models, we note
that initial income, GDP90, appears to be important in almost all models
with a high posterior probability of inclusion, meaning that, as expected,
the level of GDP per capita is fundamental in explaining the crosscountry variation in FD. Other variables in this block exhibit varying
explanatory power for FD. Models 1 and 2 present the effect of geography
on FD when policy and institutions, respectively, are controlled for. The
effect of geography on FD doesn’t seem to disappear when the institutional variables are present, implying that the usual claim that geography
works through institutions is not necesarily true in this context. The two
BMA methods show that two regional dummies (REGSSA and REGLAC)
appear to be closely related to FD, meaning that a number of developing
countries in these regions are associated with higher levels of financial
development in the 1990s, conditional on other variables. The regional
dummy REGEAP and land area (AREA) also appear to be important predictors of FD when institutions are controlled for. Similarly, policy has
a significant effect on FD in the presence of geography and institutions
(Models 1 and 3). Among others, at least EXPPRIM is significant in both

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 29 — #20

30 Determinants of Financial Development
Table 2.1 Determinants of FD by using BMA
Whole

Whole

Whole

64

64

64

1

2

Variable

PIPs

Sign

CONSTANT
GDP90
POP90
ETHPOL
ETHNIC
RELIGION
LANGUAGE
EURFRAC

1.000
0.466
0.000
0.004
0.000
1.000
0.000
0.000

(−)
(+)

REGEAP
REGMENA
REGSA
REGSSA
REGLAC
REGWENA
LANDLOCK
LATITUDE
AREA
MINDIST
POP100CR
RESPOINT

0.186
0.314
0.186
0.879
0.872
0.204
0.000
0.000
0.073
0.030
0.051

TOPEN
CTRADE
EXPMANU
EXPPRIM
SDGR
SDBMP
SDPI
SDTP
SDTT

0.850
0.099
0.000
0.409
0.000
0.252
0.076
0.000
0.329

CIVLEG
COMLEG
POLITY2
DURABLE
FREE
KKM
PCI
EURO1900
MEDSHARE
SRIGHT
CRIGHT

(+)

(+)
(−)
(−)
(−)

0.037
(+)
(−)
(−)
(−)

MC3

PIPs

Sign

0.342
0.026
0.039
0.028
0.029
0.056
0.992

1.000
0.941
0.969
0.649
0.014
0.056
0.036
0.706

(−)
(+)
(+)
(+)
0.962

0.132
0.176
0.110
0.946
0.942
0.175
0.065
0.056
0.034
0.032
0.027
0.005

0.726
0.006
0.053
0.642
0.385
0.036
0.003
0.386
0.975
0.012
0.400
0.025

0.995
(−)
(+)
(−)
(−)
0.208
(−)
(−)
(−)

3
MC3

0.488
0.839
0.070
0.000
0.071
0.000
0.040

PIPs

Sign

MC3

1.000
0.744
0.941
0.906

(−)
(+)
(+)
(+)
0.333

0.689
0.517
0.035

0.099
0.982

0.063
0.831
0.623
0.056

0.045
0.215
0.000
0.999
0.000
0.000
0.064
0.398
0.045
(−)
(−)
(+)
(+)
(+)
(−)

0.089

0.958
0.060
0.037
0.049
0.071
0.094

0.927
0.049
0.030
0.175
0.031
0.192
0.030
0.023
0.201
0.589
0.361
0.291
0.300
0.020
0.988
0.963

0.029
0.962
(−)

0.740
0.358
0.051
0.058
0.069
1.000
0.996

0.461
0.128
0.258
0.022
0.006
1.000
0.924

(+)
(−)

(+)
(−)
(+)
(+)
(−)

0.309
0.050
0.025
0.964
0.069
0.126
0.053
0.228
0.026
0.867
0.467
0.050
0.031
0.084
0.999
0.953

Note: The dependent variable FD is the aggregate index of overall financial development over period, 1990–
99. Variable description is in Appendix Table A2.1. BMA yields the posterior probabilities of inclusion (either
“PIPs” or “MC3 ”), the total posterior model probabilities for all models including a given variable and the
sign certainty index of a relationship (“Sign”). A sign is given to PIPs greater than 0.2. No sign givern means
the sign of estimated relationship being uncertain. Any MC3 greater than 0.2 is in bold. The PIPs is taken
from Raftery (1995) while the MC3 is due to Hoeting et al. (1996) who also identify the outliers.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 30 — #21

General Determinants of Financial Development 31

cases. Neither does the usual claim that policy works through institutions
by affecting their quality apply to this context. Models 2 and 3 show
that the role of institutions is not altered when geography and policy
are controlled for. Note that most of the institutional variables appear
to be significant predictors of FD, in particular, the KKM (governance
index) and PCI (political constraints index) have a posterior probability
of inclusion close to 1.
Overall, Table 2.1 has demonstrated that geography, institutions and
policy as a group are all important in the process of financial development, although their effects may be picked up by varied predictors when
conditioning on other factors is in place. These results clearly suggest that
it would be more appropriate to include all of them in the analysis.
Table 2.2 contains a thorough study of determinants of FD by using
BMA and Gets in which the above conclusion (in terms of geography,
institutions and policy all being important) is embodied. The BMA analysis reports PIPs and the sign certainty index (“Sign”) discussed above.
The Gets analysis produces the coefficients and t-values for possible
determinants in the final model. It also reports the residual sum-ofsquares (RSS), the equation standard error or residual standard deviation
2
(sigma), the squared multiple correlation
 coefficient (R ) and its val

ues adjusted for degree of freedom R2adj , the log-likelihood value and

three information criteria: the Akaike Information Criterion (AIC), the
Hannan-Quinn Criterion (HQ) and the Schwarz Criterion (SC). The output also includes three mis-specification tests (Chow test, Normality
test and Heteroscedasticity test).21 The Gets results in Table 2.2 are the
final models for three samples, respectively, in Appendix Table A2.7,
which clearly shows the variables included in the GUM and in the final
model.
In Table 2.2, the BMA analysis for the whole sample yields a subset inclusive of four “other” variables (GDP90, POP90, ETHPOL and
EURFRAC), two geographic variables (REGEAP and AREA), four policy
variables (CTRADE, EXPPRIM, SDBMP and SDPI) and five institutional
variables (CIVLEG, COMLEG, DURABLE, KKM and PCI). Given no rejection of the mis-specification tests, the Gets analysis for the whole sample
yields a subset inclusive of three “other” variables (GDP90, POP90 and
EURFRAC), two geographic variables (LATITUDE and AREA), one policy variable (SDTT ) and three institutional variables (CIVLEG, KKM and
PCI). Both the BMA and Gets analyses on the whole sample unanimously
suggest that three “other” variables (GDP90, POP90 and EURFRAC), one
geographic variable (AREA) and three institutional variables (CIVLEG,
KKM and PCI) are the main determinants for FD.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 31 — #22

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 32 — #23
PIPs
1.000
0.946
0.996
0.999
0.000
0.004
0.009
0.998
0.999
0.001
0.027
0.029
0.019
0.002
0.009
0.011
0.992
0.003
0.000
0.023
0.001
0.492
0.000
0.996
0.000
0.346

CONSTANT
GDP90
POP90
ETHPOL
ETHNIC
RELIGION
LANGUAGE
EURFRAC

REGEAP
REGMENA
REGSA
REGSSA
REGLAC
REGWENA
LANDLOCK
LATITUDE
AREA
MINDIST
POP100CR
RESPOINT

TOPEN
CTRADE
EXPMANU
EXPPRIM
SDGR
SDBMP
(−)

(−)

(+)

(−)

−2.948
−4.188

−3.694

−1.287

(−)

−0.041
−0.417

−4.074
4.403
5.856

−10.563
1.391
0.705

(−)
(+)
(+)
(+)

(+)

t-value

Gets
Coeff

Sign

BMA

Whole

Determinants of FD

Variable

Table 2.2

1.000
0.903
0.064
0.002
0.005
0.658

0.005
0.322
0.817
0.046
0.377
1.000

0.001
0.998
0.062
0.465
0.008

1.000
1.000
0.994
0.007
0.027
0.492
0.998
0.002

PIPs

(−)

(+)
(+)

(+)
(−)

(−)
(−)

(−)

0.120

1.990

3.353

3.353

1.487

(−)
(+)

(−)

−15.723
2.049
0.248

Coeff

0.004

5.192

7.548

7.548

2.941

−5.932
7.192
2.855

t-value

Gets

(−)
(+)
(+)

Sign

BMA

Developing Country

0.077
0.980
4.876

0.985
0.050

0.093
0.868
0.189
0.053
0.069
0.965
0.993
0.051
1.000
0.241
0.043
0.434

1.000
0.012
1.000
0.039
0.942
0.067
0.054
0.978

PIPs

(−)

(+)

(−)

(−)
(−)

(−)
(+)

(+)

(−)

(+)

(+)

(+)

Sign

BMA

1.854
0.034
1.246
1.416
−0.571

0.015
−2.960

−0.416

2.435
−7.789
2.562

3.512
2.421
4.264
2.597
−3.875

2.592
−4.633

−4.095

3.317
−5.614
3.979

−4.524

−3.847

−4.224
−3.000

−6.216

8.610
4.652

−3.988

1.314
3.131

t-value

Gets
Coeff

La Porta

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 33 — #24
0.529
0.205
0.058
0.764
0.002
1.000
0.995

0.700
0.033
0.130

(+)
(−)

(+)

(−)
(+)

(+)

0.66
8.46

1.846
−5.363

−0.927

−0.024

0.68
0.01

51.11
0.97
0.80
0.77
7.20
0.09
0.22
0.42

5.064
−5.184

−3.396

−3.270
1.000
1.000
0.992
0.000
0.035
0.370
1.000
0.002

1.000
0.445
0.000

(+)
(−)

(−)
(+)
(−)

(+)
(−)

1.35
1.76

−5.391

−4.562
−3.445
0.164

0.001

0.28
0.41

17.06
0.73
0.86
0.82
20.84
−0.40
−0.22
0.08

−4.026

−5.239
−3.879
4.382

2.193

0.405
0.753
0.066

0.074
0.074
0.059
0.686
0.035
0.992
0.995

0.504
0.177
0.037

(−)
(+)

(+)
(−)

(+)

(−)

1.35

−0.374

−0.453
4.849
−13.547

1.687

−0.035
−1.390
0.036

0.51

3.78
0.50
0.98
0.94
47.21
−1.11
−0.73
−0.05

−3.169

−3.033
9.156
−7.002

4.452

−5.076
−2.263
3.222

Note: The dependent variable FD is the aggregate index of overall financial development over the period, 1990–99. Variable description is in Appendix Table A2.1. There are 64 observations in the whole sample, 44 observations in the developing country sample
and 40 observations in the La Porta sample. BMA analysis yields the posterior probabilities of inclusion (PIPs) and the sign certainty index of a relationship (Sign). No sign given means the sign of estimated relationship being uncertain. Gets analysis yields
coefficients and t -values for the variables in the final model. See text for the description of PcGets output.

RSS
sigma
Rˆ2
Radjˆ2
LogLik
AIC
HQ
SC
Chow test 1
Chow test 2
Normality test
Hetero test

CIVLEG
COMLEG
POLITY2
DURABLE
FREE
KKM
PCI
EURO1900
MEDSHARE
SRIGHT
CRIGHT

SDPI
SDTP
SDTT

34 Determinants of Financial Development
Table 2.3 Top ten models and their posterior probabilities for FD

GDP90
POP90
ETHPOL
EURFRAC
REGEAP
AREA
CTRADE
EXPPRIM
SDBMP
SDPI
SDTT
CIVLEG
COMLEG
DURABLE
KKM
PCI
PMP

1

2

3

4

5

6

7

8

9

10

*
*
*
*
*
*

*
*
*
*
*
*

*
*
*
*
*
*

*
*
*

*

*

*
*
*
*
*
*
*
*
*
*

*
*
*
*
*
*
*
*

*

*
*
*
*
*
*
*
*
*
*

*
*
*
*
*
*

*

*
*
*
*
*
*
*
*
*
*

*
*
*
*
*
*

*

*
*
*
*
*
*
*
*
*
*

*

*

*

*

*
*
*

*
*

*
*
*

*
*
*

*

*

*
*
*
*
*

*
*

*

*
*
*
*

*
*
*
*

*
*
*

*
*
*
*

0.048 0.042 0.042 0.037 0.030 0.028 0.028 0.028 0.028 0.025

Note: This table presents the top ten models for FD, ranked by their posterior model probability
(PMP) in the whole sample. The variable description is in Appendix Table A2.1.

In Tables 2.1 and 2.2, the BMA procedure has yielded PIPs for all candidate variables. A natural question to ask is about the structure of the
models, especially the models with higher explanatory power. Table 2.3
lists the structure of the top ten models for FD in the whole sample
in terms of posterior model probabilities, serving as a concrete illustration of model selection. A noteworthy point is that all these models
have more than ten possible predictors with geographic variables (such
as REGEAP, AREA), policy variables (such as EXPPRIM) and institutional
variables (like KKM and PCI) and “other” variables (like GDP90, POP90,
ETHPOL and EURFRAC) present in all models. However, one should be
aware of the dramatic model uncertainty, reflected by less than 5% posterior model probabilities for all top ten “best” models, which indicates
the potential importance of the BMA and Gets procedures for model
selection as a systematic response to pervasive model uncertainty.
Moving on one step further, OLS regressions are used to estimate some
of the best performing models in Table 2.4. The best model, that is the
model with highest posterior probability in Table 2.3, is presented in
column 4. The “other” variables, like GDP90, POP90 and EURFRAC, are

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 34 — #25

General Determinants of Financial Development 35
Table 2.4 Geography, policy, institutions and FD
(1)
CONSTANT
GDP90
POP90
ETHPOL
EURFRAC
REGEAP
AREA
CTRADE
EXPPRIM
SDBMP
SDPI
SDTT

−15.159
[5.87]∗∗
1.312
[6.25]∗∗
0.521
[3.66]∗∗
1.117
[1.88]
−0.801
[2.32]∗
1.961
[4.61]∗∗
−0.177
[1.58]
0.044
[4.07]∗∗
−0.609
[1.54]
−0.001
[3.79]∗∗
0.001
[4.12]∗∗
−0.010
[1.20]

CIVLEG
COMLEG
DURABLE
KKM
PCI
Standardized coefficients
ETHPOL
EURFRAC
AREA
CTRADE
SDBMP
SDPI
SDTT
DURABLE
KKM
PCI
Observations
R-square

0.49
−0.46
−0.15
−0.04
−0.06
−0.06
−0.07

64
0.740

(2)
−8.220
[2.95]∗∗
0.990
[2.65]∗
0.584
[4.75]∗∗
1.584
[2.89]∗∗
−1.138
[3.84]∗∗
1.277
[3.29]∗∗
−0.457
[4.72]∗∗

−1.159
[2.22]∗
−0.656
[1.28]
0.018
[1.66]
1.489
[4.40]∗∗
−4.006
[4.29]∗∗
0.72
−0.62
−0.29

(3)
−10.874
[3.33]∗∗
1.000
[2.93]∗∗
0.371
[3.12]∗∗
1.029
[1.65]
−1.143
[3.68]∗∗

0.025
[2.01]
−0.970
[3.11]∗∗
0.000
[0.22]
0.001
[3.51]∗∗
−0.010
[0.93]
−1.712
[3.33]∗∗
−0.998
[1.96]
0.011
[0.73]
1.237
[3.30]∗∗
−3.791
[3.98]∗∗
0.45
−0.63

(4)
−8.056
[3.16]∗∗
0.958
[3.01]∗∗
0.512
[4.72]∗∗
1.496
[3.17]∗∗
−1.100
[4.16]∗∗
1.239
[3.92]∗∗
−0.412
[4.41]∗∗
−0.943
[4.06]∗∗

0.001
[4.50]∗∗
−0.600
[2.49]∗

0.017
[1.54]
1.445
[5.12]∗∗
−4.258
[4.90]∗∗
0.68
−0.61
−0.26

−0.05
0.68
−2.05

−0.05
−0.06
−0.06
−0.07
−0.06
0.55
−1.94

−0.05
0.66
−2.17

64
0.820

64
0.790

64
0.860

−0.06

Note: The models are estimated by OLS. The dependent variable is FD, over 1990–99. The t-values are reported
in brackets. Variable descriptions are from Appendix Table A2.1. The standardized coefficients show the
change of a standard deviation of FD due to a one standard deviation change in a variable for those other
than initial GDP and population, binary variables.
∗ , ∗∗ and ∗∗∗ significant at 10%, 5% and 1%, respectively.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 35 — #26

36 Determinants of Financial Development

found significant in every model. The regional dummy REGEAP is significant in all relevant models, showing that the East Asian and Pacific
countries are positively associated with higher FD. While AREA is significant in Models 2 and 4, the standardized coefficient for it is rather
small. For the policy variables, EXPPRIM is significant in Models 3 and
4, but not for Model 1. SDPI is significant in all relevant models, but the
standardized coefficient for it is negligible. Three institutional variables,
CIVLEG, KKM and PCI, are found to be significantly associated with
FD in all relevant models. The effects of KKM and PCI on FD are very
strong, as shown by the standardized coefficients in the lower section
of the table: a one standard deviation change in KKM translates into a
more than 0.5 standard deviation of the FD measure, and even stronger
effects for PCI.
In sum, on the one hand, the analyses above further confirm that
institutions, policy and geography, taken as a group, jointly explain
a substantial proportion of the variation in FD. On the other hand,
the above analyses show that, in comparison to policy and geography,
institutions could play a fundamental role in the process of financial
development. When taken individually, at least CIVLEG, KKM, PCI,
GDP90, POP90 and EURFRAC are found to have a significant influence on
financial development. This finding explicitly suggests that, in addition
to initial GDP and initial population, the legal origin22 and institutional
quality are the most fundamental determinants of financial development
in a country.

2.5

Empirical results (II): Specific financial developments

This section turns to study briefly the determinants of four specific
indices for financial development derived by using principal component analysis, namely, financial intermediary development (FDBANK),
stock market development (FDSTOCK), financial efficiency development
(FDEFF) and financial size development (FDSIZE). Bond market development (FDBOND) is also studied afterwards. The three samples are
investigated for each index in which EURO1900 is available only for the
developing country sample while SRIGHT , CRIGHT and MEDSHARE are
available only for the La Porta dataset sample.
As in the previous section, the Gets model search is conducted with
the relatively liberal strategy presented in Appendix Table A2.6.
The determinants of financial intermediary development (FDBANK)
are reported in Table 2.5. The whole sample has 91 observations, the
developing country sample has 70 and the La Porta sample has 40.23

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 36 — #27

General Determinants of Financial Development 37

The BMA and Gets analyses on the whole sample suggest FDBANK is positively related to initial income. East Asian and Pacific countries, Middle
Eastern and North African countries and South Asian countries witness
relative success in financial intermediary development. MINDIST is suggested to be important as well. The trade open policy index (TOPEN)
and Frankel-Romer index (CTRADE) are significantly positively signed,
suggesting financial intermediary development is boosted by more open
trade policies. Three institutional variables (POLITY2, KKM and PCI) are
suggested to be determinants for FDBANK, consistent with a conventional view that better institutions are associated with better financial
intermediary development. The analyses based on the developing country and La Porta samples in general confirm the findings for GDP90,
REGEAP, REGMENA, TOPEN, KKM and PCI. In addition, the analyses
from the La Porta sample show that shareholders’ right and creditors’
rights may be closely related to financial intermediary development.
The determinants of stock market development (FDSTOCK) are
reported in Table 2.6. The whole sample has 81 observations, the developing country sample has 50 and the La Porta sample has 49. The BMA
and Gets analyses on the whole sample indicate that FDSTOCK is positively related to the initial population and the ethnic polarization index,
while it is negatively related to the language fractionalization index
(EURFRAC).24 East Asian and Pacific countries experience a rise in stock
market development. Land area is also important for FDSTOCK. Among
other policy factors, TOPEN and SDGR are almost suggested by two methods to be in the model – this finding is also supported in the developing
country and La Porta samples. The usual claim concerning the positive impacts of open trade policy on financial development applies here.
The significantly negative effect of output volatility on FDSTOCK means
that macroeconomic mismanagement might exert an adverse effect on
FDSTOCK. Three institutional variables (DURABLE, KKM and PCI) are
suggested to be the main determinants for FDSTOCK. The analyses based
on the developing country and the La Porta samples support the idea that
more open trade policies and better institutions promote stock market
development.
The determinants of financial efficiency (FDEFF) are reported in
Table 2.7. The whole sample has 79 observations, the developing country
sample has 48 and the La Porta sample has 49. Note that the lower value
of FDEFF is associated with a higher level of financial efficiency development as discussed in Section 2.2.2. The BMA and Gets analyses on the
whole sample suggest that RELIGION is significantly related to FDEFF.
East Asian and Pacific countries, South Asian countries, Middle Eastern

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 37 — #28

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 38 — #29
PIPs
1.000
0.832
0.017
0.012
0.001
0.000
0.000
1.000
1.000
1.000
0.001
0.000
0.001
0.008
0.000
0.000
0.804
0.001
0.361
0.703
0.862
0.095
0.126
0.000
0.005

CONSTANT
GDP90
POP90
ETHPOL
ETHNIC
RELIGION
LANGUAGE
EURFRAC

REGEAP
REGMENA
REGSA
REGSSA
REGLAC
REGWENA
LANDLOCK
LATITUDE
AREA
MINDIST
POP100CR
RESPOINT

TOPEN
CTRADE
EXPMANU
EXPPRIM
SDGR
SDBMP

(+)
(+)

(−)

(−)

(+)
(+)
(+)

(−)
(+)

Sign

BMA

−2.630

−0.205

2.118
2.730

−0.817

−0.224

0.790
0.020

6.770
4.115
5.019

−1.579
2.445

−2.523
0.449

1.979
1.327
2.123

t-value

Gets
Coeff

Whole

Determinants of FDBANK

Variable

Table 2.5

0.682
0.158
0.184
0.134
0.000
0.035

0.259
0.000
0.000
0.000
0.000
0.321

0.929
0.928
0.926
0.072
0.072

1.000
0.413
0.000
0.008
0.000
0.004
0.000
0.000

PIPs

(+)

(−)

(−)

(+)
(+)
(+)

(−)
(+)

Sign

BMA

−0.680

1.879
2.025

−4.218
0.568

Coeff

−2.674

5.125
4.534

−2.614
2.598

t-value

Gets

Developing Country

(+)
(−)

(−)

0.083
0.940
0.175

(−)
(−)
(−)
(−)

(−)
(+)

(+)
(+)
(+)
(+)

(−)

(+)
(−)
(+)
(−)

(+)

Sign

0.969
0.949

0.268
1.000
0.344
0.336
0.144
0.767
0.734
0.048
1.000
0.343
0.257
0.770

1.000
0.175
0.858
0.692
0.852
0.203
0.060
0.326

PIPs

BMA

1.942

−0.078

−1.750

−0.967

4.213

−2.270

−3.295

−2.751

t-value

Gets
Coeff

La Porta

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 39 — #30
0.043
0.182
0.335
0.000
0.031
1.000
0.828

0.001
0.000
0.109

0.733
−3.433

(+)

(+)
(−)

1.63
2.94
0.65
20.25

−0.367
0.060

0.000

51.30
0.82
0.82
0.79
26.08
−0.27
−0.11
0.12
0.07
0.01
0.72
0.51

3.628
−3.622

−1.743
2.079

0.383

0.013
0.035
0.059
0.000
0.033
1.000
0.430
0.045

0.035
0.000
0.143

(+)
(−)

2.27
1.45

0.753
−4.340

0.113

0.05
0.48

51.41
0.94
0.65
0.61
8.82
−0.02
0.08
0.23

2.737
−2.944

2.709

(+)
(−)

(+)
(−)

0.116
0.107
0.037
0.028
0.024
1.000
0.832
0.130
0.662
0.729

0.165
0.129
0.062

0.93

1.349

0.691

0.63

29.21
0.93
0.76
0.72
6.29
−0.01
0.08
0.24

4.850

2.305

Note: The dependent variable FDBANK is the index of financial interdediary development over the period, 1990–99. Variable
description is in Appendix Table A2.1. There are 91 observations in the whole sample, 70 observations in the developing country
sample and 40 observations in the La Porta sample. BMA analysis yields the posterior probabilities of inclusion (PIPs) and the
sign certainty index of a relationship (Sign). No sign given means the sign of estimated relationship being uncertain. The Gets
analysis yields coefficients and t -values for the variables in the final model. See text for the description of PcGets output.

RSS
sigma
Rˆ2
Radjˆ2
LogLik
AIC
HQ
SC
Chow test 1
Chow test 2
Normality test
Hetero test

CIVLEG
COMLEG
POLITY2
DURABLE
FREE
KKM
PCI
EURO1900
MEDSHARE
SRIGHT
CRIGHT

SDPI
SDTP
SDTT

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 40 — #31
PIPs
1.000
0.183
1.000
0.985
0.000
0.008
0.039
0.210
0.937
0.062
0.035
0.022
0.000
0.063
0.000
0.000
0.985
0.120
0.037
0.007
0.191
0.000
0.901
0.003
0.141
0.014

CONSTANT
GDP90
POP90
ETHPOL
ETHNIC
RELIGION
LANGUAGE
EURFRAC

REGEAP
REGMENA
REGSA
REGSSA
REGLAC
REGWENA
LANDLOCK
LATITUDE
AREA
MINDIST
POP100CR
RESPOINT

TOPEN
CTRADE
EXPMANU
EXPPRIM
SDGR
SDBMP

−6.484

−0.245

(−)

(+)

−3.001
−4.024
−5.434
−3.245
−2.828
−3.664

−1.301
−1.653
−2.485
−1.427
−1.128
−1.716

(+)

−0.129

−3.320

2.686

−3.016

−0.611

(−)

0.578

8.414
3.379

t-value

0.435
0.791

Coeff

Gets

(+)
(+)

(−)

Sign

BMA

Whole

Determinants of FDSTOCK

Variable

Table 2.6

1.000
0.000
0.003
0.000
0.244
0.736

0.007
0.000
0.000
0.000
0.006
0.532

0.056
0.021
0.351
0.892
0.012

1.000
1.000
1.000
0.000
0.014
0.974
0.005
0.977

PIPs

(−)
(−)

(+)

(−)

(−)
(−)

0.000

0.901

−1.714
−1.623
−2.098
−1.562

0.989

(+)
(−)

−6.444
4.911
7.118

−8.131
0.645
0.290

(−)
(+)
(+)

−3.172

4.497

−2.791
−2.807
−3.419
−3.196

3.567

t-value

Gets
Coeff

Sign

BMA

Developing Country

0.098
0.081
0.165
0.065
0.555
0.252

0.016
0.008
0.721
0.034
0.000
0.033
0.004
0.040
0.935
0.722
0.000
0.000

1.000
0.045
0.919
0.832
0.009
0.232
0.046
0.783

PIPs

(−)
(−)

(−)
(+)

(−)

(−)

(+)

(+)
(+)

(−)

Sign

BMA

−2.367

−1.258

0.602
0.858
−0.351

0.647

1.976
1.948
−3.908

1.361

−4.611
2.361

−3.686

−2.367

−0.462
0.173

−1.689

−2.783

−3.836
4.701
2.897

−0.833

−0.932

−0.669
0.732
1.490

t-value

Gets
Coeff

La Porta

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 41 — #32
0.002
0.111
0.010
0.216
0.134
0.976
0.245

0.000
0.745
0.000

0.22

31.22

4.687
−4.349

1.20
0.88

0.701
−1.828

(+)
(−)

3.003

4.032

14.02
0.47
0.87
0.84
71.04
−1.33
−1.13
−0.83
0.33
0.54

0.013

0.582
(+)

(+)

0.615
0.133
0.045
0.040
0.164
0.463
0.000
0.006

0.068
0.001
0.000

(+)

(−)

1.47
9.04

−0.813

0.161

0.321

0.23
0.01

4.27
0.34
0.84
0.78
61.53
−1.90
−1.70
−1.37

−1.681

2.983

2.212

0.249
0.108
0.002

0.006
0.005
0.036
0.688
0.051
0.894
0.006

0.076
1.000
0.000

(−)

(+)

(+)

(+)

0.15
8.52

1.503

0.022

1.847

0.96
0.01

17.87
0.74
0.78
0.68
24.71
−0.36
−0.12
0.26

4.310

2.619

3.524

Note: The dependent variable FDSTOCK is the index of stock market development over the period 1990–99. Variable description
is in Appendix Table A2.1. There are 81 observations in the whole sample, 50 observations in the developing country sample and
49 observations in the La Porta sample. BMA analysis yields the posterior probabilities of inclusion (PIPs) and the sign certainty
index of a relationship (Sign). No sign given means the sign relationship being uncertain. Gets analysis yields coefficients and
t -values for the variables in the final model. See text for description of PcGets output.

RSS
sigma
Rˆ2
Radjˆ2
LogLik
AIC
HQ
SC
Chow test 1
Chow test 2
Normality test
Hetero test

CIVLEG
COMLEG
POLITY2
DURABLE
FREE
KKM
PCI
EURO1900
MEDSHARE
SRIGHT
CRIGHT

SDPI
SDTP
SDTT

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 42 — #33
PIPs
1.000
0.010
0.041
0.000
0.000
0.375
0.000
0.028
0.989
0.962
0.986
0.033
0.037
0.777
0.000
0.000
0.021
0.066
0.000
0.103
0.052
0.000
0.140
0.926
0.002
0.021

CONSTANT
GDP90
POP90
ETHPOL
ETHNIC
RELIGION
LANGUAGE
EURFRAC

REGEAP
REGMENA
REGSA
REGSSA
REGLAC
REGWENA
LANDLOCK
LATITUDE
AREA
MINDIST
POP100CR
RESPOINT

TOPEN
CTRADE
EXPMANU
EXPPRIM
SDGR
SDBMP
(+)

(−)

(−)
(−)
(−)

(−)

(+)

Sign

BMA

2.473

−5.028
−5.377
−5.240

−1.397
−1.685
−2.326

0.667

−1.839

−0.789

t-value

Gets
Coeff

Whole

Determinants of FDEFF

Variable

Table 2.7

0.996
0.088
0.007
0.000
0.089
0.000

0.021
0.055
0.851
0.024
0.003
0.996

0.029
1.000
1.000
0.006
0.045

1.000
0.996
0.856
0.987
0.027
0.061
0.996
0.336

PIPs

(−)

(+)

(+)

(−)
(−)

(−)
(−)

(+)
(−)
(−)
(+)

Sign

BMA

1.379

0.411

−2.613

−0.975

9.145
−1.456
−0.536

Coeff

5.697

4.515

−7.077

−2.279

4.200
−6.307
−5.097

t-value

Gets

Developing Country

0.030
0.000
0.270
0.001
0.755
0.000

0.075
0.716
0.707
0.029
0.225
0.071
0.031
0.001
0.289
0.015
0.028
0.019

1.000
0.243
0.672
0.034
0.004
0.479
0.170
0.894

PIPs

(+)

(−)

(+)

(+)

(−)
(−)

(+)

(−)

(+)
(+)
(−)

Sign

BMA

0.149

0.160

−0.894

1.039

0.661
−0.411

2.443

2.168

−2.535

3.902

4.845
−3.726

t-value

Gets
Coeff

La Porta

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 43 — #34
0.034
0.102
0.000
0.000
0.985
1.000
0.000

0.134
0.086
0.058

(−)
(−)

2.27
16.62

−1.119

1.773
2.127

0.000
−0.332

0.01
0.34

40.53
0.77
0.74
0.70
26.36
−0.41
−0.29
−0.11

−8.260

3.340
3.725

−0.999
−1.144

0.996
0.996
0.981
0.059
0.027
0.768
0.996
0.159

0.996
0.082
0.003

(−)
(+)

(+)
(+)
(−)

(−)

0.70
0.11

6.226

2.828
2.133
−0.144

−0.001

0.60
0.95

14.68
0.64
0.82
0.76
28.43
−0.68
−0.51
−0.22

5.693

3.756
2.789
−4.539

−4.142

0.003
0.001
0.000

0.000
0.000
0.000
0.000
0.125
1.000
0.703

0.023
0.932
0.121

(−)
(+)

(−)

0.45
2.82

−2.044

−2.084

0.77
0.24

19.89
0.70
0.82
0.79
22.09
−0.57
−0.46
−0.27

−8.788

−5.550

Note: The dependent variable FDEFF is the index of financial efficiency development over the period 1990–99. Variable description
is in Appendix Table A2.1. There are 79 observations in the whole sample, 48 observations in the developing country sample and 49
observations in the La Porta sample. BMA analysis yields the posterior probabilities of inclusion (PIPs) and the sign certainty index
of a relationship (Sign). No sign given means the sign of estimated relationship being uncertain. Gets analysis yields coefficients
and t -values for the variables in the final model. See text for the description of PcGets output.

RSS
sigma
Rˆ2
Radjˆ2
LogLik
AIC
HQ
SC
Chow test 1
Chow test 2
Normality test
Hetero test

CIVLEG
COMLEG
POLITY2
DURABLE
FREE
KKM
PCI
EURO1900
MEDSHARE
SRIGHT
CRIGHT

SDPI
SDTP
SDTT

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 44 — #35
PIPs
1.000
1.000
0.999
0.068
0.000
0.000
0.342
0.049
0.351
0.552
0.016
0.003
0.013
0.554
0.000
0.000
0.999
0.008
0.040
0.001
0.987
0.000
0.000
0.003
0.000
0.218

CONSTANT
GDP90
POP90
ETHPOL
ETHNIC
RELIGION
LANGUAGE
EURFRAC

REGEAP
REGMENA
REGSA
REGSSA
REGLAC
REGWENA
LANDLOCK
LATITUDE
AREA
MINDIST
POP100CR
RESPOINT

TOPEN
CTRADE
EXPMANU
EXPPRIM
SDGR
SDBMP
(−)

1.353

4.065

−4.605

−0.267

(−)

(+)

−3.319

−1.224

(−)

−2.043

3.708

t-value

−1.324

−0.535

0.286

Coeff

Gets

−0.568

(+)
(+)

(+)

(−)
(+)
(+)

Sign

BMA

Whole

Determinants of FDSIZE

Variable

Table 2.8

1.000
0.076
0.005
0.034
0.590
0.491

0.105
0.002
0.002
0.016
0.269
0.039

0.000
1.000
0.018
0.152
0.042

1.000
0.809
0.961
0.282
0.093
0.808
0.392
0.043

PIPs

3.456

1.558

1.313

−0.001

(−)
(−)

0.511
0.012
−0.388

0.043

−3.310

4.984

2.483
3.817
−2.045

4.284

−4.294
−2.369

−4.993
2.073
4.185

−11.170
0.356
0.247

−1.621
−0.607

t-value

Coeff

Gets

(+)

(+)

(+)

(+)
(+)

(−)
(+)
(+)
(+)

Sign

BMA

Developing Country

0.616
0.843
0.001
0.020
0.768
0.032

0.505
0.111
0.137
0.003
0.045
0.640
0.987
0.063
0.379
0.594
0.080
0.034

1.000
0.825
0.624
0.009
0.138
0.000
0.346
0.005

PIPs

(−)

(+)
(+)

(−)
(−)

(−)
(+)

(+)

(+)

(−)
(+)
(+)

Sign

BMA

1.971

−1.209

5.253

−2.479

t-value

Gets
Coeff

La Porta

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 45 — #36
0.757
0.243
0.000
0.244
0.004
0.155
0.888

0.000
0.000
0.018

(−)

0.80
1.17
8.90

3.231
−1.921

0.697
−1.348

39.78
0.79
0.65
0.61
22.15
−0.36
−0.25
−0.08
0.74
0.34
0.01

3.009

0.022

(+)

(−)
(+)

0.895
0.029
0.012
0.000
0.249
0.467
0.008
0.000

0.000
0.000
0.183

(+)
(+)

(−)

1.11
0.29

−2.359

−0.752

0.37
0.87

9.38
0.50
0.75
0.66
43.18
−1.14
−0.94
−0.61

−4.514

−4.225

0.015
1.000
0.001

0.010
0.010
0.000
0.005
0.002
0.097
0.879

0.854
0.824
0.019

(+)

(−)

(−)
(−)

8.45

−0.008

0.01

46.94
1.10
0.42
0.39
−2.34
0.25
0.30
0.38

−2.716

Note: Dependent variable FDSIZE is the index of financial size development over the period 1990–99. The variable description
is in Appendix Table A2.1. There are 73 observations in the whole sample, 51 observations in developing country sample and
42 observations in La Porta sample. The BMA analysis yields the posterior probabilities of inclusion (PIPs) and the sign certainty
index of a relationship (Sign). No sign given means the sign of estimated relationship being uncertain. The Gets analysis yields
coefficients and t -values for the variables in the final model. See text for the description of PcGets output.

RSS
sigma
Rˆ2
Radjˆ2
LogLik
AIC
HQ
SC
Chow test 1
Chow test 2
Normality test
Hetero test

CIVLEG
COMLEG
POLITY2
DURABLE
FREE
KKM
PCI
EURO1900
MEDSHARE
SRIGHT
CRIGHT

SDPI
SDTP
SDTT

46 Determinants of Financial Development

and North African countries tend to have more efficient financial markets. Financial markets are more efficient in countries where institutional
quality (captured by KKM) is higher. The results from two subsamples
show that initial GDP and population are also important for FDEFF.
The determinants of financial size development (FDSIZE), also called
financial depth, are reported in Table 2.8. The whole sample has 73 observations, the developing country sample has 51 and the La Porta sample
has 42. The BMA and Gets analyses on the whole sample suggest that
financial depth in a country is positively related to the initial population. The West European and North American countries – including most
developed countries – witnessed a decline in financial depth. Countries
with a larger land area experience relatively less financial size development. Countries with a more open trade policy are found to have better
financial development in terms of size. Financial depth is also associated
with a stable political system (captured by DURABLE) and fewer political
constraints on the executive (captured by PCI). Most of these findings
are supported by analyses based on the developing country and the La
Porta samples. In addition, the analyses from the La Porta sample show
that financial depth might be closely related to shareholders’ rights.
We now turn to the case of bond market development. Since there
are only size measures for bond market development and bond market capitalization available in the World Bank Financial Development
and Financial Structure Database (2008) with incomplete data for many
developing countries, the above financial development measures do not
include indexes of bond market development. Appendix Table A2.8
presents the specific BMA and Gets analyses for bond market development, denoted by FDBOND, which is the sum of the private and
public bonds share over GDP in 1990s. The analyses are based on
the La Porta sample of 35 countries subject to data availability. The
results show that initial GDP level (GDP90), language fractionalization
index (LANGUAGE), East Asian and Pacific countries (REGEAP), population proportion in coastal areas25 (POP100CR), terms of trade volatility
(SDTT ) and governance index (KKM) may influence bond market development. The results support previous findings in terms of institutions,
policy and geography being important for financial development, but
further study critically depends on the availability of additional data.

2.6

Conclusions

The analysis jointly applies the BMA and Gets methods to study
what drives financial development using 39 institutional, policy and

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 46 — #37

General Determinants of Financial Development 47

geographic variables. The combination of these two methods has the
potential for incorporating the merits of each method and minimizing their limits, showing advantages in mitigating arbitrary choices
and increasing precision in model selection. To explore the structural
causes of financial development, the variables considered here are either
predetermined or evolving slowly over time.
Of 39 individual variables, this research finds that the legal origin
and institutional quality are significantly associated with financial development, as are the initial income and population. These findings are
consistent with the literature.
The finding that the legal origins influence financial development supports the emphasis on the legal determinants of financial development
of La Porta et al. (1998), who argued that the origins of the legal code
substantially influence the treatment of creditors and shareholders, and
the efficiency of contract enforcement. They document that countries
with French Civil Law are said to have comparatively inefficient contract
enforcement and higher corruption, and less well-developed financial
systems, whilst countries with British legal origin achieve higher levels
of financial development.
On the role of institutions in financial development, Beck et al. (2003)
is a significant work among others. By applying the settler mortality
hypothesis of Acemoglu et al. (2001) to financial development, Beck et al.
(2003) argue that extractive colonizers in an inhospitable environment
aimed to establish institutions that privileged small elite groups rather
than private investors, while the settler colonizers in more favourable
environments were more likely to create institutions that supported
private property rights and balanced the power of the state, therefore
favouring financial development.
The importance of income levels for financial development has been
addressed in Levine (1997, 2003, 2005). In considering the banking sector development in transition economies, Jaffee and Levonian (2001)
demonstrate that the level of GDP per capita and the saving rate have positive effects on the banking system structure as measured by bank assets,
numbers, branches and employees for 23 transition economies. On the
impact of differences in culture on the process of financial development,
Stulz and Williamson (2003) provide evidence that culture, proxied by
differences in religion and language, predicts cross-country variation in
the protection and enforcement of investor rights, especially for creditor
rights.
Taken as a whole, whilst this research shows the significant roles played
by institutions, policy and geography, it highlights the dominant role of

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 47 — #38

48 Determinants of Financial Development

institutions over policy and geography in the process of financial development. The findings on the significant effects of these structural factors,
which are relatively time-invariant, tend to suggest that efforts by the
government to better institution quality, implement more open trade
and sound macroeconomic policies and improve geographic infrastructure can stimulate financial development in the long run. An efficient
and transparent institutional and legal system and a free and just society
are especially important for the development of financial markets. Further research, as in Abiad and Mody (2005) and Chapter 5, is needed to
explore what causes governments to undertake financial reforms aimed
at financial development.

Appendix text
Here is the derivation of the posterior model probability in BMA.26 We suppose
there are many models, {M1 , . . . MK } for the data D. Every model is specified by
a vector of d unknown parameters θi = (θi1, θi2, . . . θid ), i = 1, 2 . . . K. These models may be nested or not. Bayesians treat the unknown parameters as random
variables.
Let  denote a quantity of interest such as a parameter. The posterior
distribution of  given data D is derived according to

P(|D) =

K


P(|D, Mk )P(Mk |D)

(2.1)

i=1

where P(Mk |D) are the posterior model probabilities, and P(|D, Mk ) is the
posterior distribution of  given the data D and model Mk .
The equation contains all information needed to make inferences about ,
indicating that the posterior distribution of  given data D is a weighted average
of its posterior distributions given data D and a specific model. The weights are the
posterior model probabilities, P(Mk |D), which can be obtained by Bayes’ theorem
P(Mk |D) =

P(D|Mk )P(Mk )
K


(2.2)

P(D|Mi )P(Mi )

i=1

where P(Mk ) is the prior probability of model i (i = 1, 2 . . . K), and P(D|Mi ) is the
probability of the data given Mi , also called the integrated (marginal) likelihood
for model Mi or marginal (predictive) probability of the data given Mi .
To represent no prior preference for any model, each will start on an equal
1 . Therefore the posterior model
footing, that is P(M1 ) = P(M2 ) = · · · P(MK ) = K

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 48 — #39

General Determinants of Financial Development 49
probabilities P(Mk |D) can be rewritten as
P(Mk |D) =

P(D|Mk )
K


(2.3)

P(D|Mi )

i=1

To identify the value of P(D|Mk ), it is useful to compare model Mk with a baseline model. A null model (M0 ) in which no independent variables are included is
usually used as a baseline model.27
Let Bk0 be the Bayes factor for model Mk against model M0 , that is
Bk0 =

P(D|Mk )
P(D|M0 )

(2.4)

then
2 log Bk0 = 2 log P(D|Mk ) − 2 log P(D|M0 )

(2.5)

Using an approach developed by Raftery (1995), twice the log of the Bayesian
factor, “2 log Bk0 ”, can be expressed as the approximation of the difference
between BIC0 and BICk , the values of BIC for the null model, M0 , and model,
Mk , respectively
2 log Bk0 ≈ BIC0 − BICk

(2.6)

The fact that BIC0 = 0 yields the approximation for the posterior probability
P(D|Mk ), which is


1
P(D|Mk ) ∝ exp − BICk
2

(2.7)

The posterior model probabilities P(Mk |D) can then be written as


1
exp − BICk
2
P(Mk |D) ≈


K

1
exp − BICi
2

(2.8)

i=1

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 49 — #40

50

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 50 — #41

Description

Variable

Index for overall financial development. The first principal component of private
credit (PRIVO), liquidity liability (LLY), commercial-central bank (BTOT ), overhead
cost (OVC), net interest margin (NIM), stock market capitalization (MCAP), total value
traded (TVT ) and turnover ratio (TOR) in the 1990s.
Index for financial intermediary development. The first principal component of
PRIVO, LLY, BTOT, OVC and NIM in the 1990s.
Index for stock market development. The first principal component of MCAP, TVT and
TOR in the 1990s.
Index for financial efficiency development. The first principal component of OVC,
NIM, TVT and TOR in the 1990s.
Index for financial size development (financial depth). the first principal component of
LLY and MCAP in the 1990s.
Index for bond market developpment, the sum of private bond and public bond share
over GDP in the 1990s.

EXPPRIM

CTRADE
EXPMANU

TOPEN

The proportion of years that a country is open to trade during 1965–90, by the criteria
in Sachs and Warner (1995). A country is considered to be open if it meets minimum
criteria on four aspects of trade policy: average tariffs must be lower than 40%, quotas
and licensing must cover less than 40% of total imports, the black market premium
(BMP) must be less than 20%, and export taxes should be moderate.
Natural log of the Frankel-Romer measure of predisposition to external trade
Dummy for manufactured goods exporting countries Global Development Network
Database in World Bank (GDN), 2002
Dummy for fuel and non-fuel primary good exporting countries Global Development
Network Database in World Bank (GDN), 2002

Policy variables

FDBOND

FDSIZE

FDEFF

FDSTOCK

FDBANK

FD

Dependent variables

The variables

Table A2.1

Appendix tables

Frankel and Romer (1999)

Gallup et al. (1999)

FSED, 2008

FSED, 2008

FSED, 2008

FSED, 2008

World Bank’s Financial
Structure and Economic
Development Database
(FSED), 2008
FSED, 2008

Source

51

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 51 — #42

Standard deviation of annual inflation (PI), 1960–89

Standard deviation of annual black market premium (BMP), 1960–89
Standard deviation of trading partners’ GDP per capita growth (percentage weighted
average by trade share)
Standard deviation of the first log-differences of a terms of trade index for goods and
services

SDPI

SDBMP
SDTP

FREE

DURABLE

COMLEG
CIVLEG
POLITY2

The dummy for British legal origin
Legal origin dummy for French, German and Scandinavian
Index of democracy. It is called combined polity score, the democracy score minus the
autocracy score. The democracy and autocracy scores are derived from the six
authority characterics (regulation, competitiveness and openness of executive
recruitment; operational independence of chief executive or executive constraints; and
regulation and competition of participation). Based on these criteria, each country is
assigned democracy and autocracy scores ranging from 0 to 10, accordingly, the
POLITY2 ranges from −10 to 10 with higher values representing more democratic
regimes, averaged over 1960–89.
Index of political stability based on the number of years since the last (3-point or
greater) regime transition, averaged over 1960–89.
The average of indices of civil liberties and political rights over 1972–89. The basic
components of the index of civil liberties are (1) freedom of expression and belief, (2)
association and organizational rights, (3) rule of law and human rights, (4) personal
autonomy and economic rights. Rescaled from 0 to 1, with higher values indicating
better civil liberties. The basic components of the index of political rights are (1) free
and fair elections; (2) those elected rule; (3) there are competitive parties or other
competitive political groupings; (4) the opposition has an important role and power;
(5) the entities have self-determination or an extremely high degree of autonomy.
Rescaled from 0 to 1, with higher values indicating better political rights.

Institutional variables

SDTT

Standard deviation of annul growth of real, chainweighted GDP per capita, 1960–89

SDGR

(continued)

PolityIV Database (Marshall
and Jaggers, 2009)
Freedom House (FH),
www.freedomhouse.org,
2008

GDN
GDN
PolityIV Database (Marshall
and Jaggers, 2009)

GDN

Penn World Table 6.2
(PWT62) (Heston et al.,
2006)
World Development
Indicators (WDI), 2008
GDN
GDN

52

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 52 — #43

Average of six measures of institutional development: voice and accountability,
political stability and absence of violence, government effectiveness, light regulatory
burden, rule of law and freedom from graft.
Political Constraints Index is a structurally derived measure of the feasibility of policy
change (the extent to which a change in the preferences of any one actor may lead to a
change in government policy).
The percentage of the population that was European or European descent in 1900.
The index of media owned by the government, the average of the market share of
state-owned newspapers and state-owned television stations. Market share of
state-owned newspapers is the market share owned by the state out of the aggregate
market share of the five largest daily newspapers (by circulation). Market share of
state-owned television stations is the market share owned by the state out of the
aggregate market share of the five largest television stations (by viewership)
An index aggregating the shareholder rights which we labelled as “anti-director rights”.
The index is formed by adding 1 when: (1) the country allows shareholders to mail
their proxy vote to the firm, (2) shareholders are not required to deposit their shares
prior to the General Shareholders’ Meeting, (3) cumulative voting or proportional
representation of minorities in the board of directors is allowed, (4) an oppressed
minorities mechanism is in place, (5) the minimum percentage of share capital that
entitles a shareholder to call for an Extraordinary Shareholders’ Meeting is less than or
equal to 10% (the sample median) or (6) shareholders have pre-emptive rights that can
only be waived by a shareholders’ vote. The index ranges from 0 to 6.
An index aggregating creditor rights. The index is formed by adding 1 when: (1) the
country imposes restrictions, such as creditors’ consent or minimum dividends, to file
for reorganization; (2) secured creditors are able to gain possession of their security
once the reorganization petition has been approved (no automatic stay); (3) the debtor
does not retain the administration of its property pending the resolution of the
reorganization and (4) secured creditors are ranked first in the distribution of the
proceeds that result from the disposition of the assets of a bankrupt firm. The index
ranges from 0 to 4.

KKM

CRIGHT

SRIGHT

EURO1900
MEDSHARE

PCI

Description

Continued

Variable

Table 2.1

La Porta et al. (1998)

La Porta et al. (1998)

Acemoglu, et al. (2001)
Djankov et al. (2003)

Henisz (2000), 2002 version

Kaufmann et al. (2008)

Source

53

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 53 — #44

Index of ethnic fractionalization
Index of religious fractionalization
Index of language fractionalization
Index of the “first” language variables, corresponding to the fraction of the population
speaking one of the major languages of Western Europe: English, French, German,
Portuguese or Spanish.
Low income countries
Upper-middle- and lower-middle income countries
High-income OECD and non-OECD countries

ETHNIC
RELIGION
LANGUAGE
ERUFRAC

INCLOW
INCMID
INCHIGH

Log of real GDP per capita (chain) in 1990
Log of total population in 1990
Index of ethnic polarization

Dummy for point source exporting countries.

Region dummy for East Asian and Pacific countries
Region dummy for Middle Eastern and North African countries
Region dummy for South Asian countries
Region dummy for Sub-Sahara Africann countries
Region dummy for Latin American and Caribbean countries
Region dummy for West European and North American countries
Dummy for landlocked countries
Latitude–absolute distance from equator
Area (in log) in square kilometres from World Bank (1997), except for Taiwan and
Mexico from CIA (1997), with submerged land subtracted out.
Proportion of the population in 1994 within 100 km of the coastline or navigable to
the ocean river.
The log of minimum distance from three capital-goods-supplying centres plus one.

GDP90
POP90
ETHPOL

Other variables

RESPOINT

MINDIST

POP100CR

REGEAP
REGMENA
REGSA
REGSSA
REGLAC
REGWENA
LANDLOCK
LATUTUDE
AREA

Geographic variable

GDN
GDN
GDN

PWT62
PWT62
Reynal-Querol and
Montalvo (2005)
Alesina et al. (2003)
Alesina et al. (2003)
Alesina et al. (2003)
Hall and Jones (1999)

Jon Haveman’s International
trade data. www.eiit.org
Isham et al. (2002)

Gallup et al. (1999)

GDN
GDN
GDN
GDN
GDN
GDN
GDN
GDN
Gallup et al. (1999)

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 54 — #45

Institution
FD
TOPEN
CTRADE
EXPMANU
EXPPRIM
SDGR
SDBMP
SDPI
SDTP
SDTT

Geography
FD
LANDLOCK
LATITUDE
AREA
MINDIST
POP100CR
RESPOINT

Table A2.2

1.000
0.249
0.409
−0.367
−0.259
0.041
0.076
−0.209
−0.403

TOPEN

FD

1.000
0.664
0.242
0.447
−0.463
−0.322
−0.142
−0.086
−0.112
−0.411

1.000
−0.057
0.001
0.119
−0.508
0.084

LANDLOCK

1.000
−0.163
0.536
−0.098
−0.514
0.378
−0.237

FD

Descriptive statistics

1.000
0.049
−0.150
0.353
−0.154
−0.116
−0.076
−0.024

CTRADE

1.000
−0.008
−0.429
0.252
−0.255

LATITUDE

1.000
−0.274
−0.288
−0.084
−0.073
−0.196
−0.186

EXPMANU

1.000
−0.053
−0.455
−0.122

AREA

1.000
0.403
0.018
0.237
0.265
0.395

EXPPRIM

1.000
−0.293
0.242

MINDIST

1.000
0.044
0.000
0.147
0.437

SDGR

1.000
−0.053

POP100CR

1.000
0.096
0.116
0.184

SDBMP

1.000

RESPOINT

1.000
0.092
0.005

SDPI

1.000
0.128

SDTP

1.000

SDTT

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 55 — #46

Others
FD
GDP90
OPO90
ETHPOL
ETHNIC
RELIGION
LANGUAGE
EURFRAC

Policy
FD
CIVLEG
COMLEG
POLITY2
DURABLE
FREE
KKM
PCI

1.000
−0.210
−0.282
−0.499
0.036
−0.473
0.283

GDP90

FD

1.000
0.627
0.070
−0.169
−0.358
0.151
−0.161
−0.082

1.000
−0.968
−0.281
−0.143
0.015
0.100
0.048

CIVLEG

1.000
−0.071
0.037
0.332
0.455
−0.374
0.675
0.314

FD

1.000
−0.073
0.063
0.052
0.149
−0.167

POP90

1.000
0.324
0.147
−0.074
−0.072
−0.006

COMLEG

1.000
0.660
0.148
0.266
0.251

ETHPOL

1.000
0.532
−0.710
0.547
0.701

POLITY2

1.000
0.256
0.631
0.012

ETHNIC

1.000
−0.563
0.561
0.453

DURABLE

1.000
0.307
0.184

RELIGION

1.000
−0.714
−0.885

FREE

1.000
−0.397

LANGUAGE

1.000
0.665

KKM

1.000

EURFRAC

1.000

PCI

56 Determinants of Financial Development
Table A2.3 The list of countries in the full sample

East Asia & Pacific
AUS
Australia
CHN China
FJI
Fiji
HKG
Hong Kong, China
IDN
Indonesia
JPN
Japan
KOR
Korea, Rep.
MAC Macao
MNG Mongolia
MYS
Malaysia
NZL
New Zealand
PHL
Philippines
PNG
Papua New Guinea
SGP
Singapore
THA
Thailand
TWN Taiwan, China
VNM Vietnam
Sub-Saharan Africa
BDI
Burundi
BEN
Benin
BFA
Burkina Faso
BWA
Botswana
CIV
Cote d’Ivoire
CMR Cameroon
ETH
Ethiopia
GHA
Ghana
KEN
Kenya
MDG Madagascar
MLI
Mali
MOZ Mozambique
MRT
Mauritania
MUS
Mauritius
MWI Malawi
NAM Namibia
NGA
Nigeria
RWA
Rwanda
SDN
Sudan
SEN
Senegal
SLE
Sierra Leone
SWZ
Swaziland
TGO
Togo
UGA
Uganda
ZAF
South Africa
ZMB
Zambia
ZWE
Zimbabwe

Middle East &
North Africa
BHR
Bahrain
DZA
Algeria
EGY
Egypt, Arab Rep.
GRC
Greece
IRN
Iran, Islamic Rep.
ISR
Israel
JOR
Jordan
KWT Kuwait
LBN
Lebanon
MAR
Morocco
MLT
Malta
OMN Oman
PRT
Portugal
QAT
Qatar
SAU
Saudi Arabia
TUN
Tunisia
Latin America &
Caribbean
ARG
Argentina
BOL
Bolivia
BRA
Brazil
BRB
Barbados
CHL
Chile
COL
Colombia
CRI
Costa Rica
DOM Dominican Rep.
ECU
Ecuador
GTM Guatemala
GUY
Guyana
HND Honduras
HTI
Haiti
JAM
Jamaica
MEX
Mexico
NIC
Nicaragua
PAN
Panama
PER
Peru
PRY
Paraguay
SLV
El Salvador
TTO
Trinidad and Tobago
URY
Uruguay
VEN
Venezuela

South Asia
BGD Bangladesh
IND
India
LKA
Sri Lanka
NPL
Nepal
PAK
Pakistan

Western Europe &
North America
AUT Austria
BEL
Belgium
CAN Canada
CHE Switzerland
CYP
Cyprus
DEU Germany
DNK Denmark
ESP
Spain
FIN
Finland
FRA
France
GBR United Kingdom
IRL
Ireland
ISL
Iceland
ITA
Italy
LUX Luxembourg
NLD Netherlands
NOR Norway
SWE Sweden
USA
United States

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 56 — #47

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 57 — #48
3.922
3.063
1.986
2.160
1.612

FD
FDBANK
FDSTOCK
FDEFF
FDSIZE

0.490
0.613
0.662
0.540
0.806

Proportion

0.707

0.411
0.479

LLY
0.454
0.479

PRIVO
0.278
0.357

BTOT

NIM
−0.368
−0.471
0.561

OVC
−0.357
−0.437
0.546

0.676
−0.467

0.535
0.707

0.357

TVT

0.364

MCAP

0.506
−0.411

0.157

TOR

Notes: The financial development measures are described in the text. The first principal component is the linear combination of the measures selected.
The eigenvalues are the variances of the (first) principal components. The eigenvectors give the coefficients of the standardised variables.
LLY = the ratio of liquid liabilities of financial system (currency plus demand and interest-bearing liabilities of banks and non-banks) to GDP;
PRIVO = the ratio of credits issued to private sector by banks and other financial intermediaries to GDP;
OVC = the ratio of overhead costs to total assets of the banks;
NIM = the bank interest income minus interest expenses over total assets;
MCAP = the ratio of the value of domestic shares traded on domestic exchange to GDP;
TVT = the ratio of the value of domestic shares traded on domestic exchange to GDP;
TOR = the ratio of the value of domestic shares traded on domestic exchange to total value of listed domestic shares

Eigenvalue

The eigenvalue, proportion and eigenvector of each first principal component

Measure

Table A2.4

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 58 — #49
COMLEG
COMLEG
COMLEG
INCMID
INCMID
INCMID
INCMID
INCMID
INCMID

REGEAP
REGEAP
REGEAP
REGEAP
REGEAP

CIVLEG
CIVLEG
CIVLEG
CIVLEG
CIVLEG
CIVLEG

AREA
MINDIST
POP100CR
POP90
RESPOINT

POLITY2
DURABLE
FREE
KKM
PCI
EURO1900

SRIGHT
CIVLEG
CRIGHT
CIVLEG
MEDSHARE CIVLEG

GDP90
SDGR
SGBMP
SDPI
SDTP
SDTT

INCLOW
INCLOW
INCLOW
INCLOW
INCLOW
INCLOW

COMLEG
COMLEG
COMLEG
COMLEG
COMLEG
COMLEG

REGEAP
REGEAP

CTRADE
TOPEN

REGSA
REGSA
REGSA
REGSA

REGMENA
REGMENA
REGMENA
REGMENA
REGMENA

REGLAC
REGLAC
REGLAC
REGLAC

REGSSA
REGSSA
REGSSA
REGSSA
REGSSA

REGLAC
REGLAC
REGLAC
REGLAC
REGLAC

REGSSA REGLAC
REGSSA REGLAC

REGSSA
REGSSA
REGSSA
REGSSA

INCHIGH
INCHIGH
INCHIGH
INCHIGH
INCHIGH
INCHIGH

LATITUDE
LATITUDE
LATITUDE
REGEAP
REGEAP
CIVLEG
CIVLEG
REGEAP
REGEAP

REGMENA
REGMENA
COMLEG
COMLEG
REGMENA
REGMENA

RELIGION
RELIGION
RELIGION
RELIGION

REGWENA
REGWENA
REGWENA
REGWENA
REGWENA

REGSA
REGSA
LATITUDE
LATITUDE
REGSA
REGSA

REGLAC
REGLAC

REGLAC
REGLAC

REGSSA
REGSSA

REGLAC

LATITUDE

LATITUDE
LATITUDE
LATITUDE

INCMID
INCMID

REGSSA
REGSSA

REGSSA

LANDLOCK
LANDLOCK
LANDLOCK
LANDLOCK
LANDLOCK

REGWENA INCLOW
REGWENA INCLOW

REGWENA
REGWENA
REGWENA
REGWENA

LATITUDE
LATITUDE
LATITUDE
LATITUDE
LATITUDE
LATITUDE REGEAP REGMENA REGSA

REGSA
REGSA
REGSA
REGSA
REGSA

REGMENA REGSA
REGMENA REGSA

REGEAP
REGEAP
REGEAP
REGEAP

REGMENA
REGMENA
REGMENA
REGMENA

Variables used to impute the missing data

Imputation

ETHPOL
ETHNIC
LANGUAGE
EURFRAC

Variables

Table A2.5

EXPMANU EXPPRIM LANDLOCK
EXPMANU EXPPRIM LANDLOCK

REGWENA LATITUDE
REGWENA LATITUDE

REGWENA LATITUDE
REGWENA LATITUDE

REGWENA

INCHIGH
INCHIGH

General Determinants of Financial Development 59
Table A2.6 Setting for PcGets
expert significance:
0.075
0.075
expert presearch:
0.75
1
expert block search:
1
1
expert choose specific:
“HQ”
expert split sample:
0.075
0.75
expert outlier dection:
2.56
expert tests:
1
1
expert test options:
0.5
0.9
set detect outliers:
"1"
set0lagorder:
“0”
set0topdown:
“1”
set0bottomup:
“1”
setsplitsample:
“1”
setstrategy:
“expert”, 1
setreporting:
“0”
estimate:
“Gets”,
1

0.75
0.5
1

0.075 0.01
0.005
0.075 0.075 0.05
1
1
1

0.2

0.4

0.4

0
12

1
1

0
4

1

n

1

0
1

0.05
1

1
1

1
4

Note: A change has been made to the “liberal strategy” default setting by increasing the F
pre-search testing (top-down) at step 1 from 0.75 to 1. “n” denotes the sample size.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 59 — #50

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 60 — #51
−1.338
2.660
2.549
0.511
−0.557
−0.188
1.138
−0.940

−8.112
1.326
0.566
0.497
−0.774
−0.214
1.132
−0.702

0.938
0.798
0.311
−0.275
0.029
−0.388
−0.266
−0.021
−0.421
−0.036
−0.010
−0.269

0.608
0.013
−0.097
−0.378

CONSTANT
GDP90
POP90
ETHPOL
ETHNIC
RELIGION
LANGUAGE
EURFRAC

REGEAP
REGMENA
REGSA
REGSSA
REGLAC
REGWENA
LANDLOCK
LATITUDE
AREA
MINDIST
POP100CR
RESPOINT

TOPEN
CTRADE
EXPMANU
EXPPRIM

0.712
0.640
−0.198
−0.792

0.659
0.600
0.227
−0.178
0.022
−0.262
−0.325
−0.805
−2.383
−0.258
−0.824
−0.500

t-value

Coeff

GUM

−2.948
−4.188

−3.694

−1.287

−0.041
−0.417

−4.074
4.403
5.856

−10.563
1.391
0.705

t-value

Final Model
Coeff

Full

Determinants of FD by Gets

Variable

Table A2.7

1.220
1.495
−0.349
0.478

0.772
−1.241
−0.725
−0.183
0.726
−0.916

1.098
−0.046
−0.246
−0.272
0.010
−0.774
2.046
0.036
−1.975
0.362

0.334
1.011
0.457
0.309
0.079

−1.151
2.823
1.816
−0.671
0.137
−0.156
1.847
0.797

t-value

1.486
5.417
2.864
1.664
0.275

−16.154
2.250
0.768
−1.290
0.282
−0.386
2.920
2.010

Coeff

GUM

1.990

3.353

2.585

5.192

7.548

1.188

7.548

2.941

1.487

3.353

−5.932
7.192
2.855

t-value

−15.723
2.049
0.248

Coeff

Final Model

Developing Country

1.237
0.487
0.958
1.164

−0.841
−0.104
−0.853
−0.193
1.283
−1.455
3.979
0.644
−0.564
−0.508
0.471
−1.723

−4.249
−0.477
−4.670
−1.473
3.615
−9.044
2.562
0.071
−0.260
−0.232
0.015
−3.148
3.184
0.046
0.848
1.844

0.000
−0.316
1.733
0.672
0.032
0.400
0.492
−2.036

t-value

0.000
−0.878
1.044
2.592
0.111
2.183
1.469
−4.480

Coeff

GUM

3.512
2.421
4.264
2.597

2.592
−4.633

0.015
−2.960
1.854
0.034
1.246
1.416

−4.095

3.317
−5.614

2.435
−7.789
−0.416

−4.524

−3.847

−6.216

8.610
4.652

t-value

−3.000

−4.224

−3.988

1.314
3.131

Coeff

Final Model

La Porta

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 61 — #52
0.66
8.46

5.064
−5.184

1.846
−5.363

0.68
0.01

51.11
0.97
0.80
0.77
7.20
0.09
0.22
0.42

−3.396

−3.270

−0.927

−0.024

0.00
7.45
1.77

−6.899
−5.840
0.151
−0.025
0.168
0.099
−6.769
3.579

−0.126
−0.001
0.001
−0.435
−0.015

5.37
0.82
0.96
0.77
46.26
−0.47
0.08
0.99
0.00
0.39
0.41

−1.458
−1.083
1.528
−0.781
0.596
0.119
−1.870
0.633

−0.636
−1.315
1.053
−0.519
−0.720

1.35
1.76

−5.391

−4.562
−3.445
0.164

0.001

0.28
0.41

17.06
0.73
0.86
0.82
20.84
−0.40
−0.22
0.08

−4.026

−5.239
−3.879
4.382

2.193

0.00
0.00
2.59

0.152
−0.320
−1.098

0.627
−0.282
−0.413

1.03
1.02
0.99
0.77
73.12
−1.71
−1.11
−0.06
0.00
0.00
0.27

0.332
0.390
0.326
0.248
−0.424
1.245
−1.472

−1.154
1.485
−1.513
−0.825
0.827

10.563
12.554
0.083
0.010
−0.542
3.425
−13.777

−0.794
0.004
−0.031
−2.270
0.033

1.35

0.51

3.78
0.50
0.98
0.94
47.21
−1.11
−0.73
−0.05

−3.169

−3.033
9.156
−7.002

−0.453
4.849
−13.547

−0.374

4.452

−3.875
4.876
−5.076
−2.263
3.222

1.687

−0.571
0.004
−0.035
−1.390
0.036

Note: The dependent variable FD is the index of overall financial development over the period 1990–99. The variable description is in Appendix Table
A2.1. The Gets analysis yields coefficients and t -values for the variables in the final model. There are 64 observations in the whole sample, 44 observations
in the developing country sample and 40 observations in the La Porta sample.

0.00
1.34
0.17

26.44
0.97
0.90
0.77
28.28
0.24
0.72
1.46
0.00
0.28
0.92

−1.307
−1.037
0.927
0.903
0.004
1.974
−2.095

−2.353
−1.885
0.051
0.014
0.001
1.191
−4.827

CIVLEG
COMLEG
POLITY2
DURABLE
FREE
KKM
PCI
EURO1900
MEDSHARE
SRIGHT
CRIGHT

RSS
sigma
Rˆ2
Radjˆ2
LogLik
AIC
HQ
SC
Chow test 1
Chow test 2
Normality test
Hetero test

−0.203
−0.631
1.273
0.232
−0.805

−0.026
0.000
0.001
0.120
−0.008

SDGR
SDBMP
SDPI
SDTP
SDTT

62 Determinants of Financial Development
Table A2.8 Determinants of FDBOND
La Porta Sample
BMA

Gets

Variable

PIPs

Sign

CONSTANT
GDP90
POP90
ETHPOL
ETHNIC
RELIGION
LANGUAGE
EURFRAC

1.000
0.105
0.915
0.916
0.951
0.251
0.979
0.920

(−)

REGEAP
REGMENA
REGSA
REGSSA
REGLAC
REGWENA
LANDLOCK
LATITUDE
AREA
MINDIST
POP100CR
RESPOINT

0.859
0.829
0.908
0.317
0.874
0.828
0.842
0.084
0.731
0.233
0.326
0.606

(+)
(+)
(−)
(−)
(+)
(+)
(−)

TOPEN
CTRADE
EXPMANU
EXPPRIM
SDGR
SDBMP
SDPI
SDTP
SDTT

0.142
0.848
0.944
0.938
0.956
0.173
0.847
0.153
0.311

CIVLEG
COMLEG
POLITY2
DURABLE
FREE
KKM
PCI
EURO1900
MEDSHARE
SRIGHT
CRIGHT

0.465
0.449
0.944
0.877
0.124
0.825
0.850
0.267
0.197
0.838

Coeff

t-value

0.0164

3.812

0.0751

2.717

(−)
(−)
(−)
(+)
(+)

(−)
(+)
(+)

(−)
(−)
(−)
(+)
(+)

(+)
(−)
(+)
(+)
(−)
(+)

(+)
(continued)

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 62 — #53

General Determinants of Financial Development 63
Table A2.8 Continued
La Porta Sample
BMA
Variable
RSS
sigma
Rˆ2
Radjˆ2
LogLik
AIC
HQ
SC
Chow test 1
Chow test 2
Normality test
Hetero test

PIPs

Gets
Sign

Coeff

t-value
6.06
0.43
−0.08
−0.11
30.68
−1.64
−1.61
−1.55

Note: The dependent variable FDBOND is the index of bond market development over the period 1990–99. The variable description is in Appendix Table
A2.1. This study is based on La Porta sample with 35 countries. The BMA analysis yields posterior probabilities of inclusion (PIPs), the total posterior model
probabilities (PMPs) for all models including a given variable, and the sign certainty index of a relationship (Sign). No sign given means the sign of estimated
relationship being uncertain. The Gets analysis yields coefficients and t -values
for the variables in the final model.

HUANG: “CHAP02” — 2010/9/29 — 20:05 — PAGE 63 — #54

3
Private Investment and
Financial Development

3.1

Introduction

In recent decades there has been a large body of literature studying
the substantial roles that investment and financial development play in
long-run economic growth (Levine and Renelt, 1992; King and Levine,
1993 among others). This chapter aims to provide an exhaustive analysis
of the existence of and directions of causality between these two important aspects of economic activities, namely aggregate private investment
and financial development. By exploiting the time series variation in
both private investment and financial development, and allowing for
global interdependence and heterogeneity across countries, this chapter
suggests positive causal effects going in both directions.
As is well known, in the absence of asymmetric information, financial
markets can function efficiently in the sense that, for any investment
project, the financial contract provides the borrowers and investors
with expected payments determined by the prevailing economy-wide
interest rate. However, in reality, entrepreneurs are always much better
informed than investors as to the outcome of investment projects and
their actions, calling for costly state verification conducted by financial
intermediaries (Townsend, 1979),28 and the corresponding contracting
problem between financial intermediaries and entrepreneurs (Diamond,
1984; Gale and Hellwig, 1985; Williamson, 1986, 1987 and Bernanke
and Gertler, 1989). Does entrepreneurs’ investment behaviour exert
any effect on the expansion of financial systems or the reduction of
agency costs? Does the increase in private investment as a whole contribute to financial development? On the other hand, another natural
question could be whether more efficient financial markets encourage

64

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 64 — #1

Private Investment and Financial Development 65

entrepreneurs’ investment behaviour, or whether financial development
brings about a surge of private investment.
Economic theory in general predicts that private investment and financial intermediary development contribute in a significant way to each
other. On the one hand, an increase in private investment constitutes
rising demand for external finance, enlarging the extent of financial
intermediation by directly encouraging financial intermediaries to persuade savers to switch their holdings of unproductive tangible assets
to bank deposits. Levine and Renelt (1992) suggest that more investment raises the rate of economic growth, which could stimulate financial
development (Greenwood and Smith, 1997). On the other hand, the
endogenous finance-growth models (for example Diamond, 1984; Diamond and Dybvig, 1983; Greenwood and Jovanovic, 1990; Bencivenga
and Smith, 1991 and Greenwood and Smith, 1997) suggest that financial markets have an important role in channelling investment capital
to its highest valued use. Financial intermediaries tend to induce a portfolio allocation in favour of productive investment by offering liquidity
to savers, easing liquidity risks, reducing resource mobilization costs and
exerting corporate control. It seems natural to wonder if what is possible
in theory is consistent with what has happened in reality.
The causes of financial development have become an increasingly
significant research area in recent years.29 Following the renowned
Solow-Swan growth model, much research has been undertaken to examine the long-run determinants of economic growth. Levine and Renelt
(1992) emphasize the critical role of investment in growth, leading to
investment being included in most growth regressions. However, there
has been little work on the role of investment in the determination of
financial development.
Much work has been done to investigate the determinants of investment since the 1990s.30 Following the influential work of King and
Levine (1993), who find a positive effect of financial development on
various aspects of economic activity, several empirical studies provide
evidence in support of a positive impact of financial development on
capital formation in the private sector.31 However, existing research
in general assumes error independence across countries, which is a
highly restrictive assumption to make, particularly in the context of
globalization.
This background has motivated research into the interactions between
aggregate private investment and financial development in this chapter.
The econometric analysis is based on a dataset for 43 developing countries over the period 1970–98. Since commercial banks dominate the

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 65 — #2

66 Determinants of Financial Development

financial sector and stock markets play only very minor roles in most
developing countries, this research focuses on the level of financial
intermediary development, for which a new index is constructed by
using principal component analysis based on three banking development indicators32 widely used in the literature. This research has become
more important as since the 1970s many developing countries have
sought to stimulate economic growth by choosing to encourage private
investment, while abandoning import-substitution policies led by the
public sector.
It is worth noting that this analysis focuses on the period when, after
the collapse of the Bretton Woods system, the world economy has experienced “a new and deeper version of globalization” following “a gradual
liberalization of trade and capital flows” (Crafts, 2000). The increase in
global trade and financial integration33 has been found to induce closer
interdependence in the global economy through its implications for the
properties of business cycle fluctuations. Imbs (2003), using data for a
group of developed and developing countries over 1983–98, finds that
the intensity of financial linkages and the volume of intra-industry trade
have a positive impact on cross-country business cycle co-movement.
Frankel and Rose (1998) show that trading partners have a higher degree
of business cycle co-movement. Kim et al. (2003) observe a high degree of
business cycle co-movement for a set of Asian emerging market countries
over 1960–96.
The phenomenon of business cycle co-movement has often been
explained by using a common factor analysis in which macroeconomic
variables such as aggregate output, consumption and investment are
decomposed into common observed global shocks (like sharp fluctuations of oil prices), common unobserved global shocks (like technological shocks), specific regional shocks and country shocks (Gregory et al.,
1997; Kose et al., 2003 and Bai and Ng, 2004). It is these shocks that lead
to a closer real and financial interdependence across countries.
The 1990s witnessed growing research on the stochastic properties
of panel datasets where the time dimension and cross section dimension are relatively large, and, especially, the issue of cross section error
dependence has received a great deal of attention in recent years. The
application of unit root and cointegration tests to panels is motivated
by the possible increase of statistical power through pooling information across units. However, the power of tests is increased only when the
cross section units are independent, which is an assumption that may
be hard to justify given the rising degree of financial market integration

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 66 — #3

Private Investment and Financial Development 67

and business cycle synchronization. This research attempts to explore
this issue by fully taking into account the effects of global shocks causing
cross section dependence across countries.
The analysis in this chapter includes two steps. The first step is an
analysis on data for five-year averages, which is commonly used in the
literature. It applies the system GMM estimation method due to Arellano
and Bover (1995) and Blundell and Bond (1998) allowing for possible
correlations between regressors, and both individual effects and global
shocks. It then moves on to the second step, an analysis using methods on pooled annual data assuming a common factor structure in the
error term from Bai and Ng (2004). Before proceeding to estimation, the
time series properties of the panel dataset are carefully examined. The
so-called “second-generation tests” are applied, which allow for cross
section dependence, including a panel unit root test of Bai and Ng (2004)
and a panel cointegration test of Pedroni (2004) on defactored data. The
models are then estimated by the Pesaran (2006) Common Correlated
Effect approach.
The analysis on averaged data produces significant findings of positive
causal effects going in both directions, and indicates a high degree of persistence exists in the averaged data of financial development and private
investment. The annual data study suggests that the series of both private
investment and financial development are integrated, and two-way positive long-run causal effects exist in the cointegrated system. The findings
of this chapter support the view that a private investment boom typically
follows further financial development, while the demand for external
finance is reflected in the subsequent level of financial development.
This has significant policy implications for the development of financial markets and the conduct of macroeconomic policies in developing
countries in a global economy.
The remainder of the chapter proceeds in Section 3.2 to describe the
data. Section 3.3 analyses this link using system GMM estimation on data
for five-year averages. Section 3.4 employs the common factor approach
to examine this link with annual data, including panel unit root testing
panel cointegration testing and estimation. Section 3.5 concludes.

3.2

The data

This section outlines the measures and data for private investment and
financial development. Appendix Table A3.1 summarizes the variable
description and sources.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 67 — #4

68 Determinants of Financial Development

The measure of private investment, denoted by PI, is the ratio of nominal private investment to nominal GDP. The data are taken from the
World Bank Global Development Network Database (2002).34
The measure of financial development, denoted by FD. Since commercial banks dominate the financial sector and stock markets play very
minor roles in most developing countries, this research focuses on the
level of financial intermediary development, for which a new index is
constructed by using principal component analysis35 based on three
banking development indicators widely used in the literature.
The principal component analysis is based on the following three
popular banking development indicators:36
The first measure, Liquid Liabilities (LLY), is one of the major indicators used to measure the size, relative to the economy, of financial
intermediaries including three types of financial institutions: the central
bank, deposit money banks and other financial institutions. It is calculated by the ratio of liquid liabilities of banks and non-bank financial
intermediaries (currency plus demand and interest-bearing liabilities)
over GDP.
The second indicator, Private Credit (PRIVO), is defined as credit issued
to the private sector by banks and other financial intermediaries divided
by GDP. This excludes the credit issued to government, government
agencies and public enterprises, as well as the credit issued by the monetary authority and development banks. It is a general indicator of
financial intermediary activities provided to the private sector.
The third, Commercial-Central Bank (BTOT ), is the ratio of commercial bank assets to the sum of commercial bank and central bank assets.
It reflects the advantage of financial intermediaries in dealing with lending, monitoring and mobilizing saving and facilitating risk management
relative to the central bank.
Data on these financial development indicators are obtained from the
World Bank’s Financial Structure and Financial Development Database
(2008). FD is the first principal component of these three indicators
above and accounts for 74% of their variation. The weights resulting from
principal component analysis over the period 1990–98 are 0.60 for Liquid Liabilities, 0.63 for Private Credit and 0.49 for Commercial-Central
Bank.37 Since these indicators are used to measure the size of financial
intermediary development,38 the composite index, FD, mainly captures
the depth of bank-based intermediation.
Appendix Table A3.2 presents descriptive statistics for private investment, the measure of financial development, real GDP and trade
openness. The panel dataset contains 43 developing countries over the

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 68 — #5

Private Investment and Financial Development 69

period 1970–98. The countries in the full sample are listed in Appendix
Table A3.3. The transition economies are omitted. We also exclude
countries with fewer than 20 observations over 1970–98.

3.3

Analysis on data for five-year averages

To examine the relationship between private investment and financial development, this chapter conducts panel data estimation for
43 developing countries over 1970–98, based on averaged data over
non-overlapping, five-year periods in this section, and annual data in
Section 3.4. Panel data estimation tends to produce more convincing
findings than cross section analysis and classical time series analysis since
it exploits both the cross section and time dimensions of the data.39 It
allows us to control for unobserved country-specific effects and omitted
variables bias, and look at both long-run and short-run effects.
This section mainly focuses on the system GMM method proposed by
Arellano and Bover (1995) and Blundell and Bond (1998), using averaged data (with a maximum of six observations per country). As widely
used in the growth literature (Islam, 1995; Caselli et al., 1996; Levine
et al., 2000), averaging data over fixed intervals has the potential for
eliminating business cycle fluctuations and makes it easier to capture
the relationships of interest. Section 3.3.1 briefly describes the system
GMM approach, and section 3.3.2 presents the empirical results.
3.3.1

Methodology: System GMM

The following AR(1) model has been found appropriate for this
application:40
FDit = α11 FDi,t−1 + PI i,t−1 β11 + ηi1 + φ1t + vit1

(3.1)

PI it = α12 PI i,t−1 + FDi,t−1 β12 + ηi2 + φ2t + vit2

(3.2)

i = 1, 2, . . . , 43 and t = 2, . . . , 6
For the sake of convenience, denote by y the dependent variable (either
FD or PI) and by x the explanatory variables other than the lagged
dependent variable:
,

yit = αyi,t−1 + xi,t−1 β + ηi + φt + vit

(3.3)

i = 1, 2, . . . , 43 and t = 2, . . . , 6

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 69 — #6

70 Determinants of Financial Development

where ηi is an unobserved country-specific time-invariant effect not captured by xi, t−1 , and can be regarded as capturing the combined effects
of all time-invariant omitted variables.
φt captures the global shocks. Recently a large body of literature has
indicated that the existence of common factors, either global, cyclical or
seasonal effects, has the potential for causing co-movements of variables
in the world economy. Since common factors are likely to be partially
cancelled out when the data are averaged, for simplicity this section
considers only common time effects or a single global shock having an
identical effect on each cross section unit. Section 3.4 explores the effects
of common factors in more depth.
vit is the transitory disturbance term, assumed to satisfy sequential
moment conditions of the form
E(vit | yit−1 , xt−1
, ηi , φt ) = 0
i

(3.4)

= (xi1 , xi2 . . ., xi,t−1 ), .
where yit−1 = (yi1 , yi2 . . ., yi,t−1 ), , xt−1
i
This assumption implies that (1) the transient errors are serially
uncorrelated; (2) xs are predetermined variables with respect to the timevarying errors in the sense that xi, t−1 may be correlated with vi, t−1 and
earlier shocks, but is uncorrelated with vi t and subsequent shocks; (3)
the individual effects are uncorrelated with the idiosyncratic shocks, but
correlations between individual effects and lagged y and lagged x are not
ruled out and (4) the global shocks are uncorrelated with the idiosyncratic shocks, while correlations between global shocks and lagged y and
lagged x are possible.
The assumption of the explanatory variables xs being predetermined
rules out a potential endogeneity bias, but allows for feedbacks from
the past realizations of y to current xs. This assumption is believed to
be appropriate given that financial development is potentially both a
consequence and an origin of private investment, and vice versa.41
For the stability of the estimated model, the autoregressive coefficient
is assumed to lie inside the unit circle, | α| < 1.
The coefficient β reflects the existence and direction of Granger causality going from lagged x to y. According to work by Chamberlain (1984)
and Holtz-Eakin et al. (1988) on Granger non-causality tests in the
general setting of dynamic panel data estimation, the non-causality
hypothesis can be tested by checking whether the coefficients of the
lagged values of the independent variables are zero or the coefficients
on the lagged difference of independent variables in the transformed
equations are zero, that is β = 0. Given that the model is stable, a point

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 70 — #7

Private Investment and Financial Development 71

estimate for the long-run effect can be calculated as follows:
βLR =

β
(1 − α)

The standard error for the long-run effect can be approximated by
using the delta method (for example Papke and Wooldridge, 2005).
This analysis employs the system GMM method, which is proposed
by Arellano and Bover (1995) and Blundell and Bond (1998) to improve
upon the Arellano and Bond (1991) first-differenced GMM method,
which may be plagued with weak instrument problems. There have
been a number of methods proposed to estimate dynamic panel data
models with a short time dimension, in which first-differencing is
used to eliminate the individual effects. Below is Equation (3.3) in first
differences:
,

yit = αyi,t−1 + xi,t−1 β + φt + vit

(3.5)

i = 1, 2, . . . , 43 and t = 3, . . . , 6
where yit = yit − yi,t−1 , xi,t−1 = xi,t−1 − xi,t−2 , φt = φt − φt−1 and
vit = vit − vi,t−1 .
The sequential moment conditions above imply that all lagged values
of yit and xit dated from t − 2 and earlier are suitable instruments for
the differenced values of the original regressors, yi,t−1 and xi,t−1 .
While the first-differenced 2SLS estimator taken from Anderson and
Hsiao (1981, 1982) uses yit−2 and xit−2 , the first-differenced GMM
estimator uses all lagged values of yit and xit dated from t − 2 and
earlier. The moment conditions for errors in differences on which the
first-differenced GMM estimator is based can be written as:




yit−2
,
E
−
αy
−
x
β
−
φ
)
=0
(3.6)
(y
t
it
i,t−1
i,t−1
xt−2
i
t = 3, . . . , 6
= (xi1 , xi2 . . ., xi,t−2 ), .
where yit−2 = (yi1 , yi2 . . ., yi,t−2 ), and xt−2
i
Blundell and Bond (1998) argue that in the standard AR(1) model
when the time series becomes highly persistent in the sense that “the
value of the autoregressive parameter approaches unity or the variance
of the individual effects increases relative to the variance of the disturbances”, the lagged values of the series may be weak instruments for
first differences. The first-differenced GMM estimator employing these

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 71 — #8

72 Determinants of Financial Development

weak instruments has been found to have poor finite sample properties
in terms of bias and imprecision.
To tackle the weak instruments problem, Arellano and Bover (1995)
and Blundell and Bond (1998) develop a “system GMM” estimator42 by
considering a mean stationarity assumption on initial conditions in the
sense that the mean of the distribution of the initial observations coincides with the mean of the steady-state distribution of the process. For
the multivariate autoregressive model, Blundell and Bond (2000) show
that a sufficient condition for the additional moment conditions to be
valid is the joint mean stationarity of the series.
For this context the additional mean stationarity condition of (yit , xit )
enables the lagged first differences of the series (yit , xit ) dated t-1 as instruments for the untransformed equations in levels. In addition to the
moments for errors in differences described before, the system GMM estimator, denoted by SYS-GMM, is also based on the additional moments
for errors in levels as follows:




yi,t−1
,
E
(3.7)
(yit − αyi,t−1 − xi,t−1 β − φt ) = 0
xi,t−1
t = 3, . . . , 6
As suggested by Blundell and Bond (1998), combining the firstdifferenced equations using suitably lagged levels as instruments, with
levels equations using suitably lagged first differences as instruments, the
SYS-GMM estimator is expected to have much smaller finite sample bias
and greater precision in the presence of persistent data.
Apart from the orthogonality conditions (3.6) and (3.7) stated above,
the SYS-GMM estimator also makes use of the following moments for
the period-specific constants due to the existence of global shocks:
,

E(yit − αyi,t−1 − xi,t−1 β − φt ) = 0

(3.8)

t = 3, . . . , 6
To avoid the possible over-fitting bias associated with using the full
Arellano and Bond (1991) instrument set, this analysis uses restricted
instrument sets suggested by Bowsher (2002), who proposes selectively
reducing the number of moment conditions for each first-differenced
equation. More specifically, we use only lagged values of yit and xit from
t − 2 to t − 4 as instruments. Accordingly, for SYS-GMM estimators the
number of orthogonality conditions reduces to 31 in total, so that there
are 24 over-identifying restrictions. Another way to avoid the possible

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 72 — #9

Private Investment and Financial Development 73

over-fitting bias is the introduction of the two additional versions of
SYS-GMM discussed below.

3.3.2

Empirical results

This section presents the SYS-GMM estimates for Equations (3.1) and
(3.2). Two additional versions of SYS-GMM are also considered in order
to circumvent over-fitting and the possibility that the mean stationarity
assumptions may be incorrect. While SYS-GMM-1 uses only yi,t−1 as
instruments in levels, SYS-GMM-2 uses only xi,t−1 in the same way.
The OLS and within group estimates are also reported. Conventional
wisdom has revealed that, although both of them are inconsistent for
short panels, the OLS and within group (WG) estimates of the first-order
autoregressive parameter act as two extremes of the interval in which a
consistent estimate of this parameter is expected to lie.43
Three specification tests are conducted to address the consistency
of SYS-GMM estimator, which mainly depends on the validity of the
instruments. The first is a Serial Correlation test, which tests the null
hypothesis of no first-order serial correlation and no second-order serial
correlation in the residuals in the first-differenced equation. The second
is a Sargan test of over-identifying restrictions, which is used to examine
the overall validity of the instruments by comparing the moment conditions with their sample analogue. A finite sample correction is made to
the two-step covariance matrix using the method of Windmeijer (2005).
The third is a difference Sargan test, denoted by Diff-Sargan, proposed
by Blundell and Bond (1998), which examines the null hypothesis of
mean stationarity for the SYS-GMM estimator. This statistic, called an
incremental Sargan test statistic, is the difference between the Sargan
statistics for first-differenced GMM and SYS-GMM. It would be asymptotically distributed as a χ 2 with k degrees of freedom, where k is the
number of additional moment conditions.
Table 3.1 presents the results for causality going from private investment to financial development. The OLS level and WG estimates for the
lagged dependent variable form an interval in which the system GMM
estimates fall. The specification tests for the three versions of SYS-GMM
used indicate that we can reject the null that the error term in first differences exhibits no first-order serial correlation and cannot reject the
hypothesis that there is no second-order serial correlation. The Sargan
tests in three models do not signal that the instruments are invalid. The
difference Sargan for SYS-GMM cannot reject the null of the additional
moment conditions being valid. These results indicate that every model

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 73 — #10

74 Determinants of Financial Development
Table 3.1 Does private investment cause financial development? 1970–98
(five-year-average data)
Dependent
variable: FDit
FDi,t−1
PIi,t−1

M1 (p-value)
M2 (p-value)
Sargan (p-value)
Diff-Sargan
(p-value)
Granger
Causality
(p-value)
LR effect point
estimate
(Standard error)
Observations

OLS
0.880
[16.46]∗∗∗
2.785
[5.08]∗∗∗

WG

SYS-GMM

SYS-GMM-1

SYS-GMM-2

0.597
[8.32]∗∗∗
5.091
[5.62]∗∗∗

0.806
[8.87]∗∗∗
5.286
[4.27]∗∗∗

0.741
[6.87]∗∗∗
6.745
[4.58]∗∗∗

0.578
[2.82]∗∗∗
3.779
[2.21]∗∗

0.00
0.89
0.36
0.87

0.00
0.92
0.24
0.76

0.05
0.69
0.44
1.00

0.00

0.00

0.00

0.00

0.03

23.21

12.63

27.22

26.02

8.96

[9.70]∗∗
212

[2.84]∗∗∗
212

[12.53]∗∗
212

[9.04]∗∗∗
212

[7.61]
212

Notes: 43 developing countries. Robust t statistics in brackets below point estimates.
∗ , ∗∗ , ∗∗∗ significant at 10%, 5% and 1%, respectively. The system GMM results are two-step estimates
with heteroscedasticity-consistent standard errors and test statistics; the standard errors are based on
finite sample adjustment of Windmeijer (2005). The M1 and M2 test the null of no first-order and
no second-order serial correlation in first-differenced residuals. The Sargan tests the over-identifying
restrictions for GMM estimators, asymptotically X 2 . The Diff-Sargan tests the null of mean stationarity for system GMM estimators in which SYS-GMM uses standard moment conditions, while
SYS-GMM-1 only uses lagged first-differences of FD dated t − 1 as instruments in levels and SYSGMM-2 uses only lagged first-differences of PI dated t − 1 as instruments in levels. The Granger
causality test is used to examine the null hypothesis that private investment doesn’t cause financial
development. LR measures the long-run effect of private investment on financial development. Its
standard error is approximated using the delta method.

from column 3 to column 5 is well specified and the SYS-GMM estimator is indeed preferable to the first-differenced GMM estimator for this
context. SYS-GMM estimates provide strong evidence for the positive
impact of private investment on financial development. This result is
supported by the Granger non-causality test, which clearly rejects the
null hypothesis, suggesting that there is a causal effect going from private investment to financial development. The Long-Run (LR) effect
estimate of SYS-GMM indicates that this effect tends to persist into the
long run. The SYS-GMM-1 estimates further confirm the findings, while
SYS-GMM-2 estimates support the short-run effect only, not the long-run
effect. Moreover, SYS-GMM and SYS-GMM-1 estimates indicate that a
high degree of persistence exists in the averaged data.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 74 — #11

Private Investment and Financial Development 75
Table 3.2 Does financial development cause private investment? 1970–98 (fiveyear-average data)
Dependent
variable: PIit
PIi,t−1
FDi,t−1

M1 (p-value)
M2 (p-value)
Sargan (p-value)
Diff-Sargan
(p-value)
Granger
Causality
(p-value)
LR effect point
estimate
(Standard error)
Observations

OLS
0.744
[14.04]∗∗∗
0.008
[2.09]∗∗

WG
0.232
[3.12]∗∗∗
0.010
[1.67]∗

SYS-GMM

SYS-GMM-1

SYS-GMM-2

0.521
[4.27]∗∗∗
0.015
[2.32]∗∗

0.490
[3.75]∗∗∗
−0.008
[0.85]

0.424
[3.00]∗∗∗
0.022
[2.11]∗∗

0.00
0.34
0.50
0.83

0.01
0.51
0.40
0.75

0.01
0.26
0.31
0.48

0.04

0.10

0.03

0.40

0.04

0.03

0.01

0.03

−0.02

0.04

[0.01]∗∗
198

[0.01]∗
198

[0.01]∗∗
198

[0.02]
198

[0.01]∗∗
198

Notes: 43 developing countries. The Granger causality test is used to examine the null hypothesis
that financial development doesn’t cause private investment. See Table 3.1 for more notes.

In Table 3.2 we turn to whether financial development Granger causes
private investment. The specification tests indicate that the models
associated with the three types of SYS-GMM are well specified. More
specifically, we can reject no first-order serial correlation but cannot
the hypothesis that there is no second-order serial correlation. Sargan
tests and difference Sargan tests suggest that neither the instruments
and mean stationarity conditions are invalid. Both SYS-GMM and SYS–
GMM-2 show a positive causal effect going from financial development
to private investment, not only in the short run but also in the long run.
Both SYS-GMM-1 in Table 3.1 and SYS-GMM-2 in Table 3.2 produce
consistent findings with their counterparts, respectively. However, using
the lagged first differences of PI dated t–1 as instruments in levels, SYSGMM-2 in Table 2.1 and SYS-GMM-1 in Table 3.2 do not confirm the
findings by their respective SYS-GMMs, especially the latter, perhaps suggesting that the moment conditions using lagged first differences of PI
dated t–1 may not contain much information.
The SYS-GMM-1 and SYS-GMM-2 above potentially serve as the robustness tests to the SYS-GMM in the two tables. In addition, a set of

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 75 — #12

76 Determinants of Financial Development

experiments are conducted to test whether the above findings are robust
to various model specifications. We first consider including GDP per
capita in logs and trade openness separately as additional regressors,
with the results reported on Appendix Tables A3.4 and A3.5, respectively.
Second, we introduce the second lags of dependent and independent
variables into the related models and report the results in Appendix
Table A3.6.
In part A of Appendix Table A3.4, with GDP in log every model is still
well specified. Both SYS-GMM and SYS-GMM-1 estimates indicate the
positive short-run and long-run effects of private investment on financial
development. SYS-GMM-1 estimates also show a positive effect of GDP
in log on financial development. SYS-GMM-2 estimates find that both
PI and LGDP in log are significantly positively associated with FD in the
short run, but not in the long run. In part A of Appendix Table A3.4, with
GDP in log in the models SYS-GMM and SYS-GMM-2 estimates suggest
that GDP in log enters the models significantly while FD is no longer
significant. GDP in log seems to pick up the short-run effects of financial
development on private investment.
In part A of Appendix Table A3.5, when trade openness (OPENC) is
included the SYS-GMM estimates continue to show a positive effect of
private investment on financial development, not only in the short run
but also in the long run. The model for SYS-GMM-1 is not well specified. The SYS-GMM-2 estimates find that both PI and OPENC have
been found to exert significantly positive effects on financial development in the short run, but not in the long run. In part B of Appendix
Table A3.5, SYS-GMM estimates suggest that the inclusion of OPENC
doesn’t change the significantly positive effect of financial development
on private investment, in either the short run or the long run.
In Appendix Table A3.6 we investigate the causality with AR(2) models.
Models for SYS-GMM and SYS-GMM-1 in both parts A and B of Appendix
Tables A3.6 and A3.6 are well specified, as supported by the specification
tests. Both SYS-GMM and SYS-GMM-1 estimates in part A of Appendix
Table A3.6 continue to support the first lag of PI to enter the models
significantly; in addition the second lag of PI is also observed to be significantly associated with financial development. The second lag of FD
has been found to be insignificant in the models. The SYS-GMM estimates in part B of Appendix Table A3.6 show that the first lag of PI
remains significantly positive; however, the second lag of FD and PI is
insignificant.
At least the robustness tests suggest that the inclusion of trade openness in the models doesn’t affect the pattern of the findings in Tables 3.1

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 76 — #13

Private Investment and Financial Development 77

and 3.2, and nor does the inclusion of the second lags of dependent and
independent variables in the models.
In summary, by using the SYS-GMM estimation method on averaged
data over 1970–1998 and controlling for the possibility of endogeneity
and omitted variable biases, this analysis finds that the positively significant causation exists in both directions between private investment
and financial development for 43 developing countries. It also indicates
a high degree of persistence in the averaged data. The findings are robust
to various estimation methods and model specifications.
However, it is worth noting that the asymptotic properties of the SYSGMM estimator depend on having a large number of cross section units.
Concerns remain regarding the finite sample bias for this context. The
findings still await further confirmation from the analysis on pooled
annual data which will be undertaken in Section 3.4.

3.4

Analysis on annual data

Using averaged data has a number of advantages, as well documented
in the literature, but its limitations are also notable. Averaging data
over fixed intervals (typically over five or ten years) arbitrarily modifies the time series dimension so that information loss is inevitable.
Although averaging data has the potential for removing business cycle
fluctuations, it is not guaranteed that such fluctuations are eliminated
effectively given the varied length of business cycles across countries
and over time. Moreover, methods like GMM – imposing homogeneity over all slope coefficients – fail to capture potential cross sectional
heterogeneity in the parameters.
This section moves on to explore the link between private investment
and financial development by using pooled annual data. In principle,
annual data can be more informative than averaged data in examining
the relevant effect. By explicitly looking at the yearly time series variation, one can explore the existence of heterogeneity across countries
adequately and estimate the parameters of interest more precisely.
As widely pointed out, assuming cross section error independence fails
to reflect a reality in which financial market integration and business
cycle synchronization are key features of a global economy. The analysis
in this section attempts to study the causality between private investment and financial development in a world where the existence of global
shocks causes cross section dependence across countries.
The remainder of this section proceeds as follows. Subsection 3.4.1 sets
out the common factor approach of Bai and Ng (2004). Subsection 3.4.2

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 77 — #14

78 Determinants of Financial Development

contrasts the panel unit root test of Bai and Ng (2004) with the Maddala
and Wu (1999) Fisher test, which is associated with the assumption of
cross section independence. Subsection 3.4.3 conducts the panel cointegration test of Pedroni (1999, 2004) on observed data and defactored
data. Subsection 3.4.4 adopts the Pesaran (2006) Common Correlated
Effect approach to estimate the models.
3.4.1

Methodology: Common factor approach

Assuming the interactions between financial development (FD) and
private investment over GDP (PI) are represented by the unrestricted
autoregressive distributed lag ARDL(p, p) systems:
FDit =

p
p



α1ij FDi,t−j +
β1ij PI i,t−j + θ1i t + λ1i f1t + v1it
j=1

PI it =

(3.9)

j=1

p
p



α2ij PI i,t−j +
β2ij FDi,t−j + θ1i t + λ2i f2t + v2it

(3.10)

j=1

j=1

i = 1, 2, . . . , 43 and t = 2, . . . , 29
For the sake of simplicity, denoting by y the dependent variable (either
FD or PI) and by xs the explanatory variables other than the lagged
dependent variable, we have
yit =

p
p



βij xi,t−j + θ1i t + λi ft + vit
αij yi,t−j +
j=1

(3.11)

j=1

i = 1, 2, . . . , 43 and t = 2, . . . , 29
where ft is a (r×1) vector of unobserved common factors, and λi is a factor




loading vector, such that λi ft = λi1 ft1 + λi2 ft2 · · · + λir ftr (here r is the
number of common factors). The common factors could be a global trend
component, a global cyclical component, common technological shocks
or macroeconomic shocks that cause cross section dependence. vit are
errors assumed to be serially uncorrelated and independently distributed
across countries. We allow for richer dynamics in the representations to
control for business cycle influences, while the current value of x, xit , is
excluded to avoid a potential endogeneity problem.
The above representations with a factor structure are believed to be
very general. Bai (2009) points out that the interactive effects model
including the interaction between factors, ft , and factor loadings, λi , is

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 78 — #15

Private Investment and Financial Development 79

more general than an additive effects model, the traditional one-way or
two-way fixed effects model.44
Since the common factors are unobservable, standard regression methods are not applicable for an equation like (3.11). Estimation of models
with a common factor structure is still at its early stage of development.
Pesaran (2006) estimates this type of model directly by proxying the
common factors with weighted cross section averages (Subsection 3.4.4
discusses this in detail). In spite of its convenience in not involving estimation of common factors, the Pesaran (2006) approach is confined to
the single factor case. Among others, Bai and Ng (2004) and Moon and
Perron (2004) seek to estimate the common factors. Their approaches
have advantages in accommodating multiple common factors that may
coexist in the economy, effectively contributing to panel unit root testing, panel cointegration testing and estimation of models in a more
general setting. Below is a brief description of common factor analysis
resulting from Bai and Ng (2004).
To overcome possible cross section dependence in panel unit root testing, Bai and Ng (2004) propose a PANIC approach – Panel Analysis of
Non-stationarity in Idiosyncratic and Common Components. Essentially
they assume the DGP of a series zit (which could be yit or xit for this case)
has a common factor structure in the sense that the series is the sum of
an unobserved deterministic component (dit ), an unobserved common

component (λi ft ) and an idiosyncratic component (eit ) as follows:


zit = dit + λi ft + eit

(3.12)

where ft is a vector of unobserved common factors and λi is the factor loading vector as defined before. The common and idiosyncratic
components could be stationary or non-stationary and are allowed
to be integrated of different orders. The common factor (ft ) and the
idiosyncratic component (eit ) can be expressed as:
fkt = αk fk,t−1 + υit

(3.13)

eit = ρi ei,t−1 + εit

(3.14)

The factor k is stationary if αk < 1 while the idiosyncratic component
(eit ) is stationary if ρi < 1. When the idiosyncratic component (eit ) is
stationary, conventional wisdom suggests that the factors can be estimated by using principal component (PC) analysis. As a crucial step Bai
and Ng (2004) propose applying a principal components analysis on the
differenced data (when a linear trend is not allowed) or differenced and

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 79 — #16

80 Determinants of Financial Development

demeaned data (when a linear trend is allowed) to estimate the factors
for the case where eit is integrated of order one.
To estimate the factors, the following two steps should be taken.
The first step is to estimate the number of common factors, and this
is discussed by Bai and Ng (2002) and Moon and Perron (2004). Bai
and Ng (2002) suggest using a principal component analysis on the
observed data to calculate the number of factors.45 For any arbitrary
k (k < min{N, T }), the estimates of λk and f k are derived by solving the
following minimization problem (dit = 0 is assumed for simplicity):
V (k) = min (NT )−1
k , f k



s.t.

T
N 


(zit − λki ftk )2

(3.15)

i=1j=1


fk fk
k k
= Ik
= Ik or
T
N




where ft = (ft1 , ft2 , ft3, ...ftr ) , λi = (λi1 , λi2 , λi3 . . . λir ) , i = (λ1 , λ2 ,


λ3 . . . λN ) and f is the (T × r) matrix of common components. Typically


fk fk

when T < N, the normalization that T = Ik is used.46 The estimated
√
factor matrix, denoted by fk , can be expressed as T times the eigenvectors corresponding to the k largest eigenvalues of the T × T matrix

k , can
zz . Given fk , the estimated factor loading matrix, denoted by 

z fk

be computed by T .
k , Bai and Ng (2002) propose to determine the number
Given fk and 
of factors by minimizing one of the following criterion functions:
PC(k) = V (k, fk ) + kg(N, T )

(3.16)

IC(k) = ln[V (k, fk )] + kg(N, T )

(3.17)

N 
T


where V (k, fk ) = (NT )−1
(εi εi ) is a measure of fit, and g(N, T ) is a
i=1j=1

penalty function that depends on the size of panel. The criterion functions capture a trade-off between measures of fit and a penalty function.
When the number of factors increases, the fit must improve, but the
penalty goes up. Bai and Ng (2002) provide three criterion functions for
PC(k) and IC(k), respectively. In general, IC(k) is easier to use since it
does not involve the estimation of a penalty function which requires
the choice of a bounded integer (kmax).
The integer minimizing a criterion function is the estimated number
of factors.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 80 — #17

Private Investment and Financial Development 81

The second step is to estimate the common and idiosyncratic components once the true number of factors, denoted by r, has been worked
out. Let Zit be the differenced data (without a linear trend) or differenced and demeaned data (with a linear trend) of observed data zit .47
The principal component estimator of the factor matrix f , denoted by
√

f , is T − 1 times the eigenvectors corresponding to the r largest eigen
values of the (T − 1) × (T − 1) matrix ZZ . Given
f , the estimated factor


Z
f

can be computed by
loading matrix, denoted by ,
T −1 .
The approach above yields r estimated common factors
ft and associated factor loadings
λi . The estimated idiosyncratic component takes the
form of


eit = Zit −
ft
λi

(3.18)

To remove the effect of possible over-differencing, Bai and Ng (2004)
propose to re-cumulate the estimated common factors,
ft , and estimated
idiosyncratic component,
eit , yielding

Ft =

t


fs

(3.19)

s=2


Eit =

t


eis

(3.20)

s=2

t = 2, . . . T
The resulting idiosyncratic component,
Eit , is in fact the defactored
data corresponding to the observed data zit .
3.4.2

Panel unit root tests

Over recent decades a number of panel unit root testing procedures have
been proposed in the literature to increase the power of univariate unit
root tests, such as Im et al. (2003), Levine et al. (2002) and Maddala and
Wu (1999). Associated with the unrealistic assumption of cross section
independence, these testing procedures are often classified as the first
generation of panel unit root tests. Since the influential work by Banerjee
et al. (2004), testing for unit roots in heterogeneous panels under the
assumption of cross section dependence has attracted a great deal of
attention. The testing procedures proposed by Pesaran (2007), Moon and
Perron (2004) and Bai and Ng (2004) are among the second generation
of panel unit root tests.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 81 — #18

82 Determinants of Financial Development

With the common factor structure presented earlier, Bai and Ng (2004)
note that the non-stationarity of series with a factor structure originates
from the non-stationarity of either the common component or idiosyncratic component or both. Bai and Ng (2004) test for unit roots for the
common component and idiosyncratic component,
Eit , separately. For
the idiosyncratic component, Bai and Ng (2004) propose testing the following ADF equation by using the (defactored) estimated idiosyncratic
component,
Eit , with no deterministic term:

Eit = di0
Eit−1 . . . + dip 
Eit−p + µit
Eit + di1 

(3.21)

They propose to use the Fisher P-test as suggested by Maddala and Wu
(1999) on the above ADF equation.
For the non-stationarity of the common factors, Bai and Ng (2004)
distinguish two cases. When there is only one common factor, a standard
ADF test with an intercept is suggested:

Ft = Dt + θ0 
Ft−1 +

p

θj 
Ft−j + υit

(3.22)

j=1

When there is more than one common factor, Bai and Ng (2004) propose an interactive procedure, analogous to the Johansen trace test for
cointegration.
Appendix Figure AF3.1 displays the time series plots of FD and PI for
43 countries over 1970–98. The data for FD and PI are standardized to
control for common trends. More specifically, taking deviations from
year-specific means removes the common components, common technological shocks or macro shocks, which have common effects across
countries. The development process of FD was in general more gradual and growing without bounds while the development process of PI
was more volatile and subject to bounds, in particular, PI experienced
increases in the 1970s, late 1980s and early 1990s, but fell in the early
1980s.
Appendix Table A3.7 reports the values of information criterion ICp1 (k)
(Bai and Ng, 2002) for the series of FD and PI.48 When r = 1, the ICp1 (k)
values for both FD and PI are minimized, clearly suggesting that there
is only one common factor for FD and PI, respectively. The time series
of the common factors for FD and PI are presented in Appendix Figure
AF3.2.
Table 3.3 contrasts the panel unit root test proposed by Maddala and
Wu (1999) and Bai and Ng (2004). The former is related to the assumption of cross section independence while the latter is defined under the

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 82 — #19

Private Investment and Financial Development 83
Table 3.3 Unit root tests in heterogeneous panels
Maddala and Wu (1999) Fisher test
Without trend
FD

With trend

65.143
[0.95]
97.754
[0.18]

PI

71.679
[0.87]
94.101
[0.26]
Bai and Ng (2004) test
FD

PI

Without trend With trend Without trend With trend
Common
Components
(ADF)
Idiosyncratic
Components
(P test)
Unit Root

−2.713
[0.07]∗
214.555
[0.00]∗∗∗
no

−3.099
[0.11]
199.876
[0.00]∗∗∗
yes

−1.981
[0.29]

−2.202
[0.49]

79.206
[0.68]

55.067
[1.00]

yes

yes

Note: The upper panel presents the results of Maddala and Wu (1999) Fisher test on the
observed data under the null hypothesis of a unit root. The lower panel reports the Bai and
Ng (2004) test, which decomposes the errors and conducts the unit root tests for the common
components (ADF test) and idiosyncratic components (Maddala and Wu (1999) Fisher test)
separately. P -values are in brackets.

assumption of cross section dependence. The Maddala and Wu (1999)
Fisher test, which does not require a balanced panel, indicates the
series of FD and PI may be I(1) processes no matter whether a trend
is allowed.49 Controlling for the common factor, the Bai and Ng (2004)
approach suggests that the series for FD and PI are I(1) variables when
we allow for a trend.
Since PI is the ratio of nominal private investment to nominal GDP,
the evolution of PI is bounded between 0 and 1. The above finding on the
PI series being an I(1) process, even though it is constrained within the
interval between 0 and 1, is consistent with the finding in Section 3.3 on
the averaged PI series being highly persistent. However, given that the
PI series is bounded and the low power of these tests, more sophisticated
testing methods may be called for.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 83 — #20

84 Determinants of Financial Development

3.4.3

Panel cointegration tests

When both FD and PI are integrated, cointegration between the two
variables is possible. This section uses panel cointegration techniques to
investigate the existence of a long-run relationship between them. Banerjee et al. (2004) point out that “cointegration across units and within
each unit may not be easily differentiatied due to the presence of cross
section cointegration”. The analysis of panel cointegration allowing for
cross section dependence is still in its infancy. Motivated by Gengenbach
et al. (2005), who suggest the use of defactored data,
Eit , in panel cointegration testing to control for cross section dependence, this section
contrasts the Pedroni (1999, 2004) residual-based panel cointegration
tests using observed data and defactored data.
The Pedroni (2004) test, widely used in empirical research in recent
years, assumes cross section independence of panel units but allows
for some heterogeneity in the cointegrating relationships. He proposes
two classes of statistics based on individual OLS residuals of the single cointegration regression below to test the null hypothesis of no
cointegration:


yit = αi + xi,t δit + uit

(3.23)

One class is the “panel” statistics,50 which are constructed by taking
the ratio of the sum of the numerators and the sum of the denominators of individual unit root statistics across the within dimension of
the panel with a homogeneity restriction, and the other is the “group
mean” statistics,51 which are based on the averages of individual unit
root statistics along the between dimension of the panel allowing for
heterogeneity.
Pedroni (2004) shows that the ADF-based tests perform better when the
sample size is small. Table 3.4 reports the group and panel ADF statistics
of Pedroni (1999, 2004) using observed data and defactored data, both
with and without a deterministic trend. The result associated with using
observed data shows, when common factors are allowed, that the presence of cross section dependence might render the Pedroni test unable to
detect the cointegration relationship in question. However, when common factors are extracted, the null of no cointegration can always be
rejected clearly, no matter whether we allow for a trend.52 This table
indicates that a stationary long-run relationship exists between financial development and private investment, and highlights allowing for
cross section dependence as an important source of information for this
analysis.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 84 — #21

Private Investment and Financial Development 85
Table 3.4 Panel cointegration tests between FD and PI
Observed data

Defactored data

Without trend

With trend

Without trend

With trend

1.749
2.661

1.039
1.360

−3.956
−3.822

−6.311
−5.855

Panel ADF
Group ADF

Note: This table reports the Pedroni (1999, 2004) cointegration test. The number of lag truncations used in the calculation of the Pedroni statistics is four. These are one-sided tests with
an critical value of −1.64. Under the null hypothesis of no cointegration, the test statistic is
asymptotically distributed as a standard normal.

Given the low power of these tests, this chapter still reports two estimates of the long-run relationship between FD and PI. One should soon
realize that the long-run coefficients in Table 3.5 and Table 3.6 are very
similar after normalizing the coefficients.
3.4.4

Estimation on annual data

Study of the estimation of large cross section and time series panel
datasets with a common factor structure has been fairly scarce. This
section undertakes the Pesaran (2006) common correlated effects
approach for the estimation of heterogeneous panels with common factors. Section 3.4.4.1 sets out the estimation methods associated with both
cross section error independence and cross section error dependence.
Section 3.4.4.2 presents the empirical evidence.
3.4.4.1

Estimation methods

Given that the series of FD and PI appear to be cointegated, there must be
a vector error correction representation, as shown by Engle and Granger
(1987), governing the co-movements of the series of FD and PI over
time. The corresponding error correction equation to Equation (3.11) is
as follows:



β
FDit = α1i FDi,t−1 − 1i PI it
α1i
q−1

−


j=0




q






p−1

−


j=1





p






α1im  FDi,t−j 

m=j+1

β1im  PI i,t−j  + θ1i t + λi ft + v1it

(3.24)

m=j+1

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 85 — #22

86 Determinants of Financial Development




β
PI it = α2i PI i,t−1 − 2i FDit
α2i
q−1

−






j=0

q




p−1

−






j=1



p






α2im  PI i,t−j 

m=j+1



β2im  FDi,t−j  + θ2i t + λi ft + v2it

(3.25)

m=j+1

i = 1, 2, . . . , 43 and t = p + 1, . . . , 29
where
α1i =

p

α1ij − 1
j=1

α2i =

p

α2ij − 1
j=1

β1i =

q

β1ij
j=0

β2i =

q

β2ij
j=0

In Equations (3.24) and (3.25), α1i and α2i are the coefficients for the
speeds of adjustment. − αβ1i and − αβ2i are the long-run coefficients for PI it
1i

and FDit , respectively.

p


2i

α1im and

m=j+1

q


β1im are the short-run coeffi-

m=j+1

cients for FDi,t−j and PI i,t−j in Equation (3.24), respectively, whereas
p
q


α1im and
β1im are, respectively, the short-run coefficients for
m=j+1

m=j+1

PI i,t−j and FDi,t−j in Equation (3.25).
For identification, the following equation should hold:

β 
β1i
2i
=1
α1i
α2i
In the absence of common factors, the within groups (WG) approach,
mean group (MG) approach of Pesaran and Smith (1995) and pooled
mean group (PMG) approach of Pesaran et al. (1999) are especially
suited to the analysis of panels with large time and large cross section
dimensions. The consistency of the WG estimator for the dynamic

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 86 — #23

Private Investment and Financial Development 87

homogeneous model is approximately justified when T is large, as N->∞
(Nickell, 1981). In comparison to the WG method, which allows only
the intercept to vary across countries but imposes homogeneity on all
slope coefficients, the MG and PMG approaches allow for considerable
heterogeneity across countries. The MG approach applies an OLS regression for each country to obtain individual slope coefficients, and then
averages the country-specific coefficients to derive a long-run parameter for the panel.53 For small samples, the MG estimator is likely to be
inefficient although it is still consistent.
Unlike the MG approach, which imposes no restriction on slope
coefficients, the PMG approach imposes cross section homogeneity
restrictions only on the long-run coefficient, but allows short-run coefficients, the speeds of adjustment and the error variances to vary across
countries. The restriction of long-run homogeneity can be tested via
a Hausman test. Under the null hypothesis of long-run homogeneity,
the PMG estimators are consistent and more efficient than the MG estimators. Moreover, Pesaran et al. (1999) show that the PMG estimators
are consistent and asymptotically normal irrespective of whether the
underlying regressors are I(1) or I(0).
The PMG approach requires that the coefficients for long-run effects
are common across countries, that is,
α1i =

p

α1j − 1
j=1

α2i =

p

α2j − 1
j=1

β1i =

q

β1j
j=0

β2i =

q

β2j
j=0

When common factors are allowed, Pesaran (2006) suggests the use
of the (weighted) cross-sectional averages of the dependent variable
and individual specific regressors to proxy the common factors. More
specifically, he proposes augmenting the observed regressors with the
(weighted) cross-sectional averages of the dependent variable and the
individual specific regressors such that as the number of cross section

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 87 — #24

88 Determinants of Financial Development

units goes to infinity, the effects of unobserved common factors can be
eliminated.
Pesaran (2006) proposes two common correlated effect (CCE)
approaches for large heterogeneous panels whose errors contain unobserved common factors. One is the common correlated effect pooled
(CCEP) estimator, a generalization of the within groups estimator that
allows for the possibility of cross section correlation, and the other is the
common correlated effects mean group (CCEMG) estimator, a generalization of the mean group estimator of Pesaran and Smith (1995) which
is adapted for the possibility of cross section correlation. The CCEP estimator is the within groups estimator with the (weighted) cross-sectional
averages of the dependent variable and the individual specific regressors included in the model. The CCEMG approach uses OLS to estimate
an auxiliary regression for each country in which the (weighted) cross
sectional averages of the dependent variable and the individual specific
regressors are added, and then the coefficients and standard errors are
computed as usual.
The Pesaran (2006) approach exhibits considerable advantages. It does
not involve estimation of unobserved common factors and factor loadings. It allows unobserved common factors to be possibly correlated with
exogenous regressors and exert differential impacts on individual units.
It permits unit root processes amongst the observed and unobserved
common effects. The proposed estimator is still consistent, although it is
no longer efficient, when the idiosyncratic components are not serially
uncorrelated.
In this context, the cross section means of FDit , FDi,t−1 , PIit and
PIi,t−1 are considered.54 The models are augmented with the interactions between regional dummies and cross sectional means of these
variables, and time dummies. The CCEP and CCEMG estimators have
been shown to be asymptotically unbiased and consistent as N -> ∞
and T -> ∞, and to have generally satisfactory finite sample properties.
More importantly, the CCEP and CCEMG estimators hold for any number of unobserved common factors as long as the number is fixed, which
is especially attractive.
A common correlated effects pooled mean group (CCEPMG) estimator is introduced in this study, which is a generalization of the pooled
mean group estimator of Pesaran et al. (1999) which also allows for the
possibility of cross section correlation. The restriction of long-run homogeneity can also be tested via a Hausman test. Under the null hypothesis
of long-run homogeneity, the CCEPMG estimators are expected to be
consistent and more efficient than the CCEMG estimators.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 88 — #25

Private Investment and Financial Development 89

3.4.4.2

Estimation results

Table 3.5 examines whether private investment causes financial development for 43 developing countries over 1970–98, while Table 3.6 studies
causality in the reverse direction. Tables 3.5 and 3.6 contrast the CCEP,
CCEMG and CCEPMG estimates with their counterparts, the WG, MG
and PMG estimates.55 The first group of estimates is associated with the
assumption of errors being cross sectionally dependent, while the latter group assumes cross section error independence. An autoregressive
distributed lag ARDL(3, 3) system has been adopted for this analysis.56
We look first at the case of cross section error dependence. The coefficients corresponding to the speeds of adjustment in the two tables
are significantly different from zero, suggesting that two-way Granger
causalities exist between them.
Imposing homogeneity on all slope coefficients except for the intercept, the CCEP estimates in the two tables suggest that there are positive
long-run effects going in two directions. When heterogeneity is sought,
the CCEMG and CCEPMG are called for. The CCEMG estimates find
that the long-run effects are less precisely estimated for both directions.
This is of no surprise – the long-run effects become much harder to
capture when full heterogeneity is allowed. Nevertheless, it does imply
that heterogeneity is especially prominent in this context. Moving from
the CCEMG (no restriction, but potentially inefficient) to CCEPMG (a
common long-run effect required) changes the results significantly: in
particular, imposing long-run homogeneity reduces the standard errors
and the speeds of adjustment. The restriction cannot be rejected at a
conventional level by a Hausman test. The CCEPMG estimates provide
evidence in support of significant long-run effects in both directions.
The long-run coefficients in Tables 3.5 and 3.6 are actually quite similar. For example, the CCPMG and CCEMP estimates of the long-run
coefficients for FD in Table 3.6 are 0.008 and 0.028, respectively, while
their counterparts in Table 3.5 are 0.043 (1/23.055) and 0.040 (1/25.220).
This result suggests that it is very likely for a single long-run relationship
to exist in this context.
Comparing the above case with the case of cross section error independence is worthwhile. As its counterpart associated with cross section
error dependence, the WG estimates (restrictions on all slope coefficients except for the intercept) show positive long-run effects in both
directions. Allowing for heterogeneity, but no error dependence, across
countries, the MG approach finds no evidence in support of significant
long-run effects in both directions. Supported by the Hausman tests in

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 89 — #26

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 90 — #27
12.398
[3.51]∗∗∗

Long-run
coefficient PIi,t−1

987
43

987
43

−1.154
[0.31]∗∗∗
−0.513
[0.24]∗∗∗
987
43

−0.764
[0.38]∗∗
−0.229
[0.25]

25.220
[19.18]

−0.335
[0.06]∗∗∗

CCEMG

0.91

Hausman

987
43

−0.244
[0.18]
−0.269
[0.22]

12.256
[3.96]∗∗∗

−0.070
[0.02]∗∗∗

WG

987
43

−0.206
[0.18]
0.001
[0.16]

10.098
[1.33]∗∗∗

−0.077
[0.01]∗∗∗

PMG

987
43

−0.152
[0.26]
0.028
[0.19]

12.085
[7.71]

−0.142
[0.02]∗∗∗

MG

Cross section independence

0.79

Hausman

∗ , ∗∗ , ∗∗∗ significant at 10%, 5% and 1%, respectively.

Notes: This table presents the Pesaran (2006) CCEP and CCEMG estimates, and CCEPMG estimates defined in the text under the assumption of cross
section error dependence, and their counterparts associated with the assumption of cross section error independence including the within group
estimates (WG), Pesaran and Smith (1995) mean group (MG) and Pesaran et al. (1999) pooled mean group (PMG) estimates. The PMG and CCEPMG
approaches use the long-run coefficients of MG and CCEMG estimates, respectively, as initial values, and the Newton-Raphson algoithm. The Hausman
test (p-values reported) is used to examine the null hypothesis of no difference between the MG and PMG estimators, and between CCEMG and
CCEPMG estimators. The asymptotic standard errors are reported in brackets. For WG and CCEP estimates the standard errors are corrected for possible
heteroscedasticity in cross sectional error variances.

Observations
No. of countries

Short-run coefficients
PIi,t=1
−0.250
[0.18]
−0.275
PIi,t=2
[0.22]

−0.090
[0.02]∗∗∗

−0.073
[0.02]∗∗∗

Speed of
adjustment
23.055
[2.15]∗∗∗

CCEPMG

CCEP

Cross section dependence

Does private investment cause financial development? 1970–98 (Annual data)

Dependent
variable: FDit

Table 3.5

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 91 — #28
968
43

0.000
[0.01]
0.001
[0.01]

−0.422
[0.04]∗∗∗
0.008
[0.00]∗∗

Note: See Table 3.5 for notes.

Observations
No. of countries

FDi,t=2

Short-run coefficient
FDi,t=1

Speed of
adjustment
Long-run
coefficient
FDi,t=1

CCEP

968
43

[0.01]

−0.013
[0.01]

−0.921
[0.08]∗∗∗
0.008
[0.00]∗∗∗

CCEPMG

968
43

−0.016
[0.02]
−0.021
[0.01]

−1.000
[0.10]∗∗∗
0.028
[0.05]

CCEMG

Cross section dependence

0.65

Hausman

968
43

0.000
[0.01]
0.001
[0.01]

−0.418
[0.04]∗∗∗
0.008
[0.00]∗∗

WG

Does financial development cause private investment? 1970–98 (Annual data)

Dependent
variable: PIit

Table 3.6

968
43

0.003
[0.01]
0.004
[0.01]

−0.479
[0.04]∗∗∗
−0.005
[0.00]

PMG

968
43

−0.007
[0.01]
−0.003
[0.01]

−0.582
[0.05]∗∗∗
0.068
[0.07]

MG

Cross section independence

0.29

Hausman

92 Determinants of Financial Development

Tables 3.5 and 3.6, the PMG estimates indicate a significant long-run
effect going from private investment to financial development, but not
vice versa. This tends to underscore the importance of allowing for heterogeneity across countries in the sense that, compared to the PMG
approach, the WG approach – ignoring the divergent performance across
countries – is likely to produce misleading results. Moving from PMG
to CCEPMG clearly highlights the importance of controlling for error
dependence across countries.
After controlling for error dependence and heterogeneity across countries, the CCEPMG estimates clearly suggest positive long-run effects
going in both directions between private investment and financial development. A note of caution may therefore be appropriate here: taking
careful consideration of the integrated properties of the data, the error
structure and the extent of heterogeneity are always worth keeping in
mind in the econometric analysis of panel data.
In the following a set of experiments are conducted to test whether the
above findings are robust to various model specifications. This research
considers including GDP per capita in logs and trade openness separately
as additional regressors.57 Results clearly indicate that the inclusion of
either GDP in log or trade openness does not alter the pattern of the
findings.
In sum, after allowing for global interdependence and heterogeneity across countries, this analysis on annual data clearly shows positive
long-run effects going in both directions between private investment and
financial development. It is very likely that a single long-run relationship
exists in this context. The findings in general suggest that surges of private investment stimulate the deepening of financial markets, and, on
the other hand, financial development facilitates resource mobilization,
and increases the quantity of funds available for investment.

3.5

Conclusion

This chapter aims to investigate the causality between aggregate private
investment and financial development in a globalized world. Using a
panel dataset with 43 developing countries over 1970–98, the analysis
is conducted in two steps. One is system GMM estimation on data for
five-year averages, indicating positive causal effects going in both directions and a high degree of persistence in the averaged data of private
investment and financial development. The other is a common factor
approach on annual data allowing for global interdependence and heterogeneity across countries. The analysis demonstrates that the series of

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 92 — #29

Private Investment and Financial Development 93

both private investment and financial development are integrated, and
two-way positive causal effects exist in the cointegrated system. In general, the chapter implies that, in a globalized world, private investment is
both an engine and a follower of financial development, and vice versa.
This analysis has produced significant insights into the interactions
between two important aspects of economic activities, aggregate private investment and financial development, in developing countries.
The implications of the findings can be summarized in the following.
First, the finding in terms of a positive effect of private investment
on financial development has rich implications for the development of
financial markets. Since sound macroeconomic policies, and a favourable
economic and legal environment, undoubtedly facilitate private investment, any efforts by government to reduce macroeconomic policy
uncertainty, improve the regulatory framework and strengthen creditor and investor rights will be conducive to the development of financial
markets. Moreover, the finding may shed light on a possible channel
through which other variables drive financial development, for example,
the effect of democracy and political stability on the speed of financial
development (Girma and Shortland, 2008) and Chapter 4.
Second, the finding on better financial development leading to a
private investment boom has clear implications for the conduct of
macroeconomic policies in developing countries. This chapter suggests
that as the financial system in a country becomes more sophisticated,
more funds are channelled for productive investment so that firms find
it easier to get access to them. This finding is in support of the financial development framework proposed by McKinnon (1973) and Shaw
(1973), who emphasize that financial liberalization and financial development can foster economic growth by boosting investment and its
productivity, substantially influencing macroeconomic policies in developing countries since the 1970s. This research contributes to the existing
body of research on the links between financial development and economic growth, by suggesting that the former may enhance the latter
through a private investment boom. This finding suggests that financial markets may well be the channel through which macroeconomic
volatility or downturn leads to declines in private investment, which
is consistent with what has happened during the 2007–2009 financial
crisis.
Third, this research stresses the importance of taking careful account
of error structure and heterogeneity in the econometric analysis of panel
data. By considering the effects of common trends in a global economy
and allowing for heterogeneity across countries, this analysis represents

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 93 — #30

94 Determinants of Financial Development

a significant improvement in comparison to existing research, which in
general assumes error independence across countries. The results generated from existing research may deserve careful examination since the
interactions and co-movements of economic factors, and the trends of
globalization, have been central features of the world economy in recent
decades.

Appendix tables
Table A3.1 The variables
Variable

Description

FD

Index for financial development in this
paper, mainly measuring the size of
financial intermediary development. It is
the first principal component of LLY,
PRIVO and BTOT.
Liquid Liabilities, the ratio of liquid
liabilities of financial system (currency plus
demand and interestbearing liabilities of
banks and non-banks) to GDP.
Private Credit, the ratio of credits issued to
private sector by banks and other financial
intermediaries to GDP.
Commercial-central Bank, the ratio of
commercial bank assets to the sum of
commercial bank and central bank assets.
The ratio of nominal private investment to
nominal GDP. It is replaced by PI/100.
Real GDP per capita (Chain) in log.
The sum of exports and imports over GDP
(at current prices). It is replaced by
log(1 + OPENC/100).

LLY

PRIVO

BTOT

PI
LGDP
OPENC

Source

Financial Development
and Structure Database
(FDS) in World Bank,
2008
FDS, 2008

FDS, 2008

Global Development
Network (GDN), 2002
Penn World Table 6.2
Penn World Table 6.2

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 94 — #31

Private Investment and Financial Development 95
Table A3.2 Descriptive statistics
Variable

Mean

Std. Dev.

Min

Max

Observations

FD

overall
between
within

−0.52

0.91
0.75
0.52

−2.65
−2.13
−2.36

4.14
1.66
2.34

N = 1198
n = 43
T-bar = 27.86

PI

overall
between
within

0.14

0.07
0.05
0.04

0.00
0.02
0.00

0.42
0.25
0.42

N = 1183
n = 43
T-bar = 27.51

LGDP

overall
between
within

3.47

0.35
0.34
0.09

2.76
2.88
3.09

4.19
4.02
3.82

N = 1183
n = 43
T-bar = 29

OPENC

overall
between
within

0.57

0.29
0.26
0.14

0.06
0.16
0.04

2.09
1.23
1.43

N = 1247
n = 43
T-bar = 29

Note: Appendix Table A3.1 describes all variables in detail.

Table A3.3 The list of countries in the full sample
East Asia & Pacific
PHL
Philippines
MYS Malaysia
PNG Papua New Guinea
THA Thailand
KOR Korea, Rep.
South Asia
IND
India
NPL
Nepal
PAK
Pakistan
Middle East & North Africa
DZA Algeria
MAR Morocco
EGY Egypt, Arab Rep.

Sub Sahara Africa
GAB
Gabon
SEN
Senegal
NGA Nigeria
NER
Niger
MUS Mauritius
KEN
Kenya
TGO
Togo
MDG Madagascar
GHA Ghana
GMB Gambia, The
RWA Rwanda
CMR Cameroon
CIV
Cote d’Ivoire
BDI
Burundi
ZAF
South Africa

Latin America & Caribbean
HND Honduras
TTO
Trinidad and Tobago
GTM Guatemala
CRI
Costa Rica
HTI
Haiti
SLV
El Salvador
BRB
Barbados
COL
Colombia
PER
Peru
VEN
Venezuela
ECU
Ecuador
MEX Mexico
ARG
Argentina
URY
Uruguay
CHL
Chile
DOM Dominican Rep.
PRY
Paraguay

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 95 — #32

96 Determinants of Financial Development
Table A3.4 Robustness test – GDP in log included (five-year-average data)
A. Does private investment cause financial development? 1970–98
Dependent
variable: FDit
FDi,t=1

OLS
0.879
[15.21]∗∗∗
2.744
[4.17]∗∗∗
0.014
[0.12]

PIi,t=1
LGDPit

M1 (p-value)
M2 (p-value)
Sargan (p-value)
Diff-Sargan (p-value)
Granger Causality
0.00
(p-value)
LR effect point
22.61
estimate
(Standard error)
[11.89]∗
Observations
212

WG
0.427
[5.46]∗∗∗
3.845
[4.25]∗∗∗
2.215
[4.41]∗∗∗

SYS-GMM SYS-GMM-1 SYS-GMM-2
0.753
[6.38]∗∗∗
5.692
[6.70]∗∗∗
0.634
[1.30]

0.638
[6.14]∗∗∗
6.007
[4.65]∗∗∗
0.972
[1.73]∗

0.693
[3.78]∗∗∗
4.679
[3.13]∗∗∗
1.240
[2.11]∗∗

0.00

0.00
0.99
0.51
0.98
0.00

0.00
0.80
0.35
1.00
0.00

0.02
0.46
0.30
0.71
0.00

6.71

23.04

16.58

18.26

[1.81]∗∗∗
212

[10.81]∗∗
212

[5.41]∗∗∗
212

[11.57]
212

B. Does financial development cause private investment? 1970–98
Dependent
variable: PIit
PIi,t=1

OLS
0.698
[10.95]∗∗∗
0.007
[1.74]∗
0.016
[1.60]

FDi,t=1
LGDPit

M1 (p-value)
M2 (p-value)
Sargan (p-value)
Diff-Sargan (p-value)
Granger Causality
(p-value)
LR effect point
estimate
(Standard error)
Observations

WG
0.186
[2.39]∗∗
0.004
[0.55]
0.081
[1.88]∗

SYS-GMM SYS-GMM-1 SYS-GMM-2
0.512
[5.19]∗∗∗
0.004
[0.54]
0.092
[3.34]∗∗∗

0.08

0.58

0.00
0.40
0.45
0.88
0.59

0.02

0.00

0.01

[0.01]∗
198

[0.01]
198

[0.01]
198

0.498
[5.01]∗∗∗
−0.013
[1.36]
0.095
[1.19]

0.352
[3.28]∗∗∗
0.012
[1.43]
0.103
[3.08]∗∗∗

0.00
0.47
0.27
0.67
0.18

0.01
0.26
0.46
0.97
0.16

−0.03

0.02

[0.02]
198

[0.01]
198

Notes: Log GDP is included in the models to test the robustness of the findings of Tables 3.1 and 3.2.
See Table 3.1 for more notes.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 96 — #33

Private Investment and Financial Development 97
Table A3.5 Robustness test – OPENC included (five-year-average data)
A. Does private investment cause financial development? 1970–98
Dependent
variable: FDit
FDi,t=1

OLS
0.863
[15.15]∗∗∗
2.699
[4.85]∗∗∗
0.124
[0.80]

PIi,t=1
OPENCit

M1 (p-value)
M2 (p-value)
Sargan (p-value)
Diff-Sargan (p-value)
Granger Causality
(p-value)
LR effect point
estimate
(Standard error)
Observations

WG
0.565
[7.86]∗∗∗
4.206
[4.36]∗∗∗
0.746
[2.41]∗∗

SYS-GMM SYS-GMM-1 SYS-GMM-2
0.734
[8.31]∗∗∗
4.759
[3.09]∗∗∗
0.603
[1.28]

0.00

0.00

0.01
0.92
0.32
0.25
0.00

19.67

9.68

17.88

[7.87]∗∗
212

[2.59]∗∗∗
212

[8.47]∗∗
212

0.764
[6.78]∗∗∗
7.494
[4.21]∗∗∗
−0.143
[0.23]

0.478
[3.22]∗∗∗
2.713
[1.93]∗
1.305
[3.50]∗∗∗

0.00
0.90
0.25
0.09
0.00

0.06
0.90
0.36
0.30
0.06

31.74

5.20

[14.33]∗∗
212

[3.73]
212

B. Does financial development cause private investment? 1970–98
Dependent
variable: PIit
PIi,t=1

OLS
0.742
[13.87]∗∗∗
0.008
[1.80]∗
0.002
[0.15]

FDi,t=1
OPENCit

M1 (p-value)
M2 (p-value)
Sargan (p-value)
Diff-Sargan (p-value)
Granger Causality
(p-value)
LR effect point
estimate
(Standard error)
Observations

WG
0.228
[2.82]∗∗∗
0.010
[1.60]
0.004
[0.14]

SYS-GMM SYS-GMM-1 SYS-GMM-2
0.455
[3.61]∗∗∗
0.013
[1.75]∗
0.018
[0.55]

0.07

0.11

0.01
0.33
0.24
0.10
0.09

0.03

0.01

0.02

[0.02]∗
198

[0.01]
198

[0.01]∗
198

0.340
[2.24]∗∗
−0.010
[0.80]
0.071
[1.00]

0.305
[2.38]∗∗
0.019
[2.13]∗∗
0.029
[0.83]

0.01
0.39
0.36
0.13
0.43

0.02
0.21
0.15
0.03
0.04

−0.01

0.03

[0.02]
198

[0.01]∗∗
198

Notes: Trade openness (OPENC) is included in the models to test the robustness of the findings of
Tables 3.1 and 3.2. See Table 3.1 for more notes.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 97 — #34

98 Determinants of Financial Development
Table A3.6 Robustness test – two lags (five-year-average data)
A. Does private investment cause financial development? 1970–98
Dependent
variable: FDit
FDi,t=1

OLS
1.076
[10.18]∗∗∗
−0.194
[1.67]∗
3.647
[3.75]∗∗∗
−1.118
[1.00]

FDi,t=2
PIi,t=1
PIi,t=2

M1 (p-value)
M2 (p-value)
Sargan (p-value)
Diff-Sargan (p-value)
Granger Causality
0.00
(p-value)
LR effect point
21.5
estimate
(Standard error)
[11.94]∗
Observations
169

WG
0.492
[5.07]∗∗∗
−0.179
[1.94]∗
4.767
[4.20]∗∗∗
3.385
[2.88]∗∗∗

SYS-GMM SYS-GMM-1 SYS-GMM-2
0.683
[4.46]∗∗∗
−0.216
[1.54]
5.735
[2.85]∗∗∗
3.305
[1.88]∗

0.564
[2.95]∗∗∗
−0.174
[1.17]
7.524
[2.87]∗∗∗
3.983
[2.55]∗∗

0.383
[1.36]
−0.079
[0.67]
5.605
[2.88]∗∗∗
2.812
[1.76]∗

0.00

0.02
0.53
0.21
0.64
0.00

0.09
0.84
0.16
0.60
0.00

0.37
0.77
0.23
0.88
0.01

11.87

16.96

18.89

12.09

[2.48]∗∗∗
169

[6.36]∗∗
169

[5.79]∗∗∗
169

[5.52]∗∗
169

B. Does financial development cause private investment? 1970–98
Dependent
variable: PIit
PIi,t=1

OLS
0.692
[8.34]∗∗∗
0.086
[0.99]
0.010
[1.30]
−0.004
[0.50]

PIi,t=2
FDi,t=1
FDi,t=2

M1 (p-value)
M2 (p-value)
Sargan (p-value)
Diff-Sargan (p-value)
Granger Causality
(p-value)
LR effect point
estimate
(Standard error)
Observations

WG
0.087
[0.99]
−0.081
[0.93]
0.016
[2.09]∗∗
0.002
[0.28]

SYS-GMM SYS-GMM-1 SYS-GMM-2
0.506
[4.24]∗∗∗
−0.090
[0.84]
0.022
[1.96]∗
−0.005
[0.81]

0.20

0.03

0.03
0.14
0.61
0.54
0.09

0.03

0.02

0.03

[0.02]
155

[0.01]∗∗
155

[0.01]∗∗
155

0.565
[3.88]∗∗∗
−0.038
[0.34]
−0.003
[0.25]
−0.002
[0.25]

0.402
[2.82]∗∗∗
−0.064
[0.64]
0.027
[2.08]∗∗
−0.004
[0.58]

0.05
0.16
0.47
0.27
0.73

0.06
0.08
0.45
0.25
0.10

−0.01

0.03

[0.03]
155

[0.01]∗∗
155

Notes: AR(2) models are considered to test the robustness of the findings of Tables 3.1 and 3.2. See
Table 3.1 for more notes.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 98 — #35

Private Investment and Financial Development 99
Table A3.7 Determination of the numbers of common factors for FD and PI

r
r
r
r
r
r
r
r

=1
=2
=3
=4
=5
=6
=7
=8

FD

PI

2.654
3.000
3.202
3.373
3.539
3.703
3.866
4.030

3.339
3.626
3.823
4.005
4.183
4.355
4.522
4.687

Note: This table reports the values of Information Criteria
(IC1) (Bai and Ng, 2002) for different numbers of factors
(r ). The integer minimizing a criterion function, IC1 for
example, is the estimated number of factors.

Appendix figures
0.5

FD
PI

0.4
0.3
0.2
0.1
0.0
-0.1
-0.2
-0.3
-0.4
1970

1975

Figure AF3.1

1980

1985

1990

1995

2000

Time series plots of FD and PI

Note: This graph depicts the time series plots of FD and PI over 1970–98.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 99 — #36

100 Determinants of Financial Development
80

commfd
commpi

60
40
20
0
-20
-40
0

5

Figure AF3.2

10

15

20

25

30

Time series plots of common factors for FD and PI

Note: This graph depicts the time series plots of common factors for FD and PI,
identified by using the PANIC approach of Bai and Ng (2004), over 28 years (1971–
98). Here commfd denotes the common factor for the series of FD, while commpi
denotes the common factor for the series of PI.

HUANG: “CHAP03” — 2010/9/29 — 20:06 — PAGE 100 — #37

4
Political Institutions and
Financial Development

4.1

Introduction

Over the last few decades, there has been a substantial increase in financial development in many developing countries. The average ratio of
private credit to GDP increased from 23% in 1980 to 32% in 2000,
while the average ratio of liquid liabilities to GDP rose from 32% in
1980 to 42% in 2000 in the developing world. On the political front,
between 1980 and 2000 62 developing countries undertook significant
institutional reforms towards democracy.58 Do the above economic and
political events in the developing world interact in important ways?
Much work has been done to explore the relationship between institutional improvement, especially political liberalization, and economic
growth. The existing research in this field does not unanimously establish the consequences of political reform for economic development.
Instead, it is made up of one line of research supporting positive
consequences, another line stressing negative consequences and some
maintaining ambiguous views. How does democratic process to improve
institutional quality influence financial development, especially in countries with low GDP per capita, high ethnic and religious divisions or
specific legal origins?
The importance of institutional improvement for financial development has been implicitly indicated by Clague et al. (1996) and Olson
(1993), who argue that, in comparison to autocracies, democracies better
facilitate property rights protection and contract enforcement, encouraging investment directly. In recent research on the political economy
of financial development, Pagano and Volpin (2001), Rajan and Zingales
(2003) and Beck et al. (2003) highlight the role of political intervention
and institutions in financial development. In examining what forces lead
101

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 101 — #1

102 Determinants of Financial Development

governments to undertake reforms to enhance financial development,
Chapter 5 finds that the extent of democracy is one of the significant
forces. However, there has been little research which directly studies
the impact of the democratic process for institutional improvement on
financial development.
This analysis mainly carries out a dynamic panel data study, focusing on 90 developed and developing countries. It examines the impact
on financial development of the democratic process in a broader
sense, in terms of institutional improvement rather than political
transformation.59 The bias-corrected Least Square Dummy Variable
(LSDV) estimator proposed by Kiviet (1995) and recently developed by
Bruno (2005) is the central method of this study and is compared with
the system GMM estimator proposed by Arellano and Bover (1995) and
Blundell and Bond (1998).
Before proceeding to the econometric analysis, this research provides
some preliminary evidence with a before-and-after event comparison to
study probably the most important institutional change, namely political transformation from an autocratic regime to a democratic one. It
focuses on 33 countries which underwent a democratic transformation
during 1960–2000 subject to data availability for financial development. This exercise examines the responses of the level and volatility
of financial development after a regime transition.
This chapter shows that improved institutional quality is associated with increases in financial development at least in the short run,
especially for lower–income, ethnically divided and French legal origin countries. The before-and-after event study also indicates that, in
general, democratic transitions are typically preceded by low financial
development, but followed by a short-run boost in, and greater volatility of, financial development. The findings of this research underline the
influence of institutional reform over the supply side of finance and shed
light on the strong and robust relationship between institutional quality
and economic performance.
The remainder of the chapter proceeds as follows. Section 4.2 presents
a brief review of the literature on institutions, democratization and
finance. Section 4.3 describes the sample and measures that are used in
this study. The empirical results are presented in Section 4.5, following
a description of dynamic panel data methods in Section 4.4. Section 4.6
concludes.

4.2

Institutions, democratization and finance

This section briefly outlines the theoretical background and motivation of this research. It discusses the role of institutions in financial

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 102 — #2

Political Institutions and Financial Development 103

development and the possible links between the democratic process and
finance.
Research on the effect of institutional reform on general economic
performance is associated with substantial controversies. Some argue
that the democratic process enhances fundamental civil liberties, stable
politics and an open society; promotes property rights protection and
contract enforcement; discourages corruption and lawlessness and fosters economic growth (Olson, 1993; Clague et al., 1996; Minier, 1998
and Persson, 2005). On the contrary, under pressures from different
interest groups, democratic structures may suffer from inefficiency in
decision-making and difficulty in implementing viable policies for rapid
growth. “Premature” democracy in developing countries possibly lowers the economic growth rate, and even results in economic disorder,
political instability and ethnic conflict (Persson and Tabellini, 1992 and
Blanchard and Shleifer, 2000). Tavares and Wacziarg (2001) show that
“the overall effect of democracy on economic growth is moderately negative” – an increase in human capital accumulation is offset by a decrease
in physical capital accumulation in the process of democratization.
Research on the role of institutions in financial development has
been substantial, especially research on the effects of the legal and regulatory environment on the functioning of financial markets. A legal
and regulatory system involving protection of property rights, contract
enforcement and good accounting practices has been identified as essential for financial development. Most prominently, La Porta et al. (1997,
1998) have argued that the origins of the legal code substantially influence the treatment of creditors and shareholders, and the efficiency
of contract enforcement.60 Among others, Mayer and Sussman (2001)
emphasize that regulations concerning information disclosure, accounting standards, permissible banking practices, and deposit insurance do
appear to have material effects on financial development.
Another significant work in this context is Beck et al. (2003), which
extends the settler mortality hypothesis of Acemoglu et al. (2001) to
financial development. They argue that colonizers, often named as
extractive colonizers, associated with an inhospitable environment aim
to establish institutions that privilege the small elite group and potentially ignore private property rights, while colonizers, often called settler
colonizers, in more favourable environments are more likely to create
institutions that support private property and balance the power of the
state. Accordingly, institutions in the extractive environment tend to
block financial development, while those in settler colonies are more
conducive to financial development.
The recently developed “new political economy” approach regards
“regulation and its enforcement as a result of the balance of power

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 103 — #3

104 Determinants of Financial Development

between social and economic constituencies” (Pagano and Volpin, 2001).
It centres on self-interested policy-makers who can intervene in financial
markets either through overall regulation or individual cases for purposes
such as career concerns and the promotion of group interests. Rajan and
Zingales (2003) emphasize the role which the interest groups, especially
the incumbent industrial firms and the domestic financial sector, can
play in the process of financial development.61
Arguably, countries controlled by elite groups are more inclined to
protect the interests of the elite from the bulk of society, restrict participation in the political system, and so on. The more power held by
the elite groups and the more autocratic the system, the more obstacles there are for financial development. This tends to suggest that
institutional reform intending to limit the influence of elite group over
policy-making, widen suffrage in the political system, respect basic political rights and civil liberties, remove institutional obstacles and enhance
institutional efficiency is beneficial to financial development. Girma
and Shortland (2008) study the impact of democracy chrematistics and
regime change on financial development, showing that both democracy
and regime change promote financial development.62 Apart from Girma
and Shortland (2008), research directly exploring the impact of democratic process for institutional improvement on financial development
has been lacking.
This research might contribute to our understanding of the structural
determinants of financial development. Looking at this issue is also significant for examining whether institutional innovation contributes to
an improved investment climate. This is because commonly used financial development indicators such as the ratio of liquid liabilities to GDP
and the ratio of credit issued to the private sector to GDP are generally
forward-looking. Better financial development is then an early indication
of a better investment environment.

4.3
4.3.1

The measures and data
The sample

This research studies the impact of institutional improvement on financial development, controlling for GDP, trade openness, aggregate investment and the black market premium. The measures and data for financial
development and institutional improvement are explained in more
detail below. Information on the classifications of income levels, region
dummies, ethnic fractionalization and legal origins is obtained from the

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 104 — #4

Political Institutions and Financial Development 105

World Bank Global Development Network Database (GDN) (2002). The
data for GDP, trade openness and aggregate investment are from the Penn
World Table 6.2. Data for the black market premium are from the GDN
(2002).
This study focuses on a panel of 90 non-transition economies over the
period 1960–99 with five observations per country. Averaging data over
non-overlapping, eight-year periods enables us to abstract from business
cycle influences and to examine both short-run and long-run effects. The
countries included for this analysis are those undertaking some political
reforms to improve institutional quality, but not necessarily experiencing a democratic transition over 1960–99. The sample excludes the East
European countries,63 which became democracies and independent only
following the end of the Cold War. The selection of countries is based
on the Polity index, “polity2” of the PolityIV Database explained below.
We naturally use data up to the end of the twentieth century, which is
partly because of data availability for some important variables, like the
black market premium,64 and partly because annual data for 40 years are
sufficient for a dynamic panel data study.
4.3.2

The measure and data for financial development

The aggregate measure of financial development in this context is
denoted by FD. Since there is no single aggregate index in the literature,
we use principal components analysis to produce a new aggregate index.
Ideally, the principal component analysis should be based on indicators
from the banking sector, stock market and bond market so as to capture different aspects of financial development. However, data on stock
market and bond market development are rarely available for before
1975 or even later, so the analysis focuses on financial intermediary
development.
The measure is based on three widely used indicators of financial
intermediary development as follows:65
1. Liquid Liabilities (LLY), calculated as the liquid liabilities of banks
and non-bank financial intermediaries (currency plus demand and
interest-bearing liabilities) over GDP. It measures the size, relative to the
economy, of financial intermediaries including three types of financial
institutions: the central bank, deposit money banks and other financial
institutions.
2. Private Credit (PRIVO), defined as the credit issued to the private sector by banks and other financial intermediaries divided by GDP,
excluding the credit issued to government, government agencies and
public enterprises, as well as the credit issued by the monetary authority

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 105 — #5

106 Determinants of Financial Development

and development banks. This captures general financial intermediary
activities provided to the private sector.
3. Commercial-Central Bank (BTOT ), the ratio of commercial bank
assets over the sum of commercial bank and central bank assets. It proxies the advantages of financial intermediaries in channelling savings
to investment, monitoring firms, exerting corporate governance and
undertaking risk management relative to the central bank.
Since these indicators are used to measure the size of the banking
system,66 FD mainly captures the size of bank-based intermediation. FD
is the first principal component of these three indicators above, and
accounts for 72% of their variation. The weights resulting from principal component analysis over the period 1990–99 are 0.59 for Liquid
Liabilities, 0.63 for Private Credit and 0.50 for Commercial-Central Bank.
The data on these indicators are obtained from the World Bank’s
Financial Structure and Economic Development Database (2008).
4.3.3

The measure and data for institutional improvement

The research focuses on political institutions and studies their impact
on financial development. The institutional improvement index is the
Polity indicator “polity2” in the PolityIV Database (Marshall and Jaggers, 2009), denoted by POLITY2. It proxies the degree of democracy and
seeks to measure institutional quality based on the freedom of suffrage,
operational constraints and balances on executives and respect for other
basic political rights and civil liberties. It is called the “combined polity
score”,67 defined as the democracy score minus the autocracy score.68
To pick up any effect of institutional improvement on financial development, this exercise tries to incorporate all democratic reform episodes
in the sense that any increase of the annual “polity2” score for a country
will be considered even if it remains an autocratic regime or a democratic
regime over the whole period.
To select democratic transition countries for the before-and-after event
study, we also take into account the freedom index from Freedom House
Country Survey (2008).

4.4

Methodology

To assess the relationship between institutional improvement and financial development, the following model is estimated:69
,

yit = αyi,t−1 + βxi,t−1 + zi,t−1 δ + ηi + φt + vit

(4.1)

i = 1, 2, . . . 90 and t = 2 . . . 5

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 106 — #6

Political Institutions and Financial Development 107

where yit is the dependent variable FD, xit is the explanatory variable
POLITY2, zit is a vector of controlling variables including the logarithm
of the real GDP per capita (LGDP), trade openness (OPENC), aggregate
investment (CI) and the black market premium (BMP). OPENC is the
logarithm of one plus the trade share, the sum of exports and imports
over GDP (at current prices), divided by 100. CI is the ratio of investment
to real GDP per capita (using domestic prices), divided by 100. BMP is
the logarithm of one plus the black market premium divided by 100. δ
is a parameter vector, e.g. (δ1 , . . . δ4 ), . ηi is an unobserved time-invariant
country-specific effect and can be regarded as capturing the combined
effect of all omitted variables. φt is the time effect. vit is the transitory
disturbance term.
We assume that the transient errors vit are serially uncorrelated. In
system GMM estimation all x
s and z
s are assumed to be potentially
correlated with ηi and predetermined with respective to time-varying
errors.70 To avoid the potential endogeneity of explanatory variables,
lagged values of xi, t and zi, t are included in the regression equation,
which allows feedback from the past shocks onto xi, t−1 and zi, t−1
while the current and future realizations of yit do not affect them. The
assumption is inspired by Rodrik and Wacziarg (2005), who argue that
“democratisations tend to follow periods of low growth rather than
precede them”. In contrast to the GMM approach, the following biascorrected Least Squares Dummy Variable (LSDV) estimation assumes all
x
s and z
s to be strictly exogenous, which rules out the possibility of
feedbacks from the past, current and future shocks onto xi, t−1 and
zi, t−1 .
When the Ordinary Least Square (OLS) technique is used to estimate
this model, the OLS estimate of α is inconsistent and likely to be biased
upwards since the lagged values of yit are positively correlated with the
omitted fixed effects.
A number of methods have been developed to deal with the presence
of fixed effects in the dynamic panel data model. By using a within group
operator, the LSDV method eliminates any omitted variables bias created
by the unobserved individual effect and estimates the new model below
by OLS:
−

−

−

yit − yi = α(yi,t−1 − yi,−1 ) + (xi,t−1 − xi,−1 )β
−

−

+ (zi,t−1 − zi,−1 )δ + (vit − vi )
i = 1, 2, . . . 90 and t = 2 . . . 5

(4.2)

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 107 — #7

108 Determinants of Financial Development
−

−

−

−

where yi , xi and zi are the group means, that is, yi =

5



−

yit /5, xi =

t=2
5


−
xit /5 and zi =
zit /5. Since the lagged value of y is correlated with
t=2
t=2
5



the new error term, as shown by Nickell (1981), the LSDV estimate of α
can be badly downwards biased for small T , even as N goes to infinity.
Another way commonly used to wipe out the individual effects is to
apply first-differencing to Equation (3.1). By estimating the following
first-difference equation, the first-difference 2SLS estimator of Anderson
and Hsiao (1980, 1981), first-differenced GMM estimator of Arellano and
Bond (1991) and the system GMM estimator of Arellano and Bover (1995)
and Blundell and Bond (1998) are proposed among others:


yit = αyi,t−1 + xi,t−1 β + zi,t−1
δ + φt − φt−1 + vit

i = 1, 2, . . . 90 and t = 3 . . . 5

(4.3)

Conventional wisdom suggests that the first-differenced GMM estimator is consistent and asymptotically more efficient than the firstdifferenced 2SLS estimator. However, it may suffer from finite sample
bias by employing weak instruments, as argued by Blundell and Bond
(1998), that is, that “when the autoregressive parameter α is close to
unity or the variance of the individual effects (ηi ) increases relative to the
variance of the transient disturbances (vit ) in the standard AR(1) model,
the instruments available for the first-differenced equation are likely to
be weak”.
To handle the weak instrument problem, Arellano and Bover (1995)
and Blundell and Bond (1998) impose a mean stationarity assumption
on initial conditions,71 and combine the first-difference equations with
suitably lagged levels as instruments and levels equations with suitably lagged first differences as instruments. More specifically, the system
GMM estimator, one of the main focuses of this analysis, uses all lagged
values of y, x and z as instruments for yi,t−1 , xi,t−1 and zi,t−1 in
the first difference equation above,72 and the lagged first differences of
the series (yit , xit , zit ) dated t–1 as instruments for the untransformed
equations in levels.73 The system GMM estimator has been found to be
more efficient than the first-differenced GMM estimator in the presence
of persistent data and weak instruments for first differences.
The asymptotic properties of the system GMM estimator depend on
having a large number of cross section units, however. One of the main
problems in using this estimator is that it may have poor finite sample
properties in terms of bias and imprecision. Starting from Kiviet (1995),

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 108 — #8

Political Institutions and Financial Development 109

a bias correction of LSDV has recently been developed for use in short
dynamic panels. Kiviet (1995) derives an approach to approximating
the small sample bias of the LSDV estimator and suggests that the bias
approximation be evaluated at the estimates from some consistent estimates rather than the unobserved true parameter values, which makes
bias correction operationally feasible. The Monte Carlo evidence from
Kiviet (1995), Judson and Owen (1999) and Bun and Kiviet (2003) suggests that the bias-corrected LSDV estimator (LSDVC) is more efficient
than LSDV, first-differenced 2SLS, first-differenced GMM and system
GMM in terms of bias and root mean square error (RMSE) for small or
moderately large samples. Bruno (2005) derives a bias approximation of
various orders in dynamic unbalanced panels with a strictly exogenous
selection rule.74
This analysis compares the OLS, LSDV, LSDVC and SYS-GMM, standing
for the system GMM estimator, for the whole sample and three subsamples. The LSDVC estimator is regarded as the preferred estimator,
especially for subsamples, even though the independent variables other
than the lagged dependent variable are assumed to be strictly exogenous. The initial estimator for the LSDVC could be either first-differenced
GMM or the SYS-GMM estimator. However, the SYS-GMM is selected
since the Difference Sargan test of additional moments conditions could
not reject the null, and the SYS-GMM may be a more reliable estimator
than first-differenced GMM in this context.

4.5

Evidence

The econometric methods are applied to study the effect on financial
development of a broader issue, that is institutional improvement, based
on even a slight change of the Polity index, “polity2”. Before proceeding
to the econometric analysis, we look at some preliminary evidence on
the effect of the establishment of representative government on financial
development by applying a “before-and-after” approach to 33 countries
which underwent transformation from autocratic regimes to complete
or partial democracies at some point during 1960–2001.
4.5.1

Preliminary evidence

The sample selection for the “before-and-after” event study relies on both
the “polity2” index and “freedom” index from the Freedom House Country Survey (2008). Countries with either their “polity2” index increasing
from negative values to positive values or their “freedom” index jumping from “Not Free” to “Partly Free” or “Free” for at least ten years are

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 109 — #9

110 Determinants of Financial Development

considered for this analysis. In general, the “polity2” and “freedom”
indices yield similar results on the timing of democratic transition for
most cases. However, the “polity2” index excludes countries with small
populations (less than half a million) and the “freedom” index is only
available starting from 1972–73.75 For completeness, the selection of
democratic transition countries combines both indices when both are
available and relies on either of them otherwise.
The “before-and-after” approach compares an individual country’s
financial development performance under autocratic and democratic
regimes.76 To ease interpretation, the FD measure has been rescaled77
in Table 4.1. The five- or ten-year average of FD preceding democratic
transition is compared with the mean of FD during the first five or ten
years under democracy for 33 countries.
The ten-year average of standardized FD for the sample countries
increases by 0.093 on average after the initiation of a democratic transition and more than half of the sample countries exhibit an improvement
in financial development.78 It is worth noting that the majority of countries which suffered from a dramatic drop in financial development
after democratization are Latin American countries. In contrast, most
African countries underwent a pick-up in financial development after
their democratic transformations. The divergent performance in countries’ financial development implies that, apart from democratization,
the level of financial development in each country may be affected by
numerous factors including macroeconomic risks and changes in the
general investment climate.79 On average, these results tend to suggest
that the establishment of representative government is often associated
with an increase in financial development, but the effect is only sizeable
for a subset of countries.
The upper chart of Figure 4.1 displays the cross-country median FD ten
years before and after transition for the whole sample. The lower chart
of Figure 4.1 plots the coefficients on the fixed-effect estimate of 20 time
dummies before and after democratisation to reflect the dynamic effect
of a sustained democratization.80 The two figures show that the sample
countries in general experience a drop in FD prior to democratization,
which is in accordance with the view that worsened economic conditions
are associated with a subsequent democratization. After democratization,
FD appears to move slightly upwards on average in one to five years,
followed by a surge in five to ten years.
Figure 4.2 describes the standard deviation of the FD growth rate
before and after a stable democratization for whole the sample and subsamples. Democratization has led to a substantial rise in the standard

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 110 — #10

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 111 — #11
1983
1982
1985
1989
1978
1979
1994
1996
1984
1986
1980
1989
1987
1993
1991
1994
1992
1994
1990
1990
1988
1989
1979

Period covered

Argentina
Bolivia
Brazil
Chile
Dominican Rep.
Ecuador
Ethiopia
Ghana
Grenada
Guatemala
Honduras
Hungary
Korea, Rep.
Lesotho
Madagascar
Mexico
Mali
Malawi
Nicaragua
Nepal
Pakistan
Panama
Peru

(0, 10]
−0.466
−0.789
−0.341
0.305
−0.351
−0.486
−0.615
−1.042
0.635
−0.411
−0.278
−0.335
1.031
−0.266
−0.460
−0.367
−0.532
−0.737
−0.548
−0.657
−0.266
0.035
−0.351

−0.375
−1.055
−0.492
−0.136
−0.527
−0.674
−0.562
−1.295
0.247
−0.569
−0.199
−0.631
0.307
−0.300
−0.942
−0.592
−0.625
−0.814
−0.342
−0.735
−0.224
0.142
−0.300

2

[−10, 0)

1

−0.092
0.266
0.151
0.441
0.176
0.188
−0.053
0.252
0.388
0.157
−0.079
0.296
0.724
0.034
0.483
0.224
0.093
0.078
−0.206
0.077
−0.042
−0.106
−0.051

DIFF1

3

−0.266
−1.000
−0.492
−0.138
−0.337
−0.629
−0.547
−1.256
0.232
−0.645
−0.142
−0.584
0.482
−0.572
−0.983
−0.404
−0.625
−0.840
−0.667
−0.745
−0.186
0.093
−0.230

[−5, 0)

4

Change in FD standardized before and after democratization
Demo’tion
year

Countries

Table 4.1

−0.723
−1.397
−0.610
0.040
−0.311
−0.451
−0.458
−0.969
0.256
−0.532
−0.252
−0.323
0.874
−0.364
−0.808
−0.138
−0.559
−0.783
−0.757
−0.506
−0.231
0.039
−0.433

(0, 5]

5

−0.457
−0.396
−0.118
0.178
0.027
0.178
0.090
0.288
0.024
0.114
−0.110
0.261
0.393
0.208
0.176
0.267
0.066
0.056
−0.090
0.239
−0.045
−0.054
−0.203

DIFF2

6

−0.789
−0.017
−0.725
−0.262
0.190
−0.370

−0.134
−0.717
−0.719
−0.577
−1.333
0.286
−0.492
−0.255
−0.868
0.133
−0.028
−0.902
−0.779

0.603
−0.185
−0.192
−0.446
1.428
−0.213
−0.741
−0.245
−0.499
−0.750
−0.506
−0.163
−0.127
0.817
−0.554

−0.570
−0.844
−0.408
0.405
−0.356
−0.697

(5, 10]

[−10, −5)
−0.483
−1.110

8

7

continued

0.039
−0.489
0.562
0.135
0.627
−0.184

0.317
0.307
0.064
0.421
1.295
−0.185
0.160
0.534

0.539
0.361
0.022

−0.087
0.266

DIFF3

9

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 112 — #12
1986
1989
1989
1982
1987
1993
1978
1985
1994
1991

(0, 10]
−0.113
−0.167
−0.588
−0.740
−0.110
0.029
0.903
−0.523
0.514
−1.341

−0.003
−0.398
−0.467
−0.685
0.036
−0.299
−0.193
−0.145
0.453
−0.926

2

[−10, 0)

1

0.118
−0.079
0.078
0.252

−0.110
0.230
−0.121
−0.055
−0.146
0.328
1.096
−0.378
0.061
−0.415

DIFF1

3

0.034
−0.610
−0.588
−0.724
0.221
−0.275
−0.132
0.246
0.434
−0.926

[−5, 0)

4

−0.363
−0.319
−0.471
−0.876
0.405
−0.128
0.048
−0.419
0.562
−1.349

(0, 5]

5

0.015
−0.110
0.090
0.181

−0.396
0.291
0.117
−0.151
0.184
0.148
0.181
−0.666
0.128
−0.423

DIFF2

6

0.076
−0.125
−0.211
−0.925
−0.508
0.440
0.452
−0.398
0.731
−1.316

(5, 10]

[−10, −5)
−0.040
0.132
−0.346
−0.645
−0.149
−0.323
−0.254
−0.536
0.465

8

7

0.212
−0.005
0.149
0.450

0.116
−0.257
0.136
−0.280
−0.359
0.763
0.706
0.139
0.266

DIFF3

9

Notes: This table compares the financial development performance for 33 countries before and after democratization. See text for the country selection.
Columns 1 and 2 show the average of FD standardized ten years before or after transition, respectively. DIFF1 is the difference between them. Columns
4 and 5 show the average of FD standardized five years before or after transition. DIFF2 is the difference between the two columns. Columns 6 and 7
show the average of FD standardized ten to five years before transition and five to ten years after transition, respectively. DIFF3 is the difference between
columns 6 and 7. In the lower section the average, 1st Quartile, median value and 3rd Quartile are caculated for DIFF1, DIFF2 and DIFF3. The FD
measure has been divided by the cross-country standard deviation of FD in 1999.

Average
1st Quartile
Median Value
3rd Quartile

Philippines
Poland
Paraguay
El Salvador
Suriname
Seychelles
Thailand
Uruguay
South Africa
Zambia

Demo’tion
year

Continued

Period covered

Countries

Table 4.1

Political Institutions and Financial Development 113
−0.2

FD

Cross-country median financial development

−0.4
−0.6
−0.8
0

5

10

FD

15

20

Fixed effect estimates of financial development

0.50
0.25
0.00
−0.25
0

5

Figure 4.1

10

15

20

Financial development ten years before and after democratization

Note: 33 democratization countries, 1960–99. Upper figure shows the crosscountry median financial development for these countries. Lower figure plots the
coefficients of fixed-effect estimate of 20 time dummies before and after democratization. The regression is estimated by OLS, in which the country effects, time
effects, controlling variables like LGDP, OPENC, BMP and CI are included.
0.60

Pre-transition
Post-transition

0.55
0.50
0.45
0.40
0.35
0.30
0.25
1

2

Figure 4.2 Volatility
democratization

3

of

4

financial

5

6

7

development

8

9

ten

years

10

11

pre/post-

Note: 33 democratization countries, 1960–99. This figure shows the volatility
of financial development, standard deviation of FD growth rate, for the whole
sample and eight subsamples before and after democratization.

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 113 — #13

114 Determinants of Financial Development

deviation of the FD growth rate for the whole sample. Regional groups
like Latin American (LAC) and Sub-Saharan African (SSA) countries experience a higher standard deviation of the FD growth rate, but Asian
countries (ASIA) do not.81 The standard deviations of the FD growth
rate in income groups, like low-income countries (INCLOW) and middleincome countries (INCMID), and in legal origin groups, like British legal
origin countries (LEG_UK) and French legal origin countries (LEG_FR),
increases after their democratic transition. An increase in the standard
deviation of the FD growth rate may reflect the fact that the removal
of institutional obstacles after democratic transition could bring about
short-run investment booms, reflected in a more volatile FD growth rate.
4.5.2

Regression results

Section 4.5.1 does provide some interesting results on the impact of
democratic transition on financial development. However, this evidence is preliminary, and not convincing. In what follows we present
the econometric evidence, for both the whole sample and three
subsamples.82
4.5.2.1

Whole sample results

Table 4.2 reports the results for the whole sample, including estimation
by OLS, LSDV, LSDVC and SYS-GMM. For each estimate, the first column is the baseline specification in which the income level and trade
openness are present, while the second column controls for the black
market premium and aggregate investment. The point estimate and the
approximate standard error of the long-run effect for each model are
reported. Given the estimated models, the OLS, LSDV, LSDVC and SYSGMM estimates require that the long-run effect must have same sign as
the short-run effect. For the SYS-GMM estimate, the table reports serial
correlation tests, a Sargan test and a Difference Sargan test. The serial correlation tests are used to examine the null hypothesis of no first-order
serial correlation and no second-order serial correlation respectively in
residuals in first differences. Given the errors in levels are serially uncorrelated, we would expect to find significant first-order serial correlation,
but no significant second-order correlation in the first-differenced residuals. The Sargan test of over-identifying restrictions is used to examine
the overall validity of the instruments by comparing the sample moment
conditions with their population analogue. The Difference Sargan test,
proposed by Blundell and Bond (1998), is used to test the null hypothesis
that the lagged differences of the explanatory variables are uncorrelated
with the errors in the levels equations.

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 114 — #14

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 115 — #15
0.306
[0.39]
233

0.951
[0.00]***
0.015
[0.04]**
0.133
[0.02]**
0.159
[0.53]

0.110
[0.06]*
220

0.863
[0.00]***
0.015
[0.04]**
0.013
[0.85]
0.273
[0.31]
−0.240
[0.00]***
2.372
[0.00]***

0.041
[0.02]*
233

0.379
[0.00]***
0.025
[0.04]**
1.232
[0.00]***
1.818
[0.01]***

0.038
[0.02]*
220

0.320
[0.00]***
0.026
[0.05]**
1.179
[0.00]***
1.912
[0.02]**
−0.089
[0.45]
0.798
[0.49]

LSDV

0.102
[0.08]
233

0.825
[0.00]***
0.018
[0.19]
0.655
[0.02]**
1.214
[0.05]**

0.099
[0.07]
220

0.796
[0.00]***
0.020
[0.15]
0.567
[0.04]**
1.470
[0.06]*
−0.050
[0.73]
0.755
[0.48]

LSDVC

0.03
0.20
0.28
0.18
0.256
[0.23]
233

0.689
[0.01]***
0.080
[0.03]**
0.466
[0.20]
2.195
[0.17]

0.848
[0.00]***
0.028
[0.05]**
0.048
[0.74]
0.500
[0.31]
−0.236
[0.02]**
2.785
[0.10]*
0.04
0.98
0.34
0.89
0.187
[0.17]
220

SYS-GMM

3 LR measures the long-run effect of political liberalization on financial development. Its standard error is approximated using the delta method.

GMM estimator.

2 Sargan is a test of the over-identifying restrictions for GMM estimators, asymptotically ?2 . Diff-Sargan tests the null of mean stationarity for the system

1 M 1 and M 2 are tests for null of no first-order and no second-order serial correlation in the first-differenced residuals, asymptotically N (0,1).

Notes: 82 countries. p-value is reported in brackets below point estimates. Year dummies included in all models. ∗ significant at 10%; ∗∗ significant at
5%; ∗∗∗ significant at 1%.The LSDVC estimator is the corrected LSDV estimator developed by Kiviet (1995) for finite sample bias and contructed for
dynamic unbalanced panels by Bruno (2005). The SYS-GMM results are two-step estimates with heteroscedasticity-consistent standard errors and test
statistics; the standard errors are based on the finite sample adjustment of Windmeijer (2005).

M1(p-value)1
M2(p-value)1
Sargan(p-value)2
Diff-Sargan (p-value)2
LR effect point estimate3
(Standard error)
Observations

CI_(i, t − 1)

BMP_(i, t − 1)

OPENC_(i, t − 1)

LGDP_(i, t − 1)

POLITY2_(i, t − 1)

FD_(i, t − 1)

OLS

Institutional improvement and financial development (whole sample), 1960–99

Dependent variable: FD_(it)

Table 4.2

116 Determinants of Financial Development

It is worth noting that, first, the autoregressive parameter estimated by
LSDVC and SYS-GMM lies in the interval defined by the OLS levels and
LSDV estimates. Recall that, in AR(1) models, the OLS levels estimate of
the autoregessive parameter is biased upwards in the presence of fixed
effects and the LSDV estimate is biased downwards in a short panel. A
consistent estimate of the autoregressive parameter can be expected to lie
in between the OLS levels and LSDV estimates. It is a simple indication
of the presence of serious finite sample biases when particular estimates
fail to fall into this interval or are very close to the bounds.
Both OLS and LSDV estimates indicate a significant positive effect
of democratization on financial development although they are biased
in opposite directions. The LSDVC estimator suggests evidence at the
20% significance level. The SYS-GMM estimate provides strong evidence
that the improvement in institutional quality is associated with financial
development, and the diagnostic tests, including the first- and secondorder serial correlation tests, Sargan test and Difference Sargan test,
support this. In general, the coefficients on the GDP level, trade openness
and aggregate investment are positively signed, while the coefficient of
the black market premium is negatively signed. The long-run effects in
the cases of the OLS and LSDV estimates have been found to be positive
and stable. However, the long-run effects for LSDVC and SYS-GMM are
less precisely estimated.
In general, the table provides evidence, which is not due to unobserved
heterogeneity or endogeneity biases, that democratization is followed by
advances in financial development at least in the short run.
4.5.2.2

Subsamples

In principle, the system GMM and LSDVC estimates impose homogeneity on all slope coefficients. One concern over the above findings is
that these parameters may be heterogeneous across countries. A natural way to confront this problem is to investigate subsamples, which are
more homogeneous. We turn to three subsamples in this section: lowerincome countries, ethnically diverse countries and French legal origin
countries.83 Since the cross section dimensions of these samples are relatively small, LSDVC is expected to be more appropriate than SYS-GMM
for them.
Table 4.3 presents the results for the lower-income countries, made up
of low-income and lower-middle-income countries, covering the majority of the developing countries. We find strong evidence of a positive
effect of institutional improvement on financial development in the
short run for every estimator. The LSDVC should be the most reliable

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 116 — #16

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 117 — #17
0.180
[0.18]
177

0.932
[0.00]***
0.012
[0.11]
0.128
[0.05]*
−0.297
[0.33]

0.064
[0.05]
169

0.854
[0.00]***
0.009
[0.25]
0.049
[0.49]
−0.348
[0.26]
−0.244
[0.01]***
2.202
[0.01]***

Notes: 57 countries. For other notes, please see Table 4.2.

M1 (p-value)1
M2 (p-value)1
Sargan (p-value)2
Diff-Sargan (p-value)2
LR effect point estimate3
(Standard error)
Observations

CI_(i, t − 1)

BMP_(i, t − 1)

OPENC_(i, t − 1)

LGDP_(i, t − 1)

POLITY2_(i, t − 1)

FD_(i, t − 1)

OLS

0.072
[0.02]***
177

0.387
[0.00]***
0.044
[0.00]***
0.662
[0.03]**
1.676
[0.03]**

0.068
[0.02]***
169

0.292
[0.04]**
0.048
[0.00]***
0.659
[0.04]**
1.412
[0.10]*
−0.123
[0.26]
0.847
[0.50]

LSDV

0.186
[0.11]*
177

0.840
[0.00]***
0.030
[0.04]**
0.249
[0.35]
1.177
[0.14]

0.142
[0.08]*
169

0.775
[0.00]***
0.032
[0.07]*
0.245
[0.29]
1.123
[0.18]
−0.086
[0.47]
0.554
[0.64]

LSDVC

Institutional improvement and financial development (lower-income countries), 1960–99

Dependent variable: FD_(it)

Table 4.3

0.00
0.26
0.41
0.84
5.454
[141.51]
177

0.991
[0.00]***
0.049
[0.07]*
0.238
[0.52]
0.603
[0.51]

0.00
0.22
0.68
0.97
0.13
[0.09]
169

0.790
[0.00]***
0.027
[0.04]**
0.255
[0.13]
0.363
[0.63]
−0.223
[0.04]**
2.213
[0.12]

SYS-GMM

118 Determinants of Financial Development

estimator, given the above discussion. Moreover, it also indicates that
the effect of improved institutional quality on financial development is
sustained into the long run. Trade openness enters the models at the
20% significance level.
Table 4.4 shows the results for ethnically diverse countries which have
a level of ethnic fractionalization greater than the sample median. We
find strong evidence of the positive effect of institutional improvement
on financial development in the short run. The autoregressive parameter estimates from LSDVC and SYS-GMM are very close. The LSDVC
estimates suggest a positive effect of political liberalization on financial
development at the 20% significance level with GDP and trade openness
entering significantly. The SYS-GMM estimates provide much stronger
evidence, in which GDP and trade openness are present at the 20% significance level. The long-run effects and approximate standard errors
are in general less precisely estimated except for the case of the OLS and
LSDV estimates.
The results for countries with French legal origin are reported in Table
4.5. This selection is essentially inspired by the work of La Porta et al.
(1998), which regards legal origin as a main determinant of financial
development. The experiments for British, German (LEG-GE) and Scandinavian (LEG-SC) legal origin groups produce no evidence in favour of
a causal link from institutional improvement to financial development.
First it is worth noting that the autoregressive parameter estimated
by SYS-GMM in the baseline model lies outside of the interval defined
by the OLS and LSDV estimates, further implying the LSDVC may be a
more reasonable estimator in this context. The LSDVC estimates typically
show evidence in support of a positive effect of institutional improvement on financial development for French legal origin countries at the
15% significance level. The finding seems to be in line with La Porta
et al. (1998), which claims that the main characteristic for countries with
French legal origins is that private property rights are generally neglected,
while British legal origin countries care more about private property owners. The finding supports a tentative hypothesis that democratization in
French legal origin countries tends to change the status of private property owners in the national economy, and is thus conducive to financial
development.
In sum, the above studies on subsamples have produced a coherent
set of findings: improved institutional quality leads to greater financial
development, at least in the short run. In the group of lower-income
countries, a significant long-run effect is also observed. In general, we

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 118 — #18

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 119 — #19
0.200
[0.15]
220

0.913
[0.00]***
0.017
[0.02]**
0.144
[0.01]***
0.333
[0.17]

0.115
[0.06]**
211

0.840
[0.00]***
0.018
[0.01]***
0.045
[0.51]
0.388
[0.12]
−0.237
[0.00]***
1.894
[0.02]**

Notes: 67 countries. For other notes, please see Table 4.2.

M1 (p-value)1
M2 (p-value)1
Sargan (p-value)2
Diff-Sargan (p-value)2
LR effect point estimate3
(Standard error)
Observations

CI_(i, t − 1)

BMP_(i, t − 1)

OPENC_(i, t − 1)

LGDP_(i, t − 1)

POLITY2_(i, t − 1)

FD_(i, t − 1)

OLS

0.041
[0.02]**
220

0.365
[0.00]***
0.026
[0.04]**
1.193
[0.00]***
1.879
[0.01]***

0.036
[0.02]*
211

0.313
[0.01]***
0.025
[0.07]*
1.148
[0.00]***
1.959
[0.02]**
−0.091
[0.44]
0.458
[0.70]

LSDV

0.109
[0.08]
220

0.820
[0.00]***
0.020
[0.16]
0.585
[0.03]**
1.318
[0.06]*

0.103
[0.08]
211

0.794
[0.00]***
0.021
[0.17]
0.501
[0.08]*
1.529
[0.07]*
−0.055
[0.67]
0.530
[0.69]

LSDVC

Institutional improvement and financial development (ethnically diverse countries), 1960–99

Dependent variable: FD_(it)

Table 4.4

0.02
0.19
0.12
0.73
0.384
[0.69]
220

0.857
[0.00]***
0.055
[0.11]
0.378
[0.18]
1.447
[0.21]

0.03
0.54
0.24
0.61
0.175
[0.159]
211

0.807
[0.00]***
0.034
[0.06]*
0.206
[0.11]
0.816
[0.14]
0.218
[0.01]***
1.304
[0.24]

SYS-GMM

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 120 — #20
0.075
[0.05]
153

0.763
[0.00]***
0.018
[0.12]
0.144
[0.04]**
0.468
[0.23]

0.070
[0.04]*
150

0.721
[0.00]***
0.020
[0.07]*
0.042
[0.63]
0.755
[0.04]**
−0.185
[0.01]***
1.836
[0.04]**

Notes: 49 countries. For other notes, please see Table 4.2.

M1 (p-value)1
M2 (p-value)1
Sargan (p-value)2
Diff-Sargan (p-value)2
LR effect point estimate3
(Standard error)
Observations

CI_(i, t − 1)

BMP_(i, t − 1)

OPENC_(i, t − 1)

LGDP_(i, t − 1)

POLITY2_(i, t − 1)

FD_(i, t − 1)

OLS

0.034
[0.02]
153

0.214
[0.10]*
0.027
[0.11]
0.643
[0.06]*
2.691
[0.01]***

0.038
[0.02]*
150

0.214
[0.12]
0.030
[0.08]*
0.572
[0.12]
2.110
[0.06]*
−0.135
[0.35]
1.110
[0.49]

LSDV

0.094
[0.06]
153

0.708
[0.00]***
0.027
[0.12]
0.294
[0.47]
2.421
[0.03]**

0.104
[0.08]
150

0.694
[0.00]***
0.032
[0.14]
0.155
[0.65]
1.997
[0.07]*
−0.088
[0.61]
1.558
[0.38]

LSDVC

Institutional improvement and financial development (French legal origin countries), 1960–99

Dependent variable: FD_(it)

Table 4.5

0.08
0.12
0.31
0.51
0.251
[0.27]
153

0.848
[0.00]***
0.038
[0.11]
0.319
[0.10]*
0.522
[0.48]

0.06
0.19
0.91
0.95
0.144
[0.12]
150

0.709
[0.00]***
0.042
[0.04]**
0.129
[0.51]
1.250
[0.12]
−0.135
[0.06]*
1.445
[0.38]

SYS-GMM

Political Institutions and Financial Development 121

find the black market premium has a negative effect, while GDP, trade
openness and aggregate investment enter positively.

4.6

Conclusion

This research examines whether institutional improvement stimulates
financial development using a panel of 90 economies over the period
1960–99. By comparing newly developed panel data techniques, including bias-corrected LSDV and system GMM estimators, this research shows
that improved institutional quality is associated with increases in financial development at least in the short run, and this is particularly true for
lower-income, ethnically divided and French legal origin countries. For
the lower-income countries, this effect is expected to persist over longer
horizons. The preliminary evidence from a “before-and-after” approach
indicates that, in general, democratic transitions are typically preceded
by low financial development, but followed by a short-run boost in, and
greater volatility of, this.
The findings of this research highlight the influence of institutional
innovation on the supply side of financial development. They shed light
on the strong and robust relationship between institutional quality and
economic performance, and present further grounds for institutional
reform.
The findings in the panel data study on the coexistence of the effect
of institutional innovation, GDP and trade openness on financial development are very significant. First, the study enriches the evidence for
an openness-finance nexus. Huang and Temple (2005)’s cross section
and panel data study suggests that trade openness is very likely to
boost financial development, for which institutional improvement could
serve as one channel. The IMF (2003) indicates the possible existence
of such a channel by concluding that “greater openness to trade and
stronger competition are conducive to institutional improvement, and
thus to growth”. However, the findings of this research tend to suggest that there are additional channels via which more open policies
exert a positive effect on financial development. The findings are also
consistent with Rajan and Zingales (2003)’s claim that trade openness
is helpful for changing incumbents’ willingness to promote financial
development.
Second, the study has implications for economic and political reform.
Giavazzi and Tabellini (2004) argue that “studying the effects of each
reform (economic and political reform) individually can be misleading”
and there are positive feedback effects and interaction effects between

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 121 — #21

122 Determinants of Financial Development

economic and political liberalization. The findings of this chapter seem
to be consistent with their findings on the interaction effects, in the sense
that institutional reform under an open economic environment exerts
an additional boost to investment and economic growth, and thus to
financial development.

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 122 — #22

Political Institutions and Financial Development 123

Appendix tables
Table A4.1 The variables
Variable

Description

Source

FD

Index for financial development in this paper,
mainly measuring the size of financial
intermediary development. It is the first
principal component of LLY, PRIVO and BTOT .
Liquid Liabilities, the ratio of liquid liabilities
of financial system (currency plus demand and
interest-bearing liabilities of banks and
nonbanks) to GDP.

Financial Development
and Structure Database
(FDS) in World Bank,
2008

LLY

PRIVO

BTOT

POLITY2

Private Credit, the ratio of credits issued to
private sector by banks and other financial
intermediaries to GDP.
Commercial-Central Bank, the ratio of
commercial bank assets to the sum of
commercial bank and central bank assets.
The index for the degree of democracy. It is the
“polity2” in PolityIV Database.

FDS, 2008

FDS, 2008

PolityIV Database
Marshall and Jaggers
(2008)

LGDP

Real GDP per capita (Chain) in log.

Penn World Table 6.2

OPENC

The sum of exports and imports over GDP (at
current prices). The regression uses
log(1+OPENC/100).
The sum of investment over real GDP per
capita (using domestic prices). The regression
uses CI/100.
Black market premium (%, means zero). The
regression uses log(1+BMP/100).

Penn World Table 6.2

Dummy for low-income group

GDN, 2002

CI

BMP
INCLOW
INCMID

Penn World Table 6.2

Global Development
Network (GDN), 2002

Dummy for middle-income group, made up of
lower-middle-income and low-income
countries
ETHFRAC Dummy for ethnic fractionalization

GDN, 2002

LEG_UK

GDN, 2002

Dummy for British legal origin

GDN, 2002

LEG_FR

Dummy for French legal origin

GDN, 2002

LEG_GE

Dummy for German legal origin

GDN, 2002

LEG_SC

Dummy for Scandivanian legal origin

GDN, 2002

ASIA

Dummy for Asian countries

GDN, 2002

LAC

Dummy for Latin American countries

GDN, 2002

SSA

Dummy for Sub-Sarahan African countries

GDN, 2002

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 123 — #23

124 Determinants of Financial Development
Table A4.2 Descriptive statistics
Variable

Mean

Std. Dev.

Min

Max

Observations

FD

overall
between
within

−0.61

1.17
1.09
0.56

−2.91
−2.77
−2.45

1.85
2.53
4.48

N = 341
n = 90
T-bar = 3.79

POLITY2

overall
between
within

−1.83

6.39
5.37
3.56

−10.00
−9.78
−12.70

10.00
9.83
10.92

N=438
n=90
T-bar = 4.87

LGDP

overall
between
within

7.73

0.84
0.86
0.26

5.89
6.28
6.70

10.06
10.06
8.73

N = 399
n = 86
T-bar = 4.64

OPENC

overall
between
within

0.43

0.19
0.19
0.08

0.07
0.13
0.16

1.18
1.08
0.78

N = 399
n = 86
T-bar = 4.64

CI

overall
between
within

0.13

0.08
0.07
0.04

0.01
0.02
0.00

0.39
0.31
0.35

N = 399
n = 86
T-bar = 4.64

BMP

overall
between
within

0.33

0.66
0.47
0.53

−0.04
0.00
−1.65

7.64
3.17
5.88

N = 402
n = 88
T-bar = 4.57

Note: Appendix Table A4.1 describes all variables in detail.

HUANG: “CHAP04” — 2010/9/29 — 20:06 — PAGE 124 — #24

5
Financial Reforms for
Financial Development

5.1

Introduction

Financial liberalization has been one of the key trends characterizing the
post-Bretton Woods era, with decreasing capital controls and an increasing participation of developing countries in international financial markets in recent decades. More broadly, domestic financial development,
measured in terms of liquid liabilities or stock market capitalization,
has risen dramatically over the same period. By using Bayesian Model
Averaging (BMA) and General-to-specific (Gets) approaches, Chapter 2
examines the long-run determinants of financial development. However,
what are the factors directly stimulating governments to liberalize the
financial sector, aimed at enhancing financial development? Building on
the framework of Abiad and Mody (2005) (AM hereafter), this chapter
attempts to answer this question and to provide a more comprehensive
view of the political economy of financial reform.
Although financial liberalization has been criticized as increasing the
likelihood of financial crises and financial fragility, it is widely regarded
as promoting the flow of financial resources, thereby reducing capital
costs, stimulating investment and fostering financial development and
economic growth (McKinnon, 1973; Shaw, 1973; Demirgüç-Kunt and
Detragiache, 1998; Summers, 2000). In practice, governments in recent
decades have been committed to reducing direct intervention in the
financial system by easing or removing controls over interest rates, credit
allocation and financial transactions domestically and internationally,
opening up the banking system for foreign entry, and privatizing commercial banks or non-bank financial intermediaries. What are the main
factors inducing governments to take these steps?

125

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 125 — #1

126 Determinants of Financial Development

AM introduce an analytical framework to examine the factors that
induce governments to undertake financial reforms. Using an ordered
logit technique to estimate their specifications, AM argue that policy
change in a country is positively related to its level of liberalization and
any liberalization gap from the regional leader. The pace of reform is
found to be affected by shocks or discrete changes such as a balance-ofpayments crisis, a banking crisis, a new government’s first year in office,
participation in an IMF programme and a decline in US interest rates.
However, they find that ideology and political and economic structures
have “limited influence” on the likelihood of reform.
The AM analytical framework is attractive in many respects, but some
aspects of their empirical analysis merit further attention. First, the
ordered logit technique they apply may not be appropriate for this
context, although the discrete and ordinal nature of the financial liberalization level, FLi,t , and policy change,
FLi,t , may render the ordered
logit method a natural choice at first glance. In the AM analysis, the
dependent variable is not the level of financial liberalization, but the change
in the level of liberalization. AM treat a movement from a score of 1 to
3 of the underlying index the same as they do a movement from 16 to
18, among many other possibilities for a specific change (say +2). However, the lack of cardinality in the scale of their original measure implies
that movements along the scale for a specific change are not equivalent. Given this particular nature of the dependent variable, resorting
to the ordered logit technique may not lead to the expected gains.84
Second, as in most cross-country research, AM do not take into account
the effects of common trends and the possibility of error dependence
across countries and over time. The importance of error dependence
seems especially relevant when the effects of domestic learning and
regional diffusion are investigated, and is confirmed by the results of
this analysis, including a formal test of dependence following Pesaran
(2004).
In this analysis, four innovations are introduced. The first is that,
rather than their ordered logit technique, this analysis centres on the
Pesaran (2006) common correlated effect pooled (CCEP) approach that
allows for the possibility of error dependence across countries. Second,
to adjust for serial correlation in individual errors, the panel-robust standard errors after Arellano (1987) are computed for the CCEP estimates.85
Third, it adds the extent of democracy into the AM framework, by introducing the Polity indicator, “polity2”, in the PolityIV Database (Marshall
and Jaggers, 2008), seeking to measure institutional quality. The level of

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 126 — #2

Financial Reforms for Financial Development 127

democracy is a potentially important variable which reflects the political environment in which new policies are approved or rejected, and
policy changes take place. Fourth, in addition to focusing on the original dataset used by AM, it takes up a further investigation based on a
larger set of countries, in which the Abiad and Mody financial liberalization index is replaced by the Chinn-Ito index of capital account openness
(2006).
This chapter produces the following findings. In general it confirms the
negative effects of banking crises and high inflation on policy change,
as observed by AM. It is also consistent with AM in suggesting that the
effects of new governments in their first year and IMF programmes are
strong when financial sectors are highly repressed, and become weaker as
the level of financial liberalization goes up. However, this chapter points
to the following three distinct conclusions. First, it shows that some
of their findings on the effects of crises and shocks are fragile. Second,
it is at odds with AM on the effects of domestic learning and regional
diffusion. It suggests that policy change in a country is negatively rather
than positively related to its liberalization level, and the liberalization
gap from the regional leader appears less relevant than in AM. Third, this
analysis observes a significant effect of the extent of democracy, the new
variable added to the Abiad and Mody framework, on policy change. The
findings on the negative effects of domestic learning and irrelevance of
regional diffusion are supported by a larger sample of countries drawing
on the Chinn-Ito index of capital account openness.
Section 5.2 provides a brief discussion of the model specifications and
econometric methods. Section 5.3 presents the empirical results, based
on the original dataset with the AM measure, and a larger set of countries with the Chinn-Ito measure, separately. Section 5.4 discusses the
implications of the findings. Section 5.5 concludes.

5.2

Methodology

This section starts by briefly describing the models used in AM to
study how financial reform is shaped, followed by a discussion of the
econometric methods that will be applied in this chapter.
5.2.1

Model specifications

Below is the general model structure that captures the effects of domestic
learning, regional diffusion, discrete changes and ideology and structure

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 127 — #3

128 Determinants of Financial Development

on policy changes.

FLit = α(FL∗it − FLi,t−1 )
+ β1 (REG_FLi,t−1 − FLi,t−1 )


+ β2 SHOCKSit


+ β3 IDEOLOGYit


+ β4 STRUCTUREit
+ εit

(5.1)

The dependent variable,
FLit , is used to measure policy change, the
difference between the level of financial liberalization in the current
period, FLit , and the past level of financial liberalization, FLi,t−1 . FL86
it
ranges between 0 and 1, with 0 and 1 corresponding to complete financial repression and complete financial liberalization, respectively. FL∗it
is the desired level of financial liberalization. The adjustment factor, α,
measures the degree of status quo bias. A lower value of α is associated
with more resistance to reform and a greater bias towards the status
quo. The first term on the RHS is therefore used to examine domestic adjustment. The second term captures regional diffusion in which
REG_FLi,t−1 is the maximum level of financial liberalization achieved
in the region. SHOCKSit denotes discrete changes including four types
of crises – balance-of-payments crises (BOPit ), banking crises (BANKit ),
recessions (RECESSIONit ) and high inflation periods (HINFLit ) – and three
types of internal or external influences like the incumbent’s first year
in office (FIRSTYEARit ), the influence of international financial institutions reflected by a dummy for an IMF programme of lending (IMFit )
and the influence of global factors proxied by the US Treasury Bill rate
(USINTit ). IDEOLOGYit reflects political orientation including a dummy
for left-wing government (LEFTit ) and a dummy for right-wing government (RIGHTit ). STRUCTUREit represents structural variables (either
economic or political), for example the trade openness measure (OPENit )
used in AM.
Overall, the Abiad and Mody framework is appealing, covering almost
all possible aspects. However, a political structural variable, the extent
of democracy (POLITY2it ), may be relevant to the analysis and is added
to their framework. This is the Polity indicator “polity2” in the PolityIV
Database (Marshall and Jaggers 2008) and seeks to measure institutional
quality based on the freedom of suffrage, operational constraints on executives and respect for other basic political rights and civil liberties. It

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 128 — #4

Financial Reforms for Financial Development 129

is called the “combined polity score”, defined as the democracy score
minus the autocracy score.87
5.2.1.1

Benchmark specification

The benchmark specification assumes that the desired level of financial
liberalization, FL∗it , is the perfect level of financial liberalization and
the adjustment factor, α, is positively related to the level of financial
liberalization to allow for the likelihood of domestic learning. Putting
FL∗ = 1 and α = θ1 FLi,t−1 (θ1 > 0) into Equation (5.1) above and
reparameterizing, we have

FLit = θ1 FLi,t−1 (1 − FLi,t−1 )
+ θ2 (REG_FLi,t−1 − FLi,t−1 )


+ θ3 SHOCKSit


+ θ4 IDEOLOGYit


+ θ5 STRUCTUREit
+ εit

(5.2)

This equation is Equation (4) in AM.
5.2.1.2

Alternative specifications

Relaxing two assumptions used in the benchmark specification, three
alternative specifications are considered:
First, rather than assuming the desired level of financial liberalization,
FL∗it , to be full liberalization, it is natural to adopt country-specific measures of the desired extent of liberalization. When plugging FL∗ = c
(0 < c < 1) and α = θ1 FLi,t−1 into Equation (5.1) above, redefining the
coefficients yields the following equation as in Equation (5) of AM88 :

FLit = θ1 FLi,t−1 + θ2 FL2i,t−1
+ θ3 (REG_FLi,t−1 − FLi,t−1 )


+ θ4 SHOCKSit


+ θ5 IDEOLOGYit


+ θ6 STRUCTUREit
+ εit

(5.3)

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 129 — #5

130 Determinants of Financial Development

Second, the desired level of financial liberalization, FL∗it , might be
reasonably regarded to be increasing with the level of income. When
FL∗ = a + bYit and α = θ1 FLi,t−1 are considered, Equation (5.1) above
can be rearranged and reparameterized as Equation (6) in AM:89

FLit = θ1 FLi,t−1 + θ2 FL2i,t−1
+ θ3 (FLi,t−1 .Yit )
+ θ4 (REG_FLi,t−1 − FLi,t−1 )


+ θ5 SHOCKSit


+ θ6 IDEOLOGYit


+ θ7 STRUCTUREit
+ εit

(5.4)

Finally, when the possibility that shocks, ideology and structure variables may exert effects on the status quo bias is taken into account, the
previous assumption α = θ1 FLi,t−1 is replaced by the following equation:
α = γ1 FLi,t−1
+ γ2 (REG_FLi,t−1 − FLi,t−1 )


+ γ3 SHOCKSit


+ γ4 IDEOLOGYit


+ γ5 STRUCTUREit
Putting this expression as well as FL∗ = c into Equation (5.1) and
redefining the coefficients yields the third specification, Equation (8) in
AM, below:

FLit = θ1 FLi,t−1 + θ2 FL2i,t−1
+ θ3 (REG_FLi,t−1 − FLi,t−1 )
+ θ4 (REG_FLi,t−1 − FLi,t−1 ).FLi,t−1




+ θ5 SHOCKSit + θ6 SHOCKSit .FLi,t−1




+ θ7 IDEOLOGYit + θ8 IDEOLOGYit .FLi,t−1




+ θ9 STRUCTUREit + θ10 STRUCTUREit .FLi,t−1
+ εit

(5.5)

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 130 — #6

Financial Reforms for Financial Development 131

5.2.2

Econometric methods

AM use an ordered logit technique to estimate the benchmark specification and three alternative specifications with the results presented in
Tables 7, 8 and 9 of their paper. A minor problem has been detected in
their empirical results in which Singapore is misclassified as an African
country while South Africa is misclassified as an East Asian country. The
corrected results are presented in Appendix Table A5.4. In general, the
pattern of Appendix Table A5.4 (part A) is similar to that of their Table 7.
Appendix Table A5.4 (part B) presents stronger evidence for IMFit and
REG_FLi,t−1 −FLi,t−1 .90 It is worth noting that Appendix Table A5.4 (part
C) shows that FLi,t−1 , OPENit and OPENit × FLi,t−1 appear to be insignificant when country fixed effects are included, different from Table 9 of
AM, which shows these variables to be significant when country fixed
effects are included.
More importantly, the analyses conducted by AM may be questioned
in the following two aspects.
The first is that the ordered logit technique they apply may not be
natural for this context, although the discrete and ordinal nature of the
financial liberalization level, FLi,t , and policy change,
FLi,t , may render
the ordered logit method an appropriate choice at first glance. Since
the dependent variable is not the level of financial liberalization, but policy
change, financial liberalization moving from a score of 1 to 3 in terms of
their original measure91 is treated the same as moving from 16 to 18, for
example. However, given the ordinal feature of their original measure,
in reality policy change reflected by moving from a score of 1 to 3, which
could be at rather low levels, doesn’t necessarily lead to the same extent of
financial liberalization as moving from 16 to 18, which could be at much
higher levels of financial liberalization. Given this particular nature of
the dependent variable, resorting to the ordered logit technique may not
lead to the expected gains.
Second, like in most cross-country research, AM do not take into
account the effects of common trends and the possibility of error dependence across countries and over time. This seems especially relevant
when the effects of domestic learning and regional diffusion are investigated. The assumption on the error term they use implies that the
disturbances are uncorrelated between groups and over time. However,
if the error term contains one or more unobserved factors which have
different effects on every unit, as noted by Phillips and Sul (2003) among
others, “the consequences of ignoring cross section dependence can be
serious”. On the other hand, the consequences of ignoring serial correlation and heteroscedasticity can also be serious, since this may lead to

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 131 — #7

132 Determinants of Financial Development

a downwards bias in standard errors, and therefore higher significance
levels attached to the coefficients. In examining the origins of financial
openness, Quinn and Inclán (1997) argue that it is critical to consider
a common trend, such as changes in consumer tastes and technology,
that may exert substantial effects on government liberalization policies
as “fundamental but unobservable forces”.
The particular nature of the dependent variable and the possibility
of error dependence suggest that another estimation approach would
be worthwhile. The wide range of scores on the original financial liberalization index from 1 to 18 and the policy change,
FLi,t , from -1
to 1 (after transformation) makes a simpler linear regression method a
possible choice for this context. This chapter’s approach centres on the
Pesaran (2006) common correlated effect pooled (CCEP) estimator, a generalization of the fixed effects estimator which allows for the possibility
of cross-section correlation.92 To adjust for serial correlation in individual errors,93 the panel-robust standard errors from Arellano (1987) are
computed for the CCEP estimates, allowing the errors not only to be
serially correlated for a given country, but also to have variances and
covariances that vary across countries.
Pesaran (2006) proposes two common correlated effect (CCE)
approaches for large heterogeneous panels whose error contains unobserved common factors. More specifically, this approach augments the
one-way fixed effects model with the (weighted) cross-sectional means of
the dependent variable and the individual specific regressors, analogous
to a two-way fixed effects model. Including the (weighted) cross-sectional
averages of the dependent variable and individual specific regressors is
suggested by Pesaran (2006, 2007) as an effective way to filter out the
impacts of common factors, which could be common technological or
macroeconomic shocks, causing between group error dependence.
The Pesaran (2006, 2007) approach exhibits considerable advantages.
It allows unobserved common factors to be possibly correlated with
exogenous regressors and exert differential impacts on individual units.
It permits unit root processes amongst the observed and unobserved
common effects. The proposed estimator is still consistent, although it is
no longer efficient, when the idiosyncratic components are not serially
uncorrelated.
In this context, the cross sectional means of
FLit , FLi,t−1 , GDPi,t−1
and OPENit are considered since these variables may be especially likely
to reflect common effects. To allow the effects to be heterogeneous
across regions, the models are augmented with the interactions between
regional dummies and cross sectional means of the above variables, and

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 132 — #8

Financial Reforms for Financial Development 133

time dummies. The CCEP estimator has been shown to be asymptotically unbiased and consistent as N− > ∞ for both T fixed or T − > ∞,
and to have generally satisfactory finite sample properties.
Appendix Table A5.3 presents the time series properties for three continuous variables, the financial liberalization index (FLi,t ), GDP per
capita in PPP terms (GDPi,t ) and trade openness (OPENi,t ). It contrasts a
panel unit root test proposed by Pesaran (2007) in the presence of cross
section dependence with the Maddala and Wu (1999) Fisher test, which
is associated with the assumption of cross section independence of the
error term and does not require a balanced panel. The Pesaran (2007)
approach augments the standard ADF regression with cross section averages of lagged levels and first differences of individual series, to control
for cross section dependence. The Maddala and Wu (1999) Fisher test is
then applied to this more general setting. With cross sectionally independent errors, the Maddala and Wu (1999) Fisher test cannot reject the
null of non-stationarity for FLi,t , GDPi,t and OPENi,t when we do not
allow for a trend. With a trend, the series of GDPi,t and OPENi,t are close
to being found as stationary. When we allow for a trend, Pesaran’s test
shows that we can almost reject the null of non-stationarity for FLi,t ,
GDPi,t and OPENi,t at the 10% significance level94 , suggesting that FLi,t ,
GDPi,t and OPENi,t may not be I(1) variables. However, this result should
be interpreted with caution since there are reservations as to the power
and reliability of these tests.
This analysis also employs a normal within groups (WG) approach
to estimating the one-way fixed effects models (country fixed effects
included), as estimated by AM, with non-robust standard errors. How
important controlling for error dependence across countries and over
time is for this context can be examined by comparing the WG estimates
and CCEP estimates. The consistency of the one-way WG estimator for
the dynamic homogeneous model is justified by the length of the time
series,95 but this estimator is biased in small samples because of the
lagged dependent variable bias. The country fixed effects can be eliminated by an idempotent (covariance) transformation matrix as in within
groups estimation.

5.3

Empirical evidence

By applying a within groups approach to the AM framework with the
addition of the extent of democracy, this section presents empirical evidence in two steps on what shapes financial reform, an analysis on the
original dataset with the AM measure in Section 5.3.1 and an analysis

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 133 — #9

134 Determinants of Financial Development

on a larger dataset with the Chinn-Ito (2006) measure in Section 5.3.2.
In each step, the normal one-way fixed effects WG estimates with nonrobust standard errors are contrasted with Pesaran (2006) CCEP estimates
with panel-robust standard errors, with the former assuming that the
errors are serially uncorrelated and independent across countries, while
the latter approach allows for error dependence both across countries
and over time.
5.3.1

Analysis on the original dataset

This section concerns the analyses on the benchmark specification
(Equation 5.2) and three alternative specifications (Equations 5.3, 5.4
and 5.5) using AM’s original dataset. The results are presented in Tables
5.1A/B, 5.2 and 5.3 corresponding to Tables 7, 8 and 9 in AM, respectively.
Table 5.1 (part A) and 5.1 (part B) reports the WG estimates and
CCEP estimates of the benchmark specification (Equation 5.2). Table
5.1A strictly follows the model structure of AM96 while Table 5.1 (part
B) reports FLi,t−1 and FL2i,t−1 separately, presenting a direct link between
policy change,
FLit , and the level of liberalization, FLi,t−1 . In comparison to the ordered logit estimates in columns 4–6 (with country fixed
effects) of Appendix Table A5.4A, the WG estimates in Table 5.1A (country effects are included by definition) not only confirm their findings, but
also show that FIRSTYEARit and OPENit have positive effects on policy
change.
To present a direct link between policy change,
FLit , and the level
of liberalization, FLi,t−1 , Table 5.1 (part B) reports FLi,t−1 and FL2i,t−1
separately. The within R2 associated with the CCEP estimates is much
larger then those for the WG estimates, hinting at the importance of
error dependence. With satisfactory finite sample properties, the CCEP
estimates in Table 5.1 show that policy change is negatively rather
than positively associated with the lagged level of financial liberalization, FLi,t−1 , and the regional liberalization gap, REG_FLi,t−1 − FLi,t−1 .
The CCEP estimates confirm the AM finding on a negative effect of
BANKit , and positive effects of BOPit and FIRSTYEARit on policy change.
It also provides strong evidence for a negative effect of POLITY2it ,
indicating that the extent of democracy tends to hinder the pace of
reform.
Table 5.2 presents the within groups estimates, WG and CCEP, for
the alternative specifications (Equations 5 and 6 in AM). The CCEP estimates confirm the previous observations of Table 5.1 in terms of the
negative effects of the level of liberalization, regional liberalization gap,

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 134 — #10

Financial Reforms for Financial Development 135

banking crises and the extent of democracy, and the positive effects of a
balance-of-payments crisis and a new government’s first year in office. A
positive effect of USINTit is also observed.
Next we proceed to Table 5.3, which presents the within groups estimates of the most general specification (Equation 8 in AM). Note that the
corrected Table 9 in AM shows that FLi,t−1 , OPENit and OPENit × FLi,t−1
are insignificant in the presence of country fixed effects. Similarly, the

Table 5.1 Within estimates: Benchmark specification (Equation 4)
A. FLi,t−1 × (1 − FLi,t−1 ) reported
Estimators

WG

FLi,t−1
×(1 − FLi,t−1 )
REG_FLi,t−1
−FLi,t−1
BOPit

0.083
0.098
0.083
0.046
[0.038]** [0.038]*** [0.039]** [0.060]
0.076
0.070
0.083
0.109
[0.016]*** [0.016]*** [0.017]*** [0.025]***
0.017
0.013
[0.006]*** [0.006]**
−0.024
−0.022
[0.007]*** [0.007]***
−0.010
−0.009
[0.008]
[0.008]
−0.003
−0.002
[0.011]
[0.011]
0.011
[0.006]*
0.011
[0.007]*
−0.003
0.001
−0.001
[0.010]
0.000
[0.009]
0.000
[0.000]*
−0.013
[0.014]

BANKit
RECESSIONit
HINFLit
FIRSTYEARit
IMFit
USINTit
LEFTit
RIGHTit
OPENit
POLITY2it
Observations
Number of
countries
R-squared
CSD test
(p-value)

WG

WG

CCEP

CCEP

CCEP

0.070
[0.054]
0.111
[0.025]***
0.019
[0.006]***
−0.021
[0.010]**
−0.006
[0.008]
−0.009
[0.019]

0.075
[0.056]
0.121
[0.027]***
0.019
[0.006]***
−0.020
[0.009]**
−0.007
[0.008]
−0.012
[0.021]
0.011
[0.006]*
0.008
[0.008]
[0.001]***
[0.003]
0.006
[0.009]
0.005
[0.009]
0.000
[0.000]
−0.034
[0.020]*

805
35

805
35

805
35

805
35

805
35

805
35

0.03
0.00

0.05
0.00

0.07
0.00

0.13
0.03

0.15
0.01

0.17
0.01

continued

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 135 — #11

136 Determinants of Financial Development
Table 5.1 Continued
B. FLi,t−1 and (FLi,t−1 )2 reported separately
Estimators
FLi,t−1
(FLi,t−1 )2
REG_FLi,t−1
−FLi,t−1
BOPit

WG
0.081
[0.038]**
−0.104
[0.043]**
0.059
[0.022]***

BANKit
RECESSIONit
HINFLit

WG
0.096
[0.038]**
−0.113
[0.043]***
0.058
[0.022]***
0.016
[0.006]***
−0.024
[0.007]***
−0.010
[0.008]
−0.003
[0.011]

FIRSTYEARit
IMFit
USINTit
LEFTit
RIGHTit
OPENit
POLITY2it
Observations
Number of
countries
R-squared
CSD test
(p-value)

WG
0.074
[0.040]*
−0.113
[0.043]***
0.058
[0.023]**
0.011
[0.006]*
−0.020
[0.007]***
−0.009
[0.008]
−0.002
[0.011]
0.011
[0.006]*
0.012
[0.007]*
−0.003
[0.001]***
0.002
[0.010]
0.003
[0.009]
0.000
[0.000]*
−0.011
[0.014]

CCEP
−0.208
[0.058]***
−0.154
[0.066]**
−0.144
[0.042]***

CCEP
−0.178
[0.061]***
−0.175
[0.065]**
−0.133
[0.047]***
0.014
[0.006]**
−0.019
[0.010]*
−0.002
[0.007]
−0.014
[0.017]

CCEP
−0.202
[0.071]***
−0.174
[0.064]***
−0.148
[0.053]***
0.014
[0.005]**
−0.018
[0.009]*
−0.004
[0.008]
−0.012
[0.018]
0.011
[0.006]*
0.008
[0.008]
0.003
[0.004]
0.010
[0.009]
0.008
[0.008]
0.000
[0.000]
−0.038
[0.022]*

805
35

805
35

805
35

805
35

805
35

805
35

0.03
0.00

0.05
0.00

0.08
0.00

0.20
0.03

0.22
0.02

0.24
0.01

Notes: 35 countries (original dataset), 1973–96. Dependent variable is
FLit . Using normal
one-way within it groups estimator (WG) and Pesaran (2006)’s CCEP estimator, Table 5.1 A/B
presents new results corresponding to models in Table 7 in Abiad and Mody (2005) with the
addition of POLITY2it . Table 5.1A reports results for FLi,t−1 × (1 − FLi,t−1 ), while Table 5.1B
reports results for FLi,t−1 and FLi,t−1 2 separately. The within R-squared is reported. Nonrobust standard errors are reported for WG estimates, while panelrobust standard errors are
reported for CCEP estimates. CSD tests the null hypothesis of cross section independence in
the panel data models using the test following Pesaran (2004).
∗ significant at 10%; ∗∗ significant at 5%; ∗∗∗ significant at 1%.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 136 — #12

Financial Reforms for Financial Development 137
Table 5.2 Within estimates: Alternative specification (Equations 5 and 6)
Estimators
FLi,t−1
(FLi,t−1 )2

WG
0.074
[0.040]*
−0.113
[0.043]***

FLi,t−1 × Yi,t−1
REG_FLi,t−1 − FLi,t−1
BOPit
BANKit
RECESSIONit
HINFLit
FIRSTYEARit
IMFit
USINTit
LEFTit
RIGHTit
OPENit
POLITY2it
Observations
Number of countries
R-squared
CSD test (p-value)

0.058
[0.023]**
0.011
[0.006]*
−0.020
[0.007]***
−0.009
[0.008]
−0.002
[0.011]
0.011
[0.006]*
0.012
[0.007]*
−0.003
[0.001]***
0.002
[0.010]
0.003
[0.009]
0.000
[0.000]*
−0.011
[0.014]
805
35
0.08
0.00

WG
0.092
[0.040]**
−0.201
[0.053]***
0.007
[0.002]***
0.063
[0.023]***
0.011
[0.006]*
−0.023
[0.007]***
−0.010
[0.008]
−0.004
[0.011]
0.011
[0.006]*
0.012
[0.007]*
−0.003
[0.001]***
0.001
[0.010]
0.003
[0.009]
0.000
[0.000]**
−0.010
[0.014]
805
35
0.09
0.00

CCEP
−0.202
[0.071]***
−0.174
[0.064]***

−0.148
[0.053]***
0.014
[0.005]**
−0.018
[0.009]*
−0.004
[0.008]
−0.012
[0.018]
0.011
[0.006]*
0.008
[0.008]
0.003
[0.004]
0.010
[0.009]
0.008
[0.008]
0.000
[0.000]
−0.038
[0.022]*
805
35
0.24
0.01

CCEP
−0.175
[0.078]**
−0.105
[0.066]
−0.009
[0.004]**
−0.094
[0.079]
0.016
[0.005]***
−0.016
[0.009]*
−0.004
[0.008]
−0.015
[0.018]
0.011
[0.006]*
0.009
[0.008]
0.006
[0.003]**
0.011
[0.010]
0.006
[0.009]
0.000
[0.000]
−0.039
[0.018]**
805
35
0.25
0.01

Notes: This table, based on the original dataset, presents new results corresponding to models
in Table 8 in AM except for the addition of POLITY2it . See Table 5.1 for further notes.

CCEP estimates of Table 5.3 find less evidence for FLi,t−1 , OPENit and
OPENit × FLi,t−1 . It confirms the negative effect of REG_FLi,t−1 − FLi,t−1
on policy reform.97 A positive effect of FIRSTYEARit and a negative effect
of its interaction term with FLi,t−1 are observed, highlighting that new

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 137 — #13

138 Determinants of Financial Development
Table 5.3 Within estimates: Alternative specification (Equation 8)
Estimators
FLi,t−1
(FLi,t−1 )2
REG_FLi,t−1 − FLi,t−1
(REG − FLi,t−1 − FLi,t−1 ) × FLi,t−1
BOPit
BOPit × FLi,t−1
BANKit
BANKit × FLi,t−1
RECESSIONit
RECESSIONit × FLi,t−1
HINFLit
HINFLit × FLi,t−1
FIRSTYEARit
FIRSTYEARit × FLi,t−1
IMFit
IMFit × FLi,t−1
USINTit
LEFTit
LEFTit × FLi,t−1
RIGHTit
RIGHTit × FLi,t−1
OPENit

WG
−0.009
[0.072]
−0.011
[0.073]
0.025
[0.023]
0.330
[0.086]***
0.020
[0.010]**
−0.029
[0.019]
−0.023
[0.013]*
0.004
[0.027]
−0.015
[0.012]
0.020
[0.023]
0.030
[0.015]*
−0.156
[0.043]***
0.028
[0.010]***
−0.049
[0.020]**
0.020
[0.009]**
−0.050
[0.026]*
−0.003
[0.001]***
−0.025
[0.014]*
0.068
[0.034]**
0.006
[0.012]
0.020
[0.032]
0.001
[0.000]***

CCEP
−0.175
[0.121]
−0.143
[0.076]*
−0.147
[0.055]**
0.094
[0.098]
0.014
[0.010]
−0.009
[0.022]
−0.023
[0.016]
0.011
[0.026]
−0.006
[0.014]
0.008
[0.024]
0.014
[0.026]
−0.105
[0.073]
0.027
[0.012]**
−0.046
[0.027]*
0.011
[0.008]
−0.024
[0.018]
−0.001
[0.005]
−0.019
[0.014]
0.076
[0.039]*
0.008
[0.012]
0.025
[0.039]
0.001
[0.001]
continued

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 138 — #14

Financial Reforms for Financial Development 139
Table 5.3 Continued
Estimators
OPENit × FLi,t−1
POLITY2it
POLITY2it × FLi,t−1
Observations
Number of countries
R-squared
CSD test (p-value)

WG
−0.001
[0.000]***
−0.030
[0.018]*
0.002
[0.002]
805
35
0.14
0.00

CCEP
−0.001
[0.001]
−0.043
[0.031]
0.001
[0.003]
805
35
0.27
0.01

Notes: This table, based on the original dataset, presents new results corresponding
to models in Table 9 in AM except for the addition of POLITY2it . See Table 5.1 for
further notes.

governments in their first year are likely to trigger reform, especially
when the extent of financial liberalization is still at an early stage. The
effect of the interaction between LEFTit and FLi,t−1 is also shown to be
positive.
The discrepancy between the WG estimates and CCEP estimates in
the above study has pointed to the fundamental significance of relaxing
assumptions on the error term. One may wonder which is more important, controlling for serial correlation in the errors or adjusting for cross
section error dependence? To what extent does each relaxation make the
results different from those associated with error independence? Answers
may be found from Table 5.4, which reports the WG estimates with
panel-robust standard errors, controlling for serial correlation of errors
only, and the CCEP estimates with non-robust standard errors, controlling for cross section error dependence only. As it stands, both are
important. Nevertheless, the quantitatively larger effects (coefficients)
and much larger R2 associated with the CCEP estimates than with the
WG estimates may reflect that controlling for cross-country correlation
is an especially crucial step for this context. One may notice from Table
5.4 that, suggested by either the WG estimates or CCEP estimates, the
ideology and economic and political structure in general appear to have a
substantial influence on policy change, especially for LEFTit and OPENit .
This has raised a methodological concern that insufficient consideration
of error dependence could lead to misleading findings.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 139 — #15

140 Determinants of Financial Development
Table 5.4 Error dependence across countries and over time considered
separately
A. Within estimates corresponding to Table 5.1B
Estimators
FLi,t−1
(FLi,t−1 )
REG_FLi,t−1
−FLi,t−1
BOPit
BANKit
RECESSIONit
HINFLit

WG

WG

0.081
0.096
[0.049]
[0.045]**
−0.104
−0.113
[0.046]** [0.045]**
0.059
0.058
[0.025]** [0.027]**
0.016
[0.006]**
−0.024
[0.009]**
−0.010
[0.010]
−0.003
[0.019]

FIRSTYEARit
IMFit
USINTit
LEFTit
RIGHTit
OPENit
POLITY2it
Observations
Number of
countries
R-squared
CSD test
(p-value)

WG

CCEP

0.074
−0.208
[0.053]
[0.056]***
−0.113
−0.154
[0.051]** [0.049]***
0.058
−0.144
[0.027]** [0.037]***
0.011
[0.006]*
−0.020
[0.009]**
−0.009
[0.009]
−0.002
[0.020]
0.011
[0.006]*
0.012
[0.009]
−0.003
[0.001]**
0.002
[0.008]
0.003
[0.008]
0.000
[0.000]*
−0.011
[0.013]

CCEP
−0.178
[0.057]***
−0.175
[0.050]***
−0.133
[0.037]***
0.014
[0.006]**
−0.019
[0.007]***
−0.002
[0.008]
−0.014
[0.010]

CCEP
−0.202
[0.059]***
−0.174
[0.050]***
−0.148
[0.040]***
0.014
[0.006]**
−0.018
[0.007]**
−0.004
[0.008]
−0.012
[0.011]
0.011
[0.006]*
0.008
[0.007]
0.003
[0.003]
0.010
[0.011]
0.008
[0.010]
0.000
[0.000]
−0.038
[0.015]***

805
35

805
35

805
35

805
35

805
35

805
35

0.03
0.00

0.05
0.00

0.08
0.00

0.20
0.03

0.22
0.02

0.24
0.01

Notes: Panelrobust standard errors are reported for WG estimates, whilst non-robust standard
errors are reported for CCEP estimates. See Table 5.1 for further notes.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 140 — #16

Financial Reforms for Financial Development 141
Table 5.4 Continued
B. Within estimates corresponding to Table 5.2
Estimators
FLi,t−1
(FLi,t−1 )2

WG
0.074
[0.053]
−0.113
[0.051]**

FLi,t−1 × Yi,t−1
REG_FLi,t−1 − FLi,t−1
BOPit
BANKit
RECESSIONit
HINFLit
FIRSTYEARit
IMFit
USINTit
LEFTit
RIGHTit
OPENit
POLITY2it
Observations
Number of countries
R-squared
CSD test (p-value)

0.058
[0.027]**
0.011
[0.006]*
−0.020
[0.009]**
−0.009
[0.009]
−0.002
[0.020]
0.011
[0.006]*
0.012
[0.009]
−0.003
[0.001]**
0.002
[0.008]
0.003
[0.008]
0.000
[0.000]*
−0.011
[0.013]
805
35
0.08
0.00

WG
0.092
[0.053]*
−0.201
[0.068]***
0.007
[0.003]**
0.063
[0.025]**
0.011
[0.006]*
−0.023
[0.009]**
−0.010
[0.009]
−0.004
[0.020]
0.011
[0.006]*
0.012
[0.009]
−0.003
[0.001]**
0.001
[0.007]
0.003
[0.009]
0.000
[0.000]**
−0.010
[0.013]
805
35
0.09
0.00

CCEP
−0.202
[0.059]***
−0.174
[0.050]***

−0.148
[0.040]***
0.014
[0.006]**
−0.018
[0.007]**
−0.004
[0.008]
−0.012
[0.011]
0.011
[0.006]*
0.008
[0.007]
0.003
[0.003]
0.010
[0.011]
0.008
[0.010]
0.000
[0.000]
−0.038
[0.015]***
805
35
0.24
0.01

CCEP
−0.175
[0.062]***
−0.105
[0.055]*
−0.009
[0.003]***
−0.094
[0.047]**
0.016
[0.006]**
−0.016
[0.007]**
−0.004
[0.008]
−0.015
[0.011]
0.011
[0.006]*
0.009
[0.007]
0.006
[0.003]*
0.011
[0.011]
0.006
[0.010]
0.000
[0.000]
−0.039
[0.015]***
805
35
0.25
0.01

Note: Panelrobust standard errors are reported for WG estimates, whilst non-robust standard
errors are reported for CCEP estimates. See Table 5.1 for further notes.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 141 — #17

142 Determinants of Financial Development
Table 5.4 Continued
C. Within estimates corresponding to Table 5.3
Estimators
FLi,t−1
(FLi,t−1 )2
REG_FLi,t−1 − FLi,t−1
(REG − FLi,t−1 − FLi,t−1 ) × FLi,t−1
BOPit
BOPit × FLi,t−1
BANKit
BANKit × FLi,t−1
RECESSIONit
RECESSIONit × FLi,t−1
HINFLit
HINFLit × FLi,t−1
FIRSTYEARit
FIRSTYEARit × FLi,t−1
IMFit
IMFit × FLi,t−1
USINTit
LEFTit
LEFTit × FLi,t−1
RIGHTit
RIGHTit × FLi,t−1

WG
−0.009
[0.061]
−0.011
[0.068]
0.025
[0.029]
0.330
[0.082]***
0.020
[0.010]*
−0.029
[0.021]
−0.023
[0.016]
0.004
[0.025]
−0.015
[0.015]
0.020
[0.022]
0.030
[0.027]
−0.156
[0.058]**
0.028
[0.010]***
−0.049
[0.024]**
0.020
[0.011]*
−0.050
[0.022]**
−0.003
[0.001]**
−0.025
[0.014]*
0.068
[0.037]*
0.006
[0.011]
0.020
[0.034]

CCEP
−0.175
[0.105]*
−0.143
[0.104]
−0.147
[0.040]***
0.094
[0.116]
0.014
[0.010]
−0.009
[0.019]
−0.023
[0.013]*
0.011
[0.026]
−0.006
[0.012]
0.008
[0.023]
0.014
[0.015]
−0.105
[0.043]**
0.027
[0.009]***
−0.046
[0.019]**
0.011
[0.009]
−0.024
[0.026]
−0.001
[0.004]
−0.019
[0.014]
0.076
[0.034]**
0.008
[0.013]
0.025
[0.032]
continued

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 142 — #18

Financial Reforms for Financial Development 143
Table 5.4 Continued
Estimators
OPENit
OPENit × FLi,t−1
POLITY2it
POLITY2it × FLi,t−1
Observations
Number of countries
R-squared
CSD test (p-value)

WG
0.001
[0.000]**
−0.001
[0.000]**
−0.030
[0.025]
0.002
[0.002]
805
35
0.14
0.00

CCEP
0.001
[0.000]**
−0.001
[0.000]**
−0.043
[0.018]**
0.001
[0.002]
805
35
0.27
0.01

Notes: Panelrobust standard errors are reported for WG estimates, whilst
non-robust standard errors are reported for CCEP estimates. See Table 5.1
for further notes.

In sum, the above analyses based on the augmented specifications
in which POLITY2it is included, allowing for the possibility of error
dependence across countries and over time, produce interesting findings. On the one hand, this chapter confirms the significant effects of
crises and shocks on policy reform identified by AM. More specifically, it
confirms negative effects of banking crises and high inflation, and does
agree with AM that a new government in its first year and an IMF programme have a strong effect when financial sectors are highly repressed
and a weaker effect thereafter. On the other hand, it differs from AM in
the following three aspects. First, it shows that the significant effects
of balance-of-payments crises and US interest rates found by AM are
fragile. The second aspect is that it yields opposite findings to AM on the
effects of domestic learning. It shows that the extent of policy reform
is negatively rather than positively affected by the existing liberalization level, while the regional liberalization gap does not appear relevant.
Third, it addresses the importance of the extent of democracy for the
process of financial reform and identifies a negative effect of the extent
of democracy on policy change.
5.3.2

Analysis on a larger dataset

This section makes an effort to explore if the findings are robust to a
larger set of countries. It makes use of the Chinn-Ito index of financial

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 143 — #19

144 Determinants of Financial Development

openness (2006) which is available for 108 countries over 1970–2000. But
the Chinn-Ito index measures only a country’s degree of capital account
openness, one aspect of six policy dimensions on which the creation of
the AM is based. Moreover, the country coverage in this analysis is confined to the data availability of crisis variables taken from Bordo et al.
(2000) which contains only 55 countries. Since most of the added countries are OECD countries (listed in the Appendix Table A5.2), the effects of
factors like balance-of-payment crises, banking crises, IMF programmes
and the extent of democracy are expected to be weaker.98 A variable
description is presented in Appendix Table A5.1.
Tables 5.5A, 5.5B and 5.5C report the within groups estimates corresponding to Tables 5.1B, 5.2 and 5.3, respectively. As expected, these
tables show weaker evidence for the effects of shocks, crises, ideology
and economic and political structures on policy reform, except for US
interest rates and high inflation. But, since the above analysis in general
obtains findings consistent with AM on the effects of crises and shocks,
more emphasis is placed on the robustness of the new findings regarding
the negative effects of domestic learning and regional diffusion.
With a larger sample size, both the WG and CCEP estimates in these
tables clearly indicate that policy reform is negatively linked to the level
of liberalization, FLi,t−1 , at the 1% significance level. The tables further confirm that the effect of REG_FLi,t−1 − FLi,t−1 on policy change is
ambiguous. Removing the variable IMFi,t doesn’t alter the pattern of the
results, as reported in Appendix Table A5.5 (A, B, C).
Hence, the findings summarized earlier on the negative effects of
domestic learning and irrelevance of regional diffusion are largely supported by a larger sample of countries based on the Chinn-Ito index of
capital account openness.

5.4

Discussions

The above findings have several implications. The negative link between
policy change and the liberalization level suggests a convergence in the
extent of financial liberalization in the sense that countries with highly
repressed financial sectors have more potential to embark on reform,
while countries with a highly liberalized financial sector have greater
status quo bias – the reform likelihood is “saturated” (AM). Vivid examples can easily be picked up from the financial liberalization process in
East Asia in recent decades. Since the 1970s, countries or areas with levels
of liberalization much lower than those of the main developed countries
(the US or UK for example) like the Republic of Korea, Singapore, Hong

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 144 — #20

Financial Reforms for Financial Development 145
Table 5.5 Augmented dataset with Chinn-Ito measure (2006)
A. Within estimates corresponding to Table 5.1B
Estimators
FLi,t−1
(FLi,t−1 )2
REG_FLi,t−1
−FLi,t−1
BOPit

WG
−0.168
[0.044]***
0.052
[0.037]
−0.016
[0.027]

BANKit
RECESSIONit
HINFLit

WG
−0.170
[0.044]***
0.053
[0.037]
−0.018
[0.027]
0.002
[0.007]
−0.010
[0.009]
−0.001
[0.007]
−0.018
[0.012]

FIRSTYEARit
IMFit
USINTit
LEFTit
RIGHTit
OPENit
POLITY2it
Observations
Number of
countries
R-squared

WG
−0.185
[0.048]***
0.070
[0.039]*
0.007
[0.030]
0.003
[0.007]
−0.012
[0.009]
0.004
[0.007]
−0.015
[0.013]
0.000
[0.007]
0.000
[0.009]
−0.005
[0.001]***
−0.002
[0.010]
0.000
[0.010]
0.000
[0.000]
−0.003
[0.018]

CCEP
−0.204
[0.069]***
0.087
[0.049]*
0.048
[0.036]

CCEP
−0.214
[0.068]***
0.092
[0.049]*
0.044
[0.037]
−0.005
[0.007]
−0.008
[0.010]
0.001
[0.008]
−0.009
[0.017]

CCEP
−0.301
[0.086]***
0.164
[0.058]***
0.063
[0.046]
−0.006
[0.008]
−0.010
[0.011]
0.002
[0.009]
−0.007
[0.018]
0.001
[0.005]
0.007
[0.007]
−0.002
[0.002]
−0.010
[0.010]
−0.003
[0.012]
0.000
[0.000]
0.004
[0.027]

1263
55

1262
55

1150
53

1263
55

1262
55

1150
53

0.04

0.04

0.07

0.22

0.22

0.26

Notes: 55 countries, 1973–97. Dependent variable is
FLi,t . Using normal one-way within
groups estimator (WG) and Pesaran (2006)’s CCEP estimator, this table, based on a larger
dataset associated with the Chinn-Ito measure (2006), presents new results corresponding to
Table 5.1B. The within groups R-squared is reported. Variable descriptions are presented in the
Appendix Table A5.1. Countries included are listed in the Appendix Table A5.2. Non-robust
standard errors are reported for WG estimates, while panelrobust standard errors are reported
for CCEP estimates.
∗ significant at 10%; ∗∗ significant at 5%; ∗∗∗ significant at 1%.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 145 — #21

146 Determinants of Financial Development
Table 5.5 Continued
B. Within estimates corresponding to Table 5.2
Estimators
FLi,t−1
(FLi,t−1 )2

WG
−0.185
[0.048]***
0.070
[0.039]*

FLi,t−1 × Yi,t−1
REG_FLi,t−1 − FLi,t−1
BOPit
BANKit
RECESSIONit
HINFLit
FIRSTYEARit
IMFit
USINTit
LEFTit
RIGHTit
OPENit
POLITY2it
Observations
Number of countries
R-squared

0.007
[0.030]
0.003
[0.007]
−0.012
[0.009]
0.004
[0.007]
−0.015
[0.013]
0.000
[0.007]
0.000
[0.009]
−0.005
[0.001]***
−0.002
[0.010]
0.000
[0.010]
0.000
[0.000]
−0.003
[0.018]
1150
53
0.07

WG
−0.180
[0.048]***
0.028
[0.046]
0.003
[0.002]*
0.013
[0.030]
0.002
[0.007]
−0.011
[0.009]
0.005
[0.007]
−0.018
[0.013]
0.000
[0.007]
0.000
[0.009]
−0.005
[0.001]***
−0.004
[0.010]
−0.002
[0.010]
0.000
[0.000]
−0.003
[0.018]
1150
53
0.07

CCEP
−0.301
[0.086]***
0.164
[0.058]***

0.063
[0.046]
−0.006
[0.008]
−0.010
[0.011]
0.002
[0.009]
−0.007
[0.018]
0.001
[0.005]
0.007
[0.007]
−0.002
[0.002]
−0.010
[0.010]
−0.003
[0.012]
0.000
[0.000]
0.004
[0.027]
1150
53
0.26

CCEP
−0.375
[0.122]***
0.138
[0.071]*
0.002
[0.004]
0.038
[0.058]
−0.012
[0.010]
−0.002
[0.013]
0.002
[0.010]
0.006
[0.017]
0.000
[0.006]
0.010
[0.007]
−0.002
[0.002]
−0.013
[0.011]
−0.008
[0.017]
0.000
[0.001]
0.007
[0.033]
1150
53
0.33

Note: See Table 5.5A for further notes.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 146 — #22

Financial Reforms for Financial Development 147
Table 5.5 Continued
C. Within estimates corresponding to Table 5.3
Estimators
FLi,t−1
(FLi,t−1 )2
REG_FLi,t−1 − FLi,t−1
(REG − FLi,t−1 − FLi,t−1 ) × FLi,t−1
BOPit
BOPit × FLi,t−1
BANKit
BANKit × FLi,t−1
RECESSIONit
RECESSIONit × FLi,t−1
HINFLit
HINFLit × FLi,t−1
FIRSTYEARit
FIRSTYEARit × FLi,t−1
IMFit
IMFit × FLi,t−1
USINTit
LEFTit
LEFTit × FLi,t−1
RIGHTit
RIGHTit × FLi,t−1

WG
−0.360
[0.096]***
0.255
[0.089]***
−0.006
[0.031]
0.274
[0.107]**
−0.010
[0.012]
0.030
[0.020]
−0.010
[0.014]
0.003
[0.025]
0.006
[0.011]
−0.008
[0.021]
0.041
[0.018]**
−0.254
[0.054]***
−0.008
[0.011]
0.019
[0.021]
−0.002
[0.011]
0.032
[0.039]
−0.005
[0.001]***
−0.019
[0.016]
0.028
[0.031]
0.004
[0.015]
−0.011
[0.031]

CCEP
−0.681
[0.255]**
0.448
[0.232]*
−0.009
[0.057]
0.436
[0.263]
−0.013
[0.017]
0.009
[0.028]
−0.002
[0.024]
−0.002
[0.036]
0.003
[0.012]
−0.006
[0.019]
0.046
[0.033]
−0.171
[0.147]
−0.009
[0.009]
0.019
[0.017]
0.018
[0.012]
−0.006
[0.050]
−0.003
[0.002]
−0.045
[0.028]
0.068
[0.051]
−0.015
[0.031]
0.022
[0.048]
continued

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 147 — #23

148 Determinants of Financial Development
Table 5.5 Continued
Estimators
OPENit
OPENit × FLi,t−1
POLITY2it
POLITY2it × FLi,t−1
Observations
Number of countries
R-squared

WG

CCEP

0.001
[0.000]*
0.000
[0.000]
−0.010
[0.020]
0.001
[0.002]

0.000
[0.001]
0.000
[0.000]
0.008
[0.041]
0.000
[0.007]

1150
53
0.10

1150
53
0.35

Note: See Table 5.5A for further notes.

Kong, Thailand and China have actively and progressively liberalized
their financial systems.
This research finds that the significant effect of a regional liberalization
gap on policy changes is hard to identify, although two opposite views
have been proposed in the literature. AM suggest that countries with a
level of liberalization far from that of the regional leader are found to be
more likely to undertake reform, perhaps due to competitive pressure.
The larger the gap in terms of liberalization levels within a region, the
fiercer the competition amongst these countries for international capital
and technologies. In contrast, Axelrod (1997) documents that the more
similar a country is to its neighbouring nations in terms of economic,
social and political developments, the more likely it is that it “adopts
one of the neighbour’s traits” while Simmons and Elkins (2004) predict
that “governments’ liberalization policies will be influenced by the policies of their most important foreign economic competitors”. This line of
research in general predicts that a greater gap from the regional leader
tends to be associated with less incentive to compete and less chance to
catch up with the regional leader in the short run, therefore a status quo
bias is maintained.
In accordance with AM, the pattern suggested by their Table 3 that the
coefficient on REG_FLi,t−1 − FLi,t−1 is positive and the coefficient on the
interaction term is negative although insignificant, seems to be in line
with the convergence story identified earlier in the sense that countries
with lower levels of liberalization relative to that of the regional leader

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 148 — #24

Financial Reforms for Financial Development 149

are more inclined to initiate reform, while the reform momentum fades
as the liberalization gap from the regional leader shrinks. It implies that
a greater gap from the regional leader tends to be associated with more
incentives to engage in reform.
The finding concerning the negative effect of the extent of democracy
on policy change is consistent with Fernandez and Rodrik (1991), who
argue that there is uncertainty with respect to the distribution of benefits and costs from reform. They contrast democratic societies in which
the majority would vote against the reform due to the presence of this
uncertainty, just for safety, with authoritarian societies like Taiwan and
the Republic of Korea (early 1960s), Chile (1970s) and Turkey (1980s),
where “reform was imposed by the authoritarian regimes and against the
wishes of business.” The status quo appears to be more easily dislodged
in autocratic societies than in democratic societies.
Chapter 4 shows that democratization is typically followed by financial development at least in the short run, which is in line with the
argument of Rodrik and Wacziarg (2005) in terms of a short-run boost
in economic growth and a decline in growth volatility after democratization. Together with the findings of Chapter 4, a clear picture seems
to appear to us: a short-run increase in financial development emerges
after democratization; however, once democracy has been established
and enhanced, the extent of democracy may exert negative effects on
the extent to which governments undertake financial reform.
This finding tends to suggest that ideology and political structure can
have a substantial influence on policy change, contrary to some extent to
the findings of AM, who claim that ideology and economic and political
structure have a limited influence on policy change.

5.5

Conclusion

This chapter studies the forces that lead governments to undertake
reforms to enhance financial development, based on AM. Given the particular nature of the dependent variable, it suggests replacing the ordered
logit technique used by AM with a within groups approach, allowing for
the possibility of error dependence across countries and over time, which
seems of especial importance when the effects of domestic learning and
regional diffusion in the process of financial liberalization are studied.
Based on these innovations, the analysis shows that some of the AM
findings are not robust to error dependence and the estimation method.
It has produced the following significant findings, shedding new light
on the political economy of financial reform.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 149 — #25

150 Determinants of Financial Development

This chapter finds that policy change in a country is negatively rather
than positively associated with the initial extent of liberalization level,
and the distance behind the regional leader. This indicates convergence
in the extent of financial liberalization, in the sense that countries with
highly repressed financial sectors have more potential to embark on
reform, whilst countries with a highly liberalized financial sector have
greater status quo bias.
This analysis suggests that some of AM findings on the effects of shocks
and crises are robust whilst others are fragile. More specifically, it confirms the negative effects of banking crises and high inflation. It also
agrees with AM that new governments in their first year and IMF programmes have a strong effect when financial sectors are highly repressed,
and a weaker effect thereafter. But it finds no evidence in support of
the effects of balance-of-payments crises and US interest rates on policy
change.
Furthermore, it shows that economic and political structure and ideology can have a substantial influence on policy change, and the extent
of democracy, the added variable, has a significantly negative effect on
policy reform.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 150 — #26

Financial Reforms for Financial Development 151

Appendix tables
Table A5.1 The variables (mainly used with the larger dataset)
Variable

Description

Source

FL

It is the financial liberalization index,
produced by rescaling the Chinn-Ito
index to interval [0, 1]. The Chinn-Ito
index, the KAOPEN index, measures a
country’s degree of capital account
openness, taking on higher values the
more open the country is to cross-border
capital transactions.

Chinn and Ito
(2006)

Y

GDP per capita in PPP terms.

Penn World
Table 6.2

BOP

As in Abiad and Mody (2005) (originally
taken from Bordo et al. (2000)), it is the
balance-of-payments crisis variable
identified by “a forced change in parity,
abandonment of a pegged exchange rate,
or an international rescue,” or if an index
of exchange market pressure (a weighted
average of exchange rate, reserve and
interest rate changes) exceeds a critical
threshold of one and a half standard
deviations above its mean. It is set equal
to 1 if a balance of payments crisis has
occurred within the past two years, and 0
otherwise.

Bordo et al. (2000)

BANK

As in Abiad and Mody (2005) (originally
taken from Bordo et al. (2000)), it is the
bankig crisis identified by periods of
“financial distress resulting in the erosion
of most or all of aggregate banking
system capital”. It is set equal to 1 if a
banking crisis has occurred within the
past two years, and 0 otherwise.

Bordo et al. (2000)

RECESSION

As in Abiad and Mody (2005), it is the
recession dummy variable, set equal to 1
where the annual real GDP growth rate is
negative, and 0 otherwise.

Penn World
Table 6.2 (PWT62)
(Heston et al.,
2006)
continued

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 151 — #27

152 Determinants of Financial Development
Table A5.1 Continued
Variable

Description

Source

HINFL

As in Abiad and Mody (2005), it is the
high inflation dummy variable, set equal
to 1 where the annual inflation exceeds
50%, and 0 otherwise.

World Bank World
Development
Indicators (WDI),
2008

FIRSTYEAR

Based on the YRSOFFC variable (how
many years the chief executive has been
in office), it is the first year in office
dummy as in Abiad and Mody (2005).

World Bank’s
Database of
Political
Institutions (2005)

IMF

As in Abiad and Mody (2005), it is the
IMF programme dummy variable
constructed using the programme dates
from the IMF “History of Lending
Arrangements”.

Abiad and Mody
(2005), and IMF’s
“History of
Lending”.

USINT

As in Abiad and Mody (2005), it is the US
Treasury Bill rate used as the world
interest rate.

IMF’s International
Financial Statistics
(2005)

LEFT

As in Abiad and Mody (2005), it denotes a
left-wing government where its
associated party is named or described as
“communist”, “socialist”, “Social
Democratic” or “left-wing”.

World Bank’s
Database of
Political
Institutions (2005)

RIGHT

As in Abiad and Mody (2005), it denotes
the right-wing government where its
associated party is named or described as
“conservative”, or “right-wing”.

World Bank’s
Database of
Political
Institutions (2005)

OPEN

The sum of exports and imports over
GDP (at current prices), averaged over
1973–97.

Penn World Table
6.2

DEMO

Index of democracy. It is called combined
the polity score, and is the democracy
score minus the autocracy score, averaged
over 1973–97. It is also used with the
original dataset. The index has been
converted to range from 0 to 1.

PolityIV Database
(Marshall and
Jaggers 2008)

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 152 — #28

Financial Reforms for Financial Development 153
Table A5.2 The list of countries in the augmented dataset
East Asia
CHN
HKG
IDN
KOR
MYS
PHL
SGP
THA
TWN

China
Hong Kong
Indonesia*
Korea, Rep.*
Malaysia*
Philippines*
Singapore*
Thailand*
Taiwan*

Latin America
& Caribbean
ARG
Argentina*
BRA
Brazil*
CHL
Chile*
COL
Colombia*
CRI
Costa Rica
ECU
Ecuador
JAM
Jamaica
MEX
Mexico*
PER
Peru*
PRY
Paraguay
URY
Uruguay
VEN
Venezuela*

South Asia
BGD
Bangladesh*
IND
India*
LKA
Sri Lanka*
NPL
Nepal*
PAK
Pakistan*

OECD countries
AUS
Australia*
AUT
Austria
BEL
Belgium
CAN
Canada*
CHE
Switzerland
DEU
Germany*
DNK
Denmark
ESP
Spain
FIN
Finland
FRA
France*
GBR
United Kingdom*
GRC
Greece
IRL
Ireland
ISL
Iceland
ITA
Italy*
JPN
Japan*
NLD
Netherlands
NOR
Norway
NZL
New Zealand*
PRT
Portugal
SWE
Sweden
TUR
Turkey*
USA
USA*

Middle East
& Africa
EGY
Egypt*
GHA
Ghana*
ISR
Israel*
MAR
Morocco*
NGA
Nigeria
ZAF
South Africa*
ZWE
Zimbabwe*

Note: Countries with ∗ are in the original dataset of Abiad and Mody (2005).

Table A5.3 Unit root test in heterogeneous panels
Variables

FL

Trend
Maddala and
Wu (1999)’s Fisher test
Pesaran (2007)’s cross
sectionally augmented
Fisher test

GDP

OPEN

Yes

No

Yes

No

Yes

No

43.82
[0.99]
74.85

25.39
[1.00]
50.23

77.84
[0.24]
67.65

52.81
[0.94]
54.98

75.23
[0.31]
63.01

64.11
[0.68]
62.31

Notes: Maddala and Wu (1999)’s Fisher test is for the case of cross sectionally independent
error. Under the null of a unit root, the test statistic is asymptotically distributed as a standard
normal. Pesaran (2007)’s test is the Maddala and Wu (1999)’s Fisher test applied to the cross
sectionally augmented Dickey-Fuller regression. The 10% critical values provided by H.M.
Pesaran for the pair of N = 30 and T = 30 is 82.89 with a trend and 82.18 without a trend.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 153 — #29

154 Determinants of Financial Development
Table A5.4 Corrected version of Tables 7, 8 and 9 in Abiad and Mody (2005)
A. Corrected version of Table 7 in Abiad and Mody (2005)
Country
dummy
included
FLi,t−1
×(1 −
FLi,t−1 )
REG_FLi,t−1
−FLi,t−1
BOPit

No
3.933
[4.39]***
1.032
[4.18]***

BANKit
RECESSIONit
HINFLit
FIRSTYEARit
IMFit
USINTit
LEFTit
RIGHTit
OPENit
Observations
Number of
countries

805
35

No
4.562
[4.94]***

No

Yes

4.106
[4.48]***

6.794
[4.44]***

1.050
1.195
[3.76]*** [3.93]***
0.521
0.430
[2.60]*** [2.21]**
−1.020
−0.983
[2.74]*** [2.67]***
−0.018
0.002
[0.05]
[0.00]
−0.136
−0.238
[0.35]
[0.62]
0.178
[0.78]
0.327
[1.81]*
−0.071
[1.82]*
0.282
[1.14]
0.153
[0.85]
−0.001
[1.01]

2.285
[3.23]***

805
35

805
35

805
35

Yes
7.284
[4.83]***

Yes
6.574
[4.07]***

2.089
2.529
[2.71]*** [3.21]***
0.550
0.475
[2.19]**
[1.94]*
−0.995
−0.935
[2.68]*** [2.57]**
−0.055
−0.026
[0.15]
[0.07]
−0.317
−0.302
[0.50]
[0.48]
0.234
[0.87]
0.253
[0.98]
−0.090
[2.13]**
−0.035
[0.10]
−0.132
[0.39]
0.009
[1.14]
805
35

805
35

Notes: This is a corrected version of Table 7 in Abiad and Mody (2005), which treated Singapore
as an African country and South Africa as an East Asian country. Except for the difference in
magnitude, this table shows a similar pattern to Table 7 in Abiad and Mody (2005). Robust
t-statistics in brackets.
∗ significant at 10%; ∗∗ significant at 5%; ∗∗∗ significant at 1%.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 154 — #30

Financial Reforms for Financial Development 155
Table A5.4 Continued
B. Corrected version of Table 8 in Abiad and Mody (2005)
Country dummy
included
FLi,t−1

No
4.110
[4.49]***
−4.052
[3.94]***

(FLi,t−1 )2
FLi,t−1 × Yi,t−1
REG_FLi,t−1 − FLi,t−1
BOPit
BANKit
RECESSIONit
HINFLit
FIRSTYEARit
IMFit
USINTit
LEFTit
RIGHTit
OPENit
Observations
Number of countries

1.231
[2.72]***
0.429
[2.19]**
−0.985
[2.70]***
−0.002
[0.00]
−0.235
[0.63]
0.178
[0.78]
0.332
[1.74]*
−0.070
[1.80]*
0.280
[1.15]
0.146
[0.77]
−0.001
[1.00]
805
35

No
4.307
[4.69]***
−5.720
[4.19]***
0.095
[2.34]**
0.965
[1.88]*
0.476
[2.40]**
−0.976
[2.70]***
−0.005
[0.01]
−0.206
[0.53]
0.141
[0.62]
0.414
[2.12]**
−0.074
[1.87]*
0.190
[0.82]
0.153
[0.84]
0.000
[0.04]
805
35

Yes
6.546
[4.02]***
−6.638
[3.35]***

2.465
[2.09]**
0.473
[2.02]**
−0.932
[2.70]***
−0.027
[0.07]
−0.303
[0.48]
0.233
[0.86]
0.255
[0.96]
−0.090
[2.07]**
−0.029
[0.08]
−0.125
[0.38]
0.009
[1.14]
805
35

Yes
7.189
[4.34]***
−9.893
[3.90]***
0.247
[2.55]**
2.714
[2.45]**
0.457
[1.95]*
−1.007
[2.92]***
0.001
[0.00]
−0.398
[0.64]
0.245
[0.91]
0.288
[1.06]
−0.086
[1.99]**
−0.098
[0.28]
−0.072
[0.21]
0.013
[1.40]
805
35

Notes: This table corresponds to the Table 8 in Abiad and Mody (2005), which treated Singapore as an African country and South Africa as an East Asian country, and consequently
indicates that IMF in column 1 and REG_FL-FL in columns 2 and 3 are insignificant. Robust
t-statistics in brackets.
∗ significant at 10%; ∗∗ significant at 5%; ∗∗∗ significant at 1%.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 155 — #31

156 Determinants of Financial Development
Table A5.4 Continued
C. Corrected version of Table 9 in Abiad and Mody (2005)
Country dummy included
FLi,t−1
(FLi,t−1 )2
REG_FLi,t−1 − FLi,t−1
(REG − FLi,t−1 − FLi,t−1 ) × FLi,t−1
BOPit
BOPit × FLi,t−1
BANKit
BANKit × FLi,t−1
RECESSIONit
RECESSIONit × FLi,t−1
HINFLit
HINFLit × FLi,t−1
FIRSTYEARit
FIRSTYEARit × FLi,t−1
IMFit
IMFit × FLi,t−1
USINTit
LEFTit
LEFTit × FLi,t−1
RIGHTit
RIGHT × FLi,t−1

No
3.719
[2.16]**
−3.827
[2.19]**
0.508
[0.81]
2.87
[1.51]
0.811
[2.69]***
−0.892
[1.47]
−0.883
[1.65]*
−0.093
[0.09]
−0.487
[1.12]
1.235
[1.43]
0.292
[0.64]
−2.203
[1.65]*
0.566
[1.98]**
−1.163
[1.84]*
0.775
[2.94]***
−1.523
[2.26]**
−0.078
[1.93]*
−0.116
[0.29]
1.049
[1.01]
0.257
[0.87]
0.087
[0.09]

Yes
3.475
[1.61]
−1.82
[0.70]
1.459
[1.21]
10.256
[3.95]***
0.809
[1.89]*
−0.989
[1.11]
−1.043
[1.85]*
0.016
[0.01]
−0.503
[0.91]
1.164
[1.21]
0.37
[0.50]
−3.471
[2.35]**
0.592
[1.86]*
−1.055
[1.45]
0.65
[1.83]*
−1.741
[1.94]*
−0.091
[2.10]**
−0.616
[1.16]
1.282
[1.09]
0.192
[0.50]
−0.221
[0.19]
continued

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 156 — #32

Financial Reforms for Financial Development 157
Table A5.4 Continued
Country dummy included

No

OPENit
OPENit × FLi,t−1
Observations
Number of countries

Yes

3.719
[2.16]**
−3.827
[2.19]**

3.475
[1.61]
−1.82
[0.70]

805
35

805
35

Notes: This table corresponds to the Table 9 in Abiad and Mody (2005),
which treated Singapore as an African country and South Africa as an
East Asian country, and consequently indicates that (REG_FL−FL)×FL
is significant but OPEN and OPEN × FL are insignificant in column 1,
and FL, OPEN and OPEN × FL are significant in column 2. Robust
t -statistics in brackets.
∗ significant at 10%; ∗∗ significant at 5%; ∗∗∗ significant at 1%.

Table A5.5 Augmented dataset with Chinn-Ito measure (2006): IMF dropped
A. Within estimates corresponding to Table 5.1B
Estimators
FLi,t−1
(FLi,t−1 )2
REG_FLi,t−1
−FLi,t−1
BOPit
BANKit
RECESSIONit
HINFLit
FIRSTYEARit
USINTit
LEFTit
RIGHTit

WG
−0.168
[0.044]***
0.052
[0.037]
−0.016
[0.027]
0.002
[0.007]
−0.010
[0.009]
−0.001
[0.007]
−0.018
[0.012]
0.000
−0.005
[0.001]***
−0.004
[0.010]
0.000
[0.010]

WG
−0.170
[0.044]***
0.053
[0.037]
−0.018
[0.027]
0.001
[0.007]
−0.010
[0.009]
0.000
[0.007]
−0.017
[0.013]
0.001
−0.002
[0.002]
−0.008
[0.009]
0.000
[0.011]

WG
−0.174
[0.045]***
0.056
[0.038]
0.002
[0.028]
−0.005
[0.007]
−0.008
[0.010]
0.001
[0.008]
−0.009
[0.017]
[0.007]

CCEP
−0.204
[0.069]***
0.087
[0.049]*
0.048
[0.036]
−0.006
[0.008]
−0.009
[0.011]
0.001
[0.009]
−0.009
[0.017]
[0.006]

CCEP
−0.214
[0.068]***
0.092
[0.049]*
0.044
[0.037]

CCEP
−0.261
[0.084]***
0.119
[0.059]**
0.044
[0.036]

continued

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 157 — #33

158 Determinants of Financial Development
Table A5.5 Continued
Estimators
OPENit
POLITY2it
Observations
Number of
countries
R-squared

WG

WG

0.000
[0.000]*
−0.002
[0.016]

0.000
[0.000]
0.012
[0.022]

1263
55
0.04

1262
55
0.04

WG

CCEP

CCEP

CCEP

1213
53

1263
55

1262
55

1213
53

0.07

0.22

0.22

0.25

Note: See Table 5.5A for notes.

B. Within estimates corresponding to Table 5.2
Estimators
FLi,t−1
(FLi,t−1 )2

WG
−0.174
[0.045]***
0.056
[0.038]

FLi,t−1 × Yi,t−1
REG_FLi,t−1 − FLi,t−1
BOPit
BANKit
RECESSIONit
HINFLit
FIRSTYEARit
USINTit
LEFTit
RIGHTit
OPENit

0.002
[0.028]
0.001
[0.007]
−0.010
[0.009]
0.000
[0.007]
−0.017
[0.013]
0.000
[0.007]
−0.005
[0.001]***
−0.004
[0.010]
0.000
[0.010]
0.000
[0.000]*

WG
−0.169
[0.045]***
0.006
[0.044]
0.004
[0.002]**
0.007
[0.028]
0.001
[0.007]
−0.010
[0.009]
0.001
[0.007]
−0.020
[0.013]
0.000
[0.007]
−0.004
[0.001]***
−0.006
[0.010]
−0.001
[0.010]
0.000
[0.000]*

CCEP
−0.261
[0.084]***
0.119
[0.059]**
0.004
[0.004]
0.044
[0.036]
−0.006
[0.008]
−0.009
[0.011]
0.001
[0.009]
−0.009
[0.017]
0.001
[0.006]
−0.002
[0.002]
−0.008
[0.009]
0.000
[0.011]
0.000
[0.000]

CCEP
−0.343
[0.118]***
0.079
[0.081]

0.012
[0.048]
−0.011
[0.009]
0.000
[0.013]
0.002
[0.009]
0.000
[0.016]
0.000
[0.006]
−0.001
[0.002]
−0.012
[0.010]
−0.004
[0.015]
0.000
[0.000]
continued

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 158 — #34

Financial Reforms for Financial Development 159
Table A5.5 Continued
Estimators

WG

WG

CCEP

POLITY2it

−0.002
[0.016]

−0.002
[0.016]

0.012
[0.022]

0.019
[0.028]

1213
53
0.07

1213
53
0.07

1213
53
0.25

1213
53
0.31

Observations
Number of countries
R-squared

CCEP

Note: See Table 5.5A for notes.

Table A5.5 Continued
C. Within estimates corresponding to Table 5.3
Estimators
FLi,t−1
(FLi,t−1 )2
REG_FLi,t−1 − FLi,t−1
(REG − FLi,t−1 − FLi,t−1 ) × FLi,t−1
BOPit
BOPit × FLi,t−1
BANKit
BANKit × FLi,t−1
RECESSIONit
RECESSIONit × FLi,t−1
HINFLit
HINFLit × FLi,t−1
FIRSTYEARit
FIRSTYEARit × FLi,t−1

WG
−0.303
[0.089]***
0.190
[0.081]**
−0.024
[0.029]
0.216
[0.096]**
−0.010
[0.011]
0.027
[0.020]
−0.008
[0.014]
−0.003
[0.025]
0.006
[0.010]
−0.017
[0.020]
0.027
[0.017]
−0.201
[0.049]***
−0.005
[0.011]
0.010
[0.020]

CCEP
−0.599
[0.232]**
0.355
[0.208]*
−0.040
[0.053]
0.360
[0.224]
−0.010
[0.015]
0.000
[0.025]
0.002
[0.023]
−0.009
[0.035]
0.009
[0.011]
−0.023
[0.022]
0.022
[0.031]
−0.103
[0.143]
−0.005
[0.009]
0.010
[0.018]
continued

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 159 — #35

160 Determinants of Financial Development
Table A5.5 Continued
Estimators
USINTit
LEFTit × FLi,t−1
LEFTit × FLi,t−1
RIGHTit
RIGHTit × FLi,t−1
OPENit
OPENit × FLi,t−1
POLITY2it
POLITY2it × FLi,t−1
Observations
Number of countries
R-squared

WG
−0.005
[0.001]***
−0.017
[0.015]
0.022
[0.030]
0.009
[0.014]
−0.019
[0.030]
0.001
[0.000]**
0.000
[0.000]
−0.002
[0.019]
0.002
[0.002]
1213
53
0.09

CCEP
−0.002
[0.002]
−0.034
[0.025]
0.048
[0.049]
−0.004
[0.025]
0.006
[0.043]
0.000
[0.000]
0.000
[0.000]
0.020
[0.033]
0.002
[0.005]
1213
53
0.33

Note: See Table 5.5A for notes.

HUANG: “CHAP05” — 2010/9/29 — 20:06 — PAGE 160 — #36

6
Geographic Determinants of
Carbon Markets (CDM)

6.1

Introduction

Global warming has emerged as one of the most critical issues of our age,
and a key issue in the global economic and environmental debates. In
recent years, the global carbon market has become a newly developed
area for research and practice. It essentially consists of allowance-based
markets and project-based markets which use market-based mechanisms
to allocate and trade carbon credits that represent CO2 emission reductions in order for the participants to meet their compliance requirements
at the lowest possible cost. In allowance-based markets, the buyers
purchase emission allowances created and allocated (or auctioned) by
regulators under cap-and-trade regimes like Assigned Amount Units
(AAUs) under the Kyoto Protocol, or EU Allowances (EUAs) under the
EU Emissions Trading Scheme (EU ETS). Within project-based markets,
the buyers purchase emission credits from investing into a project that
can demonstrate a reduction of CO2 emissions in comparison to the
level of emissions in the absence of the project investment. The most
notable examples of such activities are the Clean Development Mechanism (CDM) and the Joint Implementation (JI) schemes under the Kyoto
Protocol.
As part of the emerging global carbon market, CDM is the only Kyoto
mechanism which involves developing countries in the climate change
negotiations. Under the Kyoto Protocol, the CDM is designed to realize
the benefits in terms of capital flow, technological transfer, sustainable
development and cost-effective emission abatement. However, the geographic distribution of CDM projects by host country and region has
been found to be highly uneven. This chapter addresses the issue of
whether the geographic endowments in the host countries matter for
161

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 161 — #1

162 Determinants of Financial Development

CDM development using recently developed spatial econometric techniques, with an aim of encouraging further research into economic,
institutional and policy determinants of CDM development.
In response to climate change, the global community adopted the
Kyoto Protocol in 1997. It came into force in February 2005 and calls
for legally binding limits on the greenhouse gas (GHG) emissions by
developed countries (or Annex I countries) by at least 5% in comparison to the 1990 levels over the first commitment period (i.e. 2008–12).
Although each Annex I country is assigned an amount of CO2 equivalents (expressed in Assigned Amount Units, AAUs) to be used over the
period 2008–12, some Annex I countries still face a projected shortfall
in GHG emission reductions. To meet their commitments, these countries usually seek emission reduction credits through the three “flexibility
mechanisms” defined under the Kyoto Protocol: International Emission
Trading (IET), Joint Implementation (JI) and the CDM.
The CDM is defined in Article 12 of the Kyoto Protocol, and is the
only such mechanism that involves developing countries. By joining in
the CDM, on the one hand, developing countries can get access to significant foreign capital flows and technology transfer to achieve more
sustainable, less GHG-intensive pathways of development. On the other
hand, the Annex I countries can purchase and utilize the emission reduction credits, called Certified Emission Reductions (CERs), generated from
CDM projects towards meeting their quantified emission targets under
the Protocol.
The geographic distribution of CDM projects by host country and
region has been observed as being lopsided, in terms of both the number
of projects and the volume of credits. More specifically, two regions, Asia
and the Pacific, and Latin America, together dominate the distribution
of CDM projects and CER flows, such that by the end of September 2008
China, India, Brazil and Mexico accounted for 45%, 23%, 5% and 1%
of CDM projects, respectively.99 Developing countries with large populations and economies are expected to account for a large number of
CDM projects and CER flows. However, do countries with particular
geographic characteristics like higher absolute latitudes, higher elevations and richer resource endowments have more CDM projects and CER
flows?
Economists have long noted the crucial role of geography in economic
development: transport costs, human health, agricultural productivity
and ownership of natural resources. The climate theory of underdevelopment has been widely recognized in the sense that certain geographic
endowments have an adverse impact on economic development. For

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 162 — #2

Geographic Determinants of Carbon Markets (CDM)

163

example, some geographic endowments (like mineral resource endowments) may influence the inputs into the production function, while
others (like tropical location) may make the production technologies
much harder to employ and affect technological development in the
very long term (Sachs, 2003; Sachs and Warner, 1995; Diamond, 1997;
Gallup et al. 1999).
While there is considerable research examining the sustainable development impacts of CDM development, much less work has aimed to
explore the fundamental determinants of CDM development across
countries. This chapter evaluates whether cross-sectional differences in
CDM development can be explained by cross-sectional differences in
geographic characteristics and resource endowments, once controlling
for other potential factors.
The cross-country experience of CDM project selection and foreign
direct investment indicates the existence of neighbourhood effects or
spillovers among countries.100 The neighbourhood effects of CDM
projects, together with “a new and deeper version of globalization” since
1970 (Crafts, 2000) which causes a closer interdependence across countries, suggest that spatial correlation is an important phenomenon to
be considered in this application. By employing the spatial econometric
method recently developed by Kelejian and Prucha (2010), this chapter
conducts a cross-country study on 48 developing countries over the
period from December 2003 up to September 2008.
This research has led to two significant findings. First, it provides evidence that positive spatial dependence among observations exists in this
context. More specifically, the CDM credit flows in a country increase by
about 0.34 to 0.48 units if those in its neighbouring countries increase
by one unit; and countries with larger CDM credit flows tend to be geographically clustered with other large CDM host countries. Second, by
allowing for spatial dependence and accounting for the size of the economy (initial population and initial GDP per capita), this research finds
that absolute latitude and elevation have positive impacts on CDM credit
flows, suggesting that countries further from the equator and having
higher elevations tend to initiate more CDM projects and issue more
CDM credit flows. Larger service exporting countries seem to have more
advantages in getting access to CDM projects, while on the contrary,
larger natural resource exporting countries have smaller CDM credit
flows, indicating that natural resource abundance may not necessarily
be attractive to CDM projects.
This finding sheds light on the geographic determinants of uneven
CDM project development across countries. It has rich implications

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 163 — #3

164 Determinants of Financial Development

for developing countries in terms of international cooperation and
national capacity building in order to access effectively the CDM for
their national sustainable development objectives. This research also suggests that the geographic considerations should be introduced into the
econometric and theoretical cross-country studies of climate change and
mitigation.
The remainder of the chapter proceeds as follows. Section 6.2 describes
the data and shows some stylized facts. The empirical results are presented in Section 6.4, following a description of econometric methods
in section 6.3. Section 6.5 concludes.

6.2

Data and stylized facts

This section outlines the measures and data for CDM, key geographic
variables and the control variables.
The dependent variable is the Clean Development Mechanism credit
flows, simply denoted by CDM. The indicator for CDM is the average of
the Certified Emission Reductions (2012 kCERs) generated by the CDM
projects in the pipeline over the period from December 2003 to September 2008.101 One country has one observation. To diminish the impacts
of outliers and measurement errors, it is taken in logs. The CDM projects
in the pipeline include not only those called “confirmed projects” which
have been at the registration stage, having either registered or requested
registration, but also those called “probable projects” which are at the
validation stage, waiting to be registered and implemented over the next
three years. One CER equals to one metric tonne of CO2 e.102 Data on
CER flows are from the UNEP Risoe Centre (2008).
To examine the impacts of particular geographic characteristics on
CDM project development, three geographic variables – absolute latitude, elevation and land area – are considered. Absolute latitude
(LATITUDE) equals the absolute distance from the equator of a country. The closer the countries are to the equator, the more tropical climate
they have. Elevation (ELEV ) is the mean elevation (metres above sea
level) calculated in geographic projection, and used in logs. The land
area (AREA) in square kilometres for each country is in logs. Data
on latitude, elevation and land area are taken from the physical factors dataset of Center for International Development (CID) at Harvard
University.103
To assess the role of natural resource endowments, this research uses
two groups of variables. One group of variables consists of dummies
for the manufactured goods exporting countries (EXPMANU ), service

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 164 — #4

Geographic Determinants of Carbon Markets (CDM)

165

exporting countries (EXPSERV ) and non-fuel primary goods exporting
countries (EXPPRIM) from the Global Development Network of the
World Bank (GDN). The other group of variables, taken from Isham
et al. (2005), includes dummies for the exporters of point source natural
resources (e.g. oil, diamonds, plantation crops) (RESPOINT ), “diffuse”
natural resources (e.g. wheat, rice, animals) (RESDIFF) and coffee/cocoa
natural resources (RESCOFF).
Control variables included in this analysis are the initial GDP per
capita (GDP03), the initial population (POP03), an ethnic fractionalization index (ETHNIC), a religious fractionalization index (RELIGION) and
legal origin dummies, COMLEG and CIVLEG.
The inclusion of the initial GDP per capita and population is to control for the size of the economy where GDP03 is the real GDP per capita
in 2003 in constant 2000 US$ (chain series), and POP03 is the population in 2003. Both GDP03 and POP03 are used in logs and taken from
the Penn World Table 6.2 in Heston et al. (2006). The variables ETHNIC
and RELIGION characterize social divisions and cultural differences. The
data on ETHNIC and RELIGION are taken from Alesina et al. (2003).104
COMLEG is the Common Law legal origin dummy for countries with
British legal origin, while CIVLEG is the Civil Law legal origin dummy
for countries with French, German or Scandinavian legal origins. Data
on CIVLEG and COMLEG are from the GDN.105
The sample includes 48 CDM host countries from Asia and the Pacific,
Latin America and the Caribbean, the Middle East and North Africa, SubSaharan Africa and Europe and Central Asia as listed in the Appendix
Table A6.1. Countries with fewer than three monthly non-zero observations (up to September 2008) in terms of credit flows (2012 kCERs) have
been removed.
Figure 6.1 presents the scatter plots between CDM credit flows and
absolute latitude and elevation, respectively. Despite the existence of
outliers such as China and Paraguay, the positive associations between
absolute latitude and CDM credit flows, and between elevation and CDM
credit flows, can be observed. Countries with higher absolute latitudes
and higher elevations are more likely to have more CDM projects as well
as CER credit flows.
Figure 6.2 demonstrates, in the upper chart, that CDM credit flows
in coffee exporters, diffuse exporters and point source exporters are
in general smaller than those in the non-exporters of the relevant
resources. The lower chart shows that manufactured goods exporters, service exporters and non-fuel primary goods exporters tend to have fewer
CDM credit flows in comparison to their counterparts.

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 165 — #5

166 Determinants of Financial Development

CDM and absolute latitude

A

CHN

CDM credit flows (in logs)

10
NGA
IND
BTN
BRA

8

KOR
MEX

MYS
IDN

COL

TZA
PAN
PER

6

KEN
SGP
ECU

EGY

UZB
PAK
ZAF

THA
VNM

PHL

ARG
CHL

AZE

JOR
ISR

NIC
SLV

BOL DOM
GTM

CRI

GEO

MDA

ARE
URY
MAR
CYP

BGD

MNG

ARM

HND

LKA
UGA
KHM

4

PRY

0

10

20
30
Absolute latitude

40

50

CDM and elevation

B

CHN

CDM credit flows (in logs)

10
NGA
IND
BTN
BRA
KOR

8

EGY

UZB
MYS
ARG

IDN

COL

THA

PAN
VNM ISR
NIC
SLV
PHL
DOM

MDA

6

ARE

SGP
BGD

CYP

URY
LKA

AZE
JOR

MEX
PAK
ZAF

CHL

TZA

PER
GEO
KEN
BOL
GTM
ECU
CRI
MNG
MAR
ARM
HND
UGA

KHM

4

PRY

2

4

6

8

Elevation
Figure 6.1

Scatter plots of CDM and geography

Note: Variables and data sources are described in the text. These figures show
scatter plots of absolute latitude and elevation against CDM credit flows (CERs).

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 166 — #6

Geographic Determinants of Carbon Markets (CDM)

167

CDM and resource exporters dummies

A

CDM credit flows (in logs)

8

6

4

2

0
RESCOFF

RESDIFF
Dummy=1

RESPOINT

Dummy=0

CDM and commodity exporters dummies

B

CDM credit flows (in logs)

8

6

4

2

0
EXPMANU

EXPPRIM
Dummy=1

Figure 6.2

EXPSERV

Dummy=0

CDM and resource endowments

Note: Variables and data sources are described in the text. These figures show the
comparisons of CDM credit flows (CERs) for different dummies of exporters.

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 167 — #7

168 Determinants of Financial Development

6.3

Econometric method: Spatial econometric approach

To study the impacts of geography on CDM project development, this
research conducts a cross-sectional study allowing for spatial correlation
on 48 countries over the period from December 2003 to September 2008.
It starts from an Ordinary Least Square (OLS) estimation on a basic model:



Yn = Xn β + n
n = 1, 2, . . . 48

(6.1)

where Yn is an n × 1 (n is the number of cross section units) vector of
observations on dependent variable CDM.
Xn is an n × k matrix of observations on k exogenous explanatory
variables which consist of geographic variables (LATITUDE, ELEV , AREA,
EXPSERV , EXPPRIM, RESPOINT , RESDIFF and RESCOFF) and the control
variables including GDP03, POP03, ETHNIC, RELIGION and legal origin
dummies (CIVLEG, COMLEG).
β is a k × 1 parameter vector. The error term n is an n × 1 vector with


E() = 0 and E( ) = δ 2 I.
The OLS specification typically follows the assumption of no spatial
interdependence or spatial correlation. However, spatial dependence
associated with social interactions or unobserved common shocks has
been widely recognized. On the one hand, considerable research has
been done to explore the implications of social or spatial interactions
in terms of neighbourhood effects, spatial spillovers or networks effects
(Manski, 2000; Brock and Durlauf, 2001). The fact that one agent’s decision variable is affected by those of other agents is typically formulated
as a spatial lagged dependent variable, or a spatial lag term to be included
in the right-hand side of the regression model. In the context of financial liberalization and reform, Abiad and Mody (2005, henceforth AM)
find that regional diffusion in terms of the liberalization gap from the
regional leader is significantly associated with the policy change.
On the other hand, in a globalized world, common shocks – either
observed global shocks like macroeconomic shocks or unobserved global
shocks like technological shocks – are believed to cause closer interdependence across countries. Andrews (2005) analyses the impact of common
shocks in the cross section regression in which the observations are i.i.d.
across population units conditional on common shocks, providing a
general framework for spatially correlated errors.106 In examining the
origins of financial openness, Quinn and Inclán (1997) argue that the
common trend, such as changes in consumer tastes and technology, may

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 168 — #8

Geographic Determinants of Carbon Markets (CDM)

169

substantially affect government liberalization policies as “fundamental
but unobservable forces”.
Obviously, the OLS estimation provides the foundation for spatial
analysis. This research incorporates the spatial correlation structure into
the basic linear model to account for both spatial lag dependence and
spatial error dependence.
A spatial lag model is a formal specification of spatial lag dependence
due to the presence of social and spatial interactions. Its basic form is the
mixed regressive, spatial autoregressive model:107



Yn = Xn β + λWn Yn + n , |λ|<1

(6.2)

where λ is the spatial autoregressive coefficient or spatial interdependence coefficient, measuring the dependence of Yi on neighbouring Yn .
Wn is an n × n spatial weighting matrix of known constants, reflecting the neighbouring relationships with zero across diagonals and a
row-standardized form. The added variable, λWn Yn , an average of
the neighbouring values, is referred to as a spatially lagged dependent
variable, or a spatial lag of Yn . The error term, n , is an n × 1 idiosyncratic error vector, assumed to be distributed independently across the
cross-sectional dimension with zero mean and constant variances σ2 .
When the spatial dependence exists in the error term due to unobserved effects of common shocks (for example, macroeconomic shocks,
political shocks or environmental shocks), a spatial error model can be
used as follows:108



Yn = Xn β + un
un = ρMn un + n , |ρ|<1

(6.3)

where ρ is the spatial autoregressive coefficient, measuring the amount of
spatial correlation in the errors. Mn is the spatial weighting matrix, may
or may not be the same as Wn . un is spatially correlated residuals and
n is the independent and identically distributed disturbances with zero
mean and constant variances σ2 . Mn un is known as a spatial lag of un .
By plugging the error term of the spatial error model (6.3) into the
spatial lag model (6.2), one can generate the spatial autoregressive model
with autoregressive disturbances of order (1,1), that is the SARAR(1,1)
model, as follows,
Yn = Xn β + λWn Yn + un ,
un = ρMn un + n ,

|ρ|<1

|λ|<1
(6.4)

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 169 — #9

170 Determinants of Financial Development

The above model is believed to be very general in the sense that
it allows for spatial spillovers stemming from endogenous variables,
exogenous variables and disturbances. It can be rewritten as:



Yn = Zn δ + un
un = ρMn un + n



(6.5)






where Zn = [Xn , Wn Yn ], δ = [β , λ]
The corresponding transformed model can be obtained by premultiplying (6.5) by In − ρMn ,



Yn∗ (ρ) = Zn∗ (ρ)δ + n

(6.6)

where Yn∗ (ρ) = Yn − ρMn Yn and Zn∗ (ρ) = Zn − ρMn Zn .
To estimate a general spatial model like (6.4), a number of approaches
have been proposed in the literature, for example, Kelejian and Prucha
(1998, 1999), Kelejian et al. (2004), Lee (2003, 2007) and Lee and Liu
(2006). However, these approaches in general assume that the innovations in the disturbance process are homoscedastic, which may not
hold in many applications. To fill this gap, Kelejian and Prucha (2010)
develop a Generalized Spatial Two-Step Least Square (GS2SLS) estimator
with a three-stage procedure of inference for the SARAR(1,1) model that
allows for unknown heteroscedasticity in the innovations. Arraiz et al.
(2010) provide simulation evidence showing that, when the disturbances
are heteroscedastic, the GS2SLS estimator produces consistent estimates
while the ML estimator produces inconsistent estimates.
This chapter examines the impacts of geography on CDM development
within a general SARAR(1,1) framework. To estimate the SARAR(1,1)
model, it employs the three-stage procedure of Kelejian and Prucha
(2010), which can be summarized in the following.
In the FIRST step, the model (6.5) is estimated by the Two-Stage Least
Square (2SLS) estimator using the instrument Hn . The instrument Hn is
the matrix of instruments which is formed as a subset of linearly independent columns of (Xn , Wn Xn , Wn2 Xn . . .). The first step 2SLS estimator
is as follows:
∼

∼




∼



δn = (Zn Zn )−1 Zn Yn

(6.7)

∼
∼
un = Yn − Zn δn

(6.8)

∼







where Zn = PH Zn = [Xn , Wn Yn ], Wn Yn = PH Wn Yn and PHn =




Hn (Hn Hn )−1 Hn .

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 170 — #10

Geographic Determinants of Carbon Markets (CDM)

171

In the SECOND step, ρn and σ2 are estimated, where ρn is the spatial
autoregressive parameter and σ2 is the variance of the innovation term
n . They are estimated by applying GMM to the model (6.5), based on
∼

the 2SLS residuals un obtained from the First step. More specifically, this
∼
estimator is ρn , defined as
∼

∼




∼ −1

∼

ρn = arg min [m(ρ, δn )  n m(ρ, δn )]

(6.9)

ρ[−aρ , aρ ]

∼

where  n is an estimator of the variance-covariance matrix of the
∼

1

limiting distribution of the normalized sample moments n 2 m(ρ, δn ).
∼

∼

∼

m(ρ, δn ) = gn (δn ) − Gn (δn )ρρ 2
 ´

∼ ∼
u
u
n
n

∼
1
 ´  
gn (δn ) =  

u
u
n n n 



∼
1
Gn (δn ) = 


n

∼´ 
u n un

∼´ 
2 u n un
´ ∼
=
2 u n un
∼
∼´ 
=´ ∼
=
u n un + u n un



∼

∼
=

∼

´ 
− u n un
∼
=´ ∼
=
− u n un
´ ∼
=
− u n un


n




Tr(Mn Mn ) 

0



un = Mn un
un = Mn2 un
In the THIRD step, δ in the transformed model (6.6) can be estimated
∼
by a generalized spatial 2SLS procedure (GS2SLS) after replacing ρ by ρn .
The GS2SLS estimator of δ is defined as:
∧ ∼

∧

∼

∼




∧

∼

∼

δn (ρn ) = [Z n∗ (ρn ) Zn∗ (ρn )]−1 [Z n∗ (ρn )Yn∗ (ρn )]
∼

∼

∼

∼

(6.10)
∧

∼

where Yn∗ (ρn ) = Yn − ρn Mn Yn , Zn∗ (ρn ) = Zn − ρn Mn Zn , and Z n∗ (ρn ) = PH
∼
Zn∗ (ρn ).

6.4

Empirical evidence

This section presents the empirical evidence for the impacts of various
geographic variables on CDM credit flows. Before proceeding to detailed

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 171 — #11

172 Determinants of Financial Development

econometric analysis, we briefly test for spatial dependence of CDM
credit flows across countries with evidence presented in Figure 6.3 and
Table 6.1.
Figure 6.3 plots the averaged CDM credit flows of all sample countries
against the distance to the country with the largest CDM credit flows in

CDM and distance to biggest host country

A
CHN

CDM credit flows (in logs)

10
NGA
IND
BTN
BRA

KOR

8

UZB
MYS
PAK

MEX

EGY

JOR
ISR
GEOMDA

VNM
PHL

SGP

6

MNG

BGD

ARG
CHL

ZAF

IDN
AZE
THA

ARE

ARM
LKA

COL
PAN
NIC
SLV
DOM
GTM
ECU
CRI

TZA
KEN
MAR

CYP

PER
BOL
URY

HND
UGA

KHM

4

PRY

0

5,000

10,000

15,000

20,000

Distance to biggest host country (km)
CDM and distance to smallest host country

B

CHN

CDM credit flows (in logs)

10
NGA
IND
BTN
BRA

8

KOR
MEX

ARG
CHL

6

EGY
ZAF

COL
PAN
PER
NIC
SLV
DOM
BOL
GTM
ECU
URY
CRI

UZB
PAK

IDN

AZE
TZA
KEN

MAR

THA

JOR
ISR
MDA GEO
ARE
CYP

HND

MYS

VNM
PHL

SGP
MNG
BGD

ARM
LKA

UGA
KHM

4

PRY

0

5,000

10,000

15,000

20,000

Distance to smallest host country (km)

Figure 6.3

CDM and distance to biggest and smallest host countries

Note: Variables and data sources are described in the text. These figures show
scatter plots of the distances to the biggest CDM host country (China) and to the
smallest host country (Paraguay) against CDM credit flows (CERs).

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 172 — #12

Geographic Determinants of Carbon Markets (CDM)

173

the upper chart, and the distance to the country with the smallest CDM
credit flows in the lower chart. Data on the great circle distances are from
Gleditsch et al. (2001). This figure clearly shows that countries closer to
the biggest CDM host country, which is China, tend to have more CDM
credit flows, whereas countries closer to the smallest CDM host country,
which is Paraguay, tend to have fewer CDM credit flows.109 Countries
with more (fewer) CDM credit flows appear to be geographically clustered
with other larger (smaller) CDM host countries.
By using two different spatial weighting matrices, an inverse-distance
spatial weighting matrix and a binary spatial weighting matrix, two
standard test statistics of spatial autocorrelation have been calculated
(Table 6.1). The inverse-distance spatial weighting matrix gives the
inverse of the distance to each sample point within a 4000 km neighbourhood, and zero otherwise, while the binary spatial weighting matrix
gives a weight of 1 to all sample points within a 4000 km neighbourhood,
and zero otherwise.110 Both matrices are row-standardized of one. Following Kelejian and Prucha (1999), the spatial weighting matrices have
been “idealized” so that each unit has the same number of neighbours
with “one neighbour ahead and one neighbour behind” in a wraparound
world.
Table 6.1 contrasts Moran’s I and Gearcy’s C statistics for CDM credit
flows. Both Moran’s I and Gearcy’s C statistics examine the null hypothesis of no spatial dependence. No matter which matrix is chosen, the
two Moran’s I statistics are greater than the expected value (−0.021)
and the two Gearcy’s C statistics are smaller than the expected value
(1.000), suggesting positive spatial dependence of CDM credit flows

Table 6.1 Moran’s I and Geary’s C for CDM
Moran’s I

E(I)

SD(I)

Inverse-distance Weights
Binary Weights

0.086
0.094

−0.021
−0.021

0.084
0.067

Inverse-distance Weights
Binary Weights

Gearcy’s C
0.902
0.870

E(C)
1.000
1.000

SD(C)
0.092
0.074

z-statistic
1.250
1.714
z-statistic
−1.064
−1.748

p-value
[0.102]
[0.043]∗∗
p-value
[0.144]
[0.040]∗∗

Notes: This table reports Moran’s I and Gearcy’s C tests for spatial autocorrelation for the
averaged CDM credit flows in logs for 48 CDM host countries listed in the Appendix Table
A6.1. The test statistics are calculated using an inverse-distance weighting matrix and a binary
weighting matrix, respectively, as described in the text.
∗ significant at 10%; ∗∗ significant at 5%; ∗∗∗ significant at 1%.

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 173 — #13

174 Determinants of Financial Development

across countries.111 Moreover, both Moran’s I and Gearcy’s C statistics
reject the null at about 10% significance level with an inverse-distance
spatial weighting matrix, and at 5% significance level with a binary spatial weighting matrix. This shows that the positive spatial dependence
of the CDM credit flows is significant across countries.
Tables 6.2 and 6.3 investigate whether countries with particular geographic endowments are more likely to attract CDM projects, for which
eight geographic endowment variables, as explained earlier, are selected
from various sources.112
Column 1 of Table 6.2 reports the OLS estimates for the non-spatial
model (6.1). Firstly, an OLS heteroscedasticity test following White
(1980) and Koenker (1981) is conducted to examine whether there is heteroscedasticity in the estimation regression which is related to any of the
geographic variables we examine.113 The White/Koenker test rejects the
null at 10% significance level, indicating that heteroscedasticity exists in
the estimations and should be taken into account for this context.
To test for which type(s) of spatial dependence, spatial lag dependence
or spatial error dependence or both, exist(s) in this context, we carry out
two simple Lagrange Multiplier tests (LM) separately. The hypothesis of
no spatially lagged dependent variable is rejected at about 10% significance level while the hypothesis of no spatially autocorrelated error term
can not be rejected. Furthermore, the p-values for the robust LM tests following Anselin et al. (1996) and the log-likelihood statistics are reported
to test for whether a spatial lag model is more appropriate than a spatial error model for this context. The evidence that the robust LM test
doesn’t reject the null hypothesis of no spatially autocorrelated error
term, but does reject the null of no spatially lagged dependent variable
(at about 10% significance level), together with the evidence that the
log-likelihood statistic for the spatial lag model (-41.03) is bigger than
that for the spatial error model (-41.61), suggest that a spatial lag model
is preferred to a spatial error model.
Columns 2 to 4 report the ML estimates for the spatial lag model
(6.2) and spatial error model (6.3), and the GS2SLS estimates following
Kelejian and Prucha (2010) for the SARAR(1,1) model (6.4). An inversedistance spatial weighting matrix has been used to calculate the ML
estimates and GS2SLS estimates.114
The spatial autocorrelation parameter, “ρ” appears to be insignificant
in both the spatial error model and the SARAR(1,1) model. For the spatial
autoregressive parameter, “λ”, ρ has been found weakly significant in the
spatial lag model and significant in the SARAR(1,1) model, with larger
coefficient in the SARAR(1,1) model. The GS2SLS estimate of “λ” in the

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 174 — #14

Geographic Determinants of Carbon Markets (CDM)

175

Table 6.2 Geography and CDM (by inverse-distance weights)
Non-spatial
model
λ

Spatial Lag
model

Spatial Error
model

0.185
[0.135]

ρ

LATITUDE
ELEVATION
AREA
EXPSERV
EXPPRIM
RESPOINT
RESDIFF
RESCOFF
GDP03
POP03
ETHNIC
REGLIGION
COMLEG
CIVLEG
Constant
Observations
R-squared
Log Likelihood
White/Koenker test
Spatial lag:
LM
Robust LM
Spatial error:
LM
Robust LM

0.016
[0.090]∗
0.276
[0.048]∗∗
0.155
[0.150]
0.965
[0.004]∗∗∗
−0.287
[0.368]
−1.587
[0.013]∗∗
−1.059
[0.013]∗∗
−1.368
[0.022]∗∗
0.258
[0.259]
0.360
[0.004]∗∗∗
1.336
[0.050]∗
2.077
[0.013]∗∗
0.557
[0.261]
1.278
[0.046]∗∗
−4.312
[0.074]∗
48
0.73
[0.105]

0.017
[0.088]∗
0.270
[0.008]∗∗∗
0.135
[0.173]
0.888
[0.002]∗∗∗
−0.320
[0.211]
−1.642
[0.000]∗∗∗
−1.098
[0.002]∗∗∗
−1.484
[0.001]∗∗∗
0.236
[0.090]∗
0.366
[0.001]∗∗∗
1.467
[0.015]∗∗
2.067
[0.000]∗∗∗
0.541
[0.117]
1.354
[0.004]∗∗∗
−5.175
[0.003]∗∗∗
48
0.74
−41.03

0.315
[0.226]
0.016
[0.111]
0.255
[0.012]∗∗
0.125
[0.219]
0.851
[0.004]∗∗∗
−0.337
[0.184]
−1.565
[0.000]∗∗∗
−0.998
[0.005]∗∗∗
−1.435
[0.001]∗∗∗
0.279
[0.056]∗
0.367
[0.001]∗∗∗
1.367
[0.031]∗∗
2.061
[0.000]∗∗∗
0.520
[0.135]
1.393
[0.003]∗∗∗
−4.064
[0.018]∗∗
48
0.72
−41.61

SARAR(1,1)

0.339
[0.033]∗∗
−0.300
[0.239]
0.018
[0.140]
0.274
[0.031]∗∗
0.118
[0.331]
0.860
[0.020]∗∗
−0.307
[0.333]
−1.678
[0.002]∗∗∗
−1.147
[0.010]∗∗∗
−1.525
[0.011]∗∗
0.185
[0.264]
0.360
[0.007]∗∗∗
1.606
[0.027]∗∗
2.001
[0.004]∗∗∗
0.552
[0.190]
1.331
[0.022]∗∗
−5.571
[0.006]∗∗∗
48

[0.107]
[0.107]
[0.572]
[0.570]

Notes: Dependent variable is the averaged CDM credit flows (2012 kCERs) in logs. Robust p-values are reported
in brackets. Variables and data sources are described in text. λ is the spatial autoregressive parameter in
dependent variable in the spatial lag model and SARAR (1,1) model, whilst ρ is the spatial autoregressive
parameter in the disturbance in spatial error model and SARAR(1,1) model. The White/Koenker test is to
examine the null of no heteroscedasticity. The spatial weighting matrix used here is a row-standardized
inverse-distance weighting matrix described in the text. Robust p-values are reported in brackets.
∗ significant at 10%; ∗∗ significant at 5%; ∗∗∗ significant at 1%.

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 175 — #15

176 Determinants of Financial Development

SARAR(1,1) model shows that the CDM credit flows in a country increase
by 0.34 units if those in its neighbouring countries increase by one unit.
The explanatory variables described in Section 6.2, except for
EXPMANU , have been found to be closely related to CDM credit flows
with the expected signs. In particular, the GS2SLS estimates show that
the the geographic variables LATITUDE and ELEV are positively associated with CDM development. For the resource and commodity exporter
dummies, EXPSERV is positively related, while RESPOINT , RESDIFF and
RESCOFF are negatively related, to CDM development. All of the control
variables including GDP03, POP03, ETHNIC, RELIGION and legal origin
dummies (CIVLEG, COMLEG) are in general found significantly associated with CDM development and should be included in the model.115
With a row-standardized binary weighting matrix, Table 6.3 in
general confirms the findings of Table 6.2 in terms of positive
impacts of LATITUDE, ELEV and EXPSERV , and negative impacts of
RESPOINT , RESDIFF and RESCOFF on CDM credit flows. Table 6.3 seems
to provide stronger evidence than Table 6.2, especially for the spatial
autoregressive coefficients, “λ” and “ρ”. According to the SARAR(1,1)
model, the degree of neighbourhood effects for the CDM credit flows
increases to 0.48.
The finding on the positive association between absolute latitude and
CDM credit flows is consistent with the literature. On the one hand,
research by Diamond (1997), Gallup et al. (1999) and Sachs (2003a) suggests that countries in a tropical location in terms of a smaller absolute
latitude are often associated with poor crop yields and production due to
adverse ecological conditions such as fragile tropical soils, unstable water
supply and prevalence of crop pests. On the other hand, tropical location
can be characterized as an inhospitable disease environment, believed
to be a primary cause for “extractive” institutions, in conjunction with
weaker institutions according to the settler mortality hypothesis of Acemoglu et al. (2001). Countries further from the Equator are more likely
to have better climate conditions and stronger institutions, which are
conducive to CDM project development.
The finding on the positive association between elevation and CDM
credit flows is in line with recent research. It is widely known that the
Earth’s average surface temperature rose by approximately 0.6◦ C in the
twentieth century and will rise a few degrees C in this century. Global
warming is likely to raise the sea level and change the land area and
elevation above sea level for many countries. Countries with higher elevations are therefore supposed to have more potential to attract CDM
projects.

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 176 — #16

Geographic Determinants of Carbon Markets (CDM)

177

Table 6.3 Geography and CDM (by binary weights)
Non-spatial
model
λ

Spatial Lag
model

Spatial Error
model

0.288
[0.068]∗

ρ

LATITUDE
ELEVATION
AREA
EXPSERV
EXPPRIM
RESPOINT
RESDIFF
RESCOFF
GDP03
POP03
ETHNIC
REGLIGION
COMLEG
CIVLEG
Constant
Observations
R-squared
Log Likelihood
White/Koenker test
Spatial lag:
LM
Robust LM
Spatial error:
LM
Robust LM

0.016
[0.090]∗
0.276
[0.048]∗∗
0.155
[0.150]
0.965
[0.004]∗∗∗
−0.287
[0.368]
−1.587
[0.013]∗∗
−1.059
[0.013]∗∗
−1.368
[0.022]∗∗
0.258
[0.259]
0.360
[0.004]∗∗∗
1.336
[0.050]∗
2.077
[0.013]∗∗
0.557
[0.261]
1.278
[0.046]∗∗
−4.312
[0.074]∗
48
0.73

0.018
[0.065]∗
0.255
[0.011]∗∗
0.115
[0.244]
0.831
[0.004]∗∗∗
−0.334
[0.187]
−1.671
[0.000]∗∗∗
−1.127
[0.001]∗∗∗
−1.515
[0.001]∗∗∗
0.220
[0.111]
0.382
[0.000]∗∗∗
1.581
[0.009]∗∗∗
1.940
[0.000]∗∗∗
0.559
[0.101]
1.407
[0.002]∗∗∗
−5.591
[0.001]∗∗∗
48
0.75
−40.56

0.495
[0.041]∗∗
0.016
[0.094]∗
0.232
[0.018]∗∗
0.118
[0.232]
0.779
[0.006]∗∗∗
−0.401
[0.118]
−1.574
[0.000]∗∗∗
−1.023
[0.003]∗∗∗
−1.529
[0.001]∗∗∗
0.267
[0.063]∗
0.358
[0.001]∗∗∗
1.395
[0.027]∗∗
2.011
[0.000]∗∗∗
0.482
[0.150]
1.408
[0.002]∗∗∗
−3.544
[0.042]∗∗
48
0.71
−40.99

SARAR(1,1)

0.476
[0.023]∗∗
−0.299
[0.205]
0.020
[0.108]
0.256
[0.047]∗∗
0.087
[0.479]
0.796
[0.034]∗∗
−0.319
[0.306]
−1.717
[0.002]∗∗∗
−1.182
[0.008]∗∗∗
−1.546
[0.009]∗∗∗
0.162
[0.325]
0.392
[0.004]∗∗∗
1.765
[0.018]∗∗
1.834
[0.006]∗∗∗
0.602
[0.155]
1.457
[0.014]∗∗
−6.221
[0.003]∗∗∗
48

[0.105]
[0.055]∗
[0.070]∗
[0.385]
[0.563]

Notes: The spatial weighting matrix used for the spatial lag model, spatial error model and SARAR(1,1) model
in this table is arow-standardized binary weighting matrix described in the text. See Table 6.2 for more notes.

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 177 — #17

178 Determinants of Financial Development

Some growth literature indicates that natural resource abundance is
connected with social and economic instability and weak institutional
quality, which hamper CDM project development. Isham et al. (2005)
find that, in comparison to manufacturing exporters, the exporting
countries of “point source” natural resources (e.g. oil, diamonds, plantation crops) and coffee/cocoa natural resources are more likely to have
severe social and economic divisions, and less likely to develop socially
cohesive mechanisms and effective institutional capacities for managing
shocks.
In sum, this research produces the following significant findings. First,
it provides evidence for the presence of positive spatial dependence
amongst observations for this context, especially the spatial lag dependence associated with neighbourhood effects and social interactions.
CDM credit flows in a country are significantly affected by those of its
neighbouring countries, more specifically, the CDM credit flows in a
country increase by about 0.34 to 0.48 units if those in its neighbouring
countries increase by one unit. Second, by allowing for spatial dependence and accounting for the size of the economy (initial population
and initial GDP per capita), this research finds that the absolute latitude
and elevation have positive impacts on CDM credit flows, suggesting that
countries further from the Equator and having a higher elevation tend to
initiate more CDM projects and issue more CDM credit flows. Countries
with more exports of services seem to have more advantages in attracting
CDM projects, whilst in contrast countries with more exports of natural
resources have smaller CDM credit flows, indicating that natural resource
abundance may not necessarily be conducive to CDM development.

6.5

Concluding remarks

Under the Kyoto Protocol, the Clean Development Mechanism (CDM)
is designed to provide the non-Annex I countries (developing countries
and transition economies) with access to the flows of technology and
capital which could contribute to their sustainable development objectives, whilst allowing Annex I countries to earn credits to meet their
Kyoto commitments by investing in GHG emission reduction projects
in non-Annex I countries.
This chapter investigates whether the cross-sectional differences in
geographic endowments can explain the cross-sectional differences in
CDM credit flows. It conducts a cross-country study allowing for both
spatial error dependence and spatial lag dependence for 48 CDM host
countries over December 2003–September 2008.

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 178 — #18

Geographic Determinants of Carbon Markets (CDM)

179

This research leads to two significant findings. First, it provides evidence for a positive relationship between CDM credit flows in a country
and those in its neighbours, more specifically, the CDM credit flows
in a country increase by about 0.34 to 0.48 units if those in its neighbours increase by one unit. Countries with larger (smaller) CDM credit
flows have been found to be geographically clustered with other larger
(smaller) CDM host countries. Second, by allowing for spatial dependence and accounting for the size of the economy (initial population
and initial GDP per capita), this research finds that absolute latitude and
elevation have positive impacts on CDM credit flows, suggesting that
countries further from the equator and having higher elevations are in
better positions to attract CDM projects. Countries with more exports of
service are more associated with larger CDM credit flows, whilst in contrast countries with more exports of natural resources have fewer CDM
credit flows, indicating that natural resource abundance doesn’t necessarily play a large role in promoting CDM development. These findings
are robust to the choices of different spatial weighting matrices – an
inverse-distance spatial weighting matrix and a binary spatial weighting
matrix. The research also controls for an ethnic fractionalization index,
a religious fractionalization index and legal origin dummies.
This finding sheds light on the geographic determinants of uneven
CDM project development across countries, and has rich implications for
developing countries in terms of international cooperation and national
capacity building to access effectively CDM for their national sustainable
development objective. This research may contribute to our understanding of the cross-country differences in CDM development and contain
some merits for the UNFCCC in terms of improving the geographic distribution of CDM project activities and capacity building. This research
also suggests that geographic considerations should be introduced into
the econometric and theoretical cross-country studies of climate change
and mitigation.

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 179 — #19

180 Determinants of Financial Development

Appendix table
Table A6.1 The list of countries in the full sample
Code

Country name

Code

Country name

ARE
ARG
ARM
AZE
BGD
BOL
BRA
BTN
CHL
CHN
COL
CRI
CYP
DOM
ECU
EGY
GEO
GTM
HND
IDN
IND
ISR
JOR
KEN

United Arab Emirates
Argentina
Armenia
Azerbaijan
Bangladesh
Bolivia
Brazil
Bhutan
Chile
China
Colombia
Costa Rica
Cyprus
Dominican Rep.
Ecuador
Egypt, Arab Rep.
Georgia
Guatemala
Honduras
Indonesia
India
Israel
Jordan
Kenya

KHM
KOR
LKA
MAR
MDA
MEX
MNG
MYS
NGA
NIC
PAK
PAN
PER
PHL
PRY
SGP
SLV
THA
TZA
UGA
URY
UZB
VNM
ZAF

Cambodia
Korea, Rep.
Sri Lanka
Morocco
Moldova, Rep.
Mexico
Mongolia
Malaysia
Nigeria
Nicaragua
Pakistan
Panama
Peru
Philippines
Paraguay
Singapore
El Salvador
Thailand
Tanzania
Uganda
Uruguay
Uzbekistan
Vietnam
South Africa

Note: This table lists the country codes and country names for 48 CDM
host countries considered in this analysis. Data are from the UNEP
Risoe Centre CDM/JI Pipeline Analysis and Database (2008).

HUANG: “CHAP06” — 2010/9/29 — 20:06 — PAGE 180 — #20

Conclusion
This research studied the fundamental issues related to financial market
development and carbon market development in the context of globalization, using recently developed econometric and statistical methods.
Chapter 1 contained an overall review of the literature on the development of financial markets. Chapters 2 to 5 examined specific issues
related to financial development whilst Chapter 6 was about the geographic determinants of CDM development, which is an important part
of carbon markets, especially for developing countries.
Chapter 2 sought to investigate the political, economic and geographic
determinants of the development of financial markets. By jointly applying two prominent tools for addressing model uncertainty, BMA and
Gets approaches, it suggested that initial GDP, initial population, legal
origin and institutional quality are fundamental determinants of the
cross-country differences in financial development.
Chapters 3 and 4, respectively specifically focused on the economic
and political determinants of financial development in the context of
globalization. By using GMM estimation on averaged data and a common factor approach on annual data, Chapter 3 indicated a positive
causal effect going from aggregate private investment to financial development, and vice versa. From a political viewpoint, Chapter 4 revealed
a positive effect of institutional improvement on financial development
at least in the short run, and an increase in financial development after
democratic transformation.
Chapter 5 analysed what induces governments to undertake financial
reforms, and what causes financial development. Starting from AM and
allowing for error dependence across countries and over time, Chapter
5 found that some of the AM findings are robust, but others are fragile.
It also identified a negative effect of the extent of democracy on policy
reform. Together with Chapter 4, it seems to indicate that a short-run
increase in financial development emerges after democratization, but
that once democracy has been established and enhanced, its extent may
exert negative effects on the likelihood of financial reform.
The last chapter found that countries with larger (smaller) CDM credit
flows tend to be geographically clustered with other larger (smaller)
CDM host countries and countries with higher absolute latitude, higher

181

HUANG: “CONCLU” — 2010/9/29 — 20:06 — PAGE 181 — #1

182 Determinants of Financial Development

elevations and more exports of services tend to have more CDM credit
flows than others.
This research to some extent contributes to our understanding of the
causes of financial development, and adds to the growing research in this
area. However, what I have done so far merely represents a start being
made in this direction, much remains to be done. A number of areas for
further research are summarized below:
1. While it is suggested that the level of financial development in a
country is determined by its institutional quality, macroeconomic policies and geographic characteristics, as well as the level of income and
cultural characteristics, further research into the more fundamental factors behind these characteristics is obviously worthwhile. To this end,
other approaches may be considered, for example, recursive methods or
dynamic programming applied to the theoretical models.
2. This research suggests that economic reforms with more open trade
policies and attractive investment policies, and political reforms aimed
at a more democratic society, should be conducive to financial development. Other research suggests that legal and regulatory reforms boost
financial development. However, how to undertake these reforms, and
in what sequence, has not yet been fully understood.
3. Although this research takes into account in Chapters 3 and 4, the
effects of globalization on financial development, much work is still
needed to explore the link between domestic and international financial
markets, the impact of financial market integration on the development of domestic financial markets and the role of foreign financial
institutions in domestic financial development.
4. As time goes, with more data available on the number of CDM
projects and/or the volume of CDM credit flows, time series studies or
panel data studies can be carried out to find whether or not a more
open trade policy is conducive to CDM development, whether or not
institutional quality is important for CDM, whether or not financial
development promotes CDM development, and so on.

HUANG: “CONCLU” — 2010/9/29 — 20:06 — PAGE 182 — #2

Notes
1. See Levine (1997, 2005) for a review.
2. The 39 potential determinants considered for this analysis are grouped under
four headings: institutions, policy, geography and others. See Section 2.2.3
for details.
3. The description of these measures relies heavily on Demirgüç-Kunt and
Levine (1996, 1999).
4. Since data for the efficiency of the bond market are not available whilst data
for the size of the bond market are mainly available for developed countries in the World Bank’s Financial Structure and Economic Development
Database (2008), to avoid resulting in smaller sample sizes in the principal
component analysis, bond market development is not included here. A simple analysis of the determinants of bond market development (for a smaller
sample) is presented in Appendix Table A2.8.
5. Measures of financial liberalization and financial openness are not used
here due to the concern that the effects of other variables on financial
development may work through them.
6. ccounting standards data in La Porta et al. (1998) forms another interesting variable, but this variable has to be excluded due to its limited country
coverage.
7. To some extent, absolute latitude serves as an alternative indicator for the
zero-one tropical dummy in the GDN.
8. The EBA proposed by Leamer (1983, 1985) regards a variable to be robust if
its extreme bounds lie strictly to one side or the other side of zero, where
the extreme bounds for the coefficients of a particular variable are defined as
“the lowest estimate of its value minus two times its standard error and the
highest estimate of its value plus two times its standard error, respectively”.
The interval formed by two extreme bounds constitutes the maximum scope
by which a variable may vary in the presence or absence of others, and
indicates the confidence one may have in the coefficient estimates.
9. A computer program for the Bayesian approach to model uncertainty has
been developed by Raftery et al. (1997).
10. A computer algorithm designed for implementing the general-to-specific
approach, called PcGets, has been developed by Krolzig and Hendry (2001),
following earlier work by Hoover and Perez (1999).
11. As argued by Granger and Hendry (2005) and echoed by Hansen (2005),
none of the model selection methods currently available is immune from
four possible conceptual errors of model selection methods: parametric
vision, the assumption of a true data generating process, evaluation based
on fit and ignoring of the impact of model uncertainty on inference.
12. Sala-i-Martin (1997a, 1997b) criticizes the standard of robustness employed
by Levine and Renelt (1992) as too restrictive, and suggests a different version
of extreme bounds analysis by saying that a variable is robust as long as 95%
or more of the distribution of estimates lies to one side of zero. By this
183

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 183 — #1

184 Notes

13.

14.

15.

16.
17.

18.

19.

20.
21.

22.
23.
24.
25.

methodology, relatively more variables are found to be robustly related to
growth. However, the methodology of Sala-i-Martin (1997a, 1997b) is not
Bayesian, although it uses weights proportional to the likelihoods of each
model.
Fernandez et al. (2001) re-examine the Sala-i-Martin (1997a, 1997b) dataset
using a full BMA explained below and Markov Chain Monte Carlo techniques to deal with the huge range of possible models. The full BMA of
Fernandez et al. (2001) requires full specification of the prior distribution
for every parameter conditional on each possible model and calculates the
average of the parameter estimates across all possible models by using corresponding posterior model probabilities as weights. Their research has
produced findings in support of the conclusions of Sala-i-Martin (1997a,
1997b). However, fully specifying the prior distribution for all potential
parameters is very difficult and “essentially arbitrary” (Sala-i-Martin et al.
2004) when the number of possible regressors is large.
The computational procedure for the Occam’s Window technique is implemented using the bicreg software for S-Plus or R written by Adrian Raftery
and revised by Chris Volinsky.
The Occam’s Window approach can be divided into two types, corresponding to two approaches. One is the symmetric Occam’s Window in which
models “much less likely than the most likely model” are excluded, the
other is the strict Occam’s Window in which the models having “more likely
submodels nested within them” are excluded from the subset left in the
symmetric Occam’s Window (Raftery, 1995).
The modification used to calculate the sign certainty index is developed by
Malik and Temple (2009) based on the original bicreg code.
The summary below is heavily drawn from Hoover and Perez (1999), Krolzig
and Hendry (2001), Hendry and Krolzig (2005) and Granger and Hendry
(2005).
Since any variable removed at the pre-search stage is permanently eliminated, the F pre-search testing (top-down) at step 1 in the “liberal strategy”
default setting has been increased from 0.75 to 1, so as not to risk omitting
any potential factor which is not significant in the GUM.
The effect of institutions, policy and geography on financial development
are also examined in isolation. The results are not reported here, but are
available from the author.
MC3 is essentially a shorthand for the Markov Chain Monte Carlo technique,
which is applied in this table only as a robustness test.
The Chow tests are F tests and used to test parameter constancy. The Normality test, a Chi-squared statistic, is used to check the normality of the
distribution of the residuals. The Heteroscedasticity test is for unconditional
heteroscedasticity.
The variable EURFRAC is closely associated with legal origins as noted earlier.
Many experiments suggest that the results are sensitive to the inclusion of
the variable EXPMANU .
It is the share of population who can speak one of the main European
languages.
It is the proportion of the population near a coast a river navigable to the
ocean.

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 184 — #2

Notes

185

26. The sketch in this section heavily relies on Raftery (1995), Sala-i-Martin et al.
(2004) and Malik and Temple (2009).
27. A saturated model (Ms ) in which each data point is fitted exactly can be also
used as a baseline model.
28. Financial intermediaries emerge endogenously under certain conditions, as
widely addressed by Diamond (1984) and Williamson (1986), to avoid the
duplication of monitoring costs (to minimize the monitoring costs by pooling projects), to channel savings from households to firms for use in the
production process and to pool risk.
29. Among others, Chapter 2 examines the long-run determinants of financial
development by using BMA and Gets approaches. That chapter suggests that
“the level of financial development in a country is determined by its institutional quality, macroeconomic policies, and geographic characteristics, as
well as the level of income and cultural characteristics”. Chapter 4 reveals
that institutional improvement is typically followed by a higher level of
financial development at least in the short run, while Chapter 5 suggests
that, once democracy has been established and enhanced, the extent of
democracy may exert negative effects on the extent to which governments
undertake reform aimed at financial development.
30. Among others, Doms and Dunne (1993) show that microeconomic lumpiness is very important for aggregate investment. Bertola and Caballero (1994)
argue that microeconomic irreversibilities play an important role in smoothing investment dynamics in the presence of idiosyncratic uncertainty. In the
industrial organization literature, Dixit (1989), Leahy (1993) and Caballero
and Pindyck (1996) discuss the consequences of the entry (creation) decision
of new (incumbent) entrepreneurs and exit decisions of some incumbents
for variation in the aggregate stock of capital.
31. Benhabib and Spiegel (2000) show that financial development positively
influences the investment rate. Schich and Pelgrin (2002) indicate a positive
effect going from financial development to private investment in 19 OECD
countries over 1970–97. Ndikumana (2000, 2005) finds that the development of banks and stock markets tends to stimulate domestic investment.
32. Details on these indicators can be found in Section 3.2.
33. Kose, Prasad and Terrones (2003) show that the overall volume of international trade and gross private capital flows has increased dramatically over
the past three decades: in particular, “the growth of world trade has been
larger than that of world income in almost all years since 1970”.
34. This source could be the most reliable one for private investment ratio, while
we can calculate it by deducting the net inflows of FDI and public investment
from the gross fixed capital formation. Although data for private investment
are only for up to 1998, they are sufficient (or long enough) to conduct this
analysis.
35. Essentially, the principal component analysis takes N specific indicators and
produces new indices (the principal components) X1 , X2 ,...XN which are
mutually uncorrelated. Each principal component, a linear combination of
the N indicators, captures a different dimension of the data. Typically the
variances of several of the principal components are low enough to be negligible, and hence the majority of the variation in the data will then be
captured by a small number of indices.

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 185 — #3

186 Notes
36. The summary below is heavily drawn from Demirgüç-Kunt and Levine
(1996, 1999).
37. The precision of the principal component analysis used to derive this new
index depends on having a relatively large number of variables. Given that
there are only three indices on which the principal component analysis is
based, the new index of financial development is almost the mean of the
three individual indices.
38. Two measures for the efficiency of financial intermediation widely used are
Overhead Costs, the ratio of overhead costs to total bank assets, and Net
Interest Margin, the difference between bank interest income and interest
expenses, divided by total assets. Due to the incompleteness of the available
data, they are not included in this analysis.
39. In the growth and convergence context, both the panel data analysis of
Caselli et al. (1996) and the cross section analysis of Mankiw et al. (1992)
find a negative effect of initial income on growth, but the former identifies a much larger effect than the latter, implying a 10% convergence rate
relative to 2–3% suggested by Mankiw et al. (1992).
40. Starting from a general model with three lags of the dependent and independent variables and testing the null hypothesis of the coefficients being
zero for the longest lag, we end up with one lagged independent variable
and one lagged dependent variable appearing in the model for this context,
given that the relevant specification tests are satisfied.
41. Caselli et al. (1996) treat some variables like the investment rate and population growth rate as predetermined and argue that these variables are
potentially both causes and effects of economic growth.
42. Alonso-Borrego and Arellano (1999) propose the symmetrically normalized GMM estimator and the Limited Information Maximum Likelihood
estimator. Recently Kruiniger (2008) has developed the Maximum Likelihood estimator and Newey and Windmeijer (2009) have proposed the new
variance estimator for the generalized empirical likelihood estimator.
43. Bond et al. (2001) and Bond (2002) illustrate that in principle the firstdifferenced GMM estimates for the AR(1) coefficient should lie between the
within group estimates (being downwards biased) and the OLS estimates
(being upwards biased) from a straightforward pooled regression.






44. For the case of r=2, when ft = (1 ηt ) and λi = (αi 1), we have λi ft = αi + ηt ,
where αi and ηt are the individual effect and time effect, respectively.
45. Bai (2004) suggests that differenced data can also be used to calculate the
number of factors.
k


k

46. The normalization that  N = Ik is used when T > N.
47. Bai and Ng (2004) recommend standardizing the data first, although the
PANIC approach does not require it.
48. The standardized FD and PI are used here and the rest of the study. The PANIC
approach essentially requires a balanced panel. To overcome the problem of
missing data, imputation within each region is conducted, since countries
in a region tend to have similar income levels, closer economic relations and
to be more dependent on each other. There are 49 observations imputed for
FD and 64 observations for PI, corresponding to 4% and 5% of complete
observations in the resulting balanced panels, respectively. Appendix Table
A3.3 presents the list of countries in each region.

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 186 — #4

Notes

187

49. When time trends are included, the presence of a deterministic trend in
the data-generating process is assumed while the absence of such a trend
is assumed when time trends are not allowed. Including time trends is less
general than excluding them. In particular, including trends improves fit
to some extent but may cause a large loss of power and possibly severe
multicollinearity in the ADF regressions.
50. Four “panel” statistics are a “variance ratio” statistic (Z

vNT ), a “panel-t”
statistic (Z
tNT ), a “panel-rho” statistic (Zρ
NT −1 ) and a “panel-ADF” statistic
(Zadf
 NT ).
51. Three “group mean” statistics are a “group-t” statistic (
Z
tNT ), a “group-rho”
statistic (
Zρ
NT −1 ) and a “group-ADF” statistic (
Z  ).
adf NT

52. The Pedroni test based on defactored data should be interpreted with caution, since the defactored data are estimated and may be subject to particular
forms of measurement errors.
53. More specifically, the MG estimator and its standard
errors are calculated as


−

N


(θi − θ )2

N


−
θi
σ (

θ)
i−1 N − 1


θMG = θ = i=1 and se(

θMG ) = √ i =
, respectively.
√
N
N
N
54. To overcome the problem of missing data, imputation within each region is
conducted, since countries in a region tend to have similar income levels,
closer economic relations and be more dependent on each other. There are
49 observations imputed for FD and 64 observations for PI, corresponding
to 4% and 5% of complete observations in the resulting balanced panels,
respectively.
55. The short-run coefficients reported in Tables 3.5 and 3.6 are in general less
informative. The CCEP and WG assume the short-run coefficients to be identical across countries, ignoring the heterogeneity widely recognized. The
CCEPMG and CCEMG (as well as PMG and MG) allow the short-run coefficients to vary across countries, which is a more realistic assumption to make.
However, the short-run coefficients reported are the cross-country averages,
and therefore they are highly influenced by the outliers.
56. The number of lags is constrained by the number of observations. As shown
by Pesaran et al. (1999), the PMG estimator seems quite robust to outliers
and the choice of ARDL order.
57. Data on GDP per capita and trade openness are taken from Heston et al.
(2006).
58. Countries are considered as experiencing a political transition when either
their “polity2” scores in the PolityIV Database by Marshall and Jaggers (2009)
change from negative values to positive values or when their “freedom”
indices, defined in this paper from the Freedom House Country Survey
(2008), change from “Not Free” to “Free” or “Partly Free”.
59. One of the channels through which democratization affects financial development is property rights protection and contract enforcement. Olson
(1993) and Clague et al. (1996) argue that democracies tend to result in
better protection of property rights and more efficient contract enforcement, which are conducive to financial development (La Porta et al., 1997,
1998).

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 187 — #5

188 Notes
60. La Porta et al. (1997, 1998) document that countries with a legal code like
Common Law tend to protect private property owners, while countries with
a legal code like Civil Law tend to care more about the rights of state and less
about the rights of the masses. Countries with French legal origins are said to
have comparatively inefficient contract enforcement and more corruption,
and less well-developed financial systems, while countries with British legal
origins enjoy higher levels of financial development.
61. They argue that incumbents have strong incentives to block the development of a more transparent and competitive financial sector, although these
incentives may be weakened by openness to external trade and international
flows of capital.
62. Based on annual data on developed and developing countries over 1975–
2000, Girma and Shortland (2008) use approaches such as the system
GMM approach from Arellano and Bover (1995) and Blundell and Bond
(1998). In contrast to their study, this research uses the system GMM and
LSDVC approaches, based on averaged data on 90 developed and developing countries over 1960–99 to see if democratization brings about financial
development.
63. The main reason for this is that, data prior to 1990 for these countries generated by the centrally planned economy are largely incomplete, while data
after 1990 are highly problematic or doubtful since most of these countries
underwent severe economic disorder for several years in the early stage of
the transformation process to a market-oriented economy. A research area
in the future will be to see if the transition countries fit the pattern observed
for the sample countries of this study.
64. Data for the black market premium from the GDN are available up to 1998.
65. The description here is mainly from Demirgüç-Kunt and Levine (1996,
1999).
66. Two measures for the efficiency of financial intermediation which are sometimes used are Overhead Costs, the ratio of overhead costs to total bank
assets, and Net Interest Margin, the difference between bank interest income
and interest expenses, divided by total assets. Due to the incompleteness of
the relevant data, they are not included in this analysis.
67. In this polity coding system, zero is the threshold by which a country with
a positive “polity2” score is regarded as a democracy whilst a country with
a negative “polity2” score is regarded as an autocracy.
68. The democracy and autocracy scores are derived from six authority characteristics (regulation, competitiveness and openness of executive recruitment;
operational independence of chief executive or executive constraints and
regulation and competition of participation). Based on these criteria, each
country is assigned a democracy score and an autocracy score ranging from
0 to 10. The larger the democracy score, the fairer the election of executive
power, the more open the political process and the higher the extent of the
constraints on executive power. In contrast, a larger autocracy score reflects
a less open political process in a country in terms of less competitiveness
and fairness in election, narrower participation and fewer constraints on
executive power.
69. By experimenting with five-year and eight-year averages, respectively, I start
from a general model with three lags of the dependent and independent

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 188 — #6

Notes

70.

71.

72.

73.

74.

75.

76.

77.
78.

189

variables and test the null hypothesis of the coefficients being zero for the
longest lag. I end up with one lagged dependent variable, one lagged independent variable, one lag of log GDP and one lag of the trade openness
measure appearing in the model with eight-year averages for this context,
given that the relevant specification tests are satisfied.
The series xi t−1 is defined as being predetermined with respective to vi, t
when xi t−1 is correlated with vi, t−1 and earlier shocks, but is uncorrelated
with vi t and subsequent shocks. The series xi t is strictly exogenous when
xi t is uncorrelated with earlier, current and future errors. See Bond (2002)
and Arellano (2003) for details.
For the multivariate autoregressive model, Blundell and Bond (2000) show
that a sufficient condition for the additional moment conditions to be valid
is the joint mean stationarity of all the series.
In this analysis the instrument set used is restricted (to avoid the possible
over-fitting bias) in the sense that all lagged values of y, x and z at dates t–2
and t–3 are used as instruments for yi,t−1 , xi,t−1 and zi,t−1 in the first
difference equation.
Note that when the instrument set is not restricted, the lagged first differences of the series (yit , xit , zit ) dated t–1 are used as instruments for the
untransformed equations in levels. Differences lagged two periods or more
are redundant as instruments for the levels equations because the corresponding moment conditions are linear combinations of those already in
use. In this analysis, the lagged first-differences of the series (yit , xit , zit ) dated
t–1 and t–2 are employed due to the use of restricted instrument set.
Essentially, in the bias approximation of Bruno (2005), the within operator
is adjusted to include an exogenous selection rule which selects only the
observations with observable current and one-time lagged values, by which
missing observations for some individuals are allowed.
Since the freedom index has data starting from the period 1972–73, it is
not used for the panel data study, but is used for selecting the democratic
transition countries.
The event identification methodology of Papaioannou and Siourounis
(2008) has been found useful for selecting the democratic transition countries, but the selection method in this analysis differs from their method
in the following ways. First, for simplicity this analysis selects the sample
exclusively depending on the changes from autocratic rule to democratic
regimes without any further divisions, while Papaioannou and Siourounis
(2008) divide democratizations into “full”, “partial” and “borderline” with
different thresholds in terms of either the “polity2” or the “freedom” index.
Second, this analysis is interested in the effect of a stable regime change
on financial development. Hence, the sample includes only the countries
whose regime changes last for at least ten years.
The FD measure has been standardized. More specifically, it is divided by the
cross-country standard deviation of FD in 1999.
When we compare the five-year averages before and after democratization,
we find that the five-year average of standardized FD post-democratization
for 33 countries is larger by 0.015 cross-country standard deviations of FD
than before their democratization, and about two-thirds of the sample countries benefit from this process. Columns 7 to 9 show that the average of

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 189 — #7

190 Notes

79.

80.

81.

82.

83.

84.

85.

86.

87.

standardized FD five to ten years post-democratisation for 33 countries is
larger by 0.212 cross-country standard deviations of FD than ten to five years
before their democratization, and much more sample countries benefit from
this process. The median values of the increase in standardized FD for three
cases of comparison are positive.
Looking at the financial development performance of each individual country, we find enormous heterogeneity across countries, ranging from an
increase of 1.096 of a cross-country standard deviation of FD in the ten-year
average of standardized FD for Thailand to a decline of 0.415 of a crosscountry standard deviation of FD for Zambia. The Republic of Korea and
Madagascar also witnessed a drastic increase in the ten-year average of standardized FD, whilst Nicaragua and Uruguay experienced a tremendous drop
in FD following their democratization. Case studies on how democratization helped the financial development process are interesting areas for future
research.
The regression is estimated by OLS in which the unobserved country specific effects, time effects and control variables such as trade openness, GDP,
aggregate investment and the black market premium are included.
The financial development performance in Asian countries and other economic performances in East Asian and Pacific countries are largely different
from those in South Asian countries.
Results regarding the impacts on specific financial development measures
such as private credit, liquidity liabilities and commercial-central bank are
available from the author upon request.
The selection of these subsamples is mainly stimulated by Rodrik and
Wacziarg (2005) in which low-income countries, ethnically diverse countries
and Sub-Saharan African countries are studied. However, I find no evidence
in support of a positive/negative link between institutional improvement
and financial development for the Sub-Saharan African countries. Experiments were also conducted for the Asian countries and Latin American
countries, again finding no evidence.
In addition, the ordered logit approach imposes strong distributional
assumptions relative to a linear model, and the estimates of individual
effects and other parameters may be inconsistent because of an incidental
parameter problem.
Other ways to address either cross section correlation or serial correlation
in this context have also been done (results are available from author upon
request).
FLit is generated by dividing the original AM financial liberalization index
by 18. The original financial liberalization index, ranging from 0 to 18, is
based on six policy dimensions (credit controls, interest rate controls, entry
barriers in the banking sector, operational restrictions, privatization in the
financial sector and restrictions on international financial transactions) with
each dimension taking on values between 0 and 3.
The democracy and autocracy scores are derived from six authority characteristics (regulation, competitiveness and openness of executive recruitment;
operational independence of chief executive or executive constraints and
regulation and competition of participation). Based on these criteria, each
country is assigned a democracy score and an autocracy score ranging from

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 190 — #8

Notes

88.
89.
90.

91.
92.

93.

94.

191

0 to 10. The larger the democracy score, the fairer the election of executive
power, the more open the political process and the higher the extent of the
constraints on executive power. In contrast, a larger autocracy score reflects
a less open political process in a country in terms of less competitiveness
and fairness in elections, narrower participation and fewer constraints on
the executive.






Here θ1 c and −θ1 are renamed as θ1 and θ2 , respectively. β1 , β2 , β3 and β4






are reparameterized as θ3 , θ4 , θ5 and θ6 , respectively.
Here θ1 c, −θ1 and bθ1 are renamed as θ1 , θ2 and θ3 , respectively. β1 , β2
, β3

and β4
are reparameterized as θ4 , θ5
, θ6
and θ7
, accordingly.
More specifically, IMFit has been found to be significant when country fixed
effects are excluded, while REG_FLi,t−1 − FLi,t−1 appears to be significant no
matter whether the country fixed effects are included or not.
Divided by 18, the original measure has been rescaled to get an index, FLi,t ,
ranging between 0 and 1.
This analysis first experiments with including time dummies in the original
AM models in the within group estimation to control for cross section correlation. However, this approach is not as general as Pesaran’s (2006) approach
which, besides other advantages, allows common factors to have differential impacts across countries. Including time dummies controls only for a
common component, whose effect is common across countries.
Although serial correlation in the errors can be alleviated once country
fixed effects are included, it may not be fully removed. The standard
robust standard errors do not allow for serial correlation in errors, only for
heteroscedasticity.
N

The test statistic takes the form of −2
ln(piT ) in which piT is the p-value
i=1

95.

96.
97.

98.

99.
100.

101.

corresponding to the unit root test of the ith individual cross section unit
for the cross-sectionally augmented DF regression. The critical values for the
Fisher P-test on a cross-sectionally augmented regression (Pesaran, 2007) are
provided by M. Hashem Pesaran.
Since the lagged dependent variable bias arising from the within group transformation can be alleviated when T is large in a dynamic panel (Nickell,
1981).
FLi,t−1 (1 − FLi,t−1 ) is reported here.
Although the coefficients on REG_FLi,t−1 − FLi,t−1 and its interaction term
are negative and positive, respectively, the range of FLi,t−1 from 0 to 1
determines the derivative of FLit with respect to REG_FLi,t−1 − FLi,t−1 ,
−0.147 + 0.094 × FLi,t−1 , is always negative.
The panel is unbalanced mainly because data on IMF programs are missing for the following six countries over period 1973–83: China, Costa Rica,
Ecuador, Jamaica, Nigeria, Portugal and Uruguay.
Data are from the UNEP Risoe Centre (2008).
For example, as the only two CDM host countries in Asia in 2003, India
and the Republic of Korea were immediately followed by four Asian host
countries in 2004 and nine other Asian host countries in 2005 (UNEP Risoe
Centre, 2008).
A country with k monthly non-zero observations (up to September 2008)
has its averaged CDM being its total CERs divided by k.

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 191 — #9

192 Notes
102. CO2 e is the Carbon Dioxide Equivalent, the unit of measurement used to
indicate the global warming potentials defined in decision 2/CP.3 of the
Marrakech Accords or as subsequently revised in accordance with Article 5
of the Kyoto Protocol.
103. Data on latitude, elevation and land area for Singapore are added to the
physical factors dataset of CID.
104. This inclusion is stimulated by the works of Alesina et al. (2003) and Stulz
and Williamson (2003), for example. Alesina et al. (2003) argue that the
ethnic and religious fractionalizations in a country are associated with its
economic success and institutional quality. Stulz and Williamson (2003)
show that culture, proxied by ethnic, religious and language differences,
explain why investor protection differs across countries and how investor
rights are enforced among countries.
105. The inclusion is due to La Porta et al. (1998) who suggest that the legal origin
of a country is helpful in explaining the extent to which investor rights are
protected in it. More specifically, countries with a Common Law tradition
tend to place more emphasis on private rights protection and less on the
rights of the state, while countries which have adopted a Civil Law tradition
do the opposite.
106. The Andrews (2005) approach is very general in the sense that the effects
of common shocks, which are ς -measurable, may differ across population
units, in a discrete or continuous fashion, and may be local or global in
nature.
107. The addition of the spatially lagged dependent variable results in a form
of endogenity, rendering the OLS an inapplicable method for spatial lag
model. To estimate the spatial lag model consistently, the Generalized 2SLS
and Maximum Likelihood approach (ML) have been proposed (Kelejian and
Prucha, 1998, 1999; Lee, 2003, 2007; Kelejian et al., 2004; Anselin, 2006).
108. Since the spatial error model is a special case of a regression specification with
a non-spherical error variance-covariance matrix, more specifically, the offdiagonal elements are non-zero. OLS estimates remain unbiased whilst the
standard errors are biased. The OLS method can therefore be applied to this
model with the standard errors adjusted to allow for error correlation. The
spatial error model can be consistently estimated by GMM or ML (Kelejian
and Prucha, 1998, 1999; Anselin, 2006).
109. This evidence is preliminary. One might find that countries like Brazil, closer
to Paraguay, have large CDM credit flows. This suggests that, apart from geographic distance, other geographic variables are also important in the process
of CDM development, and so are the institutional variables and financial
variables.
110. Data on the great circle distances are also from Gleditsch et al. (2001).
111. If Moran’s I is greater (smaller) than its expected value, E(I), and/or Gearcy’s
C is smaller (larger) than its expected value, E(C), the overall distribution
of the variable in question can be reflected by positive (negative) spatial
autocorrelation.
112. In this analysis, we also explore the impacts on CDM credit flows of other
geographic factors such as being landlocked, the minimum distance from
one of the three capital-goods-supplying centres (New York, Rotterdam and
Tokyo), mean distance to the nearest coastline or a river navigable to the

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 192 — #10

Notes

193

ocean, the proportion of a country’s total land area with 100 km of the
ocean or such a river, and the proportion of a country’s total land area in
Koeppen-Geiger temperate zones. In general we find no evidence to support
any significant associations between these factors and CDM credit flows.
This may suggest that, as more and more modern technologies have been
employed in the areas of transportation and telecommunications, and more
and more railways, automobiles, air transport and all forms of telecommunications become available, the geographic advantages in terms of easy access
to the sea and/or international trade centres tend to be diminish in the
process of economic development.
113. Under the null of no heteroscedasticity, the test statistic is distributed as
Chi-square with degree of freedom being the total number of the regressors.
114. The spatial weighting matrices, Wn and Mn , are treated as the same.
115. The GS2SLS estimates suggest that the impacts of AREA and EXPPRIM have
been less precisely estimated.

HUANG: “NOTES” — 2010/9/29 — 20:06 — PAGE 193 — #11

Bibliography
Abiad, A. and A. Mody. 2005. Financial reform: What shakes it? What shapes it?
American Economic Review, 95, 66–88.
Acemoglu, D., S. Johnson and J. Robinson. 2001. Colonial origins of comparative development: An empirical investigation. American Economic Review, 91(5),
1369-1401.
Ahn, S.C. and P. Schmidt. 1995. Efficient estimation of models for dynamic panel
data. Journal of Econometrics, 68, 5–28.
Alesina, A., A. Devleeschauwer, W. Easterly and S. Kurlat. 2003. Fractionalization.
Journal of Economic Growth, 8(2), 155–94.
Alonso-Borrego, C. and M. Arellano. 1999. Symmetrically normalised
instrumental-variable estimation using panel data. Journal of Business Economic
Statistics, 17, 36–49.
Anderson, T.W. and C. Hsiao. 1981. Estimation of dynamic models with error
components. Journal of the American Statistical Association, 76, 598–606.
Anderson, T.W. and C. Hsiao. 1982. Formulation and estimation of dynamic
models using panel data. Journal of Econometrics, 18, 47–82.
Andrews, D. W. K. 2005. Cross-section regression with common shocks. Econometrica, 73(5), 1551–85.
Anselin, L. 2006. Spatial Econometrics. In Mills, T. and K. Patterson (eds) Palgrave
Handbook of Econometrics, vol. 1, Econometric Theory, Basingstoke: Palgrave
Macmillan, 901–69.
Anselin, L., A. Bera, R. Florax and M. Yoon. 1996. Simple diagnostic tests for
spatial dependence. Regional Science and Urban Economics, 26 (1), 77–104.
Arellano, M. 1987. Computing robust standard errors for within-group estimators.
Oxford Bulletin of Economic and Statistics, 49, 431–34.
Arellano, M. 2003. Panel Data Econometrics, New York Oxford. University Press:
Advanced Texts in Econometrics.
Arellano, M. and S. Bond. 1991. Some tests of specification for panel data: Monte
carlo evidence and an application to employment equations. Review of Economic
Studies, 58, 277–97.
Arellano, M. and O. Bover. 1995. Another look at the instrumental-variable
estimation of error-components models. Journal of Econometrics, 68, 29–51.
Arraiz, Irani, D. M. Drukker, H. H. Kelejian, I. R. Prucha. 2010. A spatial Cliff-Ordtype model with heteroskedastic innovations: Small and large sample results.
Journal of Regional Science, 50(2), 592–614.
Axelrod, R. 1997. The Dissemination of Culture: A Model with Local Convergence
and Global Polarization. Journal of Conflict Resolution, 41(2), 203–26.
Bagehot, W. 1873. Lombard Street: A Description of the Money Market. London:
Henry S. King.
Bai, J. 2009. Panel data models with interactive fixed effects. Econometrica, 77(4),
1229–79.
Bai, J. and S. Ng. 2002. Determine the number of factors in approximate factor
models. Econometrica, 70, 191–221.
194

HUANG: “REF” — 2010/9/29 — 20:06 — PAGE 194 — #1

Bibliography

195

Bai, J. and S. Ng. 2004. A PANIC attack on unit roots and cointegration.
Econometrica, 72, 1127–77.
Bailliu, J. N. 2000. Private capital flows, financial development, and economic
growth in developing countries. Bank of Canada Working Paper 2000-15.
Baltagi, B. H., P. O. Demetriades and S. H. Law. 2009. Financial development and
openness: Evidence from panel data. Journal of Development Economics, 89(2),
285–96.
Banerjee, A., M. Marcellino and C. Osbat. 2004. Some cautions on the use of panel
methods for integrated series of macro-economic data. Econometrics Journal,
7(2), 322–40.
Beck, T., A. Demirguc-Kunt and R. Levine. 2003. Law, endowment and finance.
Journal of Financial Economics, 70, 37–81.
Bekaert, G., R. H. Campbell and R. Lumsdaine. 2002. Dating the integration of
world equity markets. Journal of Financial Economics, 65(2), 203–47.
Bencivenga, V. R., B. D. Smith. 1991. Financial intermediation and endogenous
growth. Review of Economic Studies, 58, 195–209.
Benhabib, J. and M. M. Spiegel. 2000. The financial development in growth and
investment. Journal of Economic Growth, 5, 341–60.
Bernanke, B. S. and M. Gertler. 1989. Agency costs, net wealth, and business
fluctuations. American Economic Review, 79, 14–31
Bertola, G. and R. J. Caballero. 1994. Irreversibility and aggregate investment.
Review of Economic Studies, 61(2), 223–46.
Blanchard, O. and A. Shleifer. 2000. Federalism with and without political
centralization; China versus Russia. NBER working paper no. 7616.
Blundell, R. and S. Bond. 1998. Initial conditions and moment restrictions in
dynamic panel data models. Journal of Econometrics, 87, 115–43.
Blundell, R. and S. Bond. 2000. GMM estimation with persistent panel data: an
application to production functions. Econometric Reviews, 19, 321–40.
Bond, S. 2002. Dynamic panel data models: a guide to micro data methods and
practice. Portuguese Economic Journal, 1, 141–62.
Bond, S., A. Hoeffler and J. Temple. 2001. GMM estimation of empirical growth
models. CEPR discussion paper no. 3048.
Bowsher, C. G. 2002. On testing overidentifying restrictions in dynamic panel
data models. Economics Letters, 77, 211–20.
Brock, W. and S. Durlauf. 2001. Discrete choice with social interactions. Review of
Economic Studies, 59, 235–60.
Bruno, G. S. F. 2005. Approximating the bias of the LSDV estimator for dynamic
unbalanced panel data models. Economics Letters, 87(3), 361–66.
Bun, M. J. G. and J. F. Kiviet. 2003. On the diminishing returns of higher order
terms in asymptotic expansions of bias. Economics Letters, 79, 145–52.
Caballero, R. J. and R. S. Pindyck. 1996. Uncertainty, investment and industry
evolution. International Economic Review, 37(3), 641–62.
Caselli, F., G. Esquivel and F. Lefort. 1996. Reopening the convergence debate:
a new look at cross-country growth empirics. Journal of Economic Growth, 1(3),
363–90.
Chamberlain, G. 1984. Panel data. Handbook of Econometrics, vol. 2, Chapter 22,
1247–1318, North-Holland: Elsevier Science.
Chinn, M. and H. Ito. 2006. What matters for financial development? Capital
controls, institutions, and interactions. Journal of Development Economics, 81(1),
163–92.

HUANG: “REF” — 2010/9/29 — 20:06 — PAGE 195 — #2

196 Bibliography
Claessens, S., A. Demirgüç-Kunt and H. Huizinga. 1998. How does foreign entry
affect the domestic banking market? World Bank Policy Research Working Paper
Series no. 1918.
Clague, C., P. Keefer, S. Knack and M. Olson. 1996. Property and contract rights
in autocracies and democracies. Journal of Economic Growth, 1, 243–76.
Crafts, N. 2000. Globalization and growth in the twentieth century. IMF working
paper no. 00/44
Demirgüç-Kunt, A. and E. Detragiache. 1998. Financial liberalisation and financial
fragility. World bank Policy Research Working Paper Series no. 1917.
Demirgüç-Kunt, A. and R. Levine. 1996. Stock markets, corporate finance and
economic growth: an overview. World Bank Economic Review, 10(2), 223–40.
Demirgüç-Kunt, A. and R. Levine. 1999. Bank-based and market-based financial
systems: Cross-country comprisions. World Bank Policy Research working paper
no. 2143.
Diamond, D.W. 1984. Financial intermediation and delegated monitoring. Review
of Economic Studies, 51, 393–414.
Diamond, D.W. and P. Dybvig. 1983. Bank runs, deposit insurance and liquidity.
Journal of Political Economy, 91(3), 401–19.
Diamond, J. 1997. Guns, germs, and steel: The fates of human societies. W. W. Norton,
New York, NY.
Dixit, A. 1989. Entry and exit decisions under uncertainty. Journal of Political
Economy, 97, 620–38.
Djankov, S., C. McLiesh, T. Nenova and A. Shleifer. 2003. Who owns the media?
Journal of Law and Economics, 46(2), 341–82.
Do, Q. T. and A. A. Levchenko. 2004. Trade and financial development. World
Bank Working Paper Series no. 3347.
Dollar, D. and A. Kraay. 2003. Institutions, trade and growth. Journal of Monetary
Economics, 32, 459–83.
Doms, M. and T. Dunne. 1998. Capital adjustment patterns in manufacturing
plants. Review of Economic Dynamics, 1, 409–29.
Easterly, W. and R. Levine. 2003. Tropics, germs, and crops: How endowments
influence economic development. Journal of Monetary Economics, 50, 3–39.
Engle, R. F. and C. W. J. Granger. 1987. Cointegration and error correction:
representation, estimation and testing. Econometrica, 55, 251–76.
Fernandez, C., E. Ley and M. Steel. 2001. Model uncertainty in cross-country
growth regressions. Journal of Applied Econometrics, 16, 563–76.
Fernandez, R. and D. Rodrik. 1991. Resistance to Reform: Status Quo Bias in the
Presence of Individual-Specific Uncertainty. American Economic Review, 81(5),
1146–55.
Frankel, J. A. and A. K. Rose. 1998. The endogeneity of the optimum currency
area criteria. Economic Journal, 108, 1009–25.
Frankel, J. A. and D. Romer. 1999. Does trade cause growth? American Economic
Review, 89, 379–98.
Freedom House. 2008. Freedom House Country Survey, available at
www.freedomhouse.org.
Gale, D. H. M. 1985. Incentive-compatible debt contracts: The one-period
problem. Review of Economic Studies, 52, 647–64.
Gale, D. and M. Hellwig. 1985. Incentive-compatible debt contracts: The oneperiod problem. Review of Economic Studies, 52, 647–64.

HUANG: “REF” — 2010/9/29 — 20:06 — PAGE 196 — #3

Bibliography

197

Gallup, J. L., J. D. Sachs and A. Mellinger. 1999. Geography and economic
development. CID at Harvard working paper no. 1.
Gengenbach, C. F. C. Palm and J. Urbain. 2005. Panel cointegration testing in
the presence of common factors. Maastricht Research School of Economics of
Technology and Organization Research Paper no. 050.
Giavazzi, F. and G. Tabellini. 2004. Economic and political liberalization. NBER
working paper no.10657.
Girma, S. and A. K. Shortland. 2008. The political economy of financial
development. Oxford Economic Papers, 60(4), 567–96.
Gleditsch, K. S. and M. D. Ward. 2001. Measuring space: A minimum-distance
database and applications to international studies. Journal of Peace Research, 38,
749–68.
Granger, C. W. J. and D. F. Hendry. 2005. A dialog concerning a new instrument
for econometric modeling. Econometrics Theory, 21(1), 278–97.
Greenwood. J. and B. Jovanovic. 1990. Financial development, growth, and the
distribution of income. Journal of Political Economy, 98, 1076–1107.
Greenwood, J. and B. D. Smith. 1997. Financial markets in development, and the
development of financial markets. Journal of Economic Dynamics and Control, 21,
145–81.
Gregory, A. W., A. C. Head and J. Raynauld. 1997. Measuring world business
cycles. International Economic Review, 38, 677–701.
Hall, R. and C. Jones. 1999. Why do some countries produce so much more output
per worker than others? Quarterly Journal of Economics, 114, 83–116.
Hansen, B. E. 2005. Challenges for econometric model selection, Econometrics
Theory, 21(1), 60–68.
Haveman, J., International trade data, Available at www.eiit.org.
Hendry, D. F. 1995. Dynamic Econometrics, Oxford, Oxford University Press.
Hendry, D. F. and H. Krolzig. 2005. The properties of automatic GETS modelling.
Economic Journal, 115(502), 32–61.
Henisz, W. J. 2000. The institutional environment for economic growth. Economics
and Politics, 12(1), 1–31.
Heston, A, R. Summers and B. Aten. 2006. Penn World Table version 6.2, Center
for International Comparisons at the University of Pennsylvania (CICUP)
Goldsmith, R. W. 1969. Financial Structure and Development, New Haven, CT: Yale
University Press.
Hicks, J. R, 1969. Automatists, Hawtreyans, and Keynesians. Journal of Money,
Credit and Banking, 1(3), 307–17.
Hoeting, J., A. Raftery and D. Madigan. 1996. A method for simultaneous variable
selection and outlier identification in linear regression. Computational Statistics
and Data Analysis, 22, 25–70.
Holtz-Eakin, D, W. Newey and H. S. Rosen. 1988. Estimation vector autoregressions with panel data. Econometrica, 56(6), 1371–95.
Hoover, K. D. and S. J. Perez. 1999. Data mining reconsidered: Encompassing and
the general-to-specific approach to specification search. Econometrics Journal, 2,
167–91.
Huybens, E. and B. D. Smith. 1999. Inflation, financial markets and long-run real
activity. Journal of Monetary Economics, 43: 283–315.
Huang, Y. and J. Temple. 2005. Does external trade promote financial development? CEPR working paper no. 5150.

HUANG: “REF” — 2010/9/29 — 20:06 — PAGE 197 — #4

198 Bibliography
Im, K., H. Pesaran and Y. Shin. 2003. Testing for unit roots in heterogeneous
panels. Journal of Econometrics, 115, 53–74.
Imbs, J. 2003. Trade, finance, specialization and synchronization. IMF working
paper no. 03/81
IMF, 2003. World Economic Outlook.
Isham, J., M. Woolcock, L. Pritchett and G. Busby. 2005. The varieties of resource
experience: How natural resource export structures affect the political economy
of economic growth. World Bank Economic Review, 19(2), 141–74.
Islam, N. 1995. Growth empirics: A panel data approach. Quarterly Journal of
Economics, 110, 1127–70.
Jaffee, D. and M. Levonian. 2001. Structure of banking systems in developed and
transition economies. European Financial Management, 7(2), 161–81.
Judson, R. A. and A. L. Owen. 1999. Estimating dynamic panel data models: A
guide for macroeconomists. Economics Letters, 65, 9–15.
Kamarck, A. M. 1976. The Tropics and Economic Development, Baltimore, MD: John
Hopkins University Press.
Kaufmann, D., A. Kraay and M. Mastruzzi. 2008. Governance Matters VII: Governance Indicators for 1996–2007. World Bank Policy Research Working Paper
no. 4654.
Kelejian, H. H. and I. R. Prucha and Y. Yuzefovich. 2004. Instrumental variable estimation of a spatial autoregressive model with autoregressive disturbances: Large
and small sample results. In J. LeSage and K. Pace (eds.) Advances in Econometrics:
Spatial and Spatiotemporal Econometrics. NewYork: Elsevier, 63–198.
Kelejian, H. H. and I. R. Prucha. 1998. A generalized spatial two-stage least squares
procedure for estimating a spatial autoregressive model with autoregressive
disturbance. Journal of Real Estate Finance and Economics, 17, 99–121.
Kelejian, H. H. and I. R. Prucha. 1999. A generalized moments estimator for the
autoregressive parameter in a spatial model. International Economic Review, 40,
509–33.
Kelejian, H. H. and I. R. Prucha. 2010. Specification and estimation of spatial
autoregressive models with autoregressive and heteroskedastic disturbances.
Journal of Econometrics, 157(1), 53–67.
Kim, S. H., M. A. Kose and M. Plummer. 2003. Dynamics of business cycles in
Asia. Review of Development Economics, 7, 462–77.
King, R. and R. Levine. 1993. Finance and growth: Schumpeter might be right.
Quarterly Journal of Economics, 108(3), 717–37.
Kiviet, J. F. 1995. On bias, inconsistency, and efficiency of various estimators in
dynamic panel data models. Journal of Econometrics, 68, 53–78.
Koenker, R. 1981. A note on studentizing a test for heteroskedasticity. Journal of
Econometrics, 17, 107–12.
Kose, M. A., C. Otrok and C. H. Whiteman. 2003. International business cycles:
world, region, and country-specific factors. American Economic Review, 93,
216–39.
Kose, M. A., E. S. Prasad and M. E. Terrones. 2003. Volatility and comovement in
a globalized world economy: an empirical exploration. IMF working paper no.
03/246
Krolzig, H. M. and D. F. Hendry. 2001. Computer automation of general-to-specific
model selection procedures. Journal of Economic Dynamics and Control, 25,
831–66.

HUANG: “REF” — 2010/9/29 — 20:06 — PAGE 198 — #5

Bibliography

199

Kruiniger, H. 2008. Maximum likelihood estimation and inference methods for
the covariance stationary panel AR(1)/unit root model. Journal of Econometrics,
144(2), 447–64.
Laeven, L. 2000. Does Financial Liberalization Reduce Financial Constraints?
World Bank Policy Research Working Paper no. 2435.
La Porta, R., F. Lopez-de-Silanes, A. Lopez-de-Silanes and R. W. Vishny. 1997. Legal
determinants of external finance. Journal of Finance, 52(3), 1131–50.
La Porta, R., F. Lopez-de-Silanes, A. Lopez-de-Silanes and R. W. Vishny. 1998. Law
and finance. Journal of Political Economy, 106, 1113–55.
Leahy, J. 1993. Investment in competitive equilibrium: the optimality of myopic
behaviour. Quarterly Journal of Economics, 108, 1105–33.
Leamer, E. E. 1983. Let’s take the con out of econometrics. American Economic
Review, 73, 31–43.
Leamer, E. E. 1985. Sensitivity analysis would help. American Economic Review, 75,
308–13.
Lee, L. 2003. Best spatial two-stage least squares estimators for a spatial autoregressive model with autoregressive disturbances. Econometric Reviews, 22, 307–35.
Lee, L. 2007. GMM and 2SLS estimation of mixed regressive, spatial autoregressive
models. Journal of Econometrics, 137, 489–514.
Lee, L. and X. Liu. 2006. Efficient GMM estimation of a spatial autoregressive model with autoregressive disturbances. Working paper, Department of
Economics, Ohio State University.
Levine, R. 1997. Financial development and economic growth: views and agenda.
Journal of Economic Literature, 35, 688–726.
Levine, R. 2003. More on finance and growth: More Finance, more growth? The
Federal Reserve Bank of St. Louis. July/August.
Levine, R. 2005. Finance and growth: Theory and evidence. In: P. Aghion and S.
N. Durlauf (eds.), Handbook of Economic Growth, North-Holland: Elsevier.
Levine, R., N. Loayza and T. Beck. 2000. Financial intermediation and growth:
causality and causes. Journal of Monetary Economics, 46(1), 31–77.
Levine, R. and D. Renelt. 1992. A sensitivity analysis of cross-country growth
regressions. American Economic Review, 82, 942–63.
Maddala, G. S. and S. Wu. 1999. A comparative study of unit root tests with panel
data and a new simple test. Oxford Bulletin of Economics and Statistics, 61, 631–52.
Malik, A. and J. R. W. Temple. 2009. The geography of output volatility. Journal of
Development Economics, 90(2), 163–78.
Mankiw, N. G., D. Romer and D. N. Weil. 1992. A contribution to the empirics of
economic growth. Quarterly Journal of Economics, 107, 407–37.
Manski, C. F. 2000. Economic analysis of social interactions. Journal of Economic
Perspectives, 14(3), 115–36.
Marshall, M.G. and Jaggers, K. 2008. Polity IV Project Country Reports, College
Park, College Park CIDUM, University of Maryland.
Marshall, M. G. and K. Jaggers. 2009. Polity IV Project Country Reports, College
Park CIDUM, University of Maryland.
Mayer, C. and O. Sussman. 2001. The assessment: finance, law and growth. Oxford
Review of Economic Policy, 17( 4), 457–66.
McKinnon, R. I. 1973. Money and Capital in Economic Development. Washington
D.C. Brookings Institution

HUANG: “REF” — 2010/9/29 — 20:06 — PAGE 199 — #6

200 Bibliography
Minier, J. A. 1998. Democracy and growth: Alternative approaches. Journal of
Economic Growth, 3(3), 241–66.
Moon, H. R. and B. Perron. 2004. Testing for a unit root in panels with dynamic
factors. Journal of Econometrics, 122, 81–126.
Ndikumana, L. 2000. Financial determinants of domestic investment in SubSaharan Africa: evidence from panel data. World Development, 28(2), 381–400.
Ndikumana, L. 2005. Financial development, financial structure, and domestic
investment: interanatioal evidence. Journal of International Money and Finance,
24, 651–73.
Newey, W. and F. Windmeijer. 2009. Generalized Method of Moments with many
weak moment conditions. Econometrica, 77(3), 687 - 719.
Nickel, S. J. 1981. Biases in dynamic models with fixed effects. Econometrica, 49,
1417–26.
Olson, M. 1993. Dictatorship, democracy and development. American Political
Science Review, 87, 567–76.
Pagano, M. and P. Volpin. 2001. The political economy of finance. Oxford Review
of Economic Policy, 17(4), 502–19.
Papaioannou, E. and G. Siourounis. 2008. Democratization and growth. Economic
Journal, 118(532), 1520–51.
Papke, L. E. and J. M. Wooldridge. 2005. A computational trick for delta-method
standard errors. Economic Letters, 86, 413–17.
Pedroni, P. 1999. Critical values for cointegration tests in heterogeneous panels
with multiple regressors. Oxford Bulletin of Economics and Statistics, 61, 653–70.
Pedroni, P. 2004. Panel cointegration, asymptotic and finite sample properties of
pooled time series tests with an application to the PPP hypothesis. Econometric
Theory, 20, 597–625.
Persson, T. 2005. Forms of democracy, policy and economic development. NBER
working paper no. 11171.
Persson, T. and G. Tabellini. 1992. Growth, income distribution and democracy.
European Economic Review, 36, 593–602.
Pesaran, M. H. 2004. General diagnostic tests for cross section dependence in
panels. CESifo Working Papers no. 1233.
Pesaran, M. H. 2006. Estimation and inference in large heterogeneous panels with
a multifactor error structure. Econometrica, 74 (4), 967–1012.
Pesaran, M. H. 2007. A simple panel unit root test in the presence of cross section
dependence. Journal of Applied Econometrics, 22(2), 265–312.
Pesaran, M. H., Y. Shin and R. P. Smith. 1999. Pooled mean group estimation of
dynamic heterogeneous panels. Journal of American Statistical Association, 94,
621–34.
Pesaran, M. H. and R. P. Smith. 1995. Estimating long-run relationships from
dynamic heterogeneous panels. Journal of Econometrics, 68, 79–113.
Phillips, P. C. B. and D. Sul. 2003. Dynamic Panel Testing and Homogeneity
Testing under Cross Section Dependence. Econometrics Journal, 6, 217–59.
Quinn, D. P. and C. Inclán. 1997. The origins of financial openness: A study of
current and capital account liberalization. American Journal of Political Science,
41(3), 771–813.
Raftery, A. E. 1995. Bayesian model selection in social research. In: P.V. Marsden
(ed.), Sociological Methodology, Cambridge: Blackwell, 111–96.

HUANG: “REF” — 2010/9/29 — 20:06 — PAGE 200 — #7

Bibliography

201

Raftery, A. E., D. Madigan and J. E. Hoeting. 1997. Bayesian model averaging for
linear regression models. Journal of American Statistical Association, 92, 179–91.
Rajan, R. G. and L. Zingales. 1998. Financial dependence and growth. American
Economic Review, 88(6), 559–86.
Rajan, R. G. and L. Zingales. 2003. The great reversals: The politics of financial
development in the twentieth century. Journal of Financial Economies, 69, 5–50.
Reynal-Querol, M. and J. G. Montalvo. 2005. Ethnic polarization, potential
conflict and civil war. American Economic Review, 95(3), 796–816.
Rodrik, D., A. Subramanian and F. Trebbi. 2004. Institutions rule: the primacy of
institutions over geography and integration in economic development. Journal
of Economic growth, 9, 131–65.
Rodrik, D. and R. Wacziarg. 2005. Do democratic transitions produce bad
economic outcomes? American Economic Review, 95(2), 50–55.
Sachs, J. D., 2003a. Institutions don’t rule: Direct effects of geography on per
capita income. NBER working paper no. 9490.
Sachs, J. 2003b. Institutions matter, but not for everything. Finance and Development, 40(2).
Sachs, J., and A. Warner. 1995a. Economic reform and the process of global
integration. Brookings Papers on Economic Activity, 1, 1–95.
Sachs, J. D. and A. M. Warner, 1995b. Natural resource abundance and economic
growth. NBER Working Paper no. W5398.
Sachs, J. and A. Warner. 1997. Fundamental sources of long-run growth. American
Economic Review Papers and Proceedings, 87, 184–188.
Saint-Paul, G. 1992, Technological Choice, Financial Markets and Economic
Development. European Economic Review, 36, 763–81.
Sala-i-Martin, X. 1997a. I have just run two million regressions. American Economic
Review, 87(2), 178–83.
Sala-i-Martin, X. 1997b. I have just run four million regressions. NBER Working
Paper No. W6252.
Sala-i-Martin, X., G. Doppelhofer and R. Miller. 2004. Determinants of long-run
growth: A Bayesian Averaging of Classical Estimates (BACE) approach. American
Economic Review, 94(4), 813–35.
Schich, S. and F. Pelgrin. 2002. Financial development and investment: panel data
evidence for OECD countries from 1970–1997. Applied Economics Letters, 9, 1–7.
Schumpeter, J. A. 1911. The Theory of Economic Development, Cambridge MA,
Harvard University Press.
Shaw, E. S. 1973. Financial Deepening in Economic Development, New York, Oxford
University Press
Simmons, B. A. and Z. Elkins. 2004. The globalization of liberalization: Policy diffusion in the international political economy. American Political Science Review,
98(1), 171–89.
Stulz, R. M. M. and R. G. Williamson, 2003. Culture, openness, and finance.
Journal of Financial Economics, 70, 313–49.
Summers, L. H. 2000. International Financial Crises: Causes, Prevention, and
Cures. American Economic Review Papers and Proceedings, 90(2), 1–16.
Tavares, J. and R. Wacziarg. 2001. How democracy affects growth. European
Economic Review, 45(8), 1341–78.
Townsend, R. M. 1979. Optimal contracts and competitive markets with costly
state verification. Journal of Economic Theory, 21, 265–93.

HUANG: “REF” — 2010/9/29 — 20:06 — PAGE 201 — #8

202 Bibliography
UNEP Risoe Centre, CDM/JI Pipeline Analysis and Database (2008).
White, H. 1980. A Heteroskedasticity-consistent covariance matrix estimator and
a direct test for heteroskedasticity. Econometrica, 48, 817–38.
Williamson, S. D. 1986. Costly monitoring, financial intermediation, and
equilibrium credit rationing. Journal of Monetary Economics, 18, 159–79.
Williamson, S. D. 1987. Costly monitoring, loan contracts, and equilibrium credit
rationing. Quarterly Journal of Economics, 102, 135–45.
Windmeijer, F. 2005. A finite sample correction for the variance of linear efficient
two-step GMM estimators. Journal of Econometrics, 126(1), 25–51.
World Bank. 1989. World Development Report. New York, Oxford University Press.
World Bank. 2002. Global Development Network.
World Bank. 2008. Financial Development and Financial Structure Database.

HUANG: “REF” — 2010/9/29 — 20:06 — PAGE 202 — #9

Index
accounting practices, 3, 103
adverse selection, 2
aggregate private investment, see
private investment
Arrow–Debreu framework, 2
Assigned Amount Units (AAUs),
161, 162
asymmetric information, 2, 64
autocracy, 17, 101, 104, 188n68

banking crises, 127, 150
banking markets
efficiency measures, 15
opening of, 5–6
Bayesian Model Averaging (BMA), 8,
10, 12, 21–2, 29, 30, 48–9, 125
black market premium, 18, 104–5,
107, 114, 116, 121, 188n64
bond market development, 46, 50,
62–3, 183n4
Brazil, 11, 162
business cycle co-movement, 66, 67

Canada, 11
cap-and-trade regimes, 161
capital account openness, 5, 127,
144, 151
capital flows, 185n33
carbon markets, 3, 9, 161–80
Chile, 10–11
China, 162
Chinn–Ito index, 127
civil law, 17, 25, 47, 118, 192n105
civil liberties, 17, 103
Clean Development Mechanism
(CDM) markets, 8, 9, 161–82,
192n112
colonization, 4, 47, 103
Commercial-Central Bank (BTOT), 15,
68, 106

common correlated effect pooled
(CCEP) approach, 126–7, 134, 139
common factor approach, 78–81
common law, 4, 17, 25, 47, 188n60,
192n105
contract enforcement, 3, 4, 47,
187n59, 188n60
countries
civil law, 17, 25, 47, 118, 192n105
common law, 4, 17, 25, 47, 188n60
cross-section dependence
across, 67
developing, 101–2
landlocked, 6, 19
creditors’ rights, 17, 103
culture, 7, 11, 20, 47

democracy, 9, 17, 101, 102, 104,
126–7, 149, 188n68
democracy index, 25
democratization, 102–4, 109–21, 149,
187n59, 189n78
deposit insurance, 103
deposit rate of interest, 2
developing countries
financial development in, 101
institutional reform in, 101–2

economic growth
determinants of, 65
financial development and, 1–3, 7
geography and, 162–3
elite groups, 104
endogenous growth models, 2
economic theory, 65
ethnicity, 20, 118, 192n104
EU Allowances (EUAs), 161
EU Emissions Trading Scheme (EU
ETS), 161
extractive colonizers, 4, 103
extreme bounds analysis, 12
203

HUANG: “INDEX” — 2010/9/29 — 20:06 — PAGE 203 — #1

204 Index
financial depth, 13, 36–46
financial development
carbon markets and, 161–80
cross-country differences in, 8,
10–63
democratization and, 102–4,
109–21, 149, 187n59, 189n78
determinants of, 3–7, 10–63
in developing countries, 101
financial liberalization and, 125–60
government reforms and, 8–9
indicators, 68
institutional improvement and, 8
measures of, 14–16
political institutions and, 101–24
private investment and, 64–100
role of, in economic growth, 1–3
financial development
determinants, 10–63
conclusions about, 46–8
data on, 13–20
descriptive statistics, 54–5
empirical results, 24–46
empirical strategy, 20–4
geographic variables, 19
institutional variables, 17–18
policy variables, 18
potential, 16–20
samples, 14
financial efficiency
development, 36–7, 42–3, 46, 50
financial intermediaries, 2, 64, 185n28
financial intermediary
development, 36–9, 50, 65, 66
financial liberalization, 2, 5–6
conclusions about, 149–50
criticism of, 125
democratization and, 149
empirical evidence on, 133–44
factors stimulating, 125–7
financial development and, 125–60
methodology of study, 127–33
study results, 144–9
variables, 141–52
financial market analysis, 1–2

financial markets
development of, 3
integration of, 66–7
financial openness, see financial
liberalization
financial repression, 2, 5
financial size development, 36–7,
44–5, 50
financial systems, 1, 46
functions of, 2
repressive policies toward, 2
France, 10
Frankel–Romer trade share, 26
French Civil Law, 4, 47, 118, 120,
188n60

GDP per capita, 20, 29, 47
Generalized Spatial Two-Step Least
Square (GS2SLS) estimator, 170
General-to-specific (Gets) approach, 8,
10, 12, 22–4, 31, 60–1, 125
geographic determinants, of carbon
markets, 161–80
geographic variables, 53
geography
economic development and, 162–3
role of, in financial
development, 6–7, 9, 11–13, 19,
27, 29, 31, 33–5, 47–8
globalization, 66, 163, 181
global shocks, 66, 67, 70, 72, 77, 168
global warming, 161
GMM estimation, 8, 67, 69–73, 108–9
government quality, 17
government reforms, 8–9, 125–7 see
also financial liberalization

income levels, 7, 11, 29, 47
India, 162
industrial revolution, 1
inflation rates, 2, 5, 11, 127, 150
information asymmetry, 2, 64
information disclosure, 103
institutional improvement, 8
conclusions about, 121–2
evidence on, 109–21
measures and data on, 104–9

HUANG: “INDEX” — 2010/9/29 — 20:06 — PAGE 204 — #2

Index
institutional improvement – continued
methodology of study, 106–9
variables, 123–4
institutional variables, 17–18, 51–2
institutions
democratization and, 102–4
political, and financial
development, 101–24
role of, in financial
development, 3–5, 12–13, 17–18,
24–9, 31, 33–5, 46–8
interest groups, 104
interest rates, 1–2, 150
investment, see private investment
investor rights, 192n105
Joint Implementation (JI)
schemes, 161
Kyoto Protocol, 161, 162, 178
landlocked countries, 6, 19
language, 20, 47, 192n104
Latin America, 10–11, 162
latitude, 6, 11, 19
Least Square Dummy Variable
(LSDV), 102
legal origin theory, 4
legal system, 3–4, 10, 11, 17, 25, 47,
103, 118, 188n60, 192n105
Liquid Liabilities (LLY), 14–15, 24–5,
28, 68, 105, 125
logarithm of the real GDP per capita
(LGDP), 107
macroeconomic policy, 5–6,
10–11, 18
macroeconomic shocks, 168
Markov Chain Monte Carlo
technique, 184n20
McKinnon–Shaw model, 2
media, state-owned, 20
Mexico, 11, 162
microeconomic lumpiness, 185n30
model uncertainty problem, 12, 20–1
moral hazard, 2

205

natural resources, 6–7, 19, 163,
167, 178
Net Interest Margin (NIM), 15, 188n66
new political economy, 4–5, 103–4

Occam’s Window, 184n13, 184n14,
184n15
Ordinary Least Squares (OLS)
technique, 107
Overhead Costs (OVC), 15, 188n66

panel cointegration tests, 84–5
panel unit root tests, 81–3
PcGets, 183n10
policy
macroeconomic policy, 10–11
role of, in financial
development, 5–6, 8–9, 11–13, 29,
31, 33–5, 46–8
policy variables, 18, 50–1
political institutions
evidence of effect, on financial
development, 109–21
financial development and, 101–24
measures and data on, 104–9
methodology of study, 106–9
variables, 123–4
political liberalization, 3, 8, 101
political rights, 17
political stability, 17
politics, 11
posterior inclusion probabilities
(PIPs), 29
posterior model probabilities, 48–9
“premature” democracy, 103
principle component
analysis, 185n35, 186n37
Private Credit (PRIVO), 15, 68,
105–6
private credit to GDP ratio, 10,
11, 101
private investment, 3, 8, 64–100,
185n33
analysis on annual data, 77–92

HUANG: “INDEX” — 2010/9/29 — 20:06 — PAGE 205 — #3

206 Index
private investment – continued
analysis on data for five-year
averages, 69–77
data on, 67–9
determinants of, 65
financial development and, 64–100
financial intermediaries and,
64–5, 66
property rights, 3, 4, 10, 103, 187n59
regime change, 104
regulatory system, 3–4, 103–4
religion, 7, 20, 47, 192n104
reserve requirements, 2
resource endowment, 6–7, 19, 163,
167, 178
savings rate, 7, 11, 47
settler colonizers, 4, 103
settler mortality hypothesis, 4, 11,
47, 103
shareholders’ rights, 17, 103
Solow–Swan growth model, 65

spatial error model, 192n108
state-owned media, 20
stock market capitalization, 125
Stock Market Capitalization
(MCAP), 15
stock market capitalization to GDP
ratio, 10, 11
stock market development, 15, 36–7,
40–1, 50
stock markets, opening of, 6

technological shocks, 168
Total Value Traded (TVT), 15
trade factors, 18
trade liberalization, 66
trade openness, 11, 107
trade policy, 24, 26
trade volume, 185n33
tropical locations, 6, 19, 163
Turnover Ratio (TOR), 15

United Kingdom, 10

HUANG: “INDEX” — 2010/9/29 — 20:06 — PAGE 206 — #4