Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Essay on monetary policy, macroprudential policy, and financial integration
(USC Thesis Other)
Essay on monetary policy, macroprudential policy, and financial integration
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Essay on Monetary Policy, Macroprudential Policy, and
Financial Integration
by
Zhou Yu
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Economics)
May 2021
Copyright 2021 Zhou Yu
I dedicate this thesis to macroeconomics and finance,
for their never ending mysteries.
ii
Acknowledgements
The author acknowledges the financial support of USC Dornsife and Econ Department
during the PhD career.
I would like to thank my committee cochairs Dr. Joshua Aizenman and Dr. Caroline
Betts. Dr. Aizenman has been very helpful and supportive since I started my research.
During my PhD career, he helped me determine my research area and provided a lot of
valuable opinions on my research so that I would like to express my sincere thanks to him.
Dr. Caroline Betts provided a lot of comments on my research and I also learned a lot from
the talk with her. I am super grateful for her help and support.
I would like to thank my committee member Dr. Wayne Ferson from USC Marshall
for his constant support. Dr. Wayne Ferson was my professor in asset pricing class and I
especially want to thank him for agreeing to join my qualifying and dissertation committee
as external member.
I would like to thank my qualifying exam committee members Dr. Robert Dekle and
Dr. Monica Morlacco for their help on my preparation of my qualifying exam defense.
I would like to thank the staffs in USC Econ department Young Miller, Alexander
Karnazes and Morgan Ponder for their help in the past few years.
Finally, I would like to thank all of my friends. Weizhao Huang helped me a lot
during the COVID-19 and helped me in data processing where I am very grateful to him.
I am also very grateful to Yu Cao for helping me in the format and process guidance
of the dissertation. I especially want to thank my friend Andrew Yimeng Xie for his
continuous support and help in my research process. Richard Yejia Xu and Chris Zhen
Chen and I are the only three Chinese students who entered USC in 2016. They have been
iii
helping me a lot in the past few years and I am super grateful. I also want to thank my
friends Rashad Ahmed, Grigory Franguridi, Usman Ghaus, Rachel Lee, Tal Roitberg, Jacob
Schneider, Chris Jeong Yoo, Nicolas Roig, Bada Han, Qin Jiang, Yinan Liu, Yiwei Qian,
Lidan Tan, Yinqi Zhang, Weining Xin, and Jingbo Wang for their help in the past few years.
Meanwhile, I am very grateful to my friend Haoqing Li for her help in summarizing the
policies in China. I especially want to thank Dr. Yuchen Wu in CITICS for his continuous
help in my global macro research and my job seeking in the industry.
iv
Table of Contents
Dedication ii
Acknowledgements iii
List of Tables viii
List of Figures x
Abstract xiii
Chapter 1: Introduction 1
Chapter 2: From SHIBOR to DR: Chinese Monetary Policy Transmission and Bench-
mark Interest Rate Reform 10
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Econometric Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.1 Interest Rate Transmission in Money Market . . . . . . . . . . . . . 26
2.4.2 Interest Rate Transmission in Bond Market . . . . . . . . . . . . . . 31
2.4.2.1 Interest Rate Transmission in NCD Yield . . . . . . . . . . 31
2.4.2.2 Interest Rate Transmission in Government Bond and CDB
Bond Yield . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4.2.3 Interest Rate Transmission in Three Factors of Government
Bond and CDB Bond Yield Curve . . . . . . . . . . . . . . . 40
2.4.2.4 Interest Rate Transmission in Enterprise Bond Yield . . . . 48
2.4.2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.4.3 Interest Rate Transmission in Credit Market . . . . . . . . . . . . . . 59
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Chapter 3: Monetary Policy and Macroprudential Policy: Evidence from China 71
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.2.1 The Response Function of Chinese Monetary Policy . . . . . . . . . 78
3.2.2 Effectiveness of Chinese Monetary Policy . . . . . . . . . . . . . . . 80
3.2.3 Relationship between Chinese Monetary Policy and Other Policies . 81
v
3.2.4 Measures of Chinese Monetary Policy . . . . . . . . . . . . . . . . . 82
3.2.5 Our Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.3 Econometric Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4 Construction of Policy Measures . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.4.1 Construction of Monetary policy Index (MPI) . . . . . . . . . . . . . 91
3.4.2 Construction of Macroprudential Policy Measure . . . . . . . . . . . 94
3.5 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.5.1 Empirical Results on Economic Growth . . . . . . . . . . . . . . . . 98
3.5.1.1 TVP-SV-V AR Model Tests . . . . . . . . . . . . . . . . . . . 99
3.5.1.2 The Empirical Results of TVP-SV-V AR Model . . . . . . . . 103
3.5.2 Empirical Results on Inflation . . . . . . . . . . . . . . . . . . . . . . 112
3.5.3 Empirical Results on Asset Prices . . . . . . . . . . . . . . . . . . . . 114
3.6 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.6.1 Households . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.6.2 Final Goods Sector . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.6.3 Intermediate Goods Sector . . . . . . . . . . . . . . . . . . . . . . . . 122
3.6.4 Banks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.6.5 Foreign Investors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.6.6 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.6.7 Analytical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.6.8 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.6.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Chapter 4: A Welfare Analysis of Financial Integration in a Risky World with Fric-
tions 132
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.3 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.3.1 Empirical Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.3.3 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.4 The Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.4.1 Benchmark Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.4.1.1 Production . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
4.4.1.2 Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
4.4.1.3 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
4.4.2 Improved Model with Frictions being Removed . . . . . . . . . . . . 167
4.4.2.1 Introduction of a Frictional World . . . . . . . . . . . . . . 168
4.4.2.2 Frictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4.4.2.3 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
4.5 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
4.6 Quantitative Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.6.1 Steady State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
4.6.2 Financial Autarky . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
vi
4.6.3 The Effect of Capital Scarcity and Risk Sharing . . . . . . . . . . . . 175
4.6.4 The Effect of Frictions . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.7 Welfare Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
4.7.1 Definition of Welfare Gains . . . . . . . . . . . . . . . . . . . . . . . 184
4.7.2 Welfare Gains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
4.7.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4.8 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
4.8.1 The Role of Country Size . . . . . . . . . . . . . . . . . . . . . . . . . 190
4.8.2 The Role of Capital Scarcity . . . . . . . . . . . . . . . . . . . . . . . 192
4.8.3 The Role of FDI Productivity . . . . . . . . . . . . . . . . . . . . . . 193
4.8.4 The Role of Speed of Frictions Being Removed . . . . . . . . . . . . 194
4.8.5 The Role of Observation Time . . . . . . . . . . . . . . . . . . . . . . 195
4.9 Extended Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
4.9.1 Model Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
4.9.1.1 Production . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
4.9.1.2 Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.9.1.3 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
4.9.2 Three-country Extended Model . . . . . . . . . . . . . . . . . . . . . 204
4.9.2.1 Production . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
4.9.2.2 Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4.9.2.3 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
4.9.3 Multi-country Extended Model . . . . . . . . . . . . . . . . . . . . . 211
4.9.3.1 Production . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
4.9.3.2 Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
4.9.3.3 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
4.9.4 Welfare Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
4.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
Chapter 5: Conclusion 232
References 238
Appendices 243
A Equations of Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
B Country List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
B.1 Country List for Empirical Analysis . . . . . . . . . . . . . . . . . . 247
B.2 Country List for Theoretical Analysis . . . . . . . . . . . . . . . . . . 248
vii
List of Tables
2.1 List of banks in the database . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.2 Variables used in the interest rate transmission analysis in credit market . 63
2.3 Regression results of lending interest rate to DR007 and SHIBOR 1W . . . 64
2.4 Regression results of debt interest rate to DR007 and SHIBOR 1W . . . . . 66
3.1 Comparison between our research and Klingelh¨ ofer and Sun (2019) . . . . 77
3.2 Result of Principle Component Analysis (PCA) . . . . . . . . . . . . . . . . 93
3.3 Coefficients of Principle Components . . . . . . . . . . . . . . . . . . . . . . 94
3.4 Parameter estimation result in TVP-SV-V AR model . . . . . . . . . . . . . . 100
3.5 Parameter calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.1 Tests for remaining nonlinearity . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.2 Determination of the number of location parameters . . . . . . . . . . . . . 154
4.3 Parameter estimates for the final PSTR model . . . . . . . . . . . . . . . . . 155
4.4 Parameters values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.5 Steady state values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
4.6 Welfare gains of financial integration 1 . . . . . . . . . . . . . . . . . . . . . 186
4.7 Welfare gains of financial integration 2 . . . . . . . . . . . . . . . . . . . . . 187
4.8 Welfare gains of financial integration for different country sizes . . . . . . . 191
4.9 Welfare gains of financial integration for different degrees of capital scarcity192
4.10 Welfare gains of financial integration for different degrees of capital scarcity194
4.11 Welfare gains of financial integration for different speed of frictions being
removed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
4.12 Welfare gains of financial integration for different observation time . . . . 196
viii
4.13 Welfare gains of financial integration in extended model 1 . . . . . . . . . . 223
4.14 Welfare gains of financial integration in extended model 2 . . . . . . . . . . 224
ix
List of Figures
1.1 Roadmap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 Correlation matrix of OMO rate and money market interest rates . . . . . . 25
2.2 Impulse response of short-term money market interest rates in the interbank
market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Impulse response of short-term money market interest rates in the exchange
market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Impulse response of interbank lending interest rates . . . . . . . . . . . . . 31
2.5 Impulse response of NCD yield to DR007 . . . . . . . . . . . . . . . . . . . 32
2.6 Impulse response of NCD yield to SHIBOR 1W . . . . . . . . . . . . . . . . 33
2.7 Impulse response of NCD yield to MLF . . . . . . . . . . . . . . . . . . . . . 33
2.8 Variance decomposition of NCD yield to DR007 . . . . . . . . . . . . . . . . 34
2.9 Variance decomposition of NCD yield to SHIBOR 1W . . . . . . . . . . . . 35
2.10 Variance decomposition of NCD yield to MLF . . . . . . . . . . . . . . . . . 35
2.11 Immediate impulse response of government bond yield to DR007 and
SHIBOR 1W . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.12 Immediate impulse response of CDB bond yield to DR007 and SHIBOR 1W 38
2.13 Max impulse response of government bond yield to DR007 and SHIBOR 1W 39
2.14 Max impulse response of CDB bond yield to DR007 and SHIBOR 1W . . . 40
2.15 Impulse response of three factors of government bond yield curve to DR007 42
2.16 Impulse response of three factors of government bond yield curve to SHI-
BOR 1W . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.17 Variance decomposition of three factors of government bond yield curve to
DR007 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
x
2.18 Variance decomposition of three factors of government bond yield curve to
SHIBOR 1W . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.19 Impulse response of three factors of CDB bond yield curve to DR007 . . . . 45
2.20 Impulse response of three factors of CDB bond yield curve to SHIBOR 1W 45
2.21 Variance decomposition of three factors of CDB bond yield curve to DR007 47
2.22 Variance decomposition of three factors of CDB bond yield curve to SHIBOR
1W . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.23 Impulse response of AAA-rated enterprise bond yield to DR007 . . . . . . 50
2.24 Impulse response of AAA-rated enterprise bond yield to SHIBOR 1W . . . 50
2.25 Impulse response of AA-rated enterprise bond yield to DR007 . . . . . . . 51
2.26 Impulse response of AA-rated enterprise bond yield to SHIBOR 1W . . . . 51
2.27 Variance decomposition of AAA-rated enterprise bond yield to DR007 . . . 53
2.28 Variance decomposition of AAA-rated enterprise bond yield to SHIBOR 1W 53
2.29 Variance decomposition of AA-rated enterprise bond yield to DR007 . . . . 54
2.30 Variance decomposition of AA-rated enterprise bond yield to SHIBOR 1W 54
2.31 Impulse response of AA minus-rated enterprise bond yield to DR007 . . . 55
2.32 Impulse response of AA minus-rated enterprise bond yield to SHIBOR 1W 56
2.33 Variance decomposition of AA minus-rated enterprise bond yield to DR007 56
2.34 Variance decomposition of AA minus-rated enterprise bond yield to SHI-
BOR 1W . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.35 Impulse response of credit market interest rates to MLF and LPR . . . . . . 59
2.36 Variance decomposition of credit market interest rates to MLF and LPR . . 60
3.1 Macroprudential policy measure, 2006-2019 . . . . . . . . . . . . . . . . . . 97
3.2 Comparison of Chinese monetary policy and macroprudential policy, 2006-
2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3 Parameter estimation result in TVP-SV-V AR model . . . . . . . . . . . . . . 101
3.4 The time-varying stochastic volatility of variables . . . . . . . . . . . . . . . 103
3.5 Time-varying simultaneous relationship among variables . . . . . . . . . . 105
xi
3.6 Impulse response function on economic growth for different lag lengths in
2006-2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.7 Impulse response function on economic growth for different selected times
in 2006-2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.8 Impulse response function on inflation for different lag lengths in 2006-2019113
3.9 Impulse response function on inflation for different selected times in 2006-
2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
3.10 Impulse response function on housing price for different lag lengths in
2006-2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.11 Impulse response function on housing price for different selected times in
2006-2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.12 Impulse response function on bond price for different lag lengths in 2006-2019118
3.13 Impulse response function on bond price for different selected times in
2006-2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.14 Impulse response function on equity price for different lag lengths in 2006-
2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.15 Impulse response function on equity price for different selected times in
2006-2019 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.1 Transition function with c=0 . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.2 Dynamics under financial autarky . . . . . . . . . . . . . . . . . . . . . . . . 176
4.3 Dynamics under full financial integration 1 . . . . . . . . . . . . . . . . . . 177
4.4 Dynamics under full financial integration 2 . . . . . . . . . . . . . . . . . . 178
4.5 Dynamics under full financial integration 1 (high risk aversion) . . . . . . . 180
4.6 Dynamics under full financial integration 2 (high risk aversion) . . . . . . . 181
4.7 Dynamics under limited financial integration 1 . . . . . . . . . . . . . . . . 182
4.8 Dynamics under limited financial integration 2 . . . . . . . . . . . . . . . . 183
4.9 Dynamics under extended model 1 . . . . . . . . . . . . . . . . . . . . . . . 221
4.10 Dynamics under extended model 2 . . . . . . . . . . . . . . . . . . . . . . . 222
xii
Abstract
As time goes on, the policies and methods used by global central banks to regulate
economic operations have also undergone tremendous changes. Facing the global financial
crisis and COVID-19’s huge damage to the economy, central banks around the world have
taken actions to support the economy. Despite global central banks use both conventional
and unconventional monetary policy to successfully support the economy, what we can’t
deny is that the monetary policy room of global central banks is getting smaller and
smaller, which may not be enough to respond to the next economic recession. But in the
next few years, the issue we should be most concerned about is the global benchmark
interest rate reform from the old London Interbank Offered Rate (LIBOR) to the new
risk-free rates (RFRs). In addition, facing the declining monetary policy space, global
central banks are bound to pay more attention to the development of macroprudential
policy. Finally, considering the huge impact of cross-border capital flows during the
crisis, the impact of financial integration on the economy will remain an important issue.
Therefore, in this paper, we conduct research on monetary policy, macroprudential policy,
and financial integration, trying to analyze their effects and provide references for their
future development.
In Chapter 2, we investigate the ongoing benchmark interest rate reform from LIBOR
to RFRs. As the global benchmark interest rate LIBOR is about to exit from the market,
now major economies have identified RFRs to replace LIBOR. However, the impact of this
reform on monetary policy transmission remains a problem. Considering China has a
clear first-mover advantage in this area, we try to learn from China’s mature experience to
provide reference for the ongoing global reform. In this chapter, we carefully analyze the
xiii
role and transmission efficiency of new and old benchmark interest rate in China: DR007
and SHIBOR 1W in money market, bond market, and credit market by using TVP-SV-V AR
model and SV AR model. As far as we know, we are the first one in this area. We are also
among the first few to obtain the CSI enterprise bond yield data to analyze the transmission
to enterprise bonds. Meanwhile, we construct a bank-level micro-database covering all 36
listed banks in mainland China. Our analysis suggests that DR007 performs very well,
whose transmission efficiency is only slightly weaker than SHIBOR 1W by about 1-7%,
and even better in some respects. This result remains robust for various interest rates in
financial market. In such case, we believe the process of global benchmark interest rate
reform may not result in a decline in the monetary policy transmission efficiency so that
we do not need to worry too much about its negative impact.
In Chapter 3, we investigate the effect of monetary policy and macroprudential policy.
We use a broad set of Chinese monetary policy instruments and principle component
analysis (PCA) approach to construct a composite monetary policy index (MPI). Com-
pared to other existing indices, our MPI has made several great improvements and can
better capture the trend of Chinese monetary policy. The time-varying parameter VAR
model with stochastic volatility (TVP-SV-V AR model) is used to estimate the effectiveness
of Chinese monetary policy and macroprudential policy. We find strong time-varying
characteristics among Chinese monetary policy, macroprudential policy and their targets.
Chinese monetary policy can promote economic growth and inflation, and its effectiveness
on economic growth is growing, especially after 2015. We also construct a model consist-
ing of state-owned enterprises (SOEs) sector and private-owned enterprises (POEs) sector
with directed lending to explain our findings. However, accommodative macroprudential
policy can cause minor damage to economic growth in the short term, and this negative
effect is increasing while it can also reduce inflation. So far macroprudential policy has
not been an independent policy, while at the same time being more independent, espe-
cially after 2015. Overall, the PBoC’s two-pillar regulation framework of monetary and
xiv
macroprudential policy is becoming increasingly effective in achieving its targets, which
also poses a new challenge to the PBoC, making it more prudent in its decision making.
In Chapter 4, we evaluate the welfare gains of financial integration empirically and
theoretically. The benefits of financial integration remain elusive in the past few decades.
To further investigate the benefits, first we use the Panel Smooth Threshold Regression
(PSTR) model that allows parameters to vary across countries and times, to conduct
empirical analysis. We find strong nonlinear relationship and threshold effects together
with heterogeneous effects across countries and times. Then we improve the two-country
neoclassical growth model in Coeurdacier et al. (2015) by adding FDI and frictions. In such
case, our framework can analyze simultaneously the welfare gains from capital scarcity
effect, risk sharing effect, and FDI scarcity effect together with how they interact. Both
the empirical and theoretical results indicate that financial integration does bring sizable
benefits that are mainly from FDI. In specific, both developed countries and emerging
economies benefit from FDI scarcity effect. Meanwhile, risk sharing effect is significant for
developed countries while capital scarcity effect is negligible for both countries. However,
since frictions are being removed slowly and available observation time is short, welfare
gains of financial integration for now are not large enough that can be explained as the
threshold effects, and we should wait for further evidence.
xv
Chapter 1
Introduction
As time goes on, the policies and methods used by global central banks to regulate
economic operations have also undergone tremendous changes. About one decade ago,
facing the huge damage caused by the global financial crisis, central banks around the
world adopted easy monetary policies to respond. As the representative of the central
banks of developed countries, the Fed has adopted the conventional monetary policy
method, that is cutting interest rate in response to the financial crisis. In specific, the Fed
has successively reduced interest rate ten times, and eventually, on December 16, 2008,
the Fed cut interest rate by 75 basis points to 0-0.25%. However, facing the unprecedented
global financial crisis, it is obviously not enough to cut interest rate alone. For this reason,
the Fed has begun to adopt a series of unconventional monetary policy instruments, such
as quantitative easing, to support the economy.
The policy effect brought about by the extraordinary monetary policy response was
also very effective. The US economy began to gradually recover, and the easy monetary
policy was maintained until 2015. During this period, considering the huge damage the
global financial crisis has brought to the global economy, global central banks also began to
think about how to prevent the next round of financial crisis from occurring. Based on the
experience during the global financial crisis, they began to explore the design and usage
of macroprudential policy to prevent systemic risks from occurring. At the same time,
considering the huge contagion of the financial crisis, countries around the world have
1
also begun to think about the pros and cons of financial integration and how to better deal
with the new challenges brought about by financial integration. But in general, economic
globalization and financial globalization are still the general trend, and the ties between
countries in the world are getting closer.
With the improvement of the economic situation, the Fed officially withdrew from
quantitative easing in October 2014. On December 17, 2015, the Fed raised interest rate
by 25 basis points to 0.25-0.5%, which is the first time after the global financial crisis. At
the same time, monetary policy normalization process has officially begun. By the end
of 2019, the Fed has adopted 9 interest rate hikes and 3 interest rate cuts to adjust the
interest rate target to 1.50-1.75%. But there is no doubt that this value is much smaller
than the 5.25% before the global financial crisis. Facing the next round of economic
recession, there seems to be little room left for global central banks. On the one hand,
considering that the effect of the negative interest rate policy is not clear, the traditional
interest rate room appears insufficient. On the other hand, though quantitative easing
seems to have a huge effect, there is ultimately a limit to the scale of assets they can
purchase. Meanwhile, the macroprudential policy that global central banks place a great
hope on is still under development and not yet mature enough to cope with the next round
of economic recession.
In 2020, with the unprecedented COVID-19, the global economy quickly entered
a recession and global central banks began to adopt easy monetary policy again. The
Fed quickly reduced interest rates twice to 0-0.25%, restarted its old policy instruments
during the global financial crisis, and began quantitative easing. At the same time, the
Fed established new policy instruments, such as Paycheck Protection Program Liquidity
Facility and Main Street Lending Program to support the economy. During this period,
cross-border capital also flowed violently, that is, international capital quickly flowed from
emerging market economies to developed countries such as the US, and then returned,
which had a huge impact on emerging market economies. As of the end of 2020, the
2
Fed’s easing policy has had the expected effect as a whole. Though the epidemic is still
developing and the economy has not fully recovered, the overall development trend is
great and the economy is slowly recovering.
But there is no doubt that the global monetary policy will remain easy in the next few
years to support economic recovery. Meanwhile, in response to the next round of economic
recession, the policy space of global central banks will further shrink. But in the next
few years, the issue we should be most concerned about is the global benchmark interest
rate reform. The old global benchmark interest rate is the London Interbank Offered
Rate (LIBOR), which is about to exit the market. At present, major economies around
the world have identified new alternatives: risk-free rates (RFRs), which are generated
based on actual transactions rather than quoting like LIBOR. However, the impact of the
global benchmark interest rate reform on monetary policy transmission still remains a
question. In addition, facing the declining monetary policy space, global central banks are
bound to pay more attention to the development of macroprudential policy. Therefore, it
is very important to study the effect of monetary policy and macroprudential policy on
the economy. Finally, considering the huge impact of cross-border capital flows during the
crisis, the impact of financial integration on the economy will remain an important issue.
Therefore, in this paper, we conduct research on monetary policy, macroprudential
policy, and financial integration, trying to analyze their effects and provide references
for their future development. Specifically, we hope to explore the following issues: Will
the ongoing benchmark interest rate reform from LIBOR to RFRs affect the transmission
efficiency of monetary policy? What is the effect of monetary policy and macroprudential
policy? Does financial integration bring sizeable benefit to countries?
The roadmap of this paper can be explained in Figure 1.1. First, the central bank will
determine policies in accordance with the economic situation. These policies include
monetary policy and macroprudential policy. Subsequently, these policies will affect
the operation of financial institutions, such as the tightness of funding. Then financial
3
institutions will affect the financial market, which will affect the real economy. Meanwhile,
with the introduction of unconventional policies, the policies also have channels that can
directly affect the financial market. For example, the central bank can directly purchase
financial assets such as government bonds and MBS to affect financial market. With
financial integration, international capital can flow in from the international financial
market, thereby affecting the domestic financial market and the real economy. At the same
time, domestic capital can also enter the international financial market in the same way to
affect the financial market and real economy of other countries. Here we mark the content
of each chapter on the roadmap. Specifically, Chapter 2 deals with the transmission of
monetary policy from policy to financial market; Chapter 3 deals with the transmission of
monetary policy and macroprudential policy from policy to the real economy; Chapter 4
deals with the impact of financial integration on the real economy.
Figure 1.1 Roadmap
In Chapter 2 we investigate the ongoing benchmark interest rate reform from LIBOR
to RFRs. Compared with other countries, China has a clear first-mover advantage in
4
developing benchmark interest rates based on actual transactions. Therefore, it can
contribute to the global money market benchmark interest rate reform and RFR benchmark
interest rate development. In such case, we try to learn from China’s mature experience in
developing benchmark interest rates based on actual transactions, and hope to provide
reference for the ongoing monetary policy benchmark interest rate reform in the world
through the analysis of DR, China’s unique benchmark interest rate based on actual
transactions. We also hope to evaluate whether the replacement of the old benchmark
interest rate LIBOR with the new benchmark interest rates RFRs will adversely affect
the monetary policy transmission. Specifically, we carefully analyze the role of DR007
and SHIBOR 1W, and compare their transmission efficiency in China’s monetary policy
transmission in money market, bond market, and credit market. Since DR007 is a relatively
new index and LIBOR is regarded as global benchmark interest rate, academia has long
regarded SHIBOR 1W as the de facto benchmark interest rate. However, DR was just
confirmed as the new benchmark interest rate on August 31, 2020, and SHIBOR will
no longer exist as a de facto benchmark interest rate. In such case, our research is very
valuable. We also deliberately obtain the data of CSI (China Securities Index) enterprise
bond yield to maturity from China Securities Index Company. Based on this data, we
investigate in detail the transmission of China’s short-term money market interest rates
to enterprise bond yields. What’s more, we construct a bank-level micro-database, which
contains a wide range of banks in order to better evaluate the transmission in credit
market.
Our main findings are distributed in five aspects. First, on the whole, the transmission
efficiency of the open market reverse repo rate to the interbank market is higher than
that of the exchange market. What’s more, the transmission efficiency to the overnight
product is significantly higher than the 7-day product. Second, the transmission efficiency
of DR007 to the bond market is significantly higher than our expectation, and the overall
transmission efficiency of DR007 is slightly weaker than that of SHIBOR 1W by about
5
1-7%, which is not a huge difference. Third, in terms of the three factors of the government
bond yield curve and the CDB bond yield curve, the response of DR007 is very close to
that of SHIBOR 1W. However, the explanatory power of DR007 is significantly greater
than that of SHIBOR 1W in several bond products, which shows that the transmission
effect of DR007 is better than SHIBOR 1W in some respects. Fourth, we find that MLF has
not a great influence on the NCD yields, which means that the MLF reform still needs to
continue. Meanwhile, we find that the response of AA minus-rated enterprise bond yields
to DR007 and SHIBOR 1W is not ideal. Fifth, according to the results of the bank-level
micro-database, the transmission efficiency of SHIBOR 1W is slightly higher than that of
DR007 in the credit market. However, the gap is very small, which is less than 1%. Our
analysis suggests that DR007 performs very well in terms of monetary policy transmission.
DR007’s transmission efficiency is only slightly weaker than SHIBOR 1W, and even better
than SHIBOR 1W in some respects. Therefore, DR007 has the ability to replace SHIBOR
1W as the new benchmark interest rate. From the comparison of DR007 and SHIBOR
1W, we believe that the process of replacing the old benchmark interest rate LIBOR with
the new benchmark interest rates RFRs may not lead to a decline in the monetary policy
transmission efficiency. What’s more, this replacement can overcome the shortcomings of
LIBOR. Therefore, we do not need to worry too much about the impact of this reform on
markets and policies.
In Chapter 3, we investigate the effect of monetary policy and macroprudential policy.
Considering that China has been continuously advancing the study and application of
macroprudential policy, it is meaningful to study from China’s experience in this area.
Meanwhile, since China has officially formed the two-pillar regulation framework consist-
ing of monetary policy and macroprudential policy, more evidence on the interaction of
monetary policy and macroprudential policy can be found. In specific, we use a broad set
of Chinese monetary policy instruments and principle component analysis (PCA) approach
to construct a composite monetary policy index that is our monetary policy index (MPI).
6
Compared to other existing Chinese monetary policy indices, our MPI has made several
great improvements and can better capture the trend of Chinese monetary policy. What’s
more, we extend the macroprudential measure in Alam et al. (2019) and the integrated
Macroprudential Policy (iMaPP) database from 2016 to 2019 using the Chinese Monetary
Policy Implementation Report. In the meantime, we use the time-varying parameter V AR
model with stochastic volatility (TVP-SV-V AR model) to investigate the effectiveness of
Chinese monetary policy and macroprudential policy over the period 2006-2019, which is
a nicer method compared with the econometric methods used in the past literature.
Our main findings are distributed in four aspects. First, there are strong time-varying
characteristics among Chinese monetary policy, macroprudential policy and their targets.
Second, Chinese monetary policy can promote economic growth, and its effectiveness
is growing, especially after 2015. Third, accommodative macroprudential policy can
cause minor damage to economic growth in the short term, and this negative effect
is increasing, which means that policy makers need to be more careful in adjusting
macroprudential policy to avoid side effect. Fourth, so far macroprudential policy has not
been an independent policy, which is consistent with the PBoC’s stated position. However,
the independence is stronger, especially after 2015. Our analysis suggests that the PBoC’s
two-pillar regulation framework of monetary and macroprudential policy is effective in
achieving its targets since these two policies have different effect on different targets. As
the effectiveness of Chinese monetary policy and macroprudential policy has continued to
increase, this also poses a new challenge to the PBoC, making the PBoC’s policy decisions
more cautious.
In Chapter 4, we evaluate the welfare gains of financial integration empirically and
theoretically. First, we conduct empirical analysis to investigate the effect of FDI, eq-
uity, and debt market integration on economic growth. Considering the effect might
be nonlinear and super complicated, an advanced econometric methodology: the Panel
Smooth Threshold Regression (PSTR) model is used. Two interpretations of PSTR model
7
are possible. On one hand PSTR model can be thought of as a regime-switching model
that allows for a small number of extreme regimes associated with the extreme value of
a transition function. Meanwhile, different from the indicator function, the transition
from one regime to another is smooth. On the other hand, PSTR model can be thought to
allow for a continuum of regimes, each one being characterized by a different value of the
transition function. Then in order to explain our empirical findings, we construct a theo-
retical model trying to better understand the effect of financial integration, especially FDI
market integration on economic growth. The model we use is the improved two-country
neoclassical growth model in Coeurdacier et al. (2015) by adding FDI and frictions. In
such case, we consider the widely accepted qualitative claims on the channels of capital
scarcity effect and risk sharing effect. Meanwhile, we also consider the importance of FDI
that can be a new channel to increase welfare and frictions being removed that may hide
the welfare gains. By using this model, we can analyze simultaneously the welfare gains
from capital scarcity effect, risk sharing effect, and FDI scarcity effect together with how
they interact. Our theoretical results are in line with the empirical findings.
Our results indicate that there is a strong nonlinear relationship existing, which means
we can’t use linear regression model to estimate the effect. We also find strong threshold
effects in the relationship between FDI, equity, and debt market integration and economic
growth. Under this circumstance, the aggregate effect of FDI, equity, and debt market
integration on economic growth is highly impacted by the threshold effects, which means
the effect is very complicated. And since the threshold effects rely on KAOPEN that
measures capital account openness, which is affected by multiple factors, strong threshold
effects mean that the aggregate effect of FDI, equity, and debt market integration on
economic growth is heterogeneous across countries and times. But generally speaking, we
are confident that the increase in FDI market integration can promote economic growth.
What’s more, the improvement of equity market integration will first promote and then
inhibit economic growth. Moreover, the increase in debt market integration can promote
8
economic growth but may face some negative effect. In such case, financial integration can
bring sizable benefits that may be mainly from FDI market integration. Finally, since the
slope parameter is not very large, the transition function is not very sharp. This means
that our regression equation cannot be reduced to the sum of a limited number of regimes
and we must use PSTR model instead of PTR model.
Theoretically, we also find financial integration does bring sizable benefits that are
mainly from FDI. In specific, financial integration can lead to a large permanent consump-
tion increase for developed countries and emerging economies. Even if we restrict the rela-
tive level and consider frictions, these numbers are still large enough. FDI scarcity effect
is super important for both developed countries and emerging economies. What’s more,
developed countries can benefit a lot from risk sharing effect but emerging economies
don’t benefit from this a lot. Finally capital scarcity effect is small for both countries.
Since frictions are being removed slowly, frictions can highly weaken the welfare gains of
financial integration especially the growth rate of capital and consumption. Meanwhile,
observation time matters a lot to the welfare gains of financial integration. In such case,
we may need to be more patient to find more convincing results. Another key finding is
that the heterogeneous effect of financial integration across countries and times is more
complicated than Coeurdacier et al. (2015).
9
Chapter 2
From SHIBOR to DR: Chinese Monetary Policy
Transmission and Benchmark Interest Rate Reform
As the global benchmark interest rate LIBOR is about to exit from the market, now major
economies have identified risk-free rates (RFRs) to replace LIBOR. However, the impact of
this reform on monetary policy transmission remains a problem. Considering China has a
clear first-mover advantage in this area, we try to learn from China’s mature experience to
provide reference for the ongoing global reform. In this chapter, we carefully analyze the
role and transmission efficiency of new and old benchmark interest rate in China: DR007
and SHIBOR 1W in money market, bond market, and credit market by using TVP-SV-V AR
model and SV AR model. As far as we know, we are the first one in this area. We are also
among the first few to obtain the CSI enterprise bond yield data to analyze the transmission
to enterprise bonds. Meanwhile, we construct a bank-level micro-database covering all 36
listed banks in mainland China. Our analysis suggests that DR007 performs very well,
whose transmission efficiency is only slightly weaker than SHIBOR 1W by about 1-7%,
and even better in some respects. This result remains robust for various interest rates in
financial market. In such case, we believe the process of global benchmark interest rate
reform may not result in a decline in the monetary policy transmission efficiency so that
we do not need to worry too much about its negative impact.
10
2.1 Introduction
In global financial market, the most widely used benchmark interest rate is the London
Interbank Offered Rate (LIBOR). Since the global financial crisis in 2008, the interbank
lending markets in various countries have shrunk, and the reference base for LIBOR quotes
has weakened. Meanwhile, many cases of LIBOR manipulation broke out, which severely
weakened the market credibility of LIBOR. Despite the regulatory authorities of various
countries have carried out drastic reforms on LIBOR and similar interbank offered rates
(IBOR), LIBOR still couldn’t gain the market recognition. In 2017, the British Financial
Conduct Authority (FCA) announced that after the end of 2021, FCA won’t require banks
to submit LIBOR, and instead will turn to develop new benchmark interest rate based on
actual transactions, which means that LIBOR may exit the market by then. Subsequently,
other major developed economies such as the US, the Eurozone, and Japan that used
LIBOR as their benchmark interest rate before also began to study exiting from LIBOR
and fostering alternative benchmark interest rates.
At present, major economies around the world have identified new alternatives: risk-
free rates (RFRs). Different from LIBOR, RFRs are independently issued by each economy
and are generated based on actual transactions. Specifically, there are two main paths for
major developed economies to promote benchmark interest rate reform. One is to use
RFRs to completely replace the old IBOR benchmark interest rates, such as the US and the
UK. The other is to reform the existing IBOR quote mechanism while introducing RFRs
to improve the reliability of IBOR quotes so that multiple benchmark interest rates can
coexist, such as the Eurozone and Japan have chosen this method.
But at the same time, the impact of the benchmark interest rate reform on the operation
framework of monetary policy cannot be ignored. After the reform, the RFRs of various
countries include a large number of non-bank financial institution transactions while such
institutions are usually not the traditional open market operation targets of the central
bank. With the increasing scale of non-bank transactions, RFRs are likely to deviate from
11
the policy rate and affect the transmission effect of monetary policy. At the same time,
RFRs based on repurchase transactions are greatly affected by changes in the supply and
demand of collaterals, which will also disturb the level of RFRs. For example, since the
introduction of SOFR, interest rates in the U.S. repo market have fluctuated significantly
and deviated significantly from the federal fund rate on several occasions. As a result, the
Fed had to expand its balance sheet again in October 2019 while this operation is not QE
as Jerome Powell, chairman of the Fed, emphasized.
Compared with other countries, China has a clear first-mover advantage in developing
benchmark interest rates based on actual transactions. Therefore, it can contribute to
the global money market benchmark interest rate reform and RFR benchmark interest
rate development. As the PBoC indicated, first, China’s benchmark interest rate based
on actual transactions has been operating for a long time. Unlike the current global
focus on developing RFRs based on actual transactions, China has been developing and
publishing actual transaction-based lending rates and repurchase rates for more than
20 years. Second, China has comprehensive, transparent and easily available market
transaction data. All transactions in China’s interbank market are completed in the trading
system of CFETS, which not only overcomes the decentralization and low transparency of
OTC market transactions, but also maintains the flexibility and transaction efficiency of
the OTC market. Third, China has always attached great importance to the supervision
and management of benchmark interest rates. Unlike the remedy after the LIBOR quote
manipulation scandal occurred, the PBoC has always attached great importance to the
development and supervision of China’s benchmark interest rate system. At the same time,
the interest rate self-regulatory mechanism has also played an important role.
In this chapter, we try to learn from China’s mature experience in developing bench-
mark interest rates based on actual transactions, and hope to provide reference for the
ongoing monetary policy benchmark interest rate reform in the world through the analysis
of DR, China’s unique benchmark interest rate based on actual transactions. We also hope
12
to evaluate whether the replacement of the old benchmark interest rate LIBOR with the
new benchmark interest rates RFRs will adversely affect the monetary policy transmission.
Specifically, we elaborate the transmission mechanism of China’s monetary policy, and
based on this, we construct the TVP-SV-V AR model, SV AR model and a micro-database of
the Chinese banking industry. Our goal is to analyze China’s monetary policy transmission
and explore the position of the new benchmark interest rate DR and the old benchmark
interest rate SHIBOR in the monetary policy transmission and their transmission efficiency.
In this process, we mainly address the following issues: Is China’s monetary policy trans-
mission effective? What is the transmission efficiency of the new benchmark interest rate
DR? What is the difference between DR and the old benchmark interest rate SHIBOR? Will
the ongoing benchmark interest rate reform from LIBOR to RFRs affect the transmission
efficiency of monetary policy?
This chapter contributes to the existing literature from several perspectives. First,
we analyze China’s new benchmark interest rate DR to provide a reference for global
benchmark interest rate reform and usage of new RFR benchmark interest rates. Currently,
LIBOR, the global benchmark interest rate, is about to exit from the market. Major
developed economies such as the US, the Eurozone, and Japan that previously used LIBOR
as benchmark interest rates have begun to reform benchmark interest rates and introduced
new RFR benchmark interest rates based on actual transactions. Relatively speaking, since
1996, China has published actual transaction-based lending rates and repurchase rates
for more than 20 years. Meanwhile, China’s new benchmark interest rate DR has a longer
history so that it can provide reference for other countries. As far as we know, we are
the first few to analyze China’s experience to provide a reference for global benchmark
interest rate reform.
Second, we carefully analyze the role of DR007 and SHIBOR 1W, and compare their
transmission efficiency in China’s monetary policy transmission. Since DR007 is a rela-
tively new index and LIBOR is regarded as global benchmark interest rate, academia has
13
long regarded SHIBOR 1W as the de facto benchmark interest rate. Therefore, few papers
focus on the effect of DR007. As far as we know, we are the first one to study DR007 and
SHIBOR 1W in detail. However, DR was just confirmed as the new benchmark interest
rate on August 31, 2020, and SHIBOR will no longer exist as a de facto benchmark interest
rate. In such case, our research is very valuable. In fact, as far as we know, we are also the
first one to analyze the transmission efficiency of the new benchmark interest rate DR007
and the old benchmark interest rate SHIBOR 1W.
Third, we deliberately obtain the data of CSI (China Securities Index) enterprise bond
yield to maturity from China Securities Index Company, which is very hard to obtain. In
such case, it is very valuable to obtain this data alone. Based on this data, we investigate in
detail the transmission of China’s short-term money market interest rates to enterprise
bond yields. As far as we know, we are the first one to use this data and conduct such
analysis. Though there is no doubt that government bonds and CDB (China Development
Bank) bonds are the benchmarks of the bond market, enterprise bonds have the largest
market share other than NCD in credit bonds. Therefore, by analyzing the transmission
efficiency of money market to enterprise bonds, we can better understand and grasp the
monetary policy transmission and its impact on the bond market.
Fourth, we use the time-varying parameter V AR model with stochastic volatility (TVP-
SV-VAR model) from Nakajima (2011). Compared to the econometric methods used in
the past literature, our method is much better where we can capture the changes of the
effects while other methods can’t. The TVP-SV-VAR model is an improved VAR model
that has been widely used in the empirical macroeconomic literature. Compared with the
traditional V AR model and other improved V AR models, the TVP-SV-V AR model has two
main improvements that enable it to better capture the characteristics of Chinese monetary
policy and macroprudential policy. First, the TVP-SV-V AR model contains time-varying
parameters that allow us to capture the underlying structure of potential temporal changes
14
in the economy flexibly and robustly. Second, the volatility in the TVP-SV-V AR model is
stochastic rather than constant, which would make our estimates more accurate.
Fifth, we construct a bank-level micro-database, which contains a wide range of banks.
We have included all 36 banks listed in mainland China into our bank-level micro-database,
which covers relevant bank-level data from 2015Q1 to 2020Q3. As far as we know, our
bank-level micro-database covers the largest number of banks among existing databases.
Taking into account the late listing of many banks, they don’t release some earlier financial
data, making it difficult to obtain detailed financial data. In order to solve this problem,
we supplement these vacant data by querying the financial data disclosed by these banks
when they issued bonds. In such case, we obtain this detailed bank-level micro-database,
which is also one of our major contribution.
Our main findings are distributed in five aspects. First, on the whole, the transmission
efficiency of the open market reverse repo rate to the interbank market is higher than that
of the exchange market. What’s more, the transmission efficiency to the overnight product
is significantly higher than the 7-day product. From the comparison of DR and SHIBOR,
the transmission efficiency of DR001 is slightly higher than that of SHIBOR ON, while
the transmission efficiency of SHIBOR 1W is slightly higher than that of DR007, but the
difference is not much.
Second, the transmission efficiency of DR007 to the bond market is significantly higher
than our expectation. In terms of the NCD (negotiable certificate of deposit) yields of
various maturities, the yields of 1-year and 10-year government bond and CDB bond, the
three factors of the yield curve of government bonds and CDB bonds, and the yields of
AAA-rated and AA-rated enterprise bonds of various maturities, the overall transmission
efficiency of DR007 is slightly weaker than that of SHIBOR 1W by about 1-7%, which is
not a huge difference.
Third, in terms of the three factors of the government bond yield curve and the CDB
bond yield curve, the response of DR007 is very close to that of SHIBOR 1W. It is interesting
15
to note that on the slope factor and level factor of the CDB bond yield curve, the response
of DR007 is greater than that of SHIBOR 1W by about 1-1.5%. This shows that DR007
is slightly better than SHIBOR 1W in the transmission of the yield curve of government
bonds and CDB bonds. Another thing needs to be paid special attention to is that the
explanatory power of DR007 is significantly greater than that of SHIBOR 1W in terms
of the NCD yields of various maturities, the yield curves of government bonds and CDB
bonds, and the yields of AAA-rated and AA-rated enterprise bonds of various maturities.
These show that the transmission effect of DR007 is better than SHIBOR 1W in some
respects.
Fourth, we find that MLF doesn’t have a great influence on the NCD yields, which
means that the MLF reform still needs to continue. Meanwhile, we find that the response
of AA minus-rated enterprise bond yields to DR007 and SHIBOR 1W is not ideal. This
may be because AA minus-rated enterprise bonds are high-yield bonds so that their yields
are more affected by other factors while relatively the effect of short-term money market
interest rate on it is not that great.
Fifth, according to the results of the bank-level micro-database, the transmission
efficiency of SHIBOR 1W is slightly higher than that of DR007 in the credit market.
However, the gap is very small, which is less than 1%. Meanwhile, the transmission
efficiency of both DR007 and SHIBOR 1W in the lending interest rate is significantly
higher than the deposit interest rate. But the transmission efficiency of both DR007 and
SHIBOR 1W in credit market is significantly lower than that in bond market, reflecting
the urgency of interest rate reform in credit market.
Our analysis suggests that DR007 performs very well in terms of monetary policy
transmission. DR007’s transmission efficiency is only slightly weaker than SHIBOR 1W,
and even better than SHIBOR 1W in some respects. Therefore, DR007 has the ability to
replace SHIBOR 1W as the new benchmark interest rate. From the comparison of DR007
and SHIBOR 1W, we believe that the process of replacing the old benchmark interest
16
rate LIBOR with the new benchmark interest rates RFRs may not lead to a decline in the
monetary policy transmission efficiency. What’s more, this replacement can overcome the
shortcomings of LIBOR. Therefore, we do not need to worry too much about the impact of
this reform on markets and policies.
The rest of the chapter is organized as follows. Section 2 describes related literature
review and our contribution. Section 3 provides econometric methodology. Section 4
shows empirical results to investigate the transmission efficiency of DR007 and SHIBOR
1W in different financial markets. Section 5 concludes.
2.2 Literature Review
This chapter lies at the intersection of five strands of the literature on the global RFR
benchmark interest rate reform, transmission mechanism of China’s monetary policy and
the transmission efficiency of benchmark interest rates.
First, this chapter contributes to the literature on the global RFR benchmark interest
rate reform. In particular, current research on monetary policy transmission mechanism
centers on the experience of the US and the Eurozone. Duffie and Stein (2015) discuss
the economic role of benchmarks in reducing market frictions and propose a two-rate
approach to mitigate LIBOR manipulation, which is consisted of LIBOR+ and low-credit-
risk reference rate. In specific, LIBOR+, the combination of IBOR and unsecured bank
borrowing from all wholesale sources including non-bank investors, serves as the reference
rate for most on-balance-sheet bank lending contracts. Low-credit-risk reference rate,
such as the treasury bill rate, serves as the reference rate for the majority of interest rate
derivatives. Then they discuss a transition strategy to deal with the large stock of legacy
contracts tied to LIBOR. Following Duffie and Stein (2015), Schrimpf and Sushko (2019)
provide an overview of RFR benchmark interest rates and compare some of their key
characteristics with LIBOR, confirming the outcome of this reform with the coexistence
17
of multiple rates. To step further, Jermann (2019) studies the consequences of replacing
a rate index within a dynamic equilibrium model in the US. He finds that under normal
conditions, an economy using SOFR behaves similarly to an economy using LIBOR while
under more extreme financial conditions, differences can be nontrivial. What’s more,
Taylor-Brill (2020) analyzes banks’ LIBOR submission from 2006 to 2008 to measure the
extent of LIBOR manipulation and suggest two benchmark regime: LIBOR+ and SOFR as
a replacement for LIBOR. Hence, there is a very high probability that multiple rates coexist
under the new normal. Jermann (2020) finds that the cumulative additional interest from
LIBOR during the crisis is estimated to be between 1% to 2% of the notional amount of
outstanding loans, depending on the tenor and type of SOFR rate used. He also finds that a
compounded SOFR reduces insurance relative to a term SOFR. Given China’s long history
of the new RFR benchmark interest rate DR, this chapter focuses on China’s experience
of developing new RFR benchmark interest rate, which has received little discussion
elsewhere in the academic literature.
Second, this chapter is also associated with the study of transmission efficiency of
monetary policy in different countries. As Cevik and Teksoz (2013) argue, the economic
structure, development of financial and capital markets, economic conditions are all
factors that affect monetary policy transmission while strengthening financial intermedia-
tion and facilitating the development of liquid domestic capital markets would advance
the effectiveness of monetary transmission mechanisms in some countries. Mizen and
Hoffman (2004) show that the pass-through for the UK during 1985-2001 is time-varying
contingent on the gap between retail and base rates. Sanusi (2010) applies Structural
Vector Autoregressive (SV AR) model to examine the pass-through of monetary policy and
interbank interest rates to lending and deposit rates in Nigeria between 2002 and 2010
and find that the pass-through of policy rate to interbank rate is substantially higher and
faster than to lending and deposit rates. Tai et al. (2012) present low-level pass-through
from market rates to deposit and credit rates in short and long run for Asian countries,
18
particularly after 2008. Illes and Lombardi (2013) observe that the difference between
lending rates and policy rates hit new lows in both Germany and the US but remain
higher in other western countries. Mishra et al. (2016) examine the strength of monetary
transmission in India and find that the pass-through efficiency from the policy rate to
banking lending rate is about 40%, using a SV AR methodology.
However, the literature on China’s transmission efficiency of monetary policy is rela-
tively outdated. Ji et al. (2016) examine the effectiveness of the interest rate pass-through
from policy rates to lending rates and find some evidence of the policy rate effect with only
20%-80% of transmission efficiency in the US. Ma and Wang (2014) and Ma et al. (2016)
use a general equilibrium model to illustrate the policy rate transmission in China and
find the positive correlation between policy rate and other interest rates with the efficiency
25% weaker than other major economies, including bond yields as well as bank deposit
and lending rates. They also conclude that loan-to-deposit ratio restriction, loan quota
and high deposit reserve may distort policy rate transmission. G. Sun and Duan (2017)
establish a micro-based monetary transmission mechanism and present evidence that
long-term SHIBOR has more effective transmission on deposit and lending rates. After the
introduction of DR, this chapter is the first to analyze the role of DR007 and SHIBOR 1W,
and compares their transmission efficiency in different markets.
Third, this chapter contributes to the recent literature by analyzing transmission
efficiency of money market to enterprise bonds. Previous studies focus on the transmission
efficiency of monetary policy through government bonds and corporate bonds while the
studies on corporate bonds in China is limited. For example, Ma et al. (2016) empirically
test the transmission efficiency of interest rate via government bond yield curves and
suggest less than perfect functioning in certain aspects of the bond market. X. He (2018)
shows that the transmission of interest rates through the government bond market has
improved remarkably in China: the longer the term of bond is, the more significant the
improvement is. Krolzig and Sserwanja (2014) analyze the response of corporate bonds to
19
monetary policy shocks in the US within a cointegrated vector-correction model framework
and find that corporate bonds react faster and deeper to a contractionary monetary shock
than treasuries. Schwalb (2017) focuses on the pass-through of large scale asset purchase
programs (LSAP) by the Fed to corporate bond yields and shows that pass-through mainly
comes from a signaling and a credit default risk channel. Burger et al. (2017) analyze
the effect of the US Federal Reserve’s monetary policy on emerging market economies
corporate bond markets, with the focus on the evolution of the structure of the bond
markets and their allocations within the bond portfolios of US investors.
Fourth, this chapter is related to the usage of a new methodology, TVP-SV-V AR model
following Primiceri (2005) and Nakajima (2011). Traditional methodologies adopted by
most empirical studies, like V AR or SVAR, don’t allow for time-varying random-walk
parameters and stochastic volatility. The use of time-varying parameters in the model
allows us to estimate the time-varying dynamics between variables while, according to
Nakajima (2011). Meanwhile, the introduction of stochastic volatility is expected to
improve the estimation of the model.
Finally, this chapter fits the literature on transmission efficiency in credit markets
using bank-level database. Bernanke and Gertler (1995) highlight the importance of
banking-lending channel for small firms in need of bank loans. The extensive literature
on bank-level micro-database helps us control for determinants of bank lending rate with
sufficient accuracy. Feng and Serletis (2010) apply regularity-constrained parametric
estimation to a panel data of 292 large banks in the US over the period from 2000 to 2005
and provide estimates of technical change, efficiency, and returns to scale. Almaqtari et al.
(2019) examine the determinants of profitability of Indian commercial banks based on
balanced panel data over the period from 2008 to 2017 for 69 commercial banks. The
results reveal the importance of bank size, the number of branches, assets management
ratio, operational efficiency, and leverage ratio in explaining the profitability measured by
ROA. Huang et al. (2019) build the quarterly panel data of 23 listed commercial banks to
20
examine the impacts of financial market shock on bank asset allocation in China while
they show that the financial characteristics of banks influence the dynamic adjustment
range of asset allocation. Gao et al. (2019) also use the non-equilibrium panel data of
16 listed banks to analyze the influence of the development of interbank business on the
stability of banks. In addition, it is found that interest rate transmission through the
banking system is much less efficient in China than in the US. According to Illes and
Lombardi (2013), the effectiveness of policy rate transmission to bank lending rates is
about 0.8 in the US. However, Ji et al. (2016) using data on listed banks show that the
elasticity of lending rates to short-term money market rates is between 0.60 and 0.67, if
the impact of benchmark interest rates determined by the PBoC is not controlled. The
elasticity falls to 0.16–0.17 after controlling the changes in these benchmark interest rates.
Compared to their previous results, our paper shows lower transmission efficiency. One
answer that can arise from the fact is that the old benchmark interest rates are gradually
losing effect and the new benchmark interest rates are still developing.
2.3 Econometric Methodology
In this section, we present the econometric methodology to investigate the monetary policy
transmission and answer the questions we mentioned earlier. The main workhorse model
we use is the time varying parameter V AR model with stochastic volatility (TVP-SV-V AR
model) from Nakajima (2011), which is developed from Primiceri (2005).
The TVP-SV-V AR model is an improved V AR model that has been widely used in the
empirical macroeconomic literature. Compared with the traditional V AR model and other
improved V AR models, the TVP-SV-V AR model has two main improvements that enable
it to better capture the characteristics of Chinese monetary policy and macroprudential
policy. First, the TVP-SV-V AR model contains time-varying parameters that follow the first-
order random walk process. In this case, the TVP-SV-V AR model allows us to capture the
21
underlying structure of potential temporal changes in the economy flexibly and robustly
and allows both temporary and permanent shifts in parameters. Second, the volatility
in the TVP-SV-V AR model is stochastic rather than constant, which would make our
estimates more accurate. This improvement is due to the fact that in many cases the
economic variables have drift coefficients and shocks of random volatility, which, if we
still use the constant volatility of the TVP-V AR model, would lead to biased estimates of
the time-varying coefficient. According to Nakajima (2011), the incorporation of stochastic
volatility into the TVP estimation significantly improves estimation performance. We
also follow Nakajima (2011)’s computational approach, using the Markov chain Monte
Carlo (MCMC) method in the context of Bayesian inference, as the likelihood function
now becomes intractable to estimate the coefficients.
To better understand the TVP-SV-V AR model, we start from the traditional structural
V AR model, where all parameters are now time-invariant. This traditional structural V AR
model can be expressed as:
Ay
t
=F
1
y
t1
+::: +F
s
y
ts
+u
t
; t =s + 1;:::;n (2.1)
wherey
t
is the k*1 vector of observed variables, andA;F
1
;:::;F
s
are k*k matrices of
coefficients. The disturbanceu
t
is a k*1 structural shock and we assume thatu
t
N(0;
2
),
where
=
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
1
0 0 0
0
2
0 0
:
:
:
:
:
:
:
:
:
:
:
:
:
:
:
0 0
k1
0
0 0 0
k
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
(2.2)
22
Since we assume it’s a structural VAR model, we specify the simultaneous relations
of the structural shock by recursive identification, assuming that A is lower-triangular
matrix:
A =
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
1 0 0 0
a
21
1 0 0
:
:
:
:
:
:
:
:
:
:
:
:
:
:
:
a
k1;1
a
k1;2
1 0
a
k1
a
k2
a
k;k1
1
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
(2.3)
Then by multiplyingA
1
to both sides in the V AR model, we can rewrite this reduced
form V AR model:
y
t
=B
1
y
t1
+::: +B
s
y
ts
+A
1
t
;
t
N(0;I
k
) (2.4)
whereB
i
=A
1
F
i
. After that, we can stack the elements in the rows of theB
i
’s to form
that is ak
2
s 1 matrix and defineX
t
=I
k
(y
0
t1
;:::;y
0
ts
), where
is the Kronecker product.
In such case, we can have:
y
t
=X
t
+A
1
t
(2.5)
Now we can extend this model to be TVP-SV-V AR model where the parameters are all
time varying now.
Following the setting we derive before, the TVP-SV-V AR model can be written as:
y
t
=X
t
t
+A
1
t
t
t
; t =s + 1;:::;n (2.6)
Please notice that now
t
;A
t
;
t
are all time varying. We denotea
t
= (a
21
;a
31
;a
32
;a
41
;:::;a
k;k1
)
0
as the stacked vector of the low-triangular elements inA
t
and denoteh
t
= (h
1t
;:::;h
kt
)
0
23
withh
jt
=log(
2
jt
) forj = 1;:::;k;t =s + 1;:::;n. Meanwhile, we assume parameters follow a
random walk process as follows:
t+1
=
t
+u
t
(2.7)
a
t+1
=a
t
+u
at
(2.8)
h
t+1
=h
t
+u
ht
(2.9)
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
t
u
t
u
at
u
ht
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
N
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
0;
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
I O O O
0
O O
O O
a
O
O O O
h
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
(2.10)
for t=s+1,...,n. We also assume the initial distribution to be
s+1
N(
0
;
0
),a
s+1
N(
a0
;
a0
) andh
s+1
N(
h0
;
h0
).
The estimation procedure for the TVP-SV-V AR model is illustrated by extending several
parts of the algorithm for the TVP regression model and the main method we use is MCMC
algorithm. Lety =fy
t
g
n
t=1
and! = (
;
a
;
h
). We set the prior probability density as(!)
for!. Given the data y, we draw samples from the posterior distribution,(;a;h;!jy), by
the following MCMC algorithm:
(1) Initialize;a;h and!;
(2) Sampleja;h;
;y;
(3) Sample
j;
(4) Sampleaj;h;
a
;y;
(5) Sample
a
ja;
(6) Samplehja;;
h
;y;
(7) Sample
h
jh;
(8) Go to (2).
24
2.4 Empirical Results
In this section, we present our main empirical results on interest rate transmission. Our
results are shown in turns based on transmission mechanism.
Figure 2.1 Correlation matrix of OMO rate and money market interest rates
From the analysis of the correlation coefficients of the OMO reverse repo rate and
interest rates in the money market of different lagging periods in Figure 2.1, it can be
concluded that: (1) The transmission of OMO reverse repo rate to the interbank market is
more synchronized and more effective than exchange market. The correlation coefficients
between the 7-day product of SHIBOR, DR and R and the OMO 7-day reverse repo rate are
all above 0.4, while the correlation coefficients of the 7-day product of GC and R- are only
0.2. (2) The effectiveness of the OMO 7-day reverse repo rate transmission to the interbank
market lending rate is higher than the repo rate. The correlation coefficient between
SHIBOR 1W and the OMO 7-day reverse repo rate lagged is the highest that is about 0.65,
followed by DR007 that is about 0.6 overall, and then R007 that is about 0.4; SHIBOR ON
25
and DR001 are about 0.2, which is slightly higher than R001. (3) The OMO 7-day reverse
repo rate is not effective for overnight transmission. The correlation coefficient with the
interbank market overnight interest rate is around 0.2, and the correlation coefficient
with the exchange market overnight interest rate is only 0.05-0.06. Therefore, we choose
SHIBOR 1W and DR007 as the starting point for the subsequent short-term money market
interest rate transmission to other markets.
It is worth noting that, unlike foreign literature, the interest rate term we choose is 7
days instead of the mainstream overnight. This result is not only derived from the above
empirical analysis, but also has practical significance. Although in theory, overnight is
better than seven days. However, China may not have the conditions to operate overnight,
because the frequency of overnight expiration is very high, and China’s liquidity gap is
very large. Disturbed by fiscal and regulatory assessments, short-term funds fluctuate
greatly. If the central bank operates a normal rollover through the open market, the daily
operation volume will be very large that is beyond the central bank’s ability to regulate
in a normal way. The seven-day operation period can be a compromise between short-
term and operational volume. And in reality, the minimum period for central bank open
market operations is 7 days. Therefore, whether starting from the empirical results or
from the reality, choosing SHIBOR 1W and DR007 as the starting point for the subsequent
short-term money market interest rate transmission to other markets is a reasonable choice.
2.4.1 Interest Rate Transmission in Money Market
In this part, we mainly study the interest rate transmission in the money market. We use the
TVP-SV-V AR model to estimate the transmission efficiency from short-term policy interest
rates to short-term money market interest rates, and obtain the changes in transmission
efficiency over time. Specifically, we select the 7-day OMO reverse repo rate as the short-
term policy rate, and select the overnight and 7-day interest rates of R, DR, SHIBOR, GC
and R- as the short-term money market interest rate. Among them, the first three belong
26
to the interbank market interest rate, and the latter two belong to the exchange market
interest rate. We select daily frequency data from July 2015 to September 2020, and take
the monthly average to avoid the problem of large interest rate fluctuations caused by
tight funding at the beginning month and the end of the month. The data sources are
the People’s Bank of China, China Foreign Exchange Trade System (CFETS), the National
Interbank Funding Center, Shanghai Stock Exchange and Shenzhen Stock Exchange.
Figure 2.2 Impulse response of short-term money market interest rates in the interbank
market
The result of TVP-SV-V AR model describes the immediate impulse response function
of various short-term money market interest rates in the interbank market at different
time nodes. Judging from the results presented in Figure 2.2, the impulse response of
the three overnight products: DR001, R001, and SHIBOR ON to the OMO interest rate
has continued to rise over time in general. However, the impulse response of R001 has
declined in 2020. This phenomenon may be caused by the structural shortage of liquidity
caused by the impact of the COVID-19 in 2020. The overnight product represented by
R001 reflects the liquidity situation of non-bank institutions while the overall liquidity
of non-bank institutions is very tight in 2020 and fluctuate greatly with policy changes.
The overall impulse response of the three 7-day products: DR007, R007, and SHIBOR 1W
27
to the OMO interest rate has continued to decline over time. This result may be related
to the continuous introduction of new policies and mechanisms by the PBoC in recent
years. Since the most common type of open market operation is 7-day reverse repo, DR007,
R007, and SHIBOR 1W react more significantly to policy changes. Meanwhile, volatile
policies cause the market gradually passivate to policy changes so that there is a decline
in transmission efficiency. However, it is worth noting that after July 2019, the impulse
response of DR007 began to rise over time. This may be related to the central bank’s
consideration of DR007 as the benchmark interest rate in the benchmark interest rate
system and continuous reform of the DR007 formation mechanism. The impulse response
of R007 is relatively stable and it only declined from 2018 to 2019, and remained stable
for the rest of the time. This may be because the central bank began to emphasize the
importance of DR007 at that time. Since R007 is the predecessor of DR007 and DR007 is
the improved version of R007, the gradual strengthening of DR007 has led to a decline
in the importance of R007 to the market. In such case, the impulse response of R007
has declined. In recent years, the impulse response of SHIBOR 1W has also remained
stable. This may be because the central bank has already prepared to gradually shift the
short-term money market benchmark interest rate from SHIBOR 1W to DR007. Therefore,
SHIBOR 1W is relatively less affected by the policy and has maintained good stability.
From the perspective of the degree of response of different products, overnight products
are obviously greater than 7-day products. This is because the overnight period is shorter
and the impact of policy changes is greater and faster. Among them, the response of
DR001 is greater than that of R001 and SHIBOR ON, and the response of R007 is greater
than that of SHIBOR 1W and DR007. However, only R007 has a significantly greater
response than others while the difference among the response of other products is not
great. This may be because R007 has a longer history and has been the only short-term
money market benchmark interest rate in fact for a long time. Therefore, the transmission
mechanism is smoother.
28
Figure 2.3 Impulse response of short-term money market interest rates in the exchange
market
The result of TVP-SV-V AR model describes the immediate impulse response function
of the short-term money market interest rate products in exchange market at different
time nodes. According to Figure 2.3, we can find that the impulse response of the two
overnight products: GC001 and R-001 on the OMO interest rate has continued to rise over
time, but the response of R-001 is greater than that of GC001. The impulse response of
the two 7-day products : GC007 and R-007 on the OMO interest rate has also continued
to rise over time, but the response of GC007 is slightly greater than that of R-007. This
result shows that at least in the exchange market, the process of interest rate marketization
has indeed significantly improved the linkage effect and transmission efficiency between
short-term money market interest rates and open market reverse repo rate.
On the whole, the transmission efficiency of the open market reverse repo rate to
the interbank market is higher than that of the exchange market, and the transmission
efficiency to the overnight product is significantly higher than the 7-day product. From
the comparison of DR and SHIBOR, the transmission efficiency of DR001 is slightly higher
than that of SHIBOR ON, while the transmission efficiency of SHIBOR 1W is slightly
29
higher than that of DR007, but the difference is not much. In summary, at this stage,
in the transmission of OMO interest rate to short-term money market rates, DR and
SHIBOR are not much different. However, considering that SHIBOR was launched earlier
and is announced daily by the quoting bank while DR is determined by actual market
transactions, theoretically, the transmission efficiency of SHIBOR should be significantly
higher, which is inconsistent with the result. This indicates that the actual transmission
efficiency of DR is higher than the expectation. Moreover, the process of promoting DR
to replace SHIBOR may not lead to a decline in the transmission efficiency of monetary
policy. On the contrary, it may lead to a further increase in the transmission efficiency
of monetary policy. This result also provides a reference for the benchmark interest rate
reform in major developed economies in the world in response to the exit of LIBOR. The
replacement of LIBOR with new RFR benchmark interest rates in the future may not
lead to a decline in the monetary policy transmission efficiency. At the same time, it can
overcome the shortcomings of LIBOR so that we don’t need to worry too much about the
impact of this reform on the market and policies.
In order to further study the situation of SHIBOR 1W, we also explored the impulse
response of DIBO007 and IBO007 to the OMO rate in Figure 2.4. Considering that DIBO,
IBO, and SHIBOR are all interest rates in the interbank lending market, this research is
very valuable for understanding SHIBOR. We find that the situations are similar to that
of SHIBOR 1W, both of which have continued to decline over time, and the response of
IBO007 is greater than that of DIBO007 and SHIBOR 1W. This result reflects that the
overall response of the lending interest rate to the OMO rate is constantly decreasing,
and the importance of lending market represented by SHIBOR is declining while the
repurchase market represented by DR is growing fast to replace the position of lending
market.
30
Figure 2.4 Impulse response of interbank lending interest rates
2.4.2 Interest Rate Transmission in Bond Market
2.4.2.1 Interest Rate Transmission in NCD Yield
In this part, we mainly study interest rate transmission in the bond market. Within one
year, the most representative body in the bond market is NCD rather than government
bond or CDB bond. Therefore, we first use the SV AR model to estimate the transmission
efficiency from the short-term money market interest rate to the NCD yields. According to
the previous analysis, we select DR007 and SHIBOR 1W as the short-term money market
interest rates, and select the 1-month, 3-month, 6-month, and 1-year AAA-rated NCD
yield to represent the NCD yields. We select daily frequency data from July 2015 to
September 2020, and take the monthly average to avoid the problem of large interest
rate fluctuations caused by tight funding at the beginning and the end of the month. The
AAA-rated NCD yield is taken from ChinaBond Commercial Bank AAA-rated NCD yield
31
to maturity. The data sources are CFETS, the National Interbank Funding Center, and
Chinabond Pricing Center. The ChinaBond NCD yield to maturity is currently the most
authoritative among China’s NCD yield index.
Figure 2.5 Impulse response of NCD yield to DR007
As shown in the figure 2.5, 2.6, and 2.7 above, the NCD yield responds significantly to
the impact of the DR007 and SHIBOR 1W. From the perspective of the direction of impulse
response, facing a monetary market interest rate shock, the NCD yield of all maturities
will have a positive change, which means the increase in money market interest rate will
lead to an increase in the NCD yield. Take the NCD yield to DR007 as an example, the
impulse response of 3-month, 6-month, and 1-year NCD yield is relatively fast where a
significant response is made in the third period. 1-month NCD yield is relatively slower
that reaches its peak in the fourth period, indicating that the timeliness and durability
of the impact of DR007 on the NCD yield of different maturities is similar, though the
response speed of the medium and long-term interbank deposit yield is slightly faster
than the short-term NCD yield.
Relatively speaking, the response of NCD yield to the SHIBOR shock is slightly stronger
than its response to the DR007 shock. Taking the 3-month NCD yield for example, facing
32
Figure 2.6 Impulse response of NCD yield to SHIBOR 1W
Figure 2.7 Impulse response of NCD yield to MLF
33
the impact of SHIBOR 1W and DR007 shock, the response lag time is basically the same.
But from the perspective of the strength and significance of the response, the 3-month
NCD yield has a slightly stronger response to SHIBOR 1W than DR007. However, it is
worth noting that the convergence rate of the NCD yield to MLF shock is significantly
faster than its response to SHIBOR 1W and DR007. Specifically, the 1-month, 3-month,
6-month, and 1-year NCD have immediate response levels of 65.04%, 55.48%, 48.78%,
and 42.32% to DR007. The maximum response levels are 86.13%, 87.04%, 80.21%, and
71.04%. The immediate response levels of 1-month, 3-month, 6-month, and 1-year NCD to
SHIBOR 1W are 76.58%, 68.51%, 58.02%, and 49.43%, and the maximum response levels
are 92.78%, 93.58%, 85.87%, and 76.16%. In general, the NCD yield of various maturities
is very sensitive to DR007 and SHIBOR 1W, but relatively speaking, the response to
SHIBOR 1W is better. In specific, the immediate response level is about 7-13% higher and
the maximum response level is about 5-7% higher. At the same time, the response level
basically weakens with the lengthening of the maturity.
Figure 2.8 Variance decomposition of NCD yield to DR007
The results of variance decomposition revealed in Figure 2.8, 2.9, and 2.10 show that
the final explanatory powers of DR007 on the fluctuations in 1-month, 3-month, 6-month,
34
Figure 2.9 Variance decomposition of NCD yield to SHIBOR 1W
Figure 2.10 Variance decomposition of NCD yield to MLF
35
and 1-year NCD are 43.75%, 42.5%, 40.15%, and 37.98%, respectively. At the same time,
SHIBOR’s final explanatory powers for the fluctuations are 36.45%, 35.23%, 32.35%, and
30.17%, respectively. The final explanatory powers of MLF are 34.49%, 37.04%, 38.32%,
and 37.79%, respectively. This further shows that short-term money market interest rates
can be transmitted to the NCD yield curve, and its transmission efficiency is very high.
However, the transmission effectiveness weakens with the extension of the period. The
response is stronger in the short and medium term, but weaker in the long term. In
addition, the explanation power of DR007 is significantly stronger on NCD than that of
SHIBOR 1W and MLF.
In summary, DR007 is weaker than SHIBOR 1W in the transmission efficiency to the
NCD yield by about 5-7% but is stronger than SHIBOR 1W in explanatory power.
2.4.2.2 Interest Rate Transmission in Government Bond and CDB Bond Yield
Among the medium- and long-term bonds, that is, bonds with maturities of more than
one year, the most representative subjects in the bond market are government bonds
and policy financial bonds. Since CDB (China Development Bank) bonds account for an
extremely high proportion of the total policy financial bonds, CDB bonds are usually
used to represent policy financial bonds. We first use the SV AR model to estimate the
transmission efficiency from short-term money market interest rates to the yields of gov-
ernment bonds and CDB bonds. According to the previous analysis, we select DR007 and
SHIBOR 1W as short-term money market interest rates, and select the most representative
1-year and 10-year government bond and CDB bond yield to represent the medium- and
long-term bond yield. We select daily frequency data from July 2015 to September 2020,
and take its monthly average to avoid the problem of large interest rate fluctuations caused
by tight funding at the beginning and the end of the month. The yields of government
bonds and CDB bonds are taken from the ChinaBond government bond yield to maturity
and ChinaBond CDB bond yield to maturity. The data sources are CFETS, the National
36
Interbank Funding Center, and Chinabond Pricing Center. The ChinaBond government
bond yield to maturity and ChinaBond CDB bond yield to maturity are currently the most
authoritative among China’s government bond yield index and CDB bond yield Index.
Figure 2.11 Immediate impulse response of government bond yield to DR007 and SHIBOR
1W
The result of TVP-SV-V AR model describes the immediate impulse response function
of each bond product at different time nodes. From the results of immediate response pre-
sented in Figure 2.11 and 2.12, the responses of government bonds to DR007 and SHIBOR
1W are relatively similar. Specifically, the impulse responses of 1-year government bond
to DR007 and SHIBOR 1W have continued to decline slowly over time, but it began to
stabilize in July 2019, while the impulse response of 10-year government bond to DR007
and SHIBOR 1W appeared to rise slowly before 2017, but then began to slowly fall. The
results of CDB bonds are similar to those of government bonds but the response is greater.
This is because CDB bonds and government bonds are closely related while CDB bonds
are larger in scale and more market-oriented than government bonds. These results reflect
that the effectiveness of transmission from short-term money market interest rates to the
bond market interest rates has not been continuously improved. On the contrary, there
37
Figure 2.12 Immediate impulse response of CDB bond yield to DR007 and SHIBOR 1W
has been a certain decline. Under the background of the continuous reform of the PBoC
to improve the monetary policy transmission efficiency, this result is different from our
expectation. The reason for that may be due to the large fluctuations in China’s policies
and frequent introduction of new mechanisms. The large fluctuations in policies hedge
against the efficiency gains brought about by reforms. It also reflects the need for the
PBoC to continue to promote reforms to improve transmission efficiency. However, as the
new policy framework becomes stable in the future, the PBoC will continue to deepen the
improvement of existing instruments under the new framework and continue to channel
the process of monetary policy transmission. We believe that the effectiveness of interest
rate transmission may increase.
From the comparison of the effect of DR007 and SHIBOR 1W, we find that there is little
difference in the immediate response of 1-year and 10-year government bond and CDB
bond to DR007 and SHIBOR 1W. In specific, the immediate response levels to DR007 are
31.20%, 13.31%, 40.22%, and 16.02%, and the immediate response levels to SHIBOR 1W
are 37.42%, 16.59%, 46.96%, and 18.89%. But relatively speaking, the impact of SHIBOR
38
1W will be greater. The difference in 1-year bond is about 6.5%, and the difference in
10-year bond is about 3%. This may be because the volatility of DR007 is relatively greater
than that of SHIBOR 1W so that relatively DR007 contains more noise and has a relatively
small immediate effect on a single bond product. However, it is worth noting that whether
it is DR007 or SHIBOR 1W, the immediate response levels of the four bond products are
not very satisfactory, and gradually weaken as the period lengthens.
Figure 2.13 Max impulse response of government bond yield to DR007 and SHIBOR 1W
From the perspective of the maximum response shown in Figure 2.13 and 2.14, the
impulse response of 1-year and 10-year government bond to DR007 and SHIBOR 1W has
continued to decline slowly over time. The results of CDB bond are similar to those of
government bond but the response is greater. From the comparison of the response from
DR007 and SHIBOR 1W, we find that the maximum response of these two to 1-year and
10-year government bond and CDB bond is very close, which is similar to the immediate
response result. Specifically, the maximum response levels to DR007 are 51.09%, 25.51%,
60.36%, and 33.51%, and the maximum response levels to SHIBOR 1W are 56.97%,
27.75%, 65.62%, and 35.27%. But relatively speaking, the impact of SHIBOR 1W is greater.
39
Figure 2.14 Max impulse response of CDB bond yield to DR007 and SHIBOR 1W
The difference in 1-year bond is about 5.5%, and the difference in 10-year bond is about
2%. In addition, the maximum response levels of the four bond products are relatively
high, and gradually weaken as the period lengthens.
In summary, the transmission efficiency of DR007 is weaker than that of SHIBOR 1W
in terms of 1-year and 10-year government bond and CDB bond.
2.4.2.3 Interest Rate Transmission in Three Factors of Government Bond and CDB
Bond Yield Curve
In order to better understand the impact of short-term money market interest rates on the
yield curve of government bonds and CDB bonds, we use the method of Diebold and Li
(2006) so that the government bond and CDB bond yield curve are broken down into three
factors: level, slope, and curvature through the Nelson-Siegel model.
40
The Diebold-Li model is a variant of the Nelson-Siegel model, obtained by reparam-
eterizing the original formulation. For observation datet and time to maturity, the
Diebold-Li model characterizes the yieldy
t
() as a function of four parameters:
y
t
() =L
t
+S
t
(
1e
) +C
t
(
1e
e
) (2.11)
whereL
t
is the long-term factor, or level,S
t
is the short-term factor, or slope, andC
t
is
the medium-term factor, or curvature. determines the maturity at which the loading on
the curvature is maximized, and governs the exponential decay rate of the model.
In specific, we use 1-month, 2-month, 3-month, 6-month, 9-month, and 1-year, 2-year,
3-year, 4-year, 5-year, 6-year, 7-year, 8-year, 9-year, 10-year, 15-year, and 20-year bond
yield data to fit Nelson-Siegel model curve. Slope factor can represent short-term impact
to measure the steepness of the yield curve. Curvature factor can represent medium-
term impact to measure the curvature of the yield curve. And level factor can represent
long-term impact to measure the overall value change of the yield curve. We select daily
frequency data from July 2015 to September 2020, and take its monthly average to avoid
the problem of large interest rate fluctuations caused by tight funding at the beginning
and the end of the month. The yields of government bonds and CDB bonds are taken
from the ChinaBond government bond yield to maturity and ChinaBond CDB bond yield
to maturity. The data sources are CFETS, the National Interbank Funding Center, and
Chinabond Pricing Center.
As shown in Figure 2.15 and 2.16 presented above, we analyze the impact on the
government bond yield curve. The slope and curvature factor respond significantly to
DR007 and SHIBOR 1W while the level factor does not respond significantly. From the
perspective of the response direction, facing one standard deviation money market interest
rate shock, all three factors will have positive changes, which means the increase in money
market interest rates will lead to an overall increase in the yield curve, a steeper curve,
and a larger curvature. Take the impact of DR007 on three factors as an example, the slope
41
Figure 2.15 Impulse response of three factors of government bond yield curve to DR007
Figure 2.16 Impulse response of three factors of government bond yield curve to SHIBOR
1W
42
and curvature factor have the fastest response speed, reaching the peak of the change to
DR007 in the second period; while the level factor does not until the seventh period. These
indicate that DR007 has a very large impact on the yield curve of government bonds in
the short and medium term, and it takes effect quickly while the long-term impact is not
obvious.
From the perspective of response, the immediate response levels of slope, curvature,
and level factor to DR007 shock are 24.65%, 43.20%, and 3.95%, and the maximum
response levels are 44.40%, 54.56%, and 16.22%; the immediate response levels to SHIBOR
1W shock are 28.89%, 56.06%, and 3.80%, and the maximum response levels are 44.17%,
61.88%, and 16.35%. Therefore, the response level of curvature factor to DR007 is
significantly less than to SHIBOR 1W, but the response levels of the other two factors to
DR007 and SHIBOR 1W are not much different. But in general, the response levels of
slope and curvature factor are great, while the response level of level factor is not ideal.
Figure 2.17 Variance decomposition of three factors of government bond yield curve to
DR007
Variance decomposition results in Figure 2.17 and 2.18 show that the final explanatory
powers of DR007 on the fluctuation of slope, curvature, and level are 42.89%, 26.30%, and
43
Figure 2.18 Variance decomposition of three factors of government bond yield curve to
SHIBOR 1W
25.44%, respectively. The final explanatory powers of SHIBOR 1W are 31.66%, 28.61%,
and 19.09%, respectively. All these show that short-term money market interest rate can
be transmitted to the yield curve of government bond, and their effectiveness is great.
However, the effectiveness of transmission weakens with the extension of the period, and
the response is strong in the short and medium term, but weak in the long term. In
addition, DR007’s ability to explain short-term and long-term changes is significantly
stronger than SHIBOR 1W, but its ability to explain the medium-term changes is slightly
weaker. This means that DR007’s transmission ability to the government bond yield curve
is generally stronger than SHIBOR 1W.
Therefore, in terms of the three factors of the government bond yield curve, the
transmission efficiency of DR007 on the curvature factor is obviously weaker than that of
SHIBOR 1W, but it is basically the same on the other two factors while DR007 is stronger
than SHIBOR 1W in explanatory power.
As shown in Figure 2.19 and 2.20 above, we also analyze the impact on CDB bond
yield curve. Since CDB bonds are relatively similar to government bonds, their credit
44
Figure 2.19 Impulse response of three factors of CDB bond yield curve to DR007
Figure 2.20 Impulse response of three factors of CDB bond yield curve to SHIBOR 1W
45
ratings are very high that is infinitely close to the credit ratings of government bonds so
that CDB bonds are also important bond products in the bond market. Therefore, similar
to the results of the analysis of the government bond yield curve, slope and curvature
factor have a significant response to DR007 and SHIBOR 1W shock, but the level factor
does not respond significantly. From the perspective of the response direction, facing one
standard deviation monetary market interest rate shock, all three factors will undergo
positive changes. This means that the increase in money market interest rates will lead
to an overall increase in the yield curve, a steeper curve, and a larger curvature. Take
the impact of three factors on DR007 as an example, the slope and curve factors have the
fastest response speed, reaching the peak of the change to DR007 in the second period;
while the level factor is not until the seventh period. These indicates that DR007 has a
very large impact on the CDB bond yield curve in the short and medium term, and it takes
effect quickly, but the long-term impact is not obvious.
The immediate response levels of slope, curvature, and level factor to DR007 shock
are 32.22%, 58.71%, and 4.22%, and the maximum response levels are 53.28%, 87.31%,
and 21.33%; the immediate response levels to SHIBOR 1W shock are 37.00%, 56.08%, and
6.53%, and the maximum response levels are 51.77%, 87.62%, and 20.19%. Therefore,
the response level of curvature factor to DR007 is not much different from to SHIBOR
1W, but the response of the other two factors to DR007 is larger than to SHIBOR 1W by
about 1-1.5%. Another interesting phenomenon is that the response of the three factors
of the CDB bond yield curve to DR007 and SHIBOR 1W is significantly stronger than
the response to the three factors of the government bond yield curve. This may be due to
the larger scale of CDB bonds compared to government bonds and the higher degree of
marketization. But in general, the response of slope and curvature factor is ideal, but the
response of level factor is small.
Variance decomposition results in Figure 2.21 and 2.22 show that the final explanatory
powers of DR007 on the fluctuation of slope, curvature, and level factor are 46.20%,
46
Figure 2.21 Variance decomposition of three factors of CDB bond yield curve to DR007
Figure 2.22 Variance decomposition of three factors of CDB bond yield curve to SHIBOR
1W
47
26.30%, and 17.07%, respectively. The final explanatory powers of SHIBOR 1W are
31.98%, 19.38%, and 11.47%, respectively. All these show that short-term money market
interest rates can be transmitted to the CDB bond yield curve effectively. However, the
effectiveness of transmission weakens with the extension of the period, and the response is
stronger in the short and medium term, but weaker in the long term. In addition, DR007’s
ability to explain short-term, medium-term, and long-term is significantly stronger than
SHIBOR 1W, which shows that the transmission ability of DR007 to the CDB bond yield
curve is generally stronger than SHIBOR 1W. At the same time, we can also find that
whether it is DR007 or SHIBOR 1W, its ultimate explanatory power is lower than the
three factors of government bond yield. This is because CDB bonds are more susceptible
to market factors than government bonds so that relatively the explanation ability of
short-term money market interest rates on CDB bonds is relatively poor.
In summary, in terms of the three factors of the CDB bond yield curve, DR007 is
basically the same as SHIBOR 1W in the transmission efficiency on the curvature factor,
but is stronger in the other two factors by about 1-1.5%. Meanwhile, DR007 performs
better in explanatory power than SHIBOR 1W.
2.4.2.4 Interest Rate Transmission in Enterprise Bond Yield
The most representative subjects in the bond market with maturity of more than one year
are government bonds and policy financial bonds, which are also the benchmark for the
entire bond market. However, they are not credit bonds while enterprise bonds are the
major subject of the credit bond market. Unfortunately, past researches often only focus
on government bonds and CDB bonds, and don’t involve the field of enterprise bonds.
Therefore, there are large deficiencies and shortcomings in the bond market transmission
research. Here we use the SV AR model to estimate the transmission efficiency from short-
term money market interest rates to enterprise bond yields in an attempt to make up for
this deficiency. According to the previous analysis, we select DR007 and SHIBOR 1W as
48
short-term money market interest rates, and select the most representative 1-year, 5-year,
and 10-year enterprise bond yield to represent enterprise bond yields. We select daily
frequency data from July 2015 to September 2020, and take its monthly average to avoid
the problem of large interest rate fluctuations caused by tight funding at the beginning
and the end of the month. The enterprise bond yield is taken from the CSI enterprise bond
yield to maturity provided by China Securities Index Company. We specifically select the
AAA-rated, AA-rated and AA minus-rated CSI enterprise bond yield to maturity. The data
sources are CFETS, China Interbank Funding Center and China Securities Index Company.
CSI enterprise bond yield to maturity is currently the most authoritative among China’s
enterprise bond yield indicators, and the acquisition of this data needs to be authorized
by CSI. Though many papers have been authorized by ChinaBond Pricing Center to use
the bond yield to maturity data, the bond yield to maturity data of China Securities
Index Company is still rarely used. The reason is that China Securities Index Company
and ChinaBond Pricing Center are two different companies. Therefore, obtaining the
data of the China Securities Index Company requires additional authorization while few
institutions in the academia can obtain its authorization to obtain data. However, in the
field of enterprise bonds, the ChinaBond Pricing Center has not launched an indicator
similar to the CSI enterprise bond yield to maturity. Therefore, it is difficult to obtain data
to investigate the situation of enterprise bonds. In such case, it is very hard to obtain the
data of CSI enterprise bond yield to maturity and the acquisition of this data is already
one of our major contribution. As far as we know, we are the first one to explore the
transmission of short-term money market interest rates to enterprise bond yields, and the
first to use the CSI enterprise bond yield to maturity in this field. This research is of great
significance to clearly understand the transmission of short-term money market interest
rates to the bond market, and it is also a major contribution.
In China, AAA-rated enterprise bonds are the highest-grade enterprise bonds, while
traditionally, AA-rated enterprise bonds are considered investment grade bonds, and
49
Figure 2.23 Impulse response of AAA-rated enterprise bond yield to DR007
Figure 2.24 Impulse response of AAA-rated enterprise bond yield to SHIBOR 1W
50
Figure 2.25 Impulse response of AA-rated enterprise bond yield to DR007
Figure 2.26 Impulse response of AA-rated enterprise bond yield to SHIBOR 1W
51
AA minus-grated enterprise bonds are high yield bonds. As shown in Figure 2.23, 2.24,
2.25, and 2.26, the AAA-rated and AA-rated enterprise bond yield have a significant
response to the impact of DR007 and SHIBOR 1W. From the perspective of the response
direction, in the face of a monetary market interest rate shock of one standard deviation,
enterprise bonds of various maturities will undergo positive changes. This means that
a rise in money market interest rates will lead to an increase in enterprise bond yields.
Take the enterprise bond yields against DR007 as an example, the yields of 1-year AA-
rated and 5-year AAA-rated enterprise bond have the fastest response speed where the
responses are significant; the 1-year AA-rated, 5-year AAA-rated, 10-year AAA-rated and
AA-rated enterprise bond follow, peaking in the third and seventh period respectively.
These indicate that the impacts of DR007 on the yields of enterprise bonds of different
maturities are different in the timeliness and durability. In specific, the response speed of
the short- and medium-term enterprise bond yield is faster than the long-term enterprise
bond yield, but the duration is shorter.
The immediate response levels of the 1-year, 5-year, and 10-year AAA enterprise
bond to DR007 are 41.74%, 34.90%, and 21.03%, and the maximum response levels
are 68.67%, 41.21%, and 32.15%; the immediate response levels to SHIBOR 1W are
50.02%, 43.64%, and 26.68%, and the maximum response levels are 74.55%, 46.87%,
and 37.36%. The immediate response levels of the 1-year, 5-year, and 10-year AA-rated
enterprise bond to DR007 are 35.09%, 25.24% and 15.79%, and the maximum response
levels are 54.48%, 29.45%, and 32.74%; the immediate response levels to SHIBOR 1W
are 43.32%, 34.48%, and 23.75%, and the maximum response levels are 57.41%, 34.48%,
and 36.42%. Therefore, the 1-year, 5-year, and 10-year AAA-rated enterprise bond and
AA-rated enterprise bond have a smaller response to DR007 than SHIBOR 1W. Among
them, the gap in AAA-rated enterprise bond is about 5-6%, and the gap in AA-rated
enterprise bond is about 3-5%. This may be because money market interest rates have
been dominated by SHIBOR for a long time, and their reactions are more sensitive.
52
Figure 2.27 Variance decomposition of AAA-rated enterprise bond yield to DR007
Figure 2.28 Variance decomposition of AAA-rated enterprise bond yield to SHIBOR 1W
53
Figure 2.29 Variance decomposition of AA-rated enterprise bond yield to DR007
Figure 2.30 Variance decomposition of AA-rated enterprise bond yield to SHIBOR 1W
54
The results of variance decomposition in Figure 2.27, 2.28, 2.29, and 2.30, show that
the final explanatory powers of DR007 for the fluctuations in the 1-year, 5-year, and
10-year AAA-rated enterprise bond are 43.10%, 30.83%, and 27.18%, respectively, and the
final explanatory powers of the fluctuations in the AA-rated enterprise bond are 29.77%,
16.72%, 28.91%, respectively. At the same time, SHIBOR’s final explanatory powers
for the fluctuations in the AAA-rated enterprise bond are 39.17%, 29.53%, and 27.20%,
respectively, and the final explanatory powers for the fluctuations in AA-rated enterprise
bond are 25.94%, 16.13%, and 29.71%, respectively. This shows that DR007’s ability to
explain short- and medium-term is significantly stronger than SHIBOR 1W, but its ability
to explain long-term is slightly weaker. This shows that DR007’s transmission ability to
enterprise bond yields is generally stronger than SHIBOR 1W.
Therefore, in terms of AAA-rated and AA-rated enterprise bond, the transmission
efficiency of DR007 is about 5-6% and 3-5% weaker than SHIBOR 1W while DR007 is
stronger than SHIBOR 1W in explanatory power.
Figure 2.31 Impulse response of AA minus-rated enterprise bond yield to DR007
But for AA minus-rated enterprise bond, the situation is quite different. As presented
in Figure 2.31, 2.32, 2.33, and 2.34, the response of the AA minus-rated enterprise bond
55
Figure 2.32 Impulse response of AA minus-rated enterprise bond yield to SHIBOR 1W
Figure 2.33 Variance decomposition of AA minus-rated enterprise bond yield to DR007
56
Figure 2.34 Variance decomposition of AA minus-rated enterprise bond yield to SHIBOR
1W
to DR007 and SHIBOR 1W is not very significant. Among them, only the 1-year and
5-year AA minus-rated enterprise bond have a significant response to the shock while the
response is positively significant only in the first period. This shows that whether it is
DR007 or SHIBOR 1W, it is difficult for them to transmit to AA minus-rated enterprise
bond, especially 10-year AA minus-rated enterprise bond. Specifically, the immediate
response levels of the 1-year, 5-year, and 10-year AA minus-rated enterprise bond to DR007
are 28.69%, 17.88%, and 6.92%, and the maximum response levels are 28.69%, 17.88%,
and 11.24%; the immediate response levels to SHIBOR 1W are 38.07%, 27.71%, and
13.65%, and the maximum response levels are 38.07%, 27.71%, and 14.48%. Therefore, no
matter it is DR007 or SHIBOR 1W, the response level is relatively low, which also confirms
that the fluctuation of DR007 and SHIBOR 1W is difficult to effectively transmit to the AA
minus-rated enterprise bond. The reason for this phenomenon is that the AA minus-rated
bond belongs to high yield bond where the credit rating is low, the probability of default
is large, and its credit spread is also large. In such case, the short-term money market
interest rate may no longer be an important determinant of its yield. On the contrary,
57
it is the default rate, default recovery rate, and liquidity premium that determine its
yield. Therefore, the relationship between short-term money market interest rates and AA
minus-rated enterprise bond is not close enough to produce effective transmission.
2.4.2.5 Summary
In summary, different from our initial expectation, the transmission efficiency of DR007 to
the bond market is significantly higher than our expectation. In terms of the NCD yields
of various maturities, the 1-year and 10-year government bond and CDB bond yields, the
yield curve of government bonds and CDB bonds, and AAA-rated and AA-rated enterprise
bond yields of various maturities, the overall transmission efficiency of DR007 is about
1-7% weaker than SHIBOR 1W. Considering that the history of DR007 is not long and the
current transmission effect is very close to that of SHIBOR 1W that is a mature product,
this result is very outstanding. Meanwhile, there is still room for further improvement for
DR007. Therefore, we do not need to worry about the negative effect of the ongoing global
benchmark interest rate reform on monetary policy transmission in the bond market. In
addition, the explanatory power of DR007 is almost significantly greater than that of
SHIBOR 1W in terms of all these bonds mentioned above. What needs to be paid special
attention to is that the response of the three factors of the government bond and CDB
bond yield curve to DR007 is very close to that to SHIBOR 1W. Especially on the slope and
level factor of the CDB bond yield curve, the response level to DR007 is greater than to
SHIBOR 1W by about 1-1.5%. This shows that DR007 is very effective in the transmission
of government bond and CDB bond yield curve. In addition, we find that the impact of
MLF on NCD yield is not great, which means that the MLF reform still needs to continue.
At the same time, we find that the response of AA minus-rated enterprise bond yields to
DR007 and SHIBOR 1W is not ideal. This may be because AA minus-rated enterprise
bonds are high yield bonds that are more affected by other factors while relatively the
effect of short-term money market interest rates is not great.
58
2.4.3 Interest Rate Transmission in Credit Market
In this part, we mainly study interest rate transmission in the credit market. We use the
SV AR model to estimate the transmission efficiency from medium-term policy rate to the
credit market interest rate. The reason why we adopt MLF rate is that in the credit market,
the most commonly used reference is not the short-term money market interest rate but
medium-term policy rate. We select the MLF rate as the medium-term policy rate, and
select the 1-year LPR rate, the weighted average of lending interest rate, and the expected
annualized rate of return of 1-year financing product (shadow banking) to represent the
credit market interest rate. We select monthly frequency data from July 2015 to September
2020. Some quarterly frequency data are interpolated to obtain monthly frequency data.
The data sources are the People’s Bank of China, China Interbank Funding Center and the
China Banking Association.
Figure 2.35 Impulse response of credit market interest rates to MLF and LPR
As can be seen in figure 2.35 and 2.36, though the impacts of the weighted average
of lending interest rate and the expected annualized rate of return of 1-year financing
product on the DR007 and the SHIBOR 1W have a positive response, the responses are
59
Figure 2.36 Variance decomposition of credit market interest rates to MLF and LPR
not significant. This shows that the transmission mechanism of the new monetary policy
represented by MLF and LPR in the credit market is still immature. Though from an
intuitive point of view, MLF and LPR have promoted the downward trend of lending
interest rates that is also affirmed by the central bank, from a statistical point of view,
this conclusion has not been verified for now. And more data may be needed to support
this conclusion in the future. In addition, the PBoC may also need to deepen reforms to
further smooth the transmission process of MLF-LPR, and perhaps this effect will be more
significant in the future.
In order to further investigate the interest rate transmission in credit market, we
construct a bank-level micro-database, which contains a wide range of banks. We have
included all 36 banks listed in mainland China into our bank-level micro-database, which
covers relevant bank-level data from 2015Q1 to 2020Q3. As far as we know, our bank-level
micro-database covers the largest number of banks among existing databases. Taking
into account the late listing of many banks, they don’t release some earlier financial data,
making it difficult to obtain detailed financial data. In order to solve this problem, we
supplement these vacant data by querying the financial data disclosed by these banks
60
when they issued bonds. In such case, we obtain this detailed bank-level micro-database,
which is also one of our major contribution. The banks in our database is listed in Table
2.1.
All variables we use are listed in Table 2.2. Following past studies such as Zhou and
Wong (2008), Georgievska et al. (2011), Mbao et al. (2014), and Nanjunga et al. (2016),
here we introduce a wide range of bank-specific controls that include bank size, return on
equity, charge-off rate, deposit-to-loan ratio, non-interest income, and capital adequacy
ratio. Meanwhile, we also include macroeconomic indicators such as GDP and M2 growth
rate since they also play important roles in lending interest rate and debt interest rate, as
increased economic activity or expansion can raise demand for loans leading to higher
lending rate.
We start by estimating a standard panel regression of the following type:
Loanrate
it
=
i
+
t
+
0
R
t1
+
1
ROE
i;t1
+
2
Chargeoff
i;t1
+
3
Depo
i;t1
+
4
Cap
i;t1
+
5
Size
it
+
6
Nonearn
it
+
7
Asset
i;t1
+
8
GDP
t1
+
9
M2
t1
+
it
(2.12)
wherei;t represents individual bank i at time t andR
t1
means the market interest rate
we use that is DR007 or SHIBOR 1W. It is believed that the impact of bank characteristics
on lending interest rate is lagged. As a result, bank-specific characteristics are lagged by
one quarter. Meanwhile, it might be argued that the state of banking sector could affect the
monetary policy rates. As a precaution, lagging one period is a good solution to a possible
endogeneity problem. We also include bank and time fixed effect to represent different
situation of different banks and different time. These variables are very important for
taking into account different bank-specific characteristics and lending interest rate shifts
due to other reasons.
The panel data regression results are reported in Table 2.3. Estimations are taken
progressively starting with bank-specific variables in column (1) and (3). The results in
column (2) and (4) are obtained by adding the macroeconomic variables.
61
Table 2.1 List of banks in the database
No. Stock Code Bank Name
1 600036.SH China Merchants Bank
2 601166.SH Industrial Bank
3 000001.SZ Ping An Bank
4 600016.SH China Minsheng Bank
5 601398.SH Industrial and Commercial Bank of China
6 002142.SZ Bank of Ningbo
7 600000.SH Shanghai Pudong Development Bank
8 601328.SH Bank of Communications
9 601288.SH Agricultural Bank of China
10 601988.SH Bank of China
11 601818.SH China Everbright Bank
12 601169.SH Bank of Beijing
13 601229.SH Bank of Shanghai
14 601939.SH China Construction Bank
15 600919.SH Bank of Jiangsu
16 601009.SH Bank of Nanjing
17 600015.SH Hua Xia Bank
18 601077.SH Chongqing Rural Commercial Bank
19 600926.SH Bank of Hangzhou
20 601997.SH Bank of Guizhou
21 601998.SH China Citic Bank
22 601128.SH Bank of Changshu
23 601838.SH Bank of Chengdu
24 601658.SH Postal Savings Bank of China
25 002958.SZ Qingdao Rural Commercial Bank
26 002966.SZ Bank of Suzhou
27 601916.SH China Zheshang Bank
28 601577.SH Bank of Changsha
29 002807.SZ Bank of Jiangyin
30 600908.SH Bank of Wuxi
31 002839.SZ Rural Commercial Bank of Zhangjiagang
32 601860.SH Jiangsu Zijin Rural Commercial Bank
33 002936.SZ Bank of Zhengzhou
34 603323.SH Jiangsu Suzhou Rural Commercial Bank
35 600928.SH Bank of Xi’an
36 002948.SZ Bank of Qingdao
62
Table 2.2 Variables used in the interest rate transmission analysis in credit market
Type of Variable Variable Name Definition
Dependent Variables
Loanrate Lending Interest Rate=Net Interest In-
come/Interest Earning Assets
Debtrate Debt Interest Rate=Interest Expense/Lia-
bility
Independent Variables
DR007 7-day Depository-Institutions Repo Rate
SHIBOR1W 1-week Shanghai Interbank Offered Rate
Micro-Control Variables
Asset Total Deposit
Nonearn Non-interest Income (logarithm)
ROE Return on Equity
Size Bank Asset Size (logarithm)
Chargeoff Charge-off Rate=Net Charge-off /Loan
Outstanding
Deporatio Deposit-to-loan Ratio=Total Deposits/To-
tal Loans
Cap Capital Adequacy Ratio=Capital/ Risk
Weighted Assets
Macro-Control Variables
GDP Real GDP Growth Rate
M2 M2 Growth Rate
The empirical results show that some bank-specific variables play a significant role in
the determination of lending interest rate. The coefficients of non-earning income growth
and total deposit asset are significant at 1% level. Specifically, non-interest income growth
is found to negatively determine the lending interest rate. The higher the bank’s income
share derived from interest income, the higher the lending interest rate. The positive effect
of deposit-to-loan ratio, which also has the positive coefficient, could be interpreted as
an indication of profit-maximizing behavior whereby banks with lower level of lending
capacity are also inclined to charge higher lending interest rate in order to increase their
profitability positions. With respect to macroeconomic variables, the impact of economic
performance as captured by GDP growth rate is statistically significant and positive, due
to high demand for loan. M2 growth effect is insignificantly positive. In general, these
results are consistent with other studies.
63
Table 2.3 Regression results of lending interest rate to DR007 and SHIBOR 1W
(1) (2) (3) (4)
Loan rate Loan rate Loan rate Loan rate
DR007 0.116*** 0.112***
(4.55) (4.40)
SHIBOR 1W 0.123*** 0.118***
(4.55) (4.40)
ROE -0.0265 -0.0265 -0.0265 -0.0265
(-0.03) (-0.03) (-0.03) (-0.03)
Chargeoff 9.18E-03 9.18E-03 9.18E-03 9.18E-03
(0.66) (0.66) (0.66) (0.66)
Deporatio -0.079* -0.079* -0.079* -0.079*
(-2.11) (-2.11) (-2.11) (-2.11)
Cap 0.298 0.298 0.298 0.298
(1.43) (1.43) (1.43) (1.43)
Nonearn -5.64E-
03***
-5.64E-
03***
-5.64E-
03***
-5.64E-
03***
(-5.73) (-5.73) (-5.73) (-5.73)
Size 0.0117 0.0117 0.0117 0.0117
(0.04) (0.04) (0.04) (0.04)
Asset 5.06*** 5.06*** 5.06*** 5.06***
(6.40) (6.40) (6.40) (6.40)
GDP 3.93E-
03*
4.12E-
03*
(1.82) (1.91)
M2 7.30E-03 6.03E-03
(0.57) (0.48)
Constant 0.123 0.0365 0.0895 0.0189
(1.10) (0.19) (0.81) (0.10)
N 641 641 641 641
R
2
32.15% 32.15% 32.15% 32.15%
Standard errors are in parentheses & probabilities are in square brackets.
***, (**, **) Significance at the 1% (5%, 10%,) level.
64
As our focus of this analysis, a rise in the short-term money market interest rates
(DR007 and SHIBOR 1W) can lead to a rise in lending interest rate at 1% significant level.
The coefficient of DR007 is 11.2% while the coefficient of SHIBOR 1W is 11.8%, which
means the transmission efficiency of SHIBOR 1W is slightly larger than that of DR007 in
lending interest rate. However, the difference is not very large, which suggests that the
effect of DR007 and SHIBOR 1W is similar in this area.
We also change our independent variable to be debt interest rate and estimate the same
standard panel regression of the following:
Debtrate
it
=
i
+
t
+
0
R
t1
+
1
ROE
i;t1
+
2
Chargeoff
i;t1
+
3
Depo
i;t1
+
4
Cap
i;t1
+
5
Size
it
+
6
Nonearn
it
+
7
Asset
i;t1
+
8
GDP
t1
+
9
M2
t1
+
it
(2.13)
As can be seen in Table 2.4, the results for the whole sample are in line with the
view: the better the liquidity of the bank, the larger the bank’s assets, the cheaper debt
interest rate. Among bank-specific characteristics, the most significant variable is capital
adequacy ratio: well capitalized banks may have easier access to debt funding and at
cheaper rates. Other variables have the expected sign but not quite significant. With
respect to macroeconomic variables, the impact of GDP growth rate though positive is
highly statistically insignificant. Most importantly, the results of short-term money market
interest rates indicate that a rise in the short-term money market interest rates (DR007
and SHIBOR 1W) can lead to a rise in debt interest rate at 1% significant level. Moreover,
the coefficient of DR007 is 5.59% while the coefficient of SHIBOR 1W is 5.92%, which
means the transmission efficiency of SHIBOR 1W is still slightly larger than that of DR007
in debt interest rate. However, the difference is not very large, which suggests that the
effect of DR007 and SHIBOR 1W is similar in this area.
Another thing needs to be paid special attention to is that the transmission efficiency
in debt interest rate is much smaller compared with lending interest rate. This may be
due to the fact that despite benchmark lending and deposit interest rate haven’t changed
65
Table 2.4 Regression results of debt interest rate to DR007 and SHIBOR 1W
(1) (2) (3) (4)
Debt rate Debt rate Debt rate Debt rate
DR007 0.0548*** 0.0559***
(7.38) (6.92)
SHIBOR 1W 0.0584*** 0.0592***
(7.38) (6.92)
ROE -0.407* -0.407* -0.407* -0.407*
(-1.80) (-1.80) (-1.80) (-1.80)
Chargeoff 0.0128* 0.0128* 0.0128* 0.0128*
(2.00) (2.00) (2.00) (2.00)
Deporatio 0.0279 0.0279 0.0279 0.0279
(1.49) (1.49) (1.49) (1.49)
Cap -0.199** -0.199** -0.199** -0.199**
(-2.14) (-2.14) (-2.14) (-2.14)
Nonearn 2.66E-04 2.66E-04 2.66E-04 2.66E-04
(0.97) (0.97) (0.97) (0.97)
Size -0.339* -0.339* -0.339* -0.339*
(-1.93) (-1.93) (-1.93) (-1.93)
Asset -0.305 -0.305 -0.305 -0.305
(-1.28) (-1.28) (-1.28) (-1.28)
GDP 7.74E-05 1.74E-04
(0.21) (0.48)
M2 5.69E-03* 5.05E-
03*
(1.92) (1.74)
Constant 0.501*** 0.436*** 0.485*** 0.427***
(13.19) (8.00) (12.87) (7.79)
N 641 641 641 641
R
2
45.94% 45.94% 45.94% 45.94%
Standard errors are in parentheses & probabilities are in square brackets.
***, (**, **) Significance at the 1% (5%, 10%,) level.
66
for a long time, the PBoC has always followed the principle of lending rate first, then
deposit rate, to continuously promote the interest rate marketization process. In this
context, after losing the old benchmark interest rate guidance, banks are more inclined
to refer to short-term money market interest rates for pricing in lending interest rates so
that the transmission efficiency of money market interest rates is higher. As for deposit
interest rates, banks still maintain the traditional practice and anchor benchmark deposit
interest rates, even if it hasn’t changed for many years. In such case, the transmission
efficiency of money market interest rates is lower. However, it is worth noting that whether
it is loan interest rate or debt interest rate, the transmission efficiency of money market
interest rate in the credit market is significantly low. This may be because that the old
benchmark interest rate is gradually losing effect and the new benchmark interest rate
is still developing. At the same time, this also means that the PBoC needs to continue
to accelerate the interest rate marketization process and improve the monetary policy
transmission efficiency.
In summary, the transmission efficiency of SHIBOR 1W is slighter higher than that of
DR007 while the difference is very small that is less than 1%. Meanwhile the transmission
from short-term money market interest rates to credit market is smoother in lending
interest rate than in debt interest rate. However, the transmission efficiency in credit
market is significantly lower than that in bond market, reflecting the urgency of interest
rate reform in credit market.
2.5 Conclusion
In the global financial market, the most widely used benchmark interest rate is LIBOR.
As LIBOR is about to exit, major developed economies such as the US, the Eurozone, the
UK, and Japan that use LIBOR as their benchmark interest rate have also begun to study
withdrawing from LIBOR and fostering alternative benchmark interest rates. At present,
67
the new benchmark interest rates that have been determined by major economies in the
world are RFRs, which are generated based on actual transactions and are different from
the old benchmark interest rate LIBOR. But at the same time, the impact of the monetary
market benchmark interest rate reform on the operational framework of monetary policy
cannot be ignored.
In this chapter, we try to learn from China’s mature experience in developing bench-
mark interest rates based on actual transactions. We hope that through the analysis of
China’s unique benchmark interest rate based on actual transactions, DR, we can pro-
vide information for the current global monetary policy benchmark interest rate reform.
Meanwhile, we also evaluate whether the replacement of the old benchmark interest rate
LIBOR by the new benchmark interest rates RFRs will adversely affect the transmission
of monetary policy. As far as we know, we are the first one to analyze China’s experience
to provide a reference for the global benchmark interest rate reform. Specifically, we
constructe TVP-SV-V AR model, SVAR model, and a micro-database of Chinese banking
industry to analyze China’s monetary policy transmission and explore the role and the
transmission efficiency of the new benchmark interest rate DR and the old benchmark
interest rate SHIBOR in this process. As far as we know, we are the first one to analyze the
transmission efficiency of the new benchmark interest rate DR007 and the old benchmark
interest rate SHIBOR 1W. The bank-level micro-database we construct covers all 36 banks
listed in mainland China, and the number of banks covered is the largest in the existing
database.
Our results show that overall, the transmission efficiency of OMO interest rate to
the interbank market is higher than that of the exchange market, and the transmission
efficiency to overnight products is significantly higher than that of 7-day products. From
the comparison of the two products of DR and SHIBOR, the transmission efficiency of
DR001 is slightly higher than that of SHIBOR ON and the transmission efficiency of
SHIBOR 1W is slightly higher than that of DR007, but there is not much difference
68
between them. But different from our initial expectation, the transmission efficiency of
DR007 to the bond market is significantly higher than our expectation. In terms of the
NCD yields of various maturities, the 1-year and 10-year government bond and CDB bond
yields, the yield curve of government bonds and CDB bonds, and AAA-rated and AA-rated
enterprise bond yields of various maturities, the overall transmission efficiency of DR007
is about 1-7% weaker than SHIBOR 1W. But on the three factors of the government bond
and CDB bond yield curve, the response to DR007 is very close to that to SHIBOR 1W.
Especially on the slope and level factor of the CDB bond yield curve, the response level
to DR007 is greater than that of SHIBOR 1W by about 1-1.5%. This shows that DR007 is
slightly better than SHIBOR 1W in the transmission of the government bond and CDB
bond yield curve. In terms of all the bonds mentioned above, the explanatory power of
DR007 is significantly greater than that of SHIBOR 1W. These show that the transmission
effect of DR007 is better than SHIBOR 1W in some respects. What’s more, according to
the results of the bank-level micro-database, the transmission efficiency of SHIBOR 1W is
slightly higher than that of DR007 in the credit market. However, the gap is very small,
which is less than 1%. Meanwhile, the transmission efficiency of both DR007 and SHIBOR
1W in the lending interest rate is significantly higher than the deposit interest rate. But the
transmission efficiency of both DR007 and SHIBOR 1W in credit market is significantly
lower than that in bond market, reflecting the urgency of interest rate reform in credit
market.
From China’s experience, DR007 has performed very well in monetary policy transmis-
sion. Its transmission efficiency is only slightly weaker than SHIBOR 1W, and even better
than SHIBOR 1W in some areas. Therefore, DR007 has the power to replace SHIBOR
1W as the new benchmark interest rate. Of course, considering that other countries’
experience in benchmark interest rates based on actual transactions is far behind China,
the performance of the new benchmark interest rates RFRs may be weaker than China’s
DR007. But over time, RFRs will eventually become mature so that we still have reasons
69
to be optimistic about the future performance of RFRs. Considering the performance of
DR007 exceeding our expectation, we also believe that the process of replacing the old
benchmark interest rate LIBOR with the new benchmark interest rates RFRS may not lead
to a decline in the monetary policy transmission efficiency. In such case, we do not need to
worry too much about the negative impact of this reform on the market and policies.
70
Chapter 3
Monetary Policy and Macroprudential Policy: Evidence
from China
We use a broad set of Chinese monetary policy instruments and principle component
analysis (PCA) approach to construct a composite monetary policy index (MPI). Com-
pared to other existing indices, our MPI has made several great improvements and can
better capture the trend of Chinese monetary policy. The time-varying parameter VAR
model with stochastic volatility (TVP-SV-V AR model) is used to estimate the effectiveness
of Chinese monetary policy and macroprudential policy. We find strong time-varying
characteristics among Chinese monetary policy, macroprudential policy and their targets.
Chinese monetary policy can promote economic growth and inflation, and its effectiveness
on economic growth is growing, especially after 2015. We also construct a model consist-
ing of state-owned enterprises (SOEs) sector and private-owned enterprises (POEs) sector
with directed lending to explain our findings. However, accommodative macroprudential
policy can cause minor damage to economic growth in the short term, and this negative
effect is increasing while it can also reduce inflation. So far macroprudential policy has
not been an independent policy, while at the same time being more independent, espe-
cially after 2015. Overall, the PBoC’s two-pillar regulation framework of monetary and
macroprudential policy is becoming increasingly effective in achieving its targets, which
also poses a new challenge to the PBoC, making it more prudent in its decision making.
71
3.1 Introduction
As a rapidly developing economy, China’s economy has undergone a number of significant
structural changes over the past few years, posing a serious threat to the governance of
the People’s Bank of China (PBoC, the central bank of China). In order to meet the new
challenges in these new situations, the PBoC has continuously improved and innovated
policy tools and regulation frameworks to ensure stable macroeconomic growth. Among
those policies of the PBoC, monetary policy and macroprudential policy are the two most
important.
Monetary policy is the traditional policy of central banks around the world, and in
China, it is also an important policy of the PBoC to regulate macroeconomy. In recent
years, Chinese monetary policy instruments and regulation framework have changed
dramatically as the PBoC has continued to push for interest rate liberalization. Many
new instruments are established, and many old instruments are declining in importance.
Meanwhile, the PBoC’s monetary policy regulation framework has begun to shift from
a quantitative to a price-based regulation framework, in which interest rates play an
increasingly important role. Under this circumstance, some of the research methods and
conclusions in the existing literature are no longer relevant for studying China’s current
monetary policy, and the effectiveness of Chinese monetary policy remains an important
issue worth exploring.
After the global financial crisis, the PBoC began to accelerate the study and application
of macroprudential policy to ensure financial stability. At the Central Economic Working
Conference in 2015, the PBoC was asked to form a two-pillar regulation framework
consisting of monetary policy and macroprudential policy. Since then, the importance of
macroprudential policy has further increased, and the main target of macroprudential
policy is to overcome the drawbacks of monetary policy and provide additional guarantees
as the PBoC considers and addresses the overall picture in an integrated manner. However,
72
there is little literature that examines the relationship between Chinese monetary policy
and macroprudential policy and the macroeconomic impact of macroprudential policy.
In this chapter, we construct two measures of Chinese monetary policy and macro-
prudential policy. In the meantime, we use the time-varying parameter V AR model with
stochastic volatility (TVP-SV-V AR model) to investigate the effectiveness of Chinese mone-
tary policy and macroprudential policy over the period 2006-2019. In this process, we
primarily address the following issues: How effective is Chinese monetary policy and
macroprudential policy? How does Chinese macroprudential policy interact with mone-
tary policy? Can Chinese macroprudential policy help monetary policy better achieve its
targets?
This chapter contributes to the existing literature from several perspectives. First, we
construct a composite monetary policy index that is our monetary policy index (MPI) using
a broad set of Chinese monetary policy instruments and principle component analysis
(PCA) approach. In such case, we can get a better grasp of Chinese monetary policy than
other literature, avoiding potential problems such as the poor quality of some Chinese
data and the short history of many series. Compared to other existing Chinese monetary
policy indices, our MPI has made several great improvements and can better capture the
trend of Chinese monetary policy. Briefly speaking, our MPI cover the time period until
2019; the frequency of MPI is monthly; MPI is a quantitative index; MPI covers much more
monetary policy instruments, including some importantly newly created instruments;
MPI can distinguish policy changes of different instruments and different magnitudes; our
construction method is more advanced.
Second, we extend the macroprudential measure in Alam et al. (2019) and the inte-
grated Macroprudential Policy (iMaPP) database from 2016 to 2019 using the Chinese
Monetary Policy Implementation Report. Since macroprudential policy is more mature in
2017-2019, our expansion is meaningful to better interpret Chinese monetary policy and
macroprudential policy.
73
Third, we use the TVP-SV-V AR model from Nakajima (2011). Compared to the econo-
metric methods used in the past literature, our method is much better. The TVP-SV-V AR
model is an improved V AR model that has been widely used in the empirical macroeco-
nomic literature. Compared with the traditional V AR model and other improved VAR
models, the TVP-SV-V AR model has two main improvements that enable it to better cap-
ture the characteristics of Chinese monetary policy and macroprudential policy. First,
the TVP-SV-VAR model contains time-varying parameters that allow us to capture the
underlying structure of potential temporal changes in the economy flexibly and robustly.
Second, the volatility in the TVP-SV-V AR model is stochastic rather than constant, which
would make our estimates more accurate.
Fourth, we improve the model in Z. Liu et al. (2020) to clarify the effect of targeted
Chinese monetary policy on the economic growth. The main difference between Z. Liu
et al. (2020) and this chapter is that we consider the favorable credit treatment required
by the PBoC to the private firms while Z. Liu et al. (2020) don’t. Meanwhile, we also
update the values of some parameters using the newest data to capture current situation.
In specific, we construct a model consisting of state-owned enterprises (SOEs) sector and
private-owned enterprises (POEs) sector with directed lending to explain our findings. By
implementing targeted monetary policy, the PBoC provides favorable credit treatment
to POEs by directing banks to lend a minimum share of their available fund to POEs at
below-market interest rate. In such case, PBoC can ease financial repression and offer
more funds to POEs that have higher productivity.
Our main findings are distributed in four aspects. First, there are strong time-varying
characteristics among Chinese monetary policy, macroprudential policy and their targets.
This conclusion illustrates the importance of the time-varying setting and the need to use
advanced econometric methods to analyze the effect of Chinese policies.
Second, Chinese monetary policy can promote economic growth, and its effectiveness
is growing, especially after 2015. This is because the PBoC is refining monetary policy so
74
that it is more targeted rather than aggregate. This also proves that the monetary policy
reform of the PBoC is indeed effective. Nevertheless, few previous studies have addressed
this area. Meanwhile, easy monetary policy can also promote inflation, housing price,
bond price, and equity price.
Third, accommodative macroprudential policy can cause minor damage to economic
growth in the short term, either because of signal channels, ”off the real to the imaginary”
situations or shadow banking, or because of greater economic volatility and poorer policy
responses. Besides this, this negative effect is increasing, which means that policy makers
need to be more careful in adjusting macroprudential policy to avoid side effect. Moreover,
accommodative macroprudential policy can also reduce inflation, and promote housing
price, bond price, and equity price.
Forth, so far macroprudential policy has not been an independent policy, which is
consistent with the PBoC’s stated position. However, the independence is stronger, espe-
cially after 2015, when the PBoC decided to accelerate marketization reforms and improve
countercyclical regulation. This result is very helpful in understanding the role of Chinese
macroprudential policy.
Fifth, as the PBoC focuses more on targeted rather than aggregate monetary policy, the
effect of Chinese monetary policy is increasing in the past few years. Meanwhile, given the
benefit of targeted monetary policy, the urgency of domestic financial reforms and capital
account liberalization may appear less compelling.
Our analysis suggests that the PBoC’s two-pillar regulation framework of monetary
and macroprudential policy is effective in achieving its targets since these two policies
have different effect on different targets. The effectiveness of Chinese monetary policy
and macroprudential policy has continued to increase, especially after 2015, as the PBoC
accelerates the pace of interest rate liberalization and policy reform. Yet, this also poses a
new challenge to the PBoC, making the PBoC’s policy decisions more cautious.
75
The closest research to ours is Klingelh¨ ofer and Sun (2019). Compared to Klingelh¨ ofer
and Sun (2019), this chapter has several important improvements and we have listed the
difference in Table 3.1. First, Klingelh¨ ofer and Sun (2019) cover the period 2000-2015,
while ours covers the period 2006-2019. Second, Klingelh¨ ofer and Sun (2019) use quarterly
data frequency, whereas ours uses monthly data frequency. Third, our monetary policy
measure is much better than theirs in many aspects. These aspects include qualitative or
quantitative index, instruments coverage, treatment towards different instruments and
policy changes of different magnitudes, and calculation method. Fourth, our macropruden-
tial policy measure is better than the one in Klingelh¨ ofer and Sun (2019). Specifically, our
measure is advanced qualitative and is based on the IMF iMaPP database created by Alam
et al. (2019) that is more comprehensive and covers 17 instruments, while their measure is
qualitative and is based on Chinese Monetary Policy Implementation Report. Finally, our
econometric model considers that Chinese monetary policy is time-varying and volatile,
whereas their model doesn’t. In specific, we use TVP-SV-V AR model to estimate the effect
of Chinese monetary policy over different time periods, while they use the SV AR model
that can only estimate the average effect over the long term.
The rest of the chapter is organized as follows. Section 2 describes related literature
review and our contribution. Section 3 provides econometric methodology. Section 4
indicates the construction of our two measures of Chinese monetary policy and macropru-
dential policy. Section 5 shows empirical results to investigate the effectiveness of Chinese
monetary policy and macroprudential policy. Section 6 presents our theoretical model to
explain our empirical findings. Section 7 concludes.
3.2 Literature Review
The existing literature on Chinese monetary policy mainly answers the following four
questions:
76
Table 3.1 Comparison between our research and Klingelh¨ ofer and Sun (2019)
Klingelh¨ ofer and Sun
(2019)
Our research
Time period 2000-2015 2006-2019
Frequency Quarterly Monthly
Monetary policy measure
Character Qualitative measure Quantitative measure
Coverage Include many instruments Include more instruments in-
cluding newly created ones
Weight of each in-
strument
Same Different
Difference of pol-
icy changes
Same Different
Consideration of
different quantity
changes
No Yes
Consideration of
different price
changes
No Yes
Calculation
method
Equal-weighted sum PCA (more advanced)
Macroprudential policy measure
Character Qualitative measure Advanced qualitative mea-
sure
Form 0-1 form Depends on action number,
can be 2/3/4
Source Chinese Monetary Policy
Implementation Report
IMF iMaPP database and
Chinese Monetary Policy Im-
plementation Report
Econometric methodology
Time-varying No Yes
Stochastic volatil-
ity
No Yes
Model SV AR TVP-SV-VAR (more ad-
vanced)
Software used Eviews Matlab
77
1. What determines Chinese monetary policy? Or what is the monetary policy response
function?
2. How effective is Chinese monetary policy?
3. What is the relationship between Chinese monetary policy and other policies?
4. How to measure Chinese monetary policy?
This chapter is related to several strands of the rapidly developing literature. The
first strand examines the effectiveness of Chinese monetary policy and macroprudential
policy. The second strand is a relatively small, but growing literature that considers the
interaction of Chinese monetary policy and macroprudential policy. The third and last
strand consists of includes a number of studies that attempt to create indices that measure
Chinese monetary policy and macroprudential policy.
3.2.1 The Response Function of Chinese Monetary Policy
Since the PBoC is not that open compared to modern central banks, Chinese monetary
policy response function remains a mystery until now. Most researchers have attempted
to estimate using either Taylor rule or McCallum rule, with varying results.
In the area of Taylor rule, researchers mainly use different improved Taylor rule
to fit China by incorporating Chinese characteristics. Y. Chen and Huo (2009) use a
modified Chinese Taylor rule with the growth rate of money supply as an intermediate
target. Different from the past literature, Zheng et al. (2012) introduce a regime-switching
forward-looking Taylor rule over the period 1992-2010. K. Chen et al. (2018) adopt the
idea in Zheng et al. (2012) and estimate an endogenous switching Taylor rule that is
based on institutional facts, which means they use M2 growth rate as an instrument.
More recently, Klingelh¨ ofer and Sun (2018) address measurement uncertainty and model
selection issues by estimating forward-looking Taylor rules using composite aggregate
indicators. They also capture the nonlinearity of the PBoC’s policy response using a
multiple regime threshold regression model.
78
In the area of McCallum rule, researchers also use improved McCallum rules, but
the usage of McCallum rule decreases as the emphasis on price-based monetary policy
increases. Burdekin and Siklos (2008) use an augmented McCallum rule that considers the
PBoC’s emphasis on targeting the growth rate of the money supply to model the PBoC’s
policy response function. S. Sun et al. (2012) also try to use McCallum rule to evaluate the
policy response function of the PBoC.
Besides this, Jawadi et al. (2014) use both a Taylor rule and a McCallum rule for
the period 1990–2008 and use a smooth transition regression model (STR) to allow for
nonlinearity. They do find evidence of nonlinearity existing in policy response function.
To summarize, previous literature has mainly used Taylor rule or McCallum rule to
estimate the response function of Chinese monetary policy. As can be seen from their
results, Chinese monetary policy response function is asymmetric and nonlinear, and the
system is always in flux. Meanwhile, their results prove that Chinese monetary policy uses
both price-based and quantitative instruments. All of these can help us better understand
Chinese monetary policy and choose the right econometric model to avoid these problems.
The main shortcoming of the previous literature is that it does not adequately consider
Chinese particular situation and still attempts to simply estimate Chinese monetary policy
response function using either the improved Taylor rule or McCallum rule, rather than
bringing the two rules more closely together. In the meantime, with the exception of
Klingelh¨ ofer and Sun (2018), none of the other literature has used a composite monetary
policy index to solve the model selection problem. Finally, as interest rate liberalization
accelerates in 2015, Taylor rule may become more important in estimating the response
function of Chinese monetary policy, whereas previous literature has mainly estimated
the period before 2015. In such case, some results may not be reliable anymore.
79
3.2.2 Effectiveness of Chinese Monetary Policy
The effectiveness of Chinese monetary policy is a super important question since it is very
important to know how good monetary policy is implemented and how to improve the
efficiency of monetary policy. Researchers have used different methods to evaluate the
effectiveness of Chinese monetary policy.
On the one hand, much of the literature indicates price-based monetary policy instru-
ments don’t have a strong effect, including Geiger (2006), Q. He et al. (2013), Berkelmans et
al. (2016), etc. On the other hand, much of the literature shows strong effect of price-based
monetary policy instruments and this result is proven by much recent literature, such as
Dickinson and Jia (2007), W. Zhang (2009), Fernald et al. (2014),etc. Recently, much of
the literature proves that Chinese monetary policy is time varying, including Deng and
Xi (2014), J. Zhang et al. (2018), C. Liu et al. (2019), etc. Other literature investigates
the effect of Chinese monetary policy in general, such as C. Zhang and Sun (2017) and
Klingelh¨ ofer and Sun (2019). With the rise of shadow banking in China, some literature
starts to investigate the impact of shadow banking on the effectiveness of monetary policy,
including K. Chen et al. (2018) and Yang et al. (2019).
To sum up, literature attempts to evaluate the effectiveness of Chinese monetary policy.
Much of the literature has focused on the effectiveness of price-based and quantitative
monetary policy, and the results remain ambiguous. Over time, price-based monetary
policy seems to be getting better and better. The main shortcoming of literature is that,
although none of these Chinese monetary policy instruments is dominant, they simply
choose measures to represent monetary policy and attempt to explore their effectiveness.
Yet, in order to get a more convincing result, they should use a composite index of
Chinese monetary policy. Another problem is that after 2013, the PBoC created many new
instruments as an important step in accelerating the interest rate liberalization, which are
still not considered in the recent literature. What’s more, after 2015 the PBoC announced
to accelerate interest rate liberalization and price-based instruments gets more important
80
while so far, most of the literature doesn’t reach the time period after 2015. Finally, most
of the literature ignores the fact that the effectiveness of Chinese monetary policy is time-
varying, that is, if this is not considered, the time period they choose will seriously affect
their conclusions.
3.2.3 Relationship between Chinese Monetary Policy and Other Policies
There is also other literature that investigates the relationship between Chinese monetary
policy and other policies. For example, Chang et al. (2015) explore the constraint of
capital controls and managed exchange rates on monetary policy options for maintaining
macroeconomic stability. Jia et al. (2015) investigate the mix usage of fiscal policy and
monetary policy.
Recently, as macroprudential policy gets more and more important, some literature
starts to investigate the relationship between monetary policy and macroprudential policy.
Wang and Sun (2013) use bank-level panel data and panel fixed effect model to study a set
of policies including monetary and macroprudential policies. They find that interest rate,
RRR and housing policy are insignificantly associated with lower loan growth. They also
point out that the current monetary policy framework and financial structure with ongoing
unfinished reforms impose too great a burden on macroprudential policies to address
financial stability concerns. More recently, Klingelh¨ ofer and Sun (2019) disentangle
and explore the effect of monetary policy and macroprudential policy. They find that
macroprudential policy can be used to retain financial stability without triggering an
economic slowdown, or as a complement to monetary policy to offset the buildup of
financial vulnerabilities arising from monetary easing. A well-designed mix of these two
policies helps to achieve both macroeconomic and financial stability objectives.
To summarize the previous literature on the three aspects of Chinese monetary policy
mentioned earlier, there are several shortcomings in the existing Chinese monetary policy
literature. First, most of the literature doesn’t use a composite index of Chinese monetary
81
policy to explore the effect s of Chinese monetary policy. Second, as the PBoC pushes for
its interest rate liberalization, most of the literature doesn’t fully consider the power of
price-based monetary policy instruments. Third, most of the literature doesn’t consider the
time-varying nature of Chinese monetary policy. Fourth, much of the existing literature
does not address the post-2015 time period, when the importance of price-based monetary
policy instruments is much higher than before. Finally, most of the literature fails to
consider that the new instruments created by the PBOC after 2013 are actually very
important in the regulation framework.
3.2.4 Measures of Chinese Monetary Policy
The final aspect of Chinese monetary policy is the measurement of Chinese monetary
policy. However, in this area, there is little available literature and index. However,
according to R. Sun (2018)’s summary, changes in the policy instruments frequently
used by the PBoC may contain information on the PBoC’s policy stance, so a composite
index that measures the monetary policy stance would be useful for monitoring a set of
indicators, not just one interest rate or one monetary aggregate. Thus, it is important to
construct a composite index to measure Chinese monetary policy, and this index must
consider China’s particular national circumstances.
There are two main methods to measure Chinese monetary policy. One is to use de jure
variable, which means we need qualitative variables to measure Chinese monetary policy.
The most used and most persuasive one is the annual policy statement of the PBoC. But
since the frequency of this index is annual, it is rarely used directly.
Other literature takes a different approach to building a composite monetary policy
index that reflects Chinese monetary policy. There are two main methods, the narrative
method and the instrument-based method.
In the narrative approach, the researcher extracts additional information by reading
a variety of materials and documents such as work reports, diaries, etc., to speculate
82
on the intentions of the policy makers. This approach is very similar to the use of the
PBoC’s policy statement mentioned earlier, which collects more information from a larger
number of sources and documents, obtains more information and generates it more
frequently. Nevertheless, this method requires strict classification criteria, and the value of
the narrative approach is weakening as the PBoC’s statement becomes increasingly vague.
In this area, Shu and Ng (2010) and R. Sun (2013) both use the documents of the PBoC to
infer the information about policy makers’ intentions while these two papers use different
methods. More recently, R. Sun (2018) constructs a narrative index series based on the
method in Shu and Ng (2010) to measure Chinese monetary policy and this index is a
five-value index that contains very tight, tight, neutral, easy and very easy. Along this line,
she finds that the PBoC is mainly relied on interest rates and RRR, and M2 can’t measure
Chinese monetary policy since there is no clear link between M2 and policy changes.
In the instrument-based method, the researcher focuses directly on monetary policy
instruments and statistics changes in those instruments. Then based on these changes,
they create a de jure composite monetary policy index. In this area, D. He and Pauwels
(2008) and Xiong (2012) both look at the monetary policy instruments and try to capture
the changes in these instruments and then assigned a value to each change. In such case,
they create monetary policy measures. The difference is that D. He and Pauwels (2008)
look at the discrete changes in RRR, benchmark interest rates, and the scale of open market
operation. But Xiong (2012) looks at different monetary policy instruments in different
periods. And mainly, he looks at credit plan for banks’ lending, various interest rates,
central bank’s refinancing to banks, RRR and open market operation. More recently, R. Sun
(2015) also uses instrument-based method but she creates a de facto measure of Chinese
monetary policy. The instruments she uses are RRR, SHIBOR, open market operation
interest rate and benchmark interest rates.
The other method is to measure Chinese monetary policy in terms of de facto variable.
In this case, we need to find a quantitative variable for usage that can be price-based or
83
quantity-based. On the other hand, now Chinese monetary policy is transitioning from a
quantitative to a price-based intermediate target, which means that if you use price-based
variables, they might not have worked a few years ago, and if you use quantitative variables,
they might not work in the near future. In the meantime, there are many instruments that
the PBoC has abandoned and new instruments that the PBoC has created, so it is difficult
to decide which one to use.
One method to solve this problem is to construct a composite index to measure Chinese
monetary policy. R. Sun (2015) uses two-stage V AR method and formulates the multiple
unconventional policy tools into a simple operating procedure model. She also points
out that Chinese monetary policy is not pure interest-rate targeting nor pure reserves
targeting, but a mixture. And in this case, Chinese monetary policy is better measured
with a composite index that considers various policy instruments together.
Other literature mainly uses several variables to measure Chinese monetary policy.
Instead of constructing a composite index of Chinese monetary policy, they insert these
variables one by one into their econometric model in an attempt to find a robust conclusion.
And among these variables, the most popular measures of Chinese monetary policy are
RRR, SHIBOR, M2 and benchmark interest rates. RRR is an important measure to reflect
liquidity change and is used in Tang et al. (2013), Fernald et al. (2014), R. Sun (2015),
Berkelmans et al. (2016), Cun and Li (2019) and K. Chen et al. (2018). SHIBOR is a
relatively new measure but gets much attention recently. SHIBOR is used in C. Zhang
and Sun (2017), Chang et al. (2015) and Tang et al. (2013). M2 is a classical measure
of money supply and is broadly used in C. Zhang and Sun (2017), Burdekin and Siklos
(2008), Fernald et al. (2014), Y. Chen and Huo (2009), Koivu et al. (2008) and K. Chen et al.
(2018). Finally, benchmark interest rates are policy interest rates determined by the PBoC.
They are used in Fernald et al. (2014) and Berkelmans et al. (2016).
To sum up, the existing composite index of Chinese monetary policy is not ideal. Most
of the existing composite indices are not quantitative and therefore do not accurately
84
reflect changes in monetary policy. Meanwhile, these indices don’t cover enough monetary
policy instruments and cover only a few major ones. What’s more, these indices do not
consider some of the new monetary policy instruments created by the PBoC, which are
actually very important. Moreover, the frequency of most indices is quarterly, while the
ideal frequency is monthly. Finally, these indices don’t cover the time period of recent
years, such as the period after 2015, which is not useful for studying Chinese monetary
policy over the past few years.
3.2.5 Our Contribution
Compared to Klingelh¨ ofer and Sun (2019), this chapter has several important improve-
ments indicated in Table 3.1 shown before. First, Klingelh¨ ofer and Sun (2019) cover the
period 2000-2015, while this chapter covers the period 2006-2019, which means that this
chapter captures trends in Chinese monetary policy. Second, Klingelh¨ ofer and Sun (2019)
use quarterly data frequency, whereas this chapter uses monthly data frequency, which
implies that this chapter uses higher data frequency. Third, in terms of Chinese monetary
policy measure, our measure is better than the one in Klingelh¨ ofer and Sun (2019). Their
measure is qualitative, while ours is quantitative. While their measure covers a lot of
monetary policy instruments, ours covers a lot more, including some important newly
created instruments. Along this line, each instrument in their measure is weighted the
same, while each instrument in our measure is weighted differently. Moreover, their mea-
sure treats each policy change the same, whereas our measure considers policy changes of
different magnitudes differently. The policy changes described here can be quantitative as
well as price-based changes. For example, our measure treats 50bps RRR cut and 100bps
RRR cut differently, while their measure treats these equally. Our measure puts a 20bps
interest rate cut on a different scale than a 10bps interest rate cut, and their measure
treats them the same. What’s more, their measure uses equal-weighted summation to
calculate the final value, while our measure uses PCA to derive the final value. In this case,
85
our method is more advanced and can extract the main information about these policy
changes. Fourth, as far as Chinese macroprudential policy measure is concerned, their
measure is qualitative, while ours is advanced qualitative. That is, their measure is in the
form of 0-1, while ours can be 2 or 3 or 4, since our measure is measured by the number of
policy actions. Meanwhile, their measure is based on the PBoC’s documents such as the
Chinese Monetary Policy Implementation Report. Our measure is based on the IMF iMaPP
database created by Alam et al. (2019). This database is more comprehensive and covers
17 instruments, which measure this database can capture Chinese macroprudential policy
more accurately. We extend this database from 2016 to 2019 using the Chinese Monetary
Policy Implementation Report. Finally, in terms of econometric methodology, our model
considers that Chinese monetary policy is time-varying and volatile, whereas their model
doesn’t. They use the SV AR model, while we use TVP-SV-V AR model. In this case, we can
estimate the effect of Chinese monetary policy over different time periods, while they can
only estimate the average effect over the long term. Moreover, they estimate the results
using Eviews, while we estimate them using Matlab.
3.3 Econometric Methodology
In this section, we present the econometric methodology to investigate the impact of
monetary policy and macroprudential policy and answer the questions we mentioned
earlier. The main workhorse model we use is the time varying parameter V AR model with
stochastic volatility (TVP-SV-V AR model). Despite we have clarified this model in Chapter
2, here we still present the details of the model.
There is little available literature on the effectiveness of Chinese monetary policy,
macroprudential policy and their interactions. Wang and Sun (2013) use panel fixed-
effects models and Klingelh¨ ofer and Sun (2019) use structural vector autoregression (SV AR)
model.
86
If we extend the study to the effectiveness of monetary policy towards China, there is
much more literature. Among them, many papers have used single-equation regression
method, such as Geiger (2006), Wang and Sun (2013), Berkelmans et al. (2016), and K.
Chen et al. (2018), etc. Single-equation regression method is a classical estimation method
that still has some application. Meanwhile, other researchers have chosen to use vector
autoregression (V AR) models. In such case, they can explore the effectiveness of Chinese
monetary policy by estimating the impulse response function and variance decomposition.
For example, papers such as Dickinson and Jia (2007) use the classical VAR model and
conduct empirical analysis. Recently, some improved V AR models have started to be
widely used. Q. He et al. (2013) and Fernald et al. (2014) use factor-augmented vector
autoregression (FAV AR) models. Meanwhile, C. Zhang and Sun (2017) and Klingelh¨ ofer
and Sun (2019) use SV AR models. Theoretically, W. Zhang (2009) and Yang et al. (2019)
build a DSGE model and use the impulse response function and variance decomposition
to derive their results.
Compared to these econometric methods, our method is much better. First, we con-
struct our MPI using a broad set of Chinese monetary policy instruments using principle
component analysis (PCA) approach. PCA is an advanced statistical method for analyzing
big data problems. By using PCA, we can reflect information on all variables with very
few variables, which is a very effective way to reduce dimensions. The logic of PCA is to
extract the primary and common information contained in the different variables, which
is what we really want to know and exploit. Using the PCA to construct the MPI, we can
get a better grasp of Chinese monetary policy than other literature, avoiding potential
problems such as the poor quality of some Chinese data and the short history of many
series. Details on the construction of the MPI will be mentioned in the next section.
Second, we use the TVP-SV-V AR model from Nakajima (2011), which is developed
from Primiceri (2005). The TVP-SV-V AR model is an improved V AR model that has been
widely used in the empirical macroeconomic literature. Compared with the traditional
87
VAR model and other improved VAR models, the TVP-SV-V AR model has two main
improvements that enable it to better capture the characteristics of Chinese monetary
policy and macroprudential policy. First, the TVP-SV-V AR model contains time-varying
parameters that follow the first-order random walk process. In this case, the TVP-SV-V AR
model allows us to capture the underlying structure of potential temporal changes in
the economy flexibly and robustly and allows both temporary and permanent shifts in
parameters. Second, the volatility in the TVP-SV-VAR model is stochastic rather than
constant, which would make our estimates more accurate. This improvement is due to the
fact that in many cases the economic variables have drift coefficients and shocks of random
volatility, which, if we still use the constant volatility of the TVP-VAR model, would
lead to biased estimates of the time-varying coefficient. According to Nakajima (2011),
the incorporation of stochastic volatility into the TVP estimation significantly improves
estimation performance. We also follow Nakajima (2011)’s computational approach, using
the Markov chain Monte Carlo (MCMC) method in the context of Bayesian inference, as
the likelihood function now becomes intractable to estimate the coefficients.
To better understand the TVP-SV-V AR model, we start from the traditional structural
V AR model, where all parameters are now time-invariant. This traditional structural V AR
model can be expressed as:
Ay
t
=F
1
y
t1
+::: +F
s
y
ts
+u
t
; t =s + 1;:::;n (3.1)
wherey
t
is the k*1 vector of observed variables, andA;F
1
;:::;F
s
are k*k matrices of
coefficients. The disturbanceu
t
is a k*1 structural shock and we assume thatu
t
N(0;
2
),
88
where
=
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
1
0 0 0
0
2
0 0
:
:
:
:
:
:
:
:
:
:
:
:
:
:
:
0 0
k1
0
0 0 0
k
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
(3.2)
Since we assume it’s a structural VAR model, we specify the simultaneous relations
of the structural shock by recursive identification, assuming that A is lower-triangular
matrix:
A =
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
1 0 0 0
a
21
1 0 0
:
:
:
:
:
:
:
:
:
:
:
:
:
:
:
a
k1;1
a
k1;2
1 0
a
k1
a
k2
a
k;k1
1
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
(3.3)
Then by multiplyingA
1
to both sides in the V AR model, we can rewrite this reduced
form V AR model:
y
t
=B
1
y
t1
+::: +B
s
y
ts
+A
1
t
;
t
N(0;I
k
) (3.4)
whereB
i
=A
1
F
i
. After that, we can stack the elements in the rows of theB
i
’s to form
that is ak
2
s 1 matrix and defineX
t
=I
k
(y
0
t1
;:::;y
0
ts
), where
is the Kronecker product.
In such case, we can have:
y
t
=X
t
+A
1
t
(3.5)
Now we can extend this model to be TVP-SV-V AR model where the parameters are all
time varying now.
89
Following the setting we derive before, the TVP-SV-V AR model can be written as:
y
t
=X
t
t
+A
1
t
t
t
; t =s + 1;:::;n (3.6)
Please notice that now
t
;A
t
;
t
are all time varying. We denotea
t
= (a
21
;a
31
;a
32
;a
41
;:::;a
k;k1
)
0
as the stacked vector of the low-triangular elements inA
t
and denoteh
t
= (h
1t
;:::;h
kt
)
0
withh
jt
=log(
2
jt
) forj = 1;:::;k;t =s + 1;:::;n. Meanwhile, we assume parameters follow a
random walk process as follows:
t+1
=
t
+u
t
(3.7)
a
t+1
=a
t
+u
at
(3.8)
h
t+1
=h
t
+u
ht
(3.9)
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
t
u
t
u
at
u
ht
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
N
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
0;
0
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B
@
I O O O
0
O O
O O
a
O
O O O
h
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
1
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
C
A
(3.10)
for t=s+1,...,n. We also assume the initial distribution to be
s+1
N(
0
;
0
),a
s+1
N(
a0
;
a0
) andh
s+1
N(
h0
;
h0
).
The estimation procedure for the TVP-SV-V AR model is illustrated by extending several
parts of the algorithm for the TVP regression model and the main method we use is MCMC
algorithm. Lety =fy
t
g
n
t=1
and! = (
;
a
;
h
). We set the prior probability density as(!)
for!. Given the data y, we draw samples from the posterior distribution,(;a;h;!jy), by
the following MCMC algorithm:
(1) Initialize;a;h and!;
(2) Sampleja;h;
;y;
(3) Sample
j;
90
(4) Sampleaj;h;
a
;y;
(5) Sample
a
ja;
(6) Samplehja;;
h
;y;
(7) Sample
h
jh;
(8) Go to (2).
3.4 Construction of Policy Measures
In this section, we first construct our monetary policy index (MPI) to measure Chinese
monetary policy. Then we extend the database in Alam et al. (2019) from 2016 to 2019
and construct an index to measure Chinese macroprudential policy.
3.4.1 Construction of Monetary policy Index (MPI)
According to R. Sun (2018)’s summary, changes in the policy instruments frequently
used by the PBoC may contain information on the PBoC’s policy orientation; therefore,
a composite index that measures the monetary policy orientation would be helpful in
monitoring a set of indicators, not just one interest rate or one monetary aggregate. Thus,
it is important to construct a composite index that measures Chinese monetary policy, and
this index must take into account China’s particular national circumstances.
Here, we use the PCA method to construct MPI using a broad set of monetary policy
instruments. PCA is an advanced statistical method for analyzing big data problems. By
using PCA, we can reflect information on all variables with very few variables, which is a
very effective way to reduce dimensions. The logic of PCA is to extract the primary and
common information contained in the different variables, which is what we really want
to know and exploit. Using the PCA to construct the MPI, we can get a better grasp of
Chinese monetary policy than other literature, avoiding potential problems such as the
poor quality of some Chinese data and the short history of many series.
91
Compared to other existing Chinese monetary policy indices, our MPI has made several
big improvements to better capture the trend of Chinese monetary policy. Most of the
existing indices do not cover the time period of recent years, such as the period after 2015,
and are not helpful for studying Chinese monetary policy in the past few years. Instead,
our MPI covers the period 2006-2019 and can be used to study Chinese recent monetary
policy. Meanwhile, the frequency of our MPI is monthly, while most indices are quarterly.
More importantly, our MPI is a quantitative index, whereas most existing indices are
qualitative. Moreover, while other indices cover a number of monetary policy instruments,
our index covers more, including some important start-up instruments. In addition, each
of the instruments in these indices is weighted primarily the same, while each of the
instruments in our MPI is weighted differently. Furthermore, our MPI considers policy
changes differently for different magnitudes, whereas their measure treats each policy
change the same. For example, our MPI can treat a 50bps RRR cut differently from a
100bps RRR and a 10bps interest rate cut differently from a 20bps interest rate cut. In
such case, our MPI can better distinguish between policy changes of different magnitudes
and more accurately capture the trend of Chinese monetary policy. Finally, our method is
more advanced since we have used PCA to extract primary and common information. On
the contrary, most of the other indices use a simple weighted average method.
We select the first three principle components to construct our MPI. The cumulative
contribution rate of these three principle components has reached 78%, which is sufficient
for our usage (Table 3.2). Though the eigenvalue of the third principle component is
slightly lower than one, we consider the difference between the third principle component
and the fourth principle component to be larger, but the difference between the third
principle component and 1 is smaller. Meanwhile, the cumulative contribution rate is the
most important index that determines how many principle components we should select.
Thus, we choose the first three principle components to construct our MPI, which we feel
is enough for our usage. The details of our principle components are revealed in Table
92
6. Then we use their eigenvalues as weights to calculate the weighted sum, which is the
value of our MPI.
Table 3.2 Result of Principle Component Analysis (PCA)
Principle
Component
number
Eigenvalue Contribution rate Cumulative
contribution rate
1 4.551 39.79% 39.79%
2 3.400 29.72% 69.51%
3 0.965 8.44% 77.95%
4 0.671 5.87% 83.82%
5 0.573 5.01% 88.82%
6 0.414 3.62% 92.44%
7 0.265 2.32% 94.76%
8 0.179 1.57% 96.33%
9 0.169 1.47% 97.81%
10 0.125 1.09% 98.90%
11 0.052 0.46% 99.36%
12 0.042 0.37% 99.72%
13 0.029 0.26% 99.98%
14 0.002 0.02% 100.00%
Total 11.438
As for the variables we use in the PCA, we select some conventional monetary policy
instruments such as OMO (7-day reverse repo rate), RRR, and interest rate policy (1-year
benchmark lending and deposit rate). Meanwhile, we also select a number of unconven-
tional monetary policy instruments, such as LPR (Loan Prime Rate, EST. 2013): 1-year LPR
rate, SLF (Standing Lending Facility, EST. 2013): the balance of SLF, MLF (Medium-term
Lending Facility, EST. 2014): the balance of MLF and 1-year MLF rate, and PSL (Pledged
Supplementary Lending, EST. 2014): the balance of PSL. These unconventional monetary
policy instruments have been super important monetary policy instruments over the years,
as the PBoC’s frequent switch in their usage has caught the eye of the entire market. More-
over, due to China’s peculiarities, unlike the central banks of other advanced countries,
the PBoC still uses some variables in the market as its own immediate target or base. In
such case, we also select several variables in the market. In the monetary market, we select
93
DR007 and SHIBOR 3M. Due to the short history of DR007, we use R007 instead in the
time period before its inception. The reason why we use R007 is that DR007 is created to
improve R007 to better reflect the liquidity situation. In the aspect of money supply, we
use M2 and total social financing (TSF) growth rate. Finally, we have added the CITIC FCI,
which is a good index of financial market conditions. The details of PCA on these fourteen
variables are shown in Table 3.3.
Table 3.3 Coefficients of Principle Components
PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 PC10
OMO -0.19 0.46 -0.06 0.11 0.36 0.00 -0.21 0.03 -0.20 -0.02
RRR -0.29 -0.07 0.71 0.14 -0.20 -0.29 0.25 -0.12 -0.27 0.20
Deposit
Benchmark
-0.32 0.34 -0.18 -0.24 -0.27 -0.13 0.25 -0.19 -0.14 -0.02
Lending
Benchmark
-0.26 0.39 -0.30 -0.28 -0.24 -0.04 0.13 -0.13 -0.04 -0.03
LPR -0.17 0.14 0.15 -0.09 0.52 0.03 -0.19 0.19 -0.44 0.16
SLF Balance -0.11 0.07 -0.04 0.46 -0.06 0.71 0.11 -0.43 -0.11 0.19
MLF -0.07 0.04 -0.08 0.11 0.56 -0.21 0.57 -0.18 0.46 0.16
MLF Balance 0.03 -0.07 -0.32 0.47 -0.23 -0.31 -0.14 0.10 -0.10 0.35
PSL Balance -0.01 -0.03 -0.35 0.42 0.07 -0.29 0.15 0.11 -0.33 -0.20
DR007 -0.42 -0.08 0.01 0.13 -0.10 0.31 0.33 0.72 0.10 -0.19
SHIBOR 3M -0.42 -0.10 -0.12 -0.08 -0.04 -0.04 -0.32 0.11 0.32 0.62
M2 0.15 0.49 0.19 0.09 -0.05 0.10 -0.16 0.18 0.28 0.03
TSF 0.15 0.45 0.24 0.37 -0.16 -0.19 -0.07 0.08 0.27 -0.09
CITICS FCI 0.52 0.18 -0.05 -0.20 -0.08 0.12 0.39 0.30 -0.25 0.53
Note: Here we just present the result of the first ten principle components.
3.4.2 Construction of Macroprudential Policy Measure
At present, the PBoC has established a regulation framework with two pillars, monetary
policy and macroprudential policy, to achieve its regulation targets. In such case, the role
of macroprudential policy is increasingly important as a pillar in the two-pillar regulation
framework. The targets of Chinese macroprudential policy are to ensure financial stability
and forestall systemic risks. Chinese macroprudential policy framework today covers
94
the macroprudential assessment (MPA) system for financial institutions, macropruden-
tial management of cross-border capital flows, and macroprudential policy for housing
finance, etc. Since monetary policy instruments and frameworks are not sufficient to
ensure stability, macroprudential policies can be implemented simultaneously to maintain
financial stability and prevent systematic financial risks. In such case, the two-pillar
regulation framework of monetary and macroprudential policy can provide additional
guarantees when the PBoC considers and addresses the overall situation. Meanwhile,
macroprudential policy can help the PBoC overcome the drawbacks of monetary policy
through countercyclical adjustments.
In early 2020, the PBoC announced the acceleration of the improvement of the macro-
prudential management framework, the first time the PBoC has specifically deployed
work on macroprudential management. Besides this, the PBoC mentioned the need to
compile macroprudential policy guidelines, build a macroprudential stress testing system,
speed up the establishment of a sound macroprudential management mechanism for
the integration of cross-border capital flows in local and foreign currencies. And more
importantly, the PBoC requested the gradual expansion of the scope of macroprudential
policies, the organization and implementation of systematically important bank assess-
ments, and promote the implementation of supervision and management measures of
financial holding companies. In this context, there is no doubt about the importance that
the PBoC attaches to macroprudential policy. However, macroprudential policy in our
country and in other advanced countries are not yet perfect today, and many central banks
in advanced countries, such as G4 countries, are also perfecting their macroprudential
policies to ward off the impending economic downturn.
In this chapter, we use the macroprudential policy measure created in Alam et al. (2019)
and, following their approach, extend the measure from 2016 to 2019 using the Chinese
Monetary Policy Implementation Report. Alam et al. (2019) construct the IMF Integrated
Macroprudential Policy (iMaPP) database and a macroprudential policy measure. This
95
measure covers 17 policy action indicators for each instrument. Each tightening event
is coded as a +1, each loosening event is coded as a -1, and no action is coded as a
0. We use the sum of the policy action indicators for all 17 instruments as a measure
of macroprudential policy. The 17 instruments include countercyclical capital buffers,
capital conservation, capital requirements, bank leverage ratio, loan loss provisions, credit
growth, loan restrictions, foreign currency lending, loan-to-value ratio, debt-service-to-
income ratio, taxes including stamp duties and capital gains, liquidity requirements,
loan-to-deposit ratio, foreign exchange positions, reserve requirements, capital/liquidity
surcharges for SIFIs and other.
The integrated Macroprudential Policy (iMaPP) database has three advantages over
other databases. First, it provides a comprehensive coverage in terms of instruments,
countries, and time periods. It combines information from five existing databases, as well
as information from the new IMF Annual Macroprudential Policy Survey and various other
sources, such as official announcements by the authorities and IMF country documents.
Second, the iMaPP database provides the average LTV limit ceiling for a given country at
any given point in time, while most other databases only provide dummy-type indicators
of policy action. Third, the iMaPP database will be updated annually using information
from the annual IMF survey.
Unfortunately, the latest version of the integrated Macroprudential Policy (iMaPP)
database was released on March 5, 2019, and only covers data prior to 2017. Following the
calculation method mentioned in Alam et al. (2019) and the integrated Macroprudential
Policy (iMaPP) database, we extend this macroprudential measure from 2016 to 2019
using the Chinese Monetary Policy Implementation Report. Though the final result may
not be as accurate as theirs, we believe our expansion can reflect the trend of Chinese
macroprudential policy in 2017-2019. The main material we use is the Chinese Monetary
Policy Implementation Report, in which the PBoC always refers to macroprudential
policy developments in the monetary policy section as a summary of its work. Since
96
macroprudential policy is more mature in 2017-2019, our expansion makes sense for
better interpreting Chinese monetary policy and macroprudential policy. Compared to
Alam et al. (2019), we also make some adjustments by converting the sign of the indicator.
In such case, a positive indicator implies that macroprudential policy is easier, while a
negative indicator implies that macroprudential policy is tighter.
Figure 3.1 demonstrates the trend in macroprudential policy measure. It can be seen
very clearly that from 2006 to 2014, the main trend of macroprudential policy has been to
tighten. This is because the PBoC wanted to offset the external shocks caused by several
financial crises, including global financial crisis and European debt crisis, to maintain
financial stability. Nevertheless, in some periods, the PBoC also adopted accommodative
policies to maintain economic stability and support economic growth. In 2015-2016, the
PBoC accelerated the process of interest rate liberalization and macroprudential policy has
fluctuated. From 2017 to 2019, the usage of macroprudential policy increased as the PBoC
spent more effort on improving macroprudential policy. In specifc, since the economy
faced greater downward pressure, macroprudential policy shifted towards easing to offset
this shock.
Figure 3.1 Macroprudential policy measure, 2006-2019
97
In Figure 3.2, we summarize the trend of Chinese monetary policy and macroprudential
policy. We find that Chinese macroprudential policy and monetary policy have moved
together over the period 2006-2016. Nonetheless, from 2017 to 2018, these two policies
went in opposite directions. Potential reasons for this include the increased usage of
macroprudential policy, the ongoing deleveraging process, and the establishment of a
countercyclical regulation framework. Then in 2019, this relationship seemed to be
blurred. In the final analysis, sometimes macroprudential policy and monetary policy go
in the same direction but sometimes don’t.
Figure 3.2 Comparison of Chinese monetary policy and macroprudential policy, 2006-
2019
3.5 Empirical Results
In this section, we present our empirical results using TVP-SV-V AR model.
3.5.1 Empirical Results on Economic Growth
Since the frequency of GDP growth rate is quarterly, here we use industrial production
growth rate to measure economic growth. The industrial production index we use is the
total value added of the industrial enterprises above designated size, which is released
98
every month. This index is one of the most important macroeconomic indices in China
that accurately reflects the economic situation. Meanwhile, many research papers have
used this index as an instrumental variable for GDP growth rate because of the close
relationship between these two indices. In such case, we select this index to measure
economic growth.
In this section, we express Chinese monetary policy in terms of our MPI, Chinese
macroprudential policy in terms of macroprudential policy measure, and economic growth
in terms of the growth rate of the total value added of the industrial enterprises above
designated size. Moreover, we denote these three variables as mn, mp, and eg for simplicity.
We use the TVP-SV-V AR model to estimate the effects of these variables, and to ensure the
robustness of our results, we run the MCMC 10,000 times per time to derive coefficients.
However, to ensure that our model fit is acceptable, our model must pass several tests,
each of which will be mentioned later.
3.5.1.1 TVP-SV-V AR Model Tests
The first test is the parameter estimation test. Table 3.4 suggests all the results of our
parameter estimates. Sb, sa and sh denote the matrix of b, a and h mentioned in
the empirical methodology section. Meanwhile, b, a and h are the coefficients of the
three variables in our model, respectively, the parameters of the stacking vector of the
low trigonometric elements in matrix A (the structural matrix that represents the error
term relationship) and the heterogeneous standard deviation of the error term. Since
these matrices are diagonal, the number 1 represents the first element and the number 2
represents the second element.
The parameter estimation quantitative test is the most important test of the TVP-SV-
V AR model, and we have listed all the quantitative indices needed for the test. In this test,
there are two subtests to pass. First, for each parameter, the Geweke index must be low
enough, the lower the better. Geweke is the convergence diagnostic value that is used to
99
test whether the parameters in our model converge to the posterior distribution. To pass
this, the Geweke index should be below 1.96 (95% confidence) to qualify. As we can see,
the Geweke index is below 1 for all parameters, not to mention below 1.96, which means
we pass this test. Moreover, the Geweke index for all parameters is small except for sa2
which has a Geweke index of 0.979, which is a good sign for model testing.
Second, for each parameter, its inefficiency index must be small enough. There are two
situations in which this test can be passed: all parameters have an inefficiency index below
the selected threshold, or only one parameter has an inefficiency index above the selected
threshold. The inefficiency index is used to measure how many samples are needed to
produce an uncorrelated sample. For example, if the inefficiency is 100, then in our model
there will be 100 uncorrelated samples out of the 10,000 that we generate. The threshold
can be 100, or 200, depending on the restriction level. Yet, sometimes it is also acceptable
if the inefficiency is higher than 200 for only two parameters. In our model, there is only
one parameter with an inefficiency above 100, namely sh2’s (128.24). Hence, our model
has passed the second test. In conclusion, our model has passed the parameter estimation
test.
Table 3.4 Parameter estimation result in TVP-SV-V AR model
Parameter Mean Stdev 95%
Upper
Bound
95%
Lower
Bound
Geweke Inefficiency
sb1 0.0023 0.0003 0.0018 0.0028 0.261 10.69
sb2 0.0023 0.0003 0.0018 0.0029 0.204 10.95
sa1 0.0053 0.0014 0.0033 0.0085 0.388 44.65
sa2 0.0055 0.0016 0.0034 0.0097 0.979 45.31
sh1 0.0061 0.0024 0.0035 0.0126 0.055 87.59
sh2 2.2511 0.2157 1.8575 2.7032 0.035 128.24
The parameter estimation qualitative test is another important test of the TVP-SV-V AR
model that is exhibited in Figure 3.3. There are three rows in the figure, each with six
parameters. In such case, we need to do three subtests. To ensure our model is valid, our
model needs to pass all three tests. The first row is the correlation test, ideally the curve
100
drops very quickly from 1. We find that first five plots drop very quickly, while the last
one does not drop very quickly, which is acceptable. The second row is the stability test,
ideally with only fluctuations and not many extreme value points. One thing needs to be
clear is that it is normal for the graph to have fluctuations up or down. As we can see, the
stability of all six graphs shows up very well, which is acceptable and normal. The last
row is the normal distribution test and ideally the graph resembles a normal distribution
curve. It is obvious that all six plots exhibit normal distribution characteristics. To sum
up, we have passed the parameter estimation qualitative test.
Figure 3.3 Parameter estimation result in TVP-SV-V AR model
So far, we have passed all the tests of the model, and our model is acceptable. Hence,
we can confidently use this TVP-SV-V AR model to derive our results. What we need to
do next is to show why the time-varying settings are appropriate and how these variables
have fluctuated over the period 2006-2019. The details are expressed in Figure 3.4. The
X-axis is the time: time 1 is January 2006 and time 168 is December 2019. The first row
101
shows the trend of the three variables. We denote monetary policy, macroprudential policy,
and economic growth as mn, mp, and eg, respectively, and from this row, we find that
these three variables fluctuate very frequently.
The second row is the stochastic volatility of three variables and the Y-axis is the
stochastic volatility. The blue line is the posterior mean and the two reading lines are 1
standard deviation bands. From the first graph, we can find the volatility of monetary
policy is relatively stable. This is because Chinese monetary policy is normal and regular,
which means that the PBoC has been always adjusting monetary policy. This framework is
also known in China as the normalized monetary policy regulation framework. Specifically,
we find an interesting fact that the volatility of monetary policy is declining after about
2013. This feature reflects the transition of Chinese monetary policy from a quantitative
regulation framework to a price-based regulation framework, as well as the increasing
usage of countercyclical regulation.
As for macroprudential policy, we can find two time periods in which volatility is
significantly higher. The first timeframe was in 2010, when the drawbacks of the 2008 RMB
4tn stimulus gradually became apparent. To offset this problem, monetary policy tightened
and the PBoC started to use macroprudential policy frequently. Another timeframe was
after 2017, when the PBoC upgraded its macroprudential assessment (MPA) to the second
pillar of its two-pillar regulation framework. Since then, macroprudential policy has been
on fire again and has been used frequently.
Finally, in terms of industrial production, we find that there are several peaks, which
are due to the business cycle and external shocks in China. The main peaks occurred
during the global financial crisis and European debt crisis, which caused huge external
shocks to Chinese manufacturing industry and economic growth. In addition to this, the
peak in January 2007 is due to the introduction of the new petroleum product pricing
system by the Chinese government, which has had a big impact on the entire industry, as
oil prices are the most important cost in most manufacturing sub-industries. The other
102
peak is in 2018, and this one is more exceptional. This peak is due to the Chinese spring
holiday being moved from January to February because of the lunar calendar. In such
case, January 2018’s growth rate is much higher, while February 2018’s growth rate turns
negative.
Figure 3.4 The time-varying stochastic volatility of variables
3.5.1.2 The Empirical Results of TVP-SV-V AR Model
In the previous section, we have shown that our model has passed all tests, which proves
that our model is valid and useful for deriving the results. In this section, we start to
present our empirical results of TVP-SV-V AR model. The biggest difference between our
results and the traditional results is that our results are characterized by time-varying
effects of variables that change over time. Along this line, we will show the clear evidence
for this feature and prove why this setting is reasonable.
103
Figure 3.5 points out the synchronous relationship among monetary policy, macro-
prudential policy and economic growth. We denote monetary policy, macroprudential
policy, and economic growth as mn, mp, and eg, respectively, and one thing that needs
to be made clear is that in this graph, the relationships mentioned here are simultaneous.
The X-axis is the time: time 1 is January 2006 and time 168 is December 2019. The first
graph presents the effect of monetary policy on macroprudential policy, an effect that is
consistently negative. Meanwhile, the effect became greater before 2014, but remained
stable after that. This is because monetary policy is the most common policy to offset dif-
ferent shocks in order to achieve multiple targets simultaneously, while macroprudential
policy is mainly used to maintain financial stability. Whenever monetary policy moves
towards easing, to guard against potential risks, the PBoC uses macroprudential policy
to offset potential drawbacks, providing double insurance to avoid excessive monetary
policy easing. Moreover, due to the short history of macroprudential policy, the effect
became larger before 2014 and reached full power levels after 2014 as the PBoC experience
is accumulated. What’s more, since the main target of macroprudential policy is not to
support economic growth, this negative effect is small.
The second graph shows that the effect of monetary policy on economic growth is
always significantly positive and stable. This is because Chinese monetary policy has its
consistency and its effectiveness is always remarkable. The third graph presents the effect
of macroprudential policy on economic growth. We can see that this effect is negative, and
small, but not significant. This result suggests that the impact of macroprudential policy
on economic growth is not significant.
Here we reveal the most important empirical result: two impulse response function
figures. Since we use TVP-SV-V AR model that contains time-varying setting, we have two
different kinds of impulse response function figures. The first kind is like Figure 3.6 that
is the impulse response function for different lag lengths in 2006-2019. This figure shows
the response to an external shock in each time point after certain time periods such as
104
Figure 3.5 Time-varying simultaneous relationship among variables
after 4 months. Here we select 4, 8 and 12 periods (months) to be our lag lengths, which
can be used to express short-, medium- and long-term effect.
The X-axis is the time from 2006 to 2019. We denote monetary policy, macroprudential
policy, and economic growth as mn, mp and eg, respectively, and if there is no time
variation, then there should be three horizontal lines in each graph, because the impulse
response function should be the same no matter which time period is chosen. In such case,
each line does not change in time, which means that each line is a horizontal line. Figure
9 indicates clearly that each line is not a horizontal line, and the variation across time is
very pronounced, indicating that the model does have time-varying properties and that
we are correct in adopting the time-varying setup.
105
First, let’s analyze the effects of monetary policy as shown in the first row. In row
1, column 2, this graph shows the effect of monetary policy on macroprudential policy.
We can find that if monetary policy is accommodative, then macroprudential policy will
tighten slightly. Particularly, this effect became smaller in the period 2006-2015 but
larger after 2015. Moreover, this negative effect was relatively consistent after 4, 8, and
12 months. All these results are in line with our expectations. As we mentioned above,
each time monetary policy moves towards easing, to guard against potential risks, the
PBoC chooses macroprudential policy to offset potential drawbacks, providing double
insurance in case of excessive monetary policy easing, making the effect negative. Because
of the short history of macroprudential policy, the efficiency of macroprudential policy has
increased with the experience of the PBoC. In this case, the PBoC did not need to adjust
macroprudential policies as frequently to achieve its targets until 2015, and the effect
became smaller. Nevertheless, after 2015, the PBoC decided to accelerate marketization
reforms and increase the use of countercyclical regulation. In this case, the PBoC has used
more aggressive macroprudential policy with greater effect. As for the consistency across
lag periods, this is because of the importance of macroprudential policy for financial
institutions, whose effects will last longer.
As for the effect of monetary policy on economic growth, in row 1, column 3, we
can find the effect is positive and very significant. This suggests that accommodative
monetary policy can significantly boost economic growth, which is consistent with theory.
More importantly, the effect of all three lags is increasing over time, especially after 2015,
because the PBoC is improving monetary policy to make it more targeted, rather than
aggregate. In such case, monetary policy is increasingly efficient and the impact of one
standard deviation shock is greater. Finally, we can see that the effect grows larger as the
lag length increases. This is because the effect of monetary policy on economic growth is
long-term, i.e., the impact of monetary policy on economic growth that is longer than one
year.
106
Second, let’s analyze the effect of macroprudential policy in row 2, column 1. This
graph is the effect of macroprudential policy on monetary policy, indicating that easier
macroprudential policy would make monetary policy easier. This is because a more
accommodative macroprudential policy is a signal that the PBoC is turning to easing,
and monetary policy will also move to be more accommodative. Meanwhile, this effect
was relatively large from 2006 to 2009, but relatively small after that. The reason for
that is that China had to relax both policies to offset the huge loss caused by the global
financial crisis, which had a larger effect. After this time, the effect was back to its original
level. Earlier we said that after 2015 the PBoC accelerated the marketization reforms and
macroprudential policy became increasingly independent, resulting in the effect becoming
smaller to almost zero, and the three lines remaining relatively consistent, as this effect
is also long-standing. Finally, the effect of monetary policy on macroprudential policy is
significantly greater than the opposite effect, which proves that monetary policy is more
important and that macroprudential policy is used to supplement monetary policy to
provide double insurance against potential damage and maintain stability.
As for the effect of macroprudential policy on economic growth, in row 2, column 3,
we can find the effect is negative and insignificant. This suggests that accommodative
macroprudential policy does not significantly boost economic growth after four months,
which is consistent with its target of maintaining financial stability. In specifc, the effect
was larger after 2009 because the PBoC was more efficient in refining its macroprudential
policy.
Finally, let’s analyze the effect of economic growth shock that is shown in row 3, column
1-2. If the economy is growing at a higher rate, then there is a risk of economic overheating.
In such case, monetary policy and macroprudential policy should be tightened, which is
consistent with the conclusions in the chart. Looking at the magnitude of the response,
we can find that its impact is larger, suggesting that the two policies are more sensitive to
shocks to economic growth. Moreover, the effect of economic growth on monetary policy
107
rose to a peak in 2009 and then declined as the PBoC adopted a very accommodative
monetary policy to offset the impact of the global financial crisis. Nonetheless, after 2015,
this effect declined to close to zero, due to the Chinese government’s increased focus on
high-quality development, which reduced the importance of GDP growth rate. What’s
more, the effect of economic growth shock on macroprudential policy declined prior to
2015 but rose after 2015. This is because until 2015, macroprudential policy development
was dominated by financial stability. After 2015, the PBoC used macroprudential policy
more frequently to refine countercyclical regulation linked to economic growth.
Figure 3.6 Impulse response function on economic growth for different lag lengths in
2006-2019
The second category, presented in Figure 3.7, is the impulse response function for
different time points selected over the period 2006-2019. Here we have selected the 40th,
108
80th, and 120th months of the time period: April 2009, August 2012, and December 2015,
and since our sample has 168 months, this selection can reflect changes over the entire
time period. This figure demonstrates the response to an external shock at the selected
time point, which is very similar to the traditional impulse response.
The X-axis is the time from 2006 to 2019. We denote monetary policy, macroprudential
policy, and economic growth as mn, mp and eg, respectively, and in the absence of time
variation, the three lines should converge to one line in each graph, since the impulse
response function should be the same regardless of the time point chosen. Figure 10
indicates clearly that the three lines do not converge in most of the graphs, and there is a
clear difference among the three lines, indicating that the model does have time-varying
properties and that we are correct in adopting the time-varying setup. We also find that
the logic of monetary policy and macroprudential policy remains constant, while the
magnitude of the impact changes.
First, let’s analyze the effect of monetary policy as shown in row 1, column 2, which
shows the effect of monetary policy on macroprudential policy, as reflected earlier. Besides
this, this negative effect is relatively consistent.
As for the effect of monetary policy on economic growth, in row1, column 3, we find
the effect to be positive and very significant, as in the previous analysis. As the lag length
increases, the effect becomes larger. This is because the effect of monetary policy on
economic growth is a long-lasting effect, that is, it lasts for a long time, more than one year.
Different from J. Zhang et al. (2018), we find that monetary policy works very fast.
Second, let’s analyze the effect of macroprudential policy, as shown in row 2, columns 1
and 3, which show the effects of macroprudential policy on monetary policy and economic
growth. The effect of macroprudential policy on monetary policy is positive as we men-
tioned earlier. Nevertheless, the impact of macroprudential policy on economic growth
has always been negative and is only significant in the early stages of economic growth.
This means that easy macroprudential policy does not promote economic growth and have
109
a negative effect at the beginning, which is not in line with our expectation that even if
easy macroprudential policy does not promote growth significantly, it should have a slight
positive effect.
There are three possible explanations: one is that the easing of macroprudential policy
is a signal for investors to invest in riskier projects that can hurt economic growth; and
two is due to the emergence of ”off the real to the imaginary” scenarios or shadow banking.
Since the financial market in China is not fully marketized, the different fund markets are
relatively independent. If macroprudential policy is more relaxed, funds will be released
into the capital markets and most will stay where they are, or will flow into the virtual
economy, as the channel from the capital markets to the real markets is blocked. In such
case, the money released does not contribute significantly to credit growth. Meanwhile,
institutions are more willing to keep their money in the virtual economy because they
are only willing to lend to large corporations with limited demand. What’s more, real
economy is less profitable, and institutions are more willing to keep their money in the
virtual economy. To make matters worse, as signals of macroprudential easing emerge,
money in the real economy will also flow into the virtual economy. These reasons all
contribute to the fact that money does not flow to credit for economic growth and the
real economy may be hurt; third, looser macroprudential policy increase volatility in the
economy and hurt growth. Recent literature evidence also suggests that greater economic
volatility leads to more policy responses that hurt economic growth.
This result also differs from Klingelh¨ ofer and Sun (2019), who argue that the impact
of macroprudential policy on economic growth is always insignificant. However, their
impulse response function is similar to ours, and their effect is much smaller than ours.
This may be due to the fact that our result is calculated on a monthly basis, while theirs is
calculated on a quarterly basis. In such case, they may have missed some important infor-
mation. Another possible explanation is that our econometric method is more advanced
110
than theirs and captures time-varying features. In this case, our method may find more
convincing evidence.
Finally, the effects of economic growth shock are shown in row 3, columns 1-2. The
effects of both policies are negative to prevent the risk of economic overheating, which is
consistent with our previous findings.
Figure 3.7 Impulse response function on economic growth for different selected times in
2006-2019
Taken together, our model shows that there are strong time-varying characteristics
among the variables. Monetary policy can boost economic growth, and the effectiveness
of monetary policy is increasing, especially after 2015, as the PBoC is refining monetary
policy to make it more targeted rather than aggregate. While monetary policy can boost
111
growth, more accommodative macroprudential policy can cause minor damage to growth
in the short run, either because of signal channels, ”off the real to the imaginary” situations
or shadow banking, or because of higher economic volatility and poorer policy responses.
Besides this, the growing negative effects of macroprudential policy on economic growth
imply that policy makers need to adjust macroprudential policy more carefully to avoid
side effects. As expected, so far macroprudential policy has not been an independent
policy and is mainly used to supplement monetary policy to provide double insurance,
which is consistent with the PBoC’s stated position. However, the independence is stronger,
especially after 2015, when the PBoC decided to accelerate marketization reforms and
improve countercyclical regulation.
3.5.2 Empirical Results on Inflation
In this section, we express Chinese monetary policy in terms of our MPI, Chinese macro-
prudential policy in terms of macroprudential policy measure, and inflation in terms of
the PPI growth rate. The reason why we use PPI rather than CPI is because the data quality
of CPI in China remains doubtful while PPI is widely accepted. Moreover, we denote
these three variables as mn, mp, and ppi for simplicity. As before, we use the TVP-SV-V AR
model to estimate the effects of these variables, and to ensure the robustness of our results,
we run the MCMC 10,000 times per time to derive coefficients. Our model has passed all
the model tests and we can be confident to analyze the results. For simplicity, we don’t
present the detail of tests here, but we are glad to provide upon request.
Here we just reveal the most important empirical result: two impulse response function
figures. From Figure 3.8 and Figure 3.9, our model shows that there are strong time-varying
characteristics among the variables. Our results on inflation are in line with Klingelh¨ ofer
and Sun (2019). Easy monetary policy can promote inflation while this effect will only be
significant after several months. Meanwhile, the effectiveness of monetary policy remains
relatively stable. The reason for that may be multilayered. Since the economic growth rate
112
in China is decreasing, the PBoC has to pay more attention to ensure economic growth and
the importance of inflation is decreasing. Another reason is that after 2010, the structure
of Chinese economy has shifted from external demand to domestic demand. In such
case, inflation is less volatile and the PBoC is not that worried about inflation anymore.
Meanwhile, accommodative macroprudential policy can reduce inflation and this negative
effect is larger during 2009 to 2015. The reason was that during that time, the PBoC was
more efficient in refining its macroprudential policy. Another interesting thing is that the
effect of macroprudential policy on inflation is relatively large, which means the PBoC
needs to be careful to adjust its macroprudential policy.
Figure 3.8 Impulse response function on inflation for different lag lengths in 2006-2019
113
Figure 3.9 Impulse response function on inflation for different selected times in 2006-2019
3.5.3 Empirical Results on Asset Prices
In this section, we investigate the effect of Chinese monetary policy and macroprudential
policy on asset prices including housing price, bond price and equity price. We express
Chinese monetary policy in terms of our MPI and Chinese macroprudential policy in
terms of macroprudential policy measure. As for the asset prices, we express housing
price, bond price, and equity price in terms of the growth rate of the price index of new
residential buildings in 70 large and medium-sized cities, the difference of ChinaBond
composite index, and the growth rate of Shanghai composite index. These three indices are
widely used to measure housing price, bond price, and equity price in China. Moreover,
114
we denote these five variables as mn, mp, hp, bond and equity for simplicity. As before,
we use the TVP-SV-V AR model to estimate the effects of these variables, and to ensure the
robustness of our results, we run the MCMC 10,000 times per time to derive coefficients.
Our model has passed all the model tests and we can be confident to analyze the results.
For simplicity, we don’t present the detail of tests here, but we are glad to provide upon
request.
Here we just reveal the most important empirical result: two impulse response func-
tion figures. From Figure 3.10 to Figure 3.15, our model shows that there are strong
time-varying characteristics among the variables. Both easy monetary policy and easy
macroprudential policy can promote housing price, bond price, and equity price very
quickly and significantly. However, the effect on housing price seems to be long-lasting,
which is because the building cycle of new residential buildings is very long. What’s more,
the effect of both policies on housing price gets weaker while the effects on bond price and
equity price remain relatively stable. This is because the housing price is increasing very
fast in the past few years. In order not to stimulate real estate market, Chinese monetary
policy and macroprudential policy is getting more targeted to ease the upward pressure
on housing price.
3.6 Model
Here we improve the model in Z. Liu et al. (2020) to clarify the effect of targeted Chinese
monetary policy on the economic growth. The main difference between Z. Liu et al. (2020)
and this chapter is that we consider the favorable credit treatment required by the PBoC
to the private firms while Z. Liu et al. (2020) don’t. Meanwhile, we also update the values
of some parameters using the newest data to capture current situation.
We consider a small open economy model with overlapping generations. There is
a continuum of households, each living for two periods. When young, the household
115
Figure 3.10 Impulse response function on housing price for different lag lengths in 2006-
2019
works, consumes, and saves. When old, the household consumes the savings. There are
two sectors: state-owned enterprises (SOEs) sector and private-owned enterprises (POEs)
sector that produce intermediate goods competitively. The final consumption good is a
composite of intermediate goods.
Firms in both sectors rely on bank loans to finance wage and rental payments. Banks
operate in a perfectly competitive market, taking interest rates on deposits and lending as
given. The government and the central bank provide favorable credit treatment to both
SOEs and POEs by directing banks to lend a minimum share of their available funds to
SOEs and POEs at below-market interest rates. The method could be direct or indirect
orders, window guidance or macroprudential assessment. However, for simplicity, we just
add this setting to our model. Banks lend their remaining funds at market interest rates to
116
Figure 3.11 Impulse response function on housing price for different selected times in
2006-2019
SOEs or POEs. Due to the capital control policy, the government imposes taxes on both
capital inflows and outflows.
3.6.1 Households
Each household lives for two periods. Young households work for firms and receive labor
income. They consume a part of their income and save the rest. Old households don’t
work and consume their savings.
A representative household born in period t has the utility function:
Efln(C
y
t
h
H
1+
t
1 +
) +ln(C
o
t+1
)g (3.11)
117
Figure 3.12 Impulse response function on bond price for different lag lengths in 2006-2019
where C
y
t
and C
o
t+1
denote the consumption when young and old, and H
t
denotes
worked hours.
The household wants to choose consumption, saving in domestic and foreign banks,
and capital investment to maximize his utility subject to the following budget constraints:
C
y
t
+D
t
+B
d
ft
+q
k
t
K
0
t
+I
t
+
k
2
(
I
t
K
o
t
¯
I
¯
K
0
)
2
K
o
t
=w
t
H
t
+T
t
+
t
(3.12)
C
o
t+1
=R
t
D
t
+ (1
d
)R
t
B
d
ft
+d
t+1
+ [q
k
t+1
(1) +r
k
t+1
](K
o
t
+I
t
)
t+1
(3.13)
When young, household consumesC
y
t
, savesD
t
andB
d
ft
in domestic and foreign banks,
purchases capitalK
o
t
from the old generation at the priceq
k
t
, and makes investment subject
to a quadratic adjustment with the scale determined by
k
. The income includes wage
118
Figure 3.13 Impulse response function on bond price for different selected times in 2006-
2019
incomew
t
H
t
, transfer from the governmentT
t
, and bequest
t
from the old generation,
which is a constant fraction of the wealth:
t
=fR
t1
D
t1
+ (1
d
)R
t1
B
d
f;t1
+d
t
+ [q
k
t
(1) +r
k
t
](K
o
t1
+I
t1
)g (3.14)
When old, the household’s wealth includes interest earnings on domestic deposits and
after-tax foreign depositR
t
D
t
and (1
d
)R
t
B
d
ft
, dividend incomed
t+1
from firms, and
gross returns from capital including incomes and its own value.
According to no arbitrage condition, the household’s optimal decisions among domestic
and foreign deposits is:
R
t
= (1
d
)R
t
(3.15)
119
Figure 3.14 Impulse response function on equity price for different lag lengths in 2006-
2019
where
d
measures the capital outflow controls.
The law of motion for capital stock is:
K
t
= (1)K
t1
+I
t
(3.16)
3.6.2 Final Goods Sector
Final goods are produced using intermediate goods from SOE and POE sector:
Y
t
= (
t
Y
m
1
m
st
+ (1
t
)Y
m
1
m
pt
)
m
m
1
(3.17)
120
Figure 3.15 Impulse response function on equity price for different selected times in
2006-2019
whereY
t
denotes final good output,Y
st
andY
pt
denote intermediate inputs in SOE and
POE sector.
m
measures the elasticity of substitution between intermediate goods, and
t
measures the expenditure share of SOE goods.
To minimize the cost of final good producer, we can have:
Y
st
=p
m
st
m
t
Y
t
(3.18)
Y
pt
=p
m
pt
(1
t
)
m
Y
t
(3.19)
wherep
st
andp
pt
are the relative price of SOE and POE intermediate goods in final
good units.
121
The zero-profit condition in the final good sector implies:
1 =
m
t
P
1
m
st
+ (1
t
)
m
P
1
m
pt
(3.20)
3.6.3 Intermediate Goods Sector
A firm in sector j (can be s or p for SOE or POE sector) produces a homogeneous in-
termediate goodY
jt
using capitalK
jt
and laborH
jt
as inputs. The production function
is:
Y
jt
=A
j
(K
jt
)
1
(H
jt
)
(3.21)
whereA
j
is the TFP in sector j, and is the labor input elasticity.
Firms in both sectors face competitive input markets and product markets. A firm
needs to obtain loans from banks at the interest rateR
jt
to pay wages and capital rents.
The firm repays the loans at the end of the period when production is completed. Then we
can have the following working capital constraint:
B
jt
=w
t
H
jt
+r
k
t
K
jt
(3.22)
To minimize the cost, we can have:
w
t
H
jt
R
jt
=Y
jt
p
jt
(3.23)
r
k
t
K
jt
R
jt
= (1)Y
jt
p
jt
(3.24)
122
3.6.4 Banks
There is a continuum of competitive banks. The representative bank takes depositD
t
from
households at deposit rateR
t
and lend to firms in SOE and POE sector in amountsB
d
st
and
B
d
pt
. Then we can have:
D
t
B
d
st
+B
d
pt
(3.25)
We assume the government and the central bank provide favorable credit treatment
to both SOEs and POEs by directing banks to lend a minimum share of their available
funds to SOEs and POEs at below-market interest rates. The minimum fraction of SOEs
and POEs are
1
and
2
while the interest rates are 1 andR
lt
(1x), whereR
lt
is the market
loan rate and x is the discount rate.
The direct lending policy implies:
B
gt
minf
1
(B
d
st
+B
d
pt
);B
d
st
g (3.26)
B
pgt
minf
2
(B
d
st
+B
d
pt
);B
d
pt
g (3.27)
whereB
gt
andB
pgt
are the amount of direct lending to SOEs and POEs.
1
and
2
mean
the largest fraction of the total bank loans SOEs and POEs can borrow at the preferential
interest rate.
The representative bank wants to maximizes its profits subject to the three constraints
mentioned before:
B
gt
+R
lt
(1x)B
pgt
+R
lt
(B
d
st
+B
d
pt
B
gt
B
pgt
)R
t
D
t
(3.28)
123
The zero-profit condition for banks implies:
R
t
=
B
gt
B
d
st
+B
d
pt
+ (1
B
gt
B
d
st
+B
d
pt
B
pgt
x
B
d
st
+B
d
pt
)R
lt
(3.29)
3.6.5 Foreign Investors
Foreign investors can lend to domestic firms at the market loan rateR
lt
with capital inflow
tax
l
. External borrowing is subject to a risk premium(
B
l
ft
Y
t
), which is an increasing
function. Then we can have:
(1
l
)R
lt
=R
t
(
B
l
ft
Y
t
) (3.30)
3.6.6 Equilibrium
An equilibrium consists of sequences of allocationsfC
y
t
,C
o
t
,I
t
,K
o
t
,Y
t
,K
st
,K
pt
,H
st
,H
pt
,K
t
,
H
t
,B
st
,B
pt
,B
gt
,B
pgt
,B
t
,B
l
ft
,NX
t
g and pricesfw
t
,R
t
,q
k
t
,r
k
t
,p
st
,p
pt
,R
st
,R
pt
,R
lt
g that solve
the optimizing problems for the households, the firms, and the banks. In the equilibrium,
the markets for the loanable funds, capital, labor, and goods all clear.
The market clearing conditions for loanable funds, capital, labor, and goods are:
B
st
=B
d
st
; B
pt
=B
d
pt
+B
l
ft
(3.31)
H
t
=H
st
+H
pt
; K
t1
=K
st
+K
pt
(3.32)
NX
t
=Y
t
C
y
t
C
o
t
I
t
k
2
(
I
t
K
o
t
¯
I
¯
K
0
)
2
K
o
t
(3.33)
3.6.7 Analytical Results
Here we present some analytical results about directed lending for steady-state resource
allocations and aggregate productivity.
124
For simplicity, we consider a special case where
m
= 1, then production function for
the final goods sector is Cobb-Douglas production function:
Y =Y
s
Y
1
p
(3.34)
Meanwhile, we can have
Y
s
p
s
=Y; Y
p
p
p
= (1)Y (3.35)
We assume that the risk-premium function is:
(
B
l
ft
Y
t
) =exp[
b
(
B
l
ft
Y
t
f
)] (3.36)
where measures the elasticity of the risk premium to the ratio of external debts to
output and
f
is a constant. Then we can have
b
f
=
B
l
ft
Y
t
=
f
+
1
b
ln[
(1
l
)R
l
R
] (3.37)
Using the cost-minimizing solution for the final goods sector and the intermediate
goods sectors, we can get the relative size of SOE sector S as:
S(
d
;
l
;
1
;
2
) =
K
s
K
p
=
R
p
R
s
1
(3.38)
Here the funding cost for POEs is not the market loan rate. Then we can have:
R = (1
d
)R
; R
l
=
(1
d
)R
1
1
1
2
x
(3.39)
125
SOE and POE firm have access to directed lending and normal lending. Thus, the
effective funding cost for SOEs and POEs are:
R
s
=
B
g
+R
l
(B
s
B
g
)
B
s
(3.40)
R
p
=
R
l
(B
p
xB
pg
)
B
p
(3.41)
Finally, we can get the relative size of SOE sector:
S(
d
;
l
;
1
;
2
) =
+b
f
1
(1R
l
)
1
(1R
l
) +R
l
R
l
[
1
(1R
l
) + (1
2
x)]
2
xR
l
+ (1)
1
(1R
l
) + 1b
f
2
xR
l
1
(1R
l
)
1
(1R
l
) +R
l
(3.42)
If
2
= 0 orx = 0, then there is no direct lending to POEs. In such case, the formula of
S returns to the formula in Z. Liu et al. (2020) that is:
S =
1
[1 +
R 1
R
1
1
1
1
(1R
l
b
f
)] (3.43)
Proposition 1. For any given values of
l
< 1,
d
< 1
1
R
, and
1
> 0, the relative size of
the SOE sectorS(
d
;
l
;
1
;
2
) decreases with
2
x.
Proof. Rewrite the formula of S, we can have:
S =A
B +R
l
(1
2
x)
C + (b
f
)R
l
2
x
+D (3.44)
where A, B, C and D are positive. Meanwhile, sinceb
f
<<, as
2
x increases, numerator
decreases and denominator increases. Thus, S decreases with
2
x, or
@S
@
2
x
< 0.
Define aggregate total factor productivity (TFP) as
˜
A =
Y
K
H
1
=A
s
A
1
p
S
1 +S
(3.45)
126
whereA
s
andA
p
denote the exogenous productivity in the SOE and POE sector.
Proposition 2. Under any given policy configurations (
d
;
l
;
1
;
2
), aggregate TFP (
˜
A)
decreases with SOE share S.
Proof. Differentiating
˜
A with respect to S, we can get:
@
˜
A
@S
=
˜
A
1
S(1 +S)
[
1
S] (3.46)
=
˜
A
1
S(1 +S)
1
(1
R
p
R
s
)
whereR
s
R
p
<
1
+(1
1
2
)R
l
1
2
R
l
(1x) =
1
(1R
l
)
1
2
+R
l
x< 1R
l
(1x)< 0. Thus, 1
R
p
R
s
< 0
and
@
˜
A
@S
< 0. This is because even with direct lending, the preferential interest rate POE
can get is always higher than SOE.
Now if we combine Proposition 1 and Proposition 2, we can find that as the PBoC
provides more favorable credit treatment to POEs, the aggregate TFP will increase, which
will lead to the increase of output. In such case, targeted monetary policy can increase the
output. However, compared with the aggregate monetary policy, the targeted monetary
policy requires smaller aggregate change in policy to achieve same effect, which means the
effect of monetary policy on economic growth has increased.
3.6.8 Calibration
In this section, we provide calibrated parameters in Table 3.5. Our principle is to match
the observed environment of Chinese economy. Hence, we keep most of the model setting
in Z. Liu et al. (2020) while we also update the values of some parameters using the newest
data.
First, we set the expenditure share of SOE goods = 0:2 rather than 0.5 or 0.3. This is
because the SOE share in China’s industrial output in 2016 was about 20% while Z. Liu
et al. (2020) use the data from 2000 to 2010.
127
Second, we set the share of directed lending to SOE
1
= 0:3 rather than 0.5. The share
of SOE current liabilities in all industrial firms was about 40% in 2018 while many SOE
bank loans were directed lending, so
1
= 0:3 seems reasonable. However, Z. Liu et al.
(2020) use the data in 2000.
Third, we set the share of directed lending to POE
2
= 0:1. The share of current loans
in inclusive financial system was about 12% in June 2020, so
2
= 0:1 seems reasonable.
Fourth, we set discount rate on directed lending to POEx = 0:2. In May 2020, the
average interest rate of loans in inclusive financial system was 6.03% while the share of
loans whose interest rate is above LPR (3.85%)+3% was more than 18%. Sox = 0:2 seems
reasonable.
Table 3.5 Parameter calibration
Parameter Description Value
Household discount rate 0.665
Inverse of labor supply elasticity 2
h
Utility weight of labor 38
Capital depreciation rate 0.651
k
Capital adjustment cost 5
r
Foreign interest rate 1.629
Transfer from old to young 0.75
Labor income share 0.5
A
p
POE TFP 1.42
Share of SOE output 0.2
m
Substitution elasticity between SOE and POE products 3
1
Share of directed lending to SOE 0.3
2
Share of directed lending to POE 0.1
x discount rate on directed lending to POE 0.2
d
Tax rate on foreign asset 15.84%
l
Tax rate on foreign debt 6.47%
b
Elasticity of risk premium to foreign debt 3
f
Desirable foreign debt-to-output ratio 0.04
128
3.6.9 Summary
By implementing targeted monetary policy, the PBoC provides favorable credit treatment
to POEs by directing banks to lend a minimum share of their available fund to POEs at
below-market interest rate. In such case, the PBoC can ease financial repression and offer
more funds to POEs that have higher productivity, which can also correct the misallocation
of resources that was in favor of SOEs so that the aggregate TFP and output will increase.
Meanwhile, targeted monetary policy requires smaller aggregate change in policy to
achieve same effect compared with aggregate monetary policy. Hence, in line with our
findings in the empirical part, as the PBoC focuses more on targeted rather than aggregate
monetary policy, the effect of Chinese monetary policy is increasing in the past few years.
Meanwhile, given the benefit of targeted monetary policy, the urgency of domestic
financial reforms and capital account liberalization may appear less compelling. Actually
the PBoC is very cautious about capital account liberalization that might increase the
systemic risk and bring more uncertainty. Compared with capital account liberalization,
targeted monetary policy is effective but less risky.
3.7 Conclusion
Based on Chinese characteristics, this chapter constructs two measures of Chinese mone-
tary policy and macroprudential policy. The measure of Chinese monetary policy (MPI) is
a composite monetary policy index that is constructed by using a broad range of Chinese
monetary policy instruments through the principle component analysis (PCA) approach.
Compared to other existing Chinese monetary policy indices, our MPI has made some
significant improvements to better reflect China’s trends in monetary policy. Further-
more, this chapter uses the Chinese Monetary Policy Implementation Report to extent
the macroprudential measure in Alam et al. (2019) and the integrated Macroprudential
Policy (iMaPP) database from 2016 to 2019. Since both measures cover the period over
129
2006-2019, we believe that they are useful in helping us better understanding the recent
changes in Chinese monetary policy and macroprudential policy. Meanwhile, we think
that these two measures are meaningful for further researches on China in the future.
Given the uniqueness of Chinese monetary policy that is the transformation from a
quantitative to a price-based regulation framework, we use the time-varying parameter
V AR model with stochastic volatility (TVP-SV-V AR model) to better capture the character-
istics of Chinese monetary policy and macroprudential policy. Compared to econometric
methods used in the past literature, our method has two main improvements. First, the
TVP-SV-V AR model incorporates time-varying parameters, which allow us to capture the
underlying structure of potential temporal changes in the economy flexibly and robustly.
Second, the volatility in the TVP-SV-V AR model is stochastic rather than constant, which
would make our estimates more accurate.
Our results suggest that there are strong time-varying characteristics among Chinese
monetary policy, macroprudential policy and their targets. This finding illustrates the
importance of the time-varying setting and the importance of using advanced econometric
methods to analyze the effect of Chinese policies. We also find evidence that Chinese
monetary policy can promote economic growth, and its effectiveness is growing, especially
after 2015. We also construct a model consisting of state-owned enterprises (SOEs) sector
and private-owned enterprises (POEs) sector with directed lending to explain our findings.
Moreover, we argue that accommodative macroprudential policy can cause minor damage
to economic growth in the short term and such negative effects are increasing, which
implies that policy makers need to be more cautious in adjusting macroprudential policy
to avoid side effects. Finally, our results confirm that, so far, macroprudential policy
has not been an independent policy, which is consistent with the PBoC’s stated position.
However, the independence is stronger, especially after 2015, which is very useful for
understanding the role of Chinese macroprudential policy. We also investigate the policy
effect on inflation, housing price, bond price, and equity price, and the results meet our
130
expectation. Our study continues the discussion on the usage of Chinese monetary policy
and macroprudential policy in Klingelh¨ ofer and Sun (2019) and makes several important
improvements. As far as we know, we are the first few to conduct research in this area.
Of course, China’s economy is far from fully liberalized, and quantitative monetary
instruments are likely to continue to play an important role. In such case, Chinese
monetary policy will remain different from that of other western economies for a long time.
Nevertheless, our analysis shows that the PBoC is improving its policies. In specific, the
PBoC’s two-pillar regulation framework of monetary and macroprudential policy has been
effective in achieving its targets since these two policies have different effect on different
targets, and the effectiveness of both policies has continued to increase, especially after
2015, as the PBoC accelerates the pace of interest rate liberalization and policy reform.
However, this also poses a new challenge to the PBoC, making the PBoC’s policy decisions
more cautious.
131
Chapter 4
A Welfare Analysis of Financial Integration in a Risky
World with Frictions
The benefits of financial integration remain elusive in the past few decades. To further
investigate the benefits, first we use the Panel Smooth Threshold Regression (PSTR) model
that allows parameters to vary across countries and times, to conduct empirical analysis.
We find strong nonlinear relationship and threshold effects together with heterogeneous
effects across countries and times. Then we improve the two-country neoclassical growth
model in Coeurdacier et al. (2015) by adding FDI and frictions. In such case, our frame-
work can analyze simultaneously the welfare gains from capital scarcity effect, risk sharing
effect, and FDI scarcity effect together with how they interact. Both the empirical and theo-
retical results indicate that financial integration does bring sizable benefits that are mainly
from FDI. In specific, both developed countries and emerging economies benefit from
FDI scarcity effect. Meanwhile, risk sharing effect is significant for developed countries
while capital scarcity effect is negligible for both countries. However, since frictions are
being removed slowly and available observation time is short, welfare gains of financial
integration for now are not large enough that can be explained as the threshold effects,
and we should wait for further evidence.
132
4.1 Introduction
In the past few decades, financial integration has achieved remarkable progress, which
leads to a worldwide impact. However, the welfare gains of financial integration is still an
important problem in international finance. According to Bekaert et al. (2002), during the
late 1980s to early 1990s, developing countries opened their domestic financial market.
And at that time, there was an opinion in the policy circle that financial integration will
be a new engine of growth. After decades of development, now the extent of financial
integration is very deep while developing countries are still chasing further financial
integration. At the same time, we understand the benefits of financial integration well
enough in theory and there are a lot of empirical researches on this topic. But from these
literature, the results are not convincing to show that financial integration is that ideal as
we expect and financial integration does bring sizable benefits to developing countries.
Meanwhile, there is few theoretical paper offering the exact estimates of the welfare gains
of financial integration. So intuitively, two questions comes to our mind: how large is the
welfare gains of financial integration? Does financial integration bring sizeable benefit?
In this chapter, we investigate the welfare gains of financial integration. First, we con-
duct empirical analysis to investigate the effect of FDI, equity, and debt market integration
on economic growth. We believe the reason why we can’t find strong evidence on the wel-
fare gains of financial integration might be the elusive relationship and poor econometric
methodology used in the previous literature. Under this circumstance, considering the
effect might be nonlinear and super complicated, an advanced econometric methodology:
Panel Smooth Threshold Regression (PSTR) model is used. Then we construct a theoret-
ical model trying to explain our empirical findings. The model we use is the improved
two-country neoclassical growth model in Coeurdacier et al. (2015) by adding FDI and
frictions. By using this model, we can analyze simultaneously the welfare gains from
capital scarcity effect, risk sharing effect, and FDI scarcity effect together with how they
interact. In specific, capital scarcity effect is the effect where emerging economies that
133
are lack of capital, can benefit from capital inflows due to the relatively higher price of
capital. Risk sharing effect is the effect where countries can benefit by sharing risk with
other countries. And FDI scarcity effect is similar to capital scarcity effect while now the
capital is FDI. Our theoretical results are in line with the empirical findings.
In specific, we conduct empirical analysis to investigate the effect of FDI, equity, and
debt market integration on economic growth. We use PSTR model to estimate the threshold
effect of FDI, equity, and debt market integration on economic growth. PSTR model allows
smooth transition and multiple regimes, which make PSTR model better than others. Two
interpretations of PSTR model are possible. On one hand PSTR model can be thought of
as a regime-switching model that allows for a small number of extreme regimes associated
with the extreme value of a transition function. Meanwhile, different from the indicator
function, the transition from one regime to another is smooth. On the other hand, PSTR
model can be thought to allow for a continuum of regimes, each one being characterized
by a different value of the transition function. The logic is then similar to that developed in
the standard univariate time series STAR (Smooth Transition Autoregressive) model. And
the main difference between these two models is that PSTR model use panel specifications
and is non dynamic.
PSTR model allows parameters to vary across countries and times. In such case,
it provides a parametric approach of the cross-country heterogeneity and of the time
instability of the effects we want to investigate, since these parameters change smoothly
as a function of a threshold variable. In such case, PSTR model can overcome two major
critics on panel data models. First, the data might be non stationary that will cause
problems in estimating coefficients. Second, there might be cross-section heterogeneity
existing. In such case, if we ignore the parameter heterogeneity among cross-sectional
units, our estimation will be biased. In such case, PSTR model has a strong advantage to
be used in our analysis.
134
Our results indicate that there is a strong nonlinear relationship existing, which means
we can’t use linear regression model to estimate the effect. We also find strong threshold
effects in the relationship between FDI, equity, and debt market integration and economic
growth. Under this circumstance, the aggregate effect of FDI, equity, and debt market
integration on economic growth is highly impacted by the threshold effects, which means
the effect is very complicated. And since the threshold effects rely on KAOPEN that
measures capital account openness, which is affected by multiple factors, strong threshold
effects mean that the aggregate effect of FDI, equity, and debt market integration on
economic growth is heterogeneous across countries and times. But generally speaking, we
are confident that the increase in FDI market integration can promote economic growth.
What’s more, the improvement of equity market integration will first promote and then
inhibit economic growth. Moreover, the increase in debt market integration can promote
economic growth but may face some negative effect. In such case, financial integration can
bring sizable benefits that may be mainly from FDI market integration. Finally, since the
slope parameter is not very large, the transition function is not very sharp. This means
that our regression equation cannot be reduced to the sum of a limited number of regimes
and we must use the PSTR model instead of the PTR model.
In order to explain our empirical findings, we construct a theoretical model trying to
better understand the effect of financial integration, especially FDI market integration on
economic growth. In specific, we consider the widely accepted qualitative claims on the
channels of capital scarcity effect and risk sharing effect. Meanwhile, we also consider
the importance of FDI that can be a new channel to increase welfare and frictions being
removed that may hide the welfare gains. However, these qualitative claims are not widely
evaluated quantitatively. In such case, in order to offer a quantitative estimation of these
channels, we improve the two-country neoclassical growth model in Coeurdacier et al.
(2015) by adding FDI and frictions. Then our framework can analyze simultaneously
the welfare gains from capital scarcity effect, risk sharing effect, and FDI scarcity effect
135
together with how they interact. In our baseline model, the world is divided into two
heterogeneous countries that can invest on other country via FDI and trade a risk-free
bond internationally. Country can be asymmetric in capital scarcity, risk, and country
size. Meanwhile, we also test the situations of different frictions, available assets that can
be traded, FDI productivity, and observation time. These can help us better capture the
welfare gains of financial integration than past literature while it can also broad the door
opened by Coeurdacier et al. (2015) for new empirical research on the growth effect of
financial integration.
Different from past literature, we find financial integration does bring sizable benefits
that are mainly from FDI. In specific, financial integration can lead to a permanent
consumption increase of 11.69% and 21.42% for developed countries and emerging
economies. Even if we restrict the relative level and consider frictions, these numbers will
be 7.42% and 8.37%, which are still large enough. FDI scarcity effect is super important for
both developed countries and emerging economies. The intuition for these results is that
FDI offers a new channel for investment where countries can not only invest in new project
to accumulate capital, but also offer better risk sharing. In our model, we just consider
the most basic benefit of FDI, and if we consider other benefits of FDI such as technology
transfer and productivity growth for domestic firms because of the competition from
foreign firms, FDI may bring more benefit than our estimation. What’s more, developed
countries can benefit a lot from risk sharing effect but emerging economies don’t benefit
from this a lot. Finally capital scarcity effect is small for both countries.
We also find that since frictions are being removed slowly, frictions can highly weaken
the welfare gains of financial integration especially the growth rate of capital and con-
sumption. Frictions may not change the aggregate welfare gains a lot while frictions can
put off the welfare gains and then affect our feeling of welfare gains especially in the early
stage of financial integration. Meanwhile, observation time matters a lot to the welfare
gains of financial integration. However, even until now, the history of financial integration
136
is not long enough and available observation time is also short. These causes us to find
that welfare gains of financial integration for now are not large enough. In specific, if we
just consider first 10 years with frictions, the welfare gains of financial integration are
2.44% and 2.77% for developed countries and emerging economies. If we select first 20
years with frictions, these numbers will be 4.21% and 4.76%. If we include more frictions,
these numbers will be smaller. Under this circumstance, we should be more patient to
wait for further evidence.
Another key finding is that the heterogeneous effect of financial integration across
countries and times is more complicated than Coeurdacier et al. (2015). The welfare
gains of financial integration may vary with heterogeneous levels of capital scarcity, risk
aversion, frictions, and available assets that can be traded. Meanwhile, country size, FDI
productivity, and observation time can also have a significant effect.
The rest of this chapter is organized as follows. Section 2 shows the literature review.
Section 3 provides our PSTR model and empirical results. Section 4 presents our baseline
model of financial integration and our improved model with fading frictions. Section 5
shows our calibration. Section 6 provides our quantitative results including steady state
values and paths of output, consumption, capital, FDI, and bond under financial autarky
and financial integration. Section 7 evaluates quantitatively the welfare gains of financial
integration. Section 8 is sensitivity analysis where we perform a wide rage of robustness
checks to many factors including country size, capital scarcity, FDI productivity, speed of
frictions being removed and observed period. Section 9 presents our extended model and
further discussion. Section 10 concludes.
4.2 Literature Review
On the empirical aspects, so far there is no consensus on the effect of financial integration
on welfare gains. Kose et al. (2006) summarize hundreds of paper in the area of financial
137
integration and growth. Their conclusion is that financial integration may have a positive
but not robust effect on growth, which means it’s a mixed effect. But using measures of
de facto integration or finer measures of de jure integration tends to find more positive
results. In specific, they believe that FDI has the strongest benefits despite the theoretical
presumption though it has not proven easy to document these benefits; the research on
the effects of equity liberalizations is quite promising and equity liberalizations should
generate positive effects; debt is close to a consensus that debt flows do not have the
positive attributes of equity-like flows.
Meanwhile, Obstfeld (2009) is also a nice summary showing that despite an abun-
dance of cross-sectional, panel, and event studies, there is strikingly little convincing
documentation of direct positive impacts of financial integration on the economic welfare
levels or growth rates of developing countries. And there is also little systematic evidence
that financial integration raises welfare indirectly by promoting collateral reforms of
economic institutions or policies. Terrones et al. (2007) also show that empirical results
pertaining to the impact of financial integration on risk sharing across countries are very
mixed. However, Quinn and Toyoda (2008) believe that these conflicting results are from
measurement errors, differing time periods and collinearity among independent variables.
They also conclude capital account openness is positively associated with growth by using
a better de jure measure of capital account openness. But despite there is still no consensus
on the effect of financial integration on welfare gains in the academia, there is no doubt in
policy circle that developing countries have never stopped their steps in chasing further
financial integration.
In specific, some papers focus on the aggregate financial integration or one type of
capital flows while others look jointly at the different types of flows including FDI, portfolio
equity, and debt.
In the aggregate part, Edison et al. (2002) conclude that they are unable to reject
the null hypothesis that international financial integration does not accelerate economic
138
growth even when controlling for particular economic, financial, institutional, and policy
characteristics. Bussiere and Fratzscher (2008) find that no empirical evidence has yet
emerged for the existence of a robust positive relationship between financial openness
and economic growth. They also find that portfolio and debt inflows can promote growth
in the short run and in the medium to longer term while FDI inflows can be a important
driving force. Klein and Olivei (2008) find a statistically significant and economically
relevant effect of open capital accounts on economic growth in a cross-section of countries
over the periods 1986–1995 and 1976–1995. Kose et al. (2009) find that de jure capital
account openness has a robust positive effect on TFP growth while the effect of de facto
financial integration on TFP growth is less clear. They also find strong evidence that FDI
and portfolio equity liabilities boost TFP growth while external debt is actually negatively
correlated with TFP growth.
Some papers evaluate multiple types of capital flows. Durham (2004) find that lagged
FDI and equity foreign portfolio investment do not have direct, unmitigated positive
effects on growth, but some data are consistent with the view that the effects of FDI
are contingent on the ”absorptive capacity” of host countries, with particular respect to
financial or institutional development. Choong et al. (2010) investigate how three types of
private capital flows can promote economic growth in recipient developed and developing
countries. The findings revealed that FDI exhibits a positive impact on growth, while both
foreign debt and portfolio investment have a negative impact on growth in all sample
countries. Aizenman et al. (2013) investigate the relationship between economic growth
and lagged international capital flows, disaggregated into FDI, portfolio investment, equity
investment, and short-term debt. They find that a large and robust relationship between
FDI, both inflows and outflows, and growth; the relationship between growth and short-
term debt is nil before the crisis, and negative during the crisis.
Meanwhile, other papers only focus on one type of capital flows. Henry (2003) finds
that when developing countries liberalize the stock market, their cost of capital falls,
139
investment booms, and the growth rate of output per worker increases. Quinn and Toyoda
(2008) find that capital account liberalization has a positive association with growth in
both developed and emerging market nations and equity market liberalization has an
independent effect on economic growth. Contessi and Weinberger (2009) conclude that
little consensus has emerged as to whether FDI is boon or bane for a country as a whole
and the evidence is still mixed. Bordo et al. (2010) study the effects of foreign currency
debt on currency and debt crises and its indirect effects on short-term growth and long-run
output effects in both 1880–1913 and 1973–2003 for 45 countries. They find that financial
crises, driven by exposure to foreign currency, result in significant permanent output
losses.
On the theoretical aspects, typically the welfare gains from financial integration come
from two channels: efficient capital allocation and international risk sharing. Efficient cap-
ital allocation means that after financial integration, capital flows from capital-abundant
countries to capital-scarce countries while the marginal product of capital is higher in the
latter than in the former. In this case, free capital movements can permit a more efficient
global allocation of savings towards their most productive use. Then financial integration
can accelerate the economic growth of developing countries and developed countries can
also benefit from the capital flows. As for the international risk sharing, since capital
can flow free from one country to another country after financial integration, developed
countries and developing countries can share the risks from the country-specific shocks
through these free capital flows. In this case, both developed countries and developing
countries can benefit from international risk sharing. Typically, we believe that financial
integration reduces the volatility of consumption rather than output.
On the one hand, in the literature of efficient capital allocation, Gourinchas and
Jeanne (2006) are the first try to provide a benchmark to estimate the benefits of financial
integration coming from the capital scarcity of developing countries. And they find that
the gains from financial integration is elusive and relatively small even for countries that
140
stand to receive a lot of capital inflows. Meanwhile, they point out that there might be
some channels not in the textbook model through which financial integration can have a
large impact on the welfare of developing countries. However, their model focuses on a
small open economy, which is not accurate enough.
On the other hand, there is an extensive literature of international risk sharing. More
recently, Colacito and Croce (2010) show that once the intertemporal distribution of
output risk altered by the long-run risks is taken into account, the implied benefits of
keeping international financial markets open can be as high as 10%, which is two times
greater than those found in the existing literature.
However, in these literature, they all estimate the welfare gains either from efficient
capital reallocation or risk sharing but never estimate from both important channels. Until
recently, Coeurdacier et al. (2015) fill this important gap in theoretical literature and
provide an integrated framework where one can study the standard neoclassical efficiency
gains together with gains from risk sharing and investigate how they interact. They find
that financial integration has very heterogeneous effects depending on the stochastic
structure of shocks, the size of countries, and their initial degree of capital scarcity. And
they also point out that financial integration doesn’t bring sizable benefits to any country.
Furthermore, they show that gains from risk sharing can be dampened by an even stronger
capital reallocation, which significantly differs from the risk sharing literature.
But in their model, the only financial asset is bond while Foreign Direct Investment
(FDI) is also an important financial asset for developing countries. Albuquerque (2003)
shows that FDI is less volatile than other capital flows and imperfect enforcement of
contracts and inalienability give a risk sharing advantage of FDI over other capital flows.
Aizenman et al. (2013) find that a large and robust relationship between FDI and growth
while the relationship between growth and short-term debt is nil before the crisis, and
negative during the crisis. Yu (2018) finds that FDI market integration, equity market
141
integration, and debt market integration all have a strong positive effect on economic
growth while FDI market integration has a very strong effect.
Meanwhile, there is a growing number of literature trying to understand why the
welfare gains from financial integration is relatively small. Some of them point out that the
existing frictions weaken the benefits of financial integration and make it relatively small.
But if these frictions are totally removed, the welfare gains from financial integration will
be much larger. However, many past researches totally ignore the existing frictions and
assume a frictionless world. Then under this assumption, they estimate the welfare gains,
which is not reasonable. And just as Colacito and Croce (2010) point out, long-run risks
has a big impact on the risk sharing. In the long run, frictions will be removed slowly.
Thus it’s necessary to consider the effect of fading frictions on financial integration.
There is no doubt that capital controls have an important effect on financial integration.
With capital controls, international capital can’t flow freely from one country to another.
And there is an extensive literature including capital controls in the analysis of financial
integration. Chanda (2005) shows that on deeper examination, capital controls do have an
important effect on economic growth. Depending on a country’s degree of heterogeneity,
capital controls can have negative or positive effects. For countries with high degrees of
heterogeneity, capital controls lead to greater inefficiencies and lower economic growth. At
the same time, for countries which don’t have such heterogeneity, it is shown that capital
controls work to enhance economic growth.
Informational frictions are also very important while the effect of informational fric-
tions is similar to the effect of capital controls. Informational frictions are proven by the
famous gravity model. Because of informational frictions, agent needs to pay more to
buy foreign assets. Portes and Rey (2005) are the first one to extend gravity model from
international trade in goods to assets. They find that the geography of information is
the main determinant of the pattern of international transactions and thus, international
capital markets are not so frictionless as is often assumed in discussions of capital mobility
142
and globalization. Aviat and Coeurdacier (2007) are a nice complement of Portes and Rey
(2005) where they argue that distance affects asset holdings mainly through its impact
on trade in goods. Thus they believe it’s necessary to model trade and financial linkages
together. Meanwhile, they also think that the explanation of informational frictions makes
sense and is a large part of the story.
What’s more, financial frictions are also very important. Bai and Zhang (2012) believe
that the key to understanding why countries don’t have obvious increased risk sharing
from widespread financial integration is that conventional wisdom assumes frictionless
international financial markets while actual markets are far from frictionless. Furthermore,
they also show that default risk is an implicit barrier to capital flows. If default risk is
eliminated, capital flows will be six times greater, and international risk sharing will
increase substantially.
With financial frictions, there is always a limit on the external borrowing and one
easy method to deal with this is using collateral constraints. Collateral constraints are
widely used in models of financial crises and recently they are used in the literature
of macroprudential policy. We believe they are also useful in the welfare analysis of
financial integration since international debt is super important to developing countries.
And because collateral constraints limit the external borrowing to make the borrowing
safer, they are actually preventing default and financial crisis. Thus we can use collateral
constraints to reflect default risk mentioned in Bai and Zhang (2012) in some aspects and
reflect financial crises in a broad sense, which will also simplify our analysis. Jeanne and
Korinek (2010) develop a model to study the optimal policy responses to booms and busts
in credit and asset prices where assets serve as collateral and debt is limited to the sum
of a fraction of the value of collateral plus an exogenous constant. Bianchi and Mendoza
(2018) try to find optimal time-consistent macroprudential policy where their collateral
constraint limits total debt, including both intertemporal debt and working capital loans
not to exceed a fraction of the market value of beginning-of-period asset holdings.
143
4.3 Empirical Analysis
4.3.1 Empirical Methodology
Here we present the econometric methodology to solve our main questions mentioned
before. The model we use is the Panel Smooth Threshold Regression (PSTR) model. Two
interpretations of PSTR model are possible. On one hand, PSTR model can be thought of
as a regime-switching model that allows for a small number of extreme regimes associated
with the extreme value of a transition function. Meanwhile, different from the indicator
function, the transition from one regime to another is smooth. On the other hand, PSTR
model can be thought to allow for a continuum of regimes, each one being characterized
by a different value of the transition function. The logic is then similar to that developed in
the standard univariate time series STAR (Smooth Transition Autoregressive) model. And
the main difference between these two models is that PSTR model use panel specifications
and is non dynamic.
PSTR model allows parameters to vary across countries and times. In such case,
it provides a parametric approach of the cross-country heterogeneity and of the time
instability of the effects we want to investigate, since these parameters change smoothly
as a function of a threshold variable. In such case, PSTR model can overcome two major
critics on panel data models. First, the data might be non stationary that will cause
problems in estimating coefficients. Second, there might be cross-section heterogeneity
existing. In such case, if we ignore the parameter heterogeneity among cross-sectional
units, our estimation will be biased. In such case, PSTR model has a strong advantage to
be used in our analysis.
Another important model that is very similar to PSTR model is the Panel Threshold
Regression (PTR) model. PTR model is also a nice option to solve the heterogeneity
problem that implies individual observations can be divided into homogeneous classes
based on the value of an observed variable that is called the threshold variable. In specific,
144
it assumes a simple transition function from one regime to another based on the value of
a threshold variable. In the simplest model with two regimes, if the threshold variable
is below a certain value that is called the threshold parameter, the regression equation is
defined in one equation or one regime. If the threshold variable exceeds the threshold
parameter, then the regression equation is defined in another equation or another regime.
However, PTR model also has its drawbacks, one of which is that it allows only for a small
number of classes or regimes. But in the field of international economics, there are so
many different countries and regimes. In such case, it is impossible to use a small number
of regimes to capture the characteristics of different countries in different time periods.
Under this circumstance, the use of PSTR model is necessary.
First, we start from the traditional regression equation:
Y
it
=
i
+
t
+
0
FDI
it
+
0
E
it
+
i
D
it
+
it
(4.1)
whereY
it
,FDI
it
,E
it
andD
it
represent GDP per capita, FDI, equity, and debt market
integration.
i
and
t
represent the fixed effect of country and time.
However, as shown in the previous literature, there is a nonlinearity existing in the
effect of FDI, equity, and debt market integration on economic growth. In such case, if we
directly use linear regression models, the coefficients and significance might be unreliable.
Meanwhile, the signal of coefficients might be wrong. One method to solve this issue is
to introduce threshold effects in a linear panel model. In this context, the first solution
requires using a simple Panel Threshold Regression (PTR) model in Hansen (1999). In this
case, the transition mechanism between extreme regimes is very simple: at each date, if the
threshold variable observed for a given country is smaller than a given value that is called
the threshold parameter, the effect of FDI, equity, and debt market integration is defined
by a particular regime; this regime is different from the regime used if the threshold
145
variable is larger than the threshold parameter. Then let’s improve the regression model
to be a PTR model with two extreme regimes:
Y
it
=
i
+
t
+
0
0
W
it
+
0
1
W
it
h(q
it
;c) +
it
(4.2)
where
j
= (
j
;
j
;
j
)
0
for j = (0;1); W
it
= (FDI
it
;E
it
;D
it
)
0
; q
it
denotes a threshold
variable with c as a threshold parameter. The transition functionh(q
it
;c) corresponds to
the indicator function:
h(q
it
;c) =
8
>
>
>
<
>
>
>
:
1; q
it
>c
0; otherwise
(4.3)
The threshold variable we choose for our empirical model is the Chinn-Ito index
(KAOPEN). KAOPEN is an index measuring a country’s degree of capital account open-
ness. The index was initially introduced in Chinn and Ito (2006). KAOPEN is based on
the binary dummy variables that codify the tabulation of restrictions on cross-border
financial transactions reported in the IMF’s Annual Report on Exchange Arrangements
and Exchange Restrictions (AREAER). Now, the index is available online for the time
period of 1970-2015 for 182 countries. The data source is Chinn and Ito (2006). KAOPEN
can measure the aggregate financial openness and is a de jure index. The reason why we
can use KAOPEN as a threshold variable is multi-layered. First, since what we need to
choose is the threshold variable rather than independent variables, we don’t require this
variable to be micro-based. On the contrary, we prefer to choose a relatively aggregate
variable to be the threshold variable. This logic is related to the design of threshold vari-
able since threshold effect relies on threshold variable and independent variables together.
Under this circumstance, threshold variable shouldn’t be very micro-based, otherwise
it will affect the regression coefficients of independent variables. Since KAOPEN is an
aggregate variable to measure financial openness, this is a nice option of threshold variable.
146
Second, this threshold variable needs to be related to the dependent variable that is the
economic growth in the empirical analysis. Since we do believe financial integration can
have an impact on economic growth, KAOPEN can satisfy this requirement. Meanwhile,
this threshold variable can’t be that closely related to the dependent variable since we
focus more on the effect of independent variables. Under this circumstance, since we have
added de facto indices to be the independent variables as we will talk about later, which is
closely related to the dependent variable, the choice of KAOPEN that is less related to the
dependent variable to be threshold variable is reasonable. Third, since the independent
variables we choose are FDI market integration index, equity market integration index,
and debt market integration index, which can represent different dimensions of financial
integration, our empirical model can capture the subtle effect of financial integration on
economic growth. In such case, we don’t need to worry about KAOPEN is too aggregated
to account properly for the various dimensions of financial integration. Fourth, we are
not the first one to use KAOPEN to be threshold variable. On the contrary, other research
papers have adopted similar method. Following the experience in the past literature, we
are confident of our setting. In summary, the choice of our threshold variable is reasonable.
Another thing we need to know is that PSTR model is an advanced econometric model
while this model is also very sensitive to the choice of threshold variables, independent
variables, and dependent variables. Meanwhile, in order to make sure that our choice of
threshold variables, independent variables, and dependent variables are reasonable, and
our empirical model setting is correct, we need to pass several model tests as we will show
later. Since our empirical model has passed all model tests, we should be confident of our
choice of threshold variable. What’s more, since PSTR model is very subtle, sometimes
even if our empirical model has passed several model tests and our model setting is
correct, the regression results may not be ideal. The reason why this situation exists is
that PSTR model is a complex model that uses advanced econometric methodology to
derive final results. In such case, the program will report error if the relationship among
147
variables is not strong. Actually, in the area of the usage of PSTR model, there are many
cases where authors have constructed several carefully designed PSTR models, trying
to get a reasonable results. But despite they have tried their best to complete empirical
regressions, the final results are not ideal and they have to give up using PSTR model and
turn to use other empirical methodology. These examples further prove that PSTR model
is very subtle. However, on the other hand, if we can use PSTR model to get a correct and
reasonable result, we will be very confident that our empirical model is reasonable. This
means our choice of threshold variables, independent variables, and dependent variables
are reasonable, and our model setting is also reasonable. Since our empirical analysis
indicates our usage of PSTR model is successful, we don’t need to worry too much if the
choice of threshold variable is reasonable.
With such a model, the coefficients ofW
it
that is the effect of FDI, equity, and debt
market integration on economic growth, is equal to
0
0
if the threshold variable is smaller
than c and is equal to
0
0
+
0
1
if the threshold variable is larger than c.
This model can be extended to a more general specification with r regimes. However
even in this case, PTR model indicates that the effect can be divided into two classes and
the transition between two regimes is not smooth, which is unrealistic. One method to
solve this problem is to improve the assumption and use a model with a smooth transition
function. This type of model has been extended to panel data with PSTR model proposed
by Gonzalez et al. (2017) and Fok et al. (2005).
Let us first consider the simplest case with two extreme regimes and a single transition
function:
Y
it
=
i
+
t
+
0
0
W
it
+
0
1
W
it
h(q
it
;
;c) +
it
(4.4)
148
In this case, the transition function is a continuous and bounded function of the
threshold variable. We follow the method used in Gonzalez et al. (2017) for the time series
STAR models, and use the following logistic transition function:
h(q
it
;
;c) =
1
1 +exp[
(q
it
c)]
(4.5)
where c denotes a location parameter and parameter
determines the slope of the
transition function. In Figure 4.1, the transition function is displayed for various values of
the slope parameter
. When
tends to infinity, the transition functionh(q
it
;
;c) tends
towards the indicator function so that PSTR model corresponds to PTR model. When
tends to zero, the transition functionh(q
it
;
;c) is constant and the model corresponds to
the standard linear model.
Figure 4.1 Transition function with c=0
149
PSTR model allows parameters to vary across countries and times. In such case,
it provides a parametric approach of the cross-country heterogeneity and of the time
instability of the effects we want to investigate, since these parameters change smoothly as
a function of a threshold variable. To be more specific, the effect is defined as a weighted
average of parameters
0
0
and
0
1
.
However, the number of threshold variables may not be one all the time. Hence, we
can extend this assumption of threshold variables to be more generalized as following:
h(q
it
;
;c) =
1
1 +exp[
Q
m
z=1
(q
it
c
z
)]
(4.6)
wherec = (c
1
;:::;c
m
) denotes a m-dimensional vector of location parameters.
Meanwhile, as Gonzalez et al. (2017) indicate, this PSTR model can also be generalized
tor + 1 extreme regimes rather than 2 extreme regimes. In such case, the PSTR model can
be called general additive PSTR model as shown below:
Y
it
=
i
+
t
+
0
0
W
it
+
r
X
j=1
0
j
W
it
h
j
(q
it
;
j
;c
j
) +
it
(4.7)
where the r transition functionsh
j
(q
it
;
j
;c
j
) depend on the slope parameters
j
and on
location parametersc
j
. In this generalization, if the threshold variableq
it
is different from
W
it
, the aggregate effect of FDI, equity, and debt market integration on economic growth
for the i-th country at time t is defined by the weighted average of the r+1 parameters
0
j
associated to the r+1 extreme regimes:
e
it
=
@y
it
@W
it
(4.8)
=
0
0
+
r
X
j=1
0
j
h
j
(q
it
;
j
;c
j
)
150
4.3.2 Data
After clarifying the theoretical part of PSTR model, we start our empirical analysis. We use
panel data that covers 70 countries (33 advanced economies and 37 developing economies)
from 2000 to 2015. The detailed list of countries is listed in the Appendix. As for the
data source and data processing,Y
it
is the GDP per capita and the data source is World
Bank national accounts data, and OECD National Accounts data files. Moreover, We use
similar definition method of trade integration index and create FDI market integration
index, equity market integration index, and debt market integration index:
FDI
it
=
FDIA
it
+FDIL
it
GDP
it
(4.9)
E
it
=
EA
it
+EL
it
GDP
it
(4.10)
D
it
=
DEA
it
+DOA
it
+DEL
it
+DOL
it
GDP
it
(4.11)
where FDIA is the stock of FDI assets and FDIL is the stock of FDI liabilities; EA is the
stock of equity and investment fund shares assets in portfolio equity assets and FDIL is
the stock of equity and investment fund shares assets in portfolio equity liabilities; DEA is
the stock of debt securities assets in portfolio equity assets and DEL is the stock of debt
securities liabilities in portfolio equity liabilities; DOA is the stock of debt instruments
assets in other investment assets and DOL is the stock of debt instruments liabilities in
other investment liabilities. The data source is Balance of Payments and International
Investment Position Statistics (BOP/IIP) database from IMF.
4.3.3 Empirical Results
In this part, we use PSTR model to analyze the effect of FDI, equity, and debt market
integration on economic growth. Similar to previous studies, we also believe that the
effects of FDI, equity, and bond market integration on economic growth may not be linear.
151
On the contrary, these effects may be nonlinear, and we need to consider the coexistence
of multiple regimes and add threshold effects. As we mentioned before, PSTR model
not only includes the assumption of the transition from one regime to another based
on the threshold variable in PTR model, but also allows for a small number of extreme
regimes associated with the extreme value of a transition function and where the transition
from one regime to another is smooth. In addition, PSTR model allows for a continuum
of regimes, each one being characterized by a different value of the transition variable.
Therefore, PSTR model can help us better understand the effects of FDI, equity, and debt
market integration on economic growth.
Table 4.1 Tests for remaining nonlinearity
Number of Loca-
tion Parameters
Type of Tests H0: r=0 vs H1: r=1 H0: r=1 vs H1: r=2
m=1
Wald Tests (LM) 9.556** 6.153
(0.023) (0.104)
Fisher Tests (LMF) 2.963** 1.888
(0.031) (0.130)
LRT Tests (LRT) 9.606** 6.174
(0.022) (0.103)
m=2
Wald Tests (LM) 20.444*** 6.58
(0.002) (0.361)
Fisher Tests (LMF) 3.197*** 1.006
(0.004) (0.42)
LRT Tests (LRT) 20.674*** 6.604
(0.002) (0.359)
m=3
Wald Tests (LM) 21.213** 5.684
(0.012) (0.771)
Fisher Tests (LMF) 2.205** 0.577
(0.020) (0.817)
LRT Tests (LRT) 21.462** 5.701
(0.011) (0.769)
We follow the method in Colletaz and Hurlin (2006) and Fouquau et al. (2008) to
conduct empirical analysis and introduce our main results. As shown in Table 4.1, we
first test whether there is a nonlinear relationship in the regression equation. If the
152
nonlinear assumption is rejected, the regression equation should be back to the traditional
linear regression equation, and PSTR model cannot be used for processing. Here we use
Wald Tests (LM), Fisher Tests (LMF), and LRT Tests (LRT) to perform nonlinearity tests
for the cases where location parameters are 1, 2, and 3, respectively. Among them, the
test we are most concerned about is Fisher Tests (LMF). We find that regardless of the
number of location parameters (m), the p-value of the nonlinear test is significant at the
5% significance level. When the number of location parameters is 2, the p-value of the
nonlinear test is significant at the 1% significance level. These show that our regression
equation has a strong nonlinear relationship so that PSTR model must be used for analysis.
In addition, we also test the number of transition functions for various situations where
the number of location parameters is 1, 2, and 3. The test results indicate that in the
three cases, no matter which test method is used, the p-value of the nonlinear test is not
significant at the 10% significance level. This also means that the hypothesis that the
number of transition functions is 2 is rejected, which also reflects that the number of
transition functions of 1 is the optimal choice in the three cases. In such case, only two
extreme regimes are sufficient to capture the nonlinear effects in the regression equation.
But it is worth noting that even if the number of transition functions is only 1, there are
still two extreme regimes. However, in PSTR model, even if there is only one transition
function, it can be considered as a smooth transition model with an infinite number of
intermediate regimes. Therefore, the use of PSTR model is still necessary, and we cannot
simply use the threshold model to replace it.
Meanwhile, this test result suggests that the effect of FDI, equity, and debt market
integration on economic growth does have a significant nonlinear nature. Therefore, if
the traditional linear regression method used in the past literature is used, this signifi-
cant nonlinearity will make the obtained result completely ignore the regime transition,
making the result very unreliable. At the same time, the result may be that no significant
effect can be found, or the effect is hugely heterogeneous, which will lead to the wrong
153
conclusion that FDI, equity, and debt market integration have no significant positive effect
on economic growth. What’s more, if only a single threshold model is used for analysis,
there are obvious shortcomings since the transition function assumed by the threshold
model is too single that may not be able to capture the different characteristics of each
country and the different situations in different time periods. Therefore, we must use
PSTR model to analyze the effect of FDI, equity, and debt market integration on economic
growth.
Table 4.2 Determination of the number of location parameters
Number of Location Parameters m=1 m=2 m=3
Optimal Number of Transition Functionsr
(m) 1 1 1
Residual Sum of Squares 35571.41 35507.71 35477.56
Number of Parameters 8 9 10
AIC Criterion 3.4806* 3.4815 3.4833
BIC Criterion 3.5164* 3.5218 3.5281
Table 4.2 describes the determination process of the number of location parameters.
We have determined the optimal number of transition functions for different numbers of
location parameters through previous analysis. Here for these three cases, we have listed
Residual Sum of Squares, number of parameters, AIC criterion, and BIC criterion. Our
method to determine the optimal number of location parameters is based on AIC and BIC
criteria. It can be found from the table that the results of AIC Criterion and BIC Criterion
both show that the optimal number of location parameters is one so that the best (m;r
)
combination should be (1,1).
Table 4.3 presents the parameter estimates for the final PSTR model. What needs to be
paid special attention to is that in PSTR model, the values of parameters cannot be directly
interpreted, but their signals can be interpreted. Meanwhile, according to our regression
model, we can still get enough information from the parameter values. Though according
to the previous analysis, the combination of (m;r
) = (1;1) is the optimal combination, we
still show the results of (m;r
) = (2;1) and (m;r
) = (3;1) combination. Combined with the
154
Table 4.3 Parameter estimates for the final PSTR model
(m;r
) (1,1) (2,1) (3,1)
Parameters
0
= (
0
;
0
;
0
)
FDI Parameter
0
0.0004 -0.0259*** -.0267***
(0.0020) (0.0085) (0.0856)
Equity Parameter
0
0.0269** 0.1141*** .09099***
(0.0122) (0.0287) (0.0280)
Debt Parameter
0
0.0083** 0.035** .0216*
(0.0042) (0.0147) (0.0114)
Parameters
1
= (
1
;
1
;
1
)
FDI Parameter
1
0.0327*** 0.0494*** .0533***
(0.0099) (0.0138) (0.0171)
Equity Parameter
1
-0.1284*** -0.1754*** -.1820***
(0.0392) (0.042) (0.0551)
Debt Parameter
1
-0.0175 -0.0422** -.0432*
(0.0122) (0.0206) (0.0228)
Location Parametersc
j
First Transition Function 1.0589 0.4005 0.2591
Second Transition Function 0.7085 0.0561
Third Transition Function 0.9376
Slopes Parameters
j
11.654 5.9616 7.7268
previous analysis, we believe that the fitting effect of the combination of (m;r
) = (2;1)
is also very good, and the significance of the parameters is obviously stronger than the
optimal combination (m;r
) = (1;1). Therefore, we also take it into consideration to provide
a reference for our analysis.
For FDI market integration, the value of
0
is positive but not significant, and the value
of
1
is positive and significant at the 1% significant level. This shows that the increase in
FDI market integration will promote economic growth while this effect is mainly generated
by the transition function. Therefore, if we simply use linear regression to estimate the
parameters, we may get an insignificant positive effect. At the same time, referring to
the situation of (m;r
) = (2;1), we find that though the value of
1
is still positive and
significant at the 1% significance level that is even larger, the value of
0
is negative and
155
significant at the 1% significance level. But it is worth noting that this time the absolute
value of
0
is less than
1
, which means that when the threshold variable is small, the
increase in FDI market integration will inhibit economic growth, but when the threshold
variable is large, the increase in FDI market integration will promote economic growth.
Therefore, in general, we believe that the increase in FDI market integration will promote
economic growth while this may be related to the threshold variable and the transition
function has a great influence on this positive effect.
For equity market integration, the value of
0
is positive and significant at the 5%
significance level, and the value of
1
is negative and significant at the 1% significance
level. The situation of (m;r
) = (2;1) is similar while
0
is significant at the 1% significance
level. Meanwhile, we find that
0
is less than the absolute value of
1
, which shows
that when the threshold variable is small, the increase of equity market integration will
promote economic growth. But when the threshold variable is large, the increase of equity
market integration will inhibit economic growth. Therefore, in general, we believe that the
improvement of equity market integration will first promote and then inhibit economic
growth, and the transition function will have a great impact on it.
For debt market integration, the value of
0
is positive and significant at the 5%
level of significance, and the value of
1
is negative but not significant. The situation of
(m;r
) = (2;1) is similar while
1
is significant at the 5% level. At the same time, we find
that
0
is less than the absolute value of
1
in the two cases, which shows that when the
threshold variable is small, the increase of debt market integration will promote economic
growth. But when the threshold variable is large, the increase of debt market integration
will inhibit economic growth. However, considering the optimal combination, the value
of theta1 is not significant so that the negative effect of debt market integration may
not be significant enough. Therefore, in general, we believe that the increase in debt
market integration will promote economic growth, and the transition function has a great
influence on it while in the case of a large threshold variable, it may have a negative effect.
156
Finally, we find that the transition function is not very sharp in the three cases. Con-
sidering that when the slope parameter tends to infinity, the transition function tends to
be an indicator function as in the threshold model without smooth transition. In these
three cases, the values of the slope parameter are 11.654, 5.9616, and 7.7268, respectively.
Therefore, the values are not very large, which means that the transition functions are all
largely different from indicator functions. This conclusion is very important since it shows
that our regression equation cannot be reduced to the sum of a limited number of regimes.
At the same time, considering that different from PTR model, PSTR model assumes that
the transition function is smooth, this result also suggests that we must use PSTR model
instead of the traditional PTR model to analyze this problem. In addition, since PSTR
model can be interpreted as a model that allows continuum of regimes, we need to realize
that continuum of regimes is necessary to be used to analyze the effect of FDI, equity, and
debt market integration on economic growth.
In summary, our results indicate that there is a strong nonlinear relationship existing,
which means we can’t use linear regression model to estimate the effect. We also find strong
threshold effects in the relationship between FDI, equity, and debt market integration and
economic growth. Under this circumstance, the aggregate effect of FDI, equity, and debt
market integration on economic growth is highly impacted by the threshold effects, which
means the effect is very complicated. And since the threshold effects rely on KAOPEN
that measures capital account openness, which is affected by multiple factors, strong
threshold effects mean that the aggregate effect of FDI, equity, and debt market integration
on economic growth is heterogeneous across countries and times. But generally speaking,
we are confident that the increase in FDI market integration can promote economic growth.
What’s more, the improvement of equity market integration will first promote and then
inhibit economic growth. Moreover, the increase in debt market integration can promote
economic growth but may face some negative effect. In such case, financial integration can
bring sizable benefits that may be mainly from FDI market integration. Finally, since the
157
slope parameter is not very large, the transition function is not very sharp. This means
that our regression equation cannot be reduced to the sum of a limited number of regimes
and we must use PSTR model instead of PTR model.
4.4 The Models
In this section, we introduce our models for the welfare analysis of financial integration.
We consider three different scenarios: (1) financial autarky, (2) full financial integration
with a non-state contingent bond and FDI, (3) limited financial integration with frictions
being removed. We will first introduce our benchmark model covering the first two
scenarios where we use a very standard method to conduct theoretical analysis. Then we
will introduce our improved model covering scenario 1 and scenario 3 where the financial
integration is limited with frictions being removed, which is much closer to the reality and
more reasonable.
4.4.1 Benchmark Model
In this section, we will introduce our benchmark model. In this model, the world is
divided into two countries: D and E where D represents the developed country and E
represents the emerging economy. Each country consists of a benevolent government and
a continuum of identical households. There is only one numeraire good in this world,
which is used for investment and consumption. Each country starts with a given initial
capital stockk
i;0
.
158
4.4.1.1 Production
There are two different methods of production in each country. The first method is the
standard method that is to use capital and labor with a Cobb-Douglas production function
to produce the output:
y
1
i;t
=A
i;t
(k
it
)
(l
i;t
)
1
(4.12)
whereA
i;t
is the stochastic total factor productivity (TFP), which can be decomposed
into two components: a transitory country-specific componenta
i;t
and a persistent world
componenta
W;t
. In such case, we can have the following formula forA
i;t
:
A
i;t
=a
W;t
a
i;t
(4.13)
The short-run transitory country-specific componenta
i;t
is a standard setting to present
the fact that each country is hit by transitory country-specific shocks in TFP growth. And
the long-run persistent world componenta
W;t
presents the fact that the world is hit by
persistent shocks in TFP growth. Bothlog(a
W;t
) andlog(a
i;t
) follow AR(1) process:
log(a
W;t+1
) = (1
W
)log(a
W;0
) +
W
log(a
W;t
) +
W;t
(4.14)
log(a
i;t+1
) = (1)log(a
i;0
) +log(a
i;t
) +
i;t
(4.15)
where
W
is the persistence of the growth rate of the long-run persistent world compo-
nenta
W;t
and is the persistence of the short-run transitory country-specific component
a
i;t
.
W;t
is an i.i.d normal distributed process with volatility
W
. And
i;t
= (
D;t
;
E;t
)
0
is
an i.i.d normal distributed process with covariance matrix =
2
D
D
E
D
E
2
E
. a
i;0
is the
initial value of total factor productivity in each country.
W;t
and
i;t
are independent.
159
The second method is to use FDI (Foreign Direct Investment) or in alienable capital
only to produce the output:
y
2
i;t
=A
i;t
A
f
(k
f
i;t
)
f
(4.16)
wherek
f
it
is the level of FDI or in alienable capital andA
f
is the relative scale factor.
Both methods of production are impacted by the same shocks so that the stochastic TFP is
the same. And what needs to be paid special attention to is that the goods produced via
FDI fully belong to the domestic country. This means domestic country can get full return
from FDI production.
In terms of production, we assume there are two different methods of production
to produce two different intermediate goods in each country. The first method is the
standard method that is to use capital and labor with a Cobb-Douglas production function
to produce the domestic intermediate good. The second method is to use FDI only to
produce another foreign intermediate good. Just as Baldwin (2012) mentions, 21st century
trade is much more complex since global supply chains internationalize the complex
two-way flows that was solved within factories. And the flows include trade, investment,
services, and IP . In such case, the production abroad is quite different from the traditional
production that relies more on the technology and knowledge. Moreover, the demand
of labor is not that large since the share of foreign intermediate goods in the production
of final goods is small. Under this circumstance, the usage of labor in the production of
foreign intermediate goods is also small. Meanwhile, multinational corporations can use
robots to produce goods, which makes the demand of labor decrease further and the usage
of robots will be counted in capital. In such case, the demand of labor in the production
of foreign intermediate goods is very small and we can choose to simplify labor here. On
the other hand, relatively speaking, the knowledge brought with FDI is more important.
And the usage of robot makes capital more important and reduce the importance of labor.
Under this circumstance, the assumption that we can use FDI only to produce foreign
160
intermediate goods fits the reality better. In summary, our assumption of using FDI only
to produce foreign intermediate goods is more realistic and reasonable.
Of course, if we pursue a more generalized model, we can add labor into the production
of foreign intermediate goods, which can be a nice point for improvement later. However,
we can also choose to keep the simplicity of the model and ensure the accuracy at the
same time. Actually this setting we adopt here is a simplified version of modelling
the production of foreign intermediate goods that is used in previous literature such as
Albuquerque (2003). And one thing needs to be clear is that even if we use the more
generalized model, the welfare gains won’t change very much. This is because even if
we add the labor input setting into the model, since the production share of foreign
intermediate goods is not large, the labor income is not large either. Meanwhile, this kind
of the production of foreign intermediate goods is always two-way, which means if we
assume we can use FDI and labor together to produce foreign intermediate goods, then
this new setting will have an impact on both sides that includes domestic and foreign
country. Compared with the baseline assumption, of course domestic country can have
more labor income domestically and the added labor income comes from the production
of foreign intermediate goods domestically. But on the other hand, compared with the
baseline assumption, domestic country will also lose FDI income abroad since now the
production of foreign intermediate goods need to pay wage to workers abroad. To sum
up, the aggregate effect of these two mechanisms is very small. Under this circumstance,
even if we add labor input setting into the production of foreign intermediate goods, the
total income of households in all countries won’t change very much and then the welfare
gains of financial integration won’t change very much either. This means our results are
robust and the need to use more generalized model is limited. Actually, we tried to add
labor input setting in the benchmark model before and found the difference of welfare
gains is very small that can be ignored.
161
Then we can have the following formula for the total production:
y
i;t
=y
1
i;t
+y
2
i;t
=A
i;t
(k
it
)
(l
i;t
)
1
+A
i;t
A
f
(k
f
it
)
f
(4.17)
Here we want to clarify the logic and intuition of FDI channel. To be more specific,
we want to make it clear why firms invest on other countries via FDI and why they don’t
choose international trade instead. As summarized by Buckley et al. (2010), in general
firms use FDI to achieve two goals: (1) internalize missing or imperfect external markets
in intermediate products and knowledge; (2) choose locations for constituent activities
to minimize the overall costs of their operations. And for the location aspect, there are
three motivations: foreign market seeking FDI, efficiency (cost reduction) seeking FDI
and resource seeking FDI. However, these two goals can’t be achieved if firms only use
international trade.
Considering the different situation in the emerging economies, there are three special
reasons to use FDI as summarized in Buckley et al. (2010): capital market imperfections,
the special ownership advantages of multinational enterprises and institutional factors.
Capital market imperfections includes four imperfections: (1) state-owned (and state-
associated) firms may have capital made available to them at below market rates; (2)
inefficient banking systems may make soft loans to potential outward investors, either as
policy or through inefficiency; (3) conglomerate firms may operate an inefficient internal
capital market that effectively subsidizes FDI; (4) family-owned firms may have access to
cheap capital from family members. The special ownership advantages of multinational
enterprises may include flexibility, economizing on the use of capital (or resources),
benefits accruing from home country embeddedness and the ability to engage in beneficial
relations with firms and other actors in order to provide access to resources controlled by
others. Finally, the institution factors are based on the fact that firms’ strategy is shaped
by the home institutional environment which is formally and informally enforced by
government and its agents and which bear upon the norms and cognitions. At the same
162
time, government seek to influence the amount, direction, and scope of capital flows.
The advantages include high levels of government support and highly bureaucratic and
burdensome administrative FDI approval procedures.
Meanwhile, there are other reasons supporting the usage of FDI. Holmes et al. (2015)
highlight the existence of quid pro quo policy (a policy that makes technology transfer a
precondition of foreign firms selling) in emerging economies so that firms have to use FDI
to open foreign market. Holmes et al. (2015) also point out the existence of localization
and believe that each unit of technology capital has a potential use in production at each
location and to realize the potential use, the capital must be deployed locally. In this part,
we improve the theory in Holmes et al. (2015) and believe that investors need to localize
their investment to open market and get profit. If they just use international trade, they
can’t capture the need of foreign market and then can’t make foreign market accept their
goods nicely because of different taste, habit and culture, etc. Meanwhile, they can’t make
customers trust their goods so that they can’t sell their goods successfully. We also use
game theory to explain the usage of FDI. We assume there are several agents and if one
agent doesn’t choose to invest via FDI, then others will take his market share he could
have and prevent him from entering the market. Then this agent will lose potential profit
from foreign market in the future and lose domestic competition since his total profit is
less than others. Under this circumstance, each agent has to invest via FDI to prevent the
potential failure in the future.
In summary, the reasons mentioned above make firms invest on other countries via
FDI and FDI has several special advantages that can’t be replaced by international trade.
The law of motion of capital is same as the law of motion of FDI in each country:
k
i;t+1
= (1)k
i;t
+k
i;t
(
i
i;t
k
i;t
) (4.18)
k
f
i;t+1
= (1)k
f
i;t
+k
f
i;t
(
i
f
i;t
k
f
i;t
) (4.19)
163
where 0<< 1 is the depreciation rate of capital and FDI.i
it
andi
f
it
are the gross
investment in capital and FDI.(x) is a standard quadratic adjustment cost function:
(x) =x
2
(x)
2
(4.20)
where is the degree of adjustment costs.
Labor market and capital markets are perfectly competitive and the returns for do-
mestic capital are their marginal productivity while for FDI, marginal productivity is
not equal to the return. The reason why we still adopt this notation is to simplify our
equations later. Thus, we can have the following formulas:
w
i;t
l
i;t
= (1)y
1
i;t
(4.21)
r
i;t
k
i;t
=y
1
i;t
(4.22)
r
f
i;t
k
f
i;t
=
f
y
2
i;t
(4.23)
For simplicity, we can normalize population in domestic capital sector to be unity in
each country. Then country size is homogeneous to the total factor productivityA
it
.
In summary, the production formulas become:
y
1
i;t
=A
i;t
(k
it
)
(4.24)
y
2
i;t
=A
i;t
A
f
(k
f
it
)
f
(4.25)
y
i;t
=y
1
i;t
+y
2
i;t
=A
i;t
(k
it
)
+A
i;t
A
f
(k
f
it
)
f
(4.26)
164
4.4.1.2 Demand
The benevolent government in each country chooses consumption, investment, borrowing
or lending to maximize utility of the domestic consumers given by
U
i
=E
0
1
X
t=1
t
U(c
i;t
) =E
0
1
X
t=1
t
c
1
i;t
(4.27)
where is the degree of relative risk aversion and the utility function is a CRRA utility
function.
Budget constraints and market clear conditions depend on the different cases where
the availability of different assets is different.
And in all cases, the stochastic discount factor (SDF) in country i is:
M
i;t+1
=(
c
i;t+1
c
i;t
)
(4.28)
Under financial autarky, the only available asset is domestic capital and there is no
foreign debt or FDI available. Then we can treatk
f
i;t
to be 0 in each country all the time
and the only choice for the household made by the benevolent government is to consume
or invest in domestic capital with the incomes. Then we can have the following budget
constraint for household:
c
i;t
+i
i;t
=w
i;t
+r
i;t
k
i;t
(4.29)
We can also have the following Euler equations:
E
t
fM
i;t+1
[r
i;t+1
0
i;t
+
0
i;t
0
i;t+1
(1 +
i;t+1
i
i;t+1
k
i;t+1
0
i;t+1
)]g = 1 (4.30)
where
i;t
=(
i
i;t
k
i;t
) and
0
i;t
is the first derivative of the function(x).
165
The goods market clear conditions in country i are:
c
i;t
+i
i;t
=A
i;t
(k
i;t
)
(4.31)
Under full financial integration, there are two assets under financial integration case.
First, there is a riskless international bond that will pay one unit of good back in the next
period and the price at time t isp
t
. Second, there is FDI that can pay back the capital
return to the home country. And one thing needs to be clear: the whole return from the
production using FDI will be back to the home country.
Then the budget constraint in country i becomes:
c
i;t
+i
i;t
+i
f
i
0
;t
=w
i;t
+r
i;t
k
i;t
+y
2
i
0
;t
+b
i;t1
b
i;t
p
t
(4.32)
where i’ represents the other country andb
i;t
represents the new bond country i buys
in time t.
Meanwhile, we have a non-Ponzi condition:
b
i;t
>b
i
(4.33)
whereb
i
is the debt limit. In our real simulation, we set it to be a very small value.
Then we can have the same Euler equation for domestic capital as before. Meanwhile,
we can have the following Euler equations for FDI:
E
t
fM
i;t+1
[r
f
i
0
;t+1
0
i
0
;t
+
0
i
0
;t
0
i
0
;t+1
(1 +
i
0
;t+1
i
f
i
0
;t+1
k
f
i
0
;t+1
0
i
0
;t+1
)]g = 1 (4.34)
where
i
0
;t
=(
i
f
i
0
;t
k
f
i
0
;t
) and
0
i
0
;t
is the first derivative of the function(x).
166
What’s more, we can have the following Euler equations for bond:
p
t
=E
t
(M
i;t+1
) +
i;t
(4.35)
where
i;t
is the Lagrange multiplier for the non-Ponzi condition.
Finally, we can have the following market clear conditions for goods market and bonds
market:
c
D;t
+i
D;t
+i
f
E;t
+c
E;t
+i
E;t
+i
f
D;t
=y
D;t
+y
E;t
(4.36)
b
D;t
+b
E;t
= 0 (4.37)
4.4.1.3 Equilibrium
Now, we can define an equilibrium under financial autarky. An equilibrium in country i is
a sequence of consumption and capitalfc
i;t
,k
i;t
g such that the Euler equation for capital
(4.30) and goods market clear condition (4.31) are both satisfied.
Meanwhile we can define an equilibrium under full financial integration. An equilib-
rium is a sequence of consumption, capital, FDI and bond in both countriesfc
i;t
,k
i;t
,k
f
i
0
;t
,
b
it
g fori =D;E and a sequence of bond pricesp
t
such that the Euler equations for capital,
FDI, and bond (4.30), (4.34), and (4.35) are satisfied while goods market and bond market
clear conditions (4.36) and (4.37) are also satisfied.
4.4.2 Improved Model with Frictions being Removed
In this section, we will introduce our improved model covering scenario 1 and scenario
3 where the financial integration is limited with frictions being removed, which is much
closer to the reality and more reasonable compared with the benchmark model. But the
basic setting of the improved model is similar to the benchmark model.
167
4.4.2.1 Introduction of a Frictional World
There are two main frictions: capital controls and financial frictions in the international
financial market. According to some past literature (like Bai and Zhang (2012)), these
frictions weaken the welfare gains from financial integration so that we may find financial
integration is not that good. But if we totally remove theses frictions, the welfare gains
from financial integration will be much larger and non-negligible. In such case, financial
integration will bring sizeable benefit.
However, there are three main problems about frictions in the existing literature. First,
many researches totally ignore the existing frictions and assume a frictionless world. After
that, they estimate the welfare gains, which is not reasonable given the background that
frictions have a big impact on the welfare gains. Second, many researches assume that
financial integration is just a event and after a certain point, suddenly these countries
are totally open to international capital and there is a full financial integration while
both conditions are impossible. Finally, many researches assume there is full financial
integration and the world is already frictionless now or many years ago. After that, they
compare the difference and estimate the welfare gains. However, even now there is no full
financial integration and many emerging countries like China still have capital controls.
Facing these problems, in this chapter we try to adopt a more reasonable setting in
frictions. In specific, we assume that the frictions are being removed slowly and it will take
many years to perfectly remove all frictions. In our minds, this approximation is much
closer to the reality and after calibration, our estimation of the welfare gains of financial
integration will be much more accurate and reasonable compared with the past literature.
4.4.2.2 Frictions
In the model, we assume that there are two frictions. The first friction is capital controls.
Since developing countries adopt capital controls, the international capital can’t flow free.
Here we use price control that takes the form of taxes on international investment to
168
measure capital controls. In such case, we can use
t
to be the tax rate on international
investment. Meanwhile, informational frictions are also an very important frictions,
which is proven from the famous gravity model (see Portes and Rey (2005)). Because of
informational frictions, agent needs to pay more to buy foreign assets. Here we can use a
iceberg cost
t
to measure informational frictions. It’s obvious that informational frictions
and capital controls have a similar impact so that we can use the same method to represent
both capital controls and informational frictions at the same time.
In such case, the budget constraint in country i becomes:
c
i;t
+i
i;t
+ (1 +
t
)i
f
i
0
;t
=w
i;t
+r
i;t
k
i;t
+y
2
i
0
;t
+b
i;t1
(1 +
t
)b
i;t
p
t
(4.38)
And we can have the new Euler equations for FDI:
E
t
fM
i;t+1
1
1 +
t
[r
f
i
0
;t+1
0
i
0
;t
+
0
i
0
;t
0
i
0
;t+1
(1 +
i
0
;t+1
i
f
i
0
;t+1
k
f
i
0
;t+1
0
i
0
;t+1
)]g = 1 (4.39)
Similarly, we can have the new Euler equations for bond:
(1 +
t
)p
t
=E
t
(M
i;t+1
) +
i;t
(4.40)
Since we assume that the frictions are being removed slowly, here we assume that
t
is
decreasing slowly to zero.
The second friction is financial frictions. With financial frictions, there is always a
limit on the external borrowing and one easy method to deal with this is using collateral
constraints. Collateral constraints are widely used in the models of financial crises and
recently they are used in the literature of macroprudential policy. We believe they are
also useful in the welfare analysis of financial integration since international debt is super
important for emerging economies. Meanwhile, since collateral constraints limit the
external borrowing to make the borrowing safer, they are actually preventing default
169
and financial crisis. In such case, collateral constraints can be used to reflect default risk
mentioned in Bai and Zhang (2012) in some aspects and financial crises in a broad sense,
which will simplify our analysis. In the model, normal periods are when the constraint is
slack, and crisis periods are when the constraint binds.
Under this setting, we assume that the amount of bond one country can borrow can’t
be higher than a fraction of the value of its capital:
b
i;t+1
t
q
i;t
k
i;t
(4.41)
where
t
is a stochastic borrowing limit fraction between 0 and 1. q
t
is the value of
domestic capital, which is also the Lagrange multiplier on the law of motion of capital
since it has the meaning of how many goods the country would give up for an additional
unit of capital.
From the FOC conditions for investment, it’s easy to find that:
q
i;t
=
1
0
i;t
(4.42)
And we can have the new Euler equations for capital:
E
t
fM
i;t+1
[r
i;t+1
0
i;t
+
0
i;t
0
i;t+1
(1 +
i;t+1
i
i;t+1
k
i;t+1
0
i;t+1
+
t
c
i;t
)]g = 1 (4.43)
Similarly, we can have the new Euler equations for bond:
(1 +
t
)p
t
=E
t
(M
i;t+1
) +
i;t
+
c
i;t
(4.44)
where
c
i;t
is the Lagrange multiplier for collateral constraint.
Since we assume that the frictions are being removed slowly, here we assume that
t
is
increasing slowly. This means the collateral constraint is becoming looser.
170
4.4.2.3 Equilibrium
Now we can define an equilibrium under financial autarky. An equilibrium in country i is
a sequence of consumption and capitalfc
i;t
,k
i;t
g such that the Euler equation for capital
(4.30) and goods market clear condition (4.31) are both satisfied.
Meanwhile, we can define an equilibrium under limited financial integration. An
equilibrium is a sequence of consumption, capital, FDI and bond in both countriesfc
i;t
,
k
i;t
,k
f
i
0
;t
,b
it
g fori =D;E and a sequence of bond pricesp
t
and iceberg cost
i;t
such that
the Euler equations for capital, FDI, and bond (4.43), (4.39), and (4.44) are satisfied while
goods market and bond market clear conditions (4.36) and (4.37) are also satisfied.
4.5 Calibration
In this section, we present our calibrated parameters and values with their sources. Since
Coeurdacier et al. (2015) is the first one and the newest one to provide an integrated
framework where we can study the standard neoclassical efficiency gains together with
gains from risk sharing, we will continue to use their parameters setting. However, as
for the parameters in FDI activity and frictions, we have to choose from other related
literature.
Table 4.4 summarizes the parameters values and their sources. The period in the model
is one year and the data is annual data. Almost all the choices of parameters are standard
in the literature of international business cycle. Since we want to explore the welfare gains
of big economies, and in such case, only the relative size is important, we assumeA
D;0
,
A
E;0
to be 1. And as for the initial capital, we will choose the capital of country D in steady
state under financial autarky to be the initial capital of country D. Moreover, the initial
capital of country E is 50% of the initial capital of country D. Meanwhile, though we can
choose high degree of relative risk aversion to be 10 because of the world shocks, we
171
Table 4.4 Parameters values
Parameters Values Sources/Targets
Initial TFP in country D A
D;0
= 1 Coeurdacier et al. (2015)
Initial TFP in country E A
E;0
= 1 Coeurdacier et al. (2015)
Relative initial capital rate k
D;0
=k
E;0
= 2 Coeurdacier et al. (2015)
Capital share in production = 0:3 Coeurdacier et al. (2015)
Persistence in world shocks
W
= 0:999 Coeurdacier et al. (2015)
Persistence in country shocks = 0:9 Coeurdacier et al. (2015)
Volatility of world shocks
W
= 0:2% Coeurdacier et al. (2015)
Volatility of shocks in country D
D
= 2:5% Coeurdacier et al. (2015)
Volatility of shocks in country E
E
= 5% Coeurdacier et al. (2015)
Correlation of cross-country shocks = 0 Coeurdacier et al. (2015)
Relative scale factor of FDI A
f
= 0:3 FDI stock/GDP = 30%-40%
Capital share of FDI in production
f
= 0:3 Albuquerque (2003)
Capital depreciation rate = 0:08 Coeurdacier et al. (2015)
Capital adjustment costs = 3 Bai and Zhang (2012)
Utility discount rate = 0:96 Coeurdacier et al. (2015)
Degree of relative risk aversion Low: = 4 Coeurdacier et al. (2015)
High: = 40 Coeurdacier et al. (2015)
Initial cost of capital controls
0
= 5:9% Bai and Zhang (2012)
Initial borrowing limit
0
= 0:75 Bianchi and Mendoza (2018)
1
= 0:90 Bianchi and Mendoza (2018)
1
Initial values of TFP and capital are set to represent the initial ratio of country D
and E.
2
Frictions will be removed slowly.
still choose to be 40 similar in other macro-finance papers. This is because we want to
capture the welfare gains of risk sharing better.
As for the FDI activity, we setA
f
to be 0.3 to match FDI stock of GDP for the US
and the whole world that is about 30%-40%. The reason why we don’t adopt the choice
of Albuquerque (2003) is because in reality, FDI share is not as high as he expects in
Albuquerque (2003). And following his method where he sets the capital share of capital
and FDI in production to be the same, we also make
f
to be 0.3, which is equal to.
Meanwhile, according to Bai and Zhang (2012), we choose
0
to be 5.9% considering
default risk. Finally according to Bianchi and Mendoza (2018), we set
0
to be 0.75 and
1
to be 0.90 where they make this choice based on the loan-to-value (LTV) ratios for both
172
households and firms in the US and abroad during the financial crisis and prior to the
crisis. We believe that at the beginning of financial integration, the extent of collateral
constraint is similar to the extent during the financial crisis, which is under tight credit
regime. This is because at that time, financial integration is completely a new thing and
people are cautious. But later, the extent of collateral constraint will converge to the
normal extent that is similar to the extent prior to financial crisis, which is under normal
credit regime.
Finally, we need to decide the initial values of capital, FDI, and bond. Here we follow
the method in Coeurdacier et al. (2015) to take the steady state rate of capital in developed
country under financial autarky to be the benchmark. However, this steady state is taken
when = 0:94 since before financial integration, the real interest rate is about 6% that
is different from the period after financial integration. In such case, we should choose
a different rather than a standard to get the initial value of capital. Therefore, no
matter under financial autarky or financial integration case, the initial value of capital
in developed country is the steady state of capital in developed country under financial
autarky while the initial value of capital in emerging economy will be half of the initial
value of capital in developed country. And since there is no FDI or bond under financial
autarky, we don’t need to worry about the initial values of FDI and bond. Meanwhile, under
financial integration including full financial integration and limited financial integration,
it’s obvious that the initial values of FDI and bond are zero.
4.6 Quantitative Results
In this section, we will present the simulation results of our models. To be more specific,
we will show the steady state values and paths of output, consumption, capital, FDI, and
bond of each countries under financial autarky and financial integration. Then these
simulation results will be used for welfare analysis in Section 5.
173
4.6.1 Steady State
First, let’s focus on the steady state values in our models. Steady state is very important
since it can tell us the welfare gains and changes in the long run. Though sometimes it
will take a long time for macroeconomic variables to converge to the steady state, we can
still use steady state as a nice assistant tool. Different from the global solution method
mentioned in Coeurdacier et al. (2015), here we use a standard global solution method
while the steady state values are deterministic rather than changeable. In such case, we
will only have one steady state.
Table 4.5 Steady state values
Financial autarky
Consumption c Capital k Total output y FDI kf Bond b
Country D 1.18 3.63 1.47 0.00 0.00
Country E 1.18 3.63 1.47 0.00 0.00
Full financial integration (FDI and bond)
Consumption c Capital k Total output y FDI kf Bond b
Country D 1.26 3.66 1.75 0.66 -3.65
Country E 1.55 3.66 1.75 0.66 3.65
Table 4.5 lists the steady state of consumption, capital, total output, FDI, and bond.
Since country D and country E are almost symmetric, the steady state values of many
variables in each country will be the same. It’s obvious that the steady state level of
consumption has increased a lot under full financial integration for both countries. This
proves that FDI scarcity effect can be an important driver of the welfare gains of financial
integration. Meanwhile, it hints that perhaps the welfare gains of financial integration
may be much larger than our expectation. And we can notice that consumption in country
E is higher than country D, which is because country E is bond lender and country D is
bond borrower. In this case, country E can have interest from bond while country D has to
pay the interest. As for capital, the steady state level of capital is slightly higher under full
financial integration for both countries, which may be because the productivity of FDI is
174
not high enough and both countries have to invest more in homeland. With the help of
FDI, the steady state level of total output is much larger under full financial integration.
The steady state level of FDI under full financial integration is same for both countries,
which is because the two countries are symmetric who will make equal investment abroad.
Finally, country E is bond buyer because country E wants to share risk by buying bond
from country D and country D is bond seller since country D can tolerate this risk.
In a word, from steady state values, it seems that the welfare gains of financial integra-
tion will be pretty large that is much larger than our expectation. But if we want to get a
detailed answer, we still need to do more work that will be presented later in this chapter.
4.6.2 Financial Autarky
In this section, we will present the paths of output, consumption, and capital under
financial autarky. Though this part is not the main focus of the whole chapter, we will give
a brief introduction of the situation under financial autarky.
Figure 4.2 shows the dynamics of consumption, capital, and output under financial
autarky. The upper panel shows the dynamics for developed country while the lower
panel shows the dynamics for emerging economy. Since FDI and bond are both 0 under
financial autarky, we don’t draw figure for FDI and bond.
Since the initial capital of each country is below its steady state level especially country
E, capital in both countries will increase that leads to the increase of consumption and
output. After about 75 years, it will almost converge to its steady state.
4.6.3 The Effect of Capital Scarcity and Risk Sharing
Here we will use Case 1 and Case 2 to clarify the effect of capital scarcity and risk sharing.
Now the world is risky and our models become stochastic models, which is closer to the
real world. In such a risky world, the effect of capital scarcity and risk sharing exist
together and we will explore the interactions between these two effects following the
175
Figure 4.2 Dynamics under financial autarky
analysis in Coeurdacier et al. (2015). Compared with Coeurdacier et al. (2015), here we
add FDI in the environment to make the world more realistic. Then the capital scarcity
effect in this chapter is different since there is not only the standard capital scarcity effect
but also the capital scarcity effect for FDI.
Case 1: A risky world.
In Case 1, there is a risky and stochastic world (
E
= 2
D
= 5%). In specific, this case
corresponds to the financial integration of a capital scarce and risky emerging economy
and a safe developed country, which captures the liberalization episode of emerging
economies during the late 1980s to early 1990s.
Figure 4.3 shows the dynamics of consumption, capital, and total output while Figure
4.4 shows the dynamics of FDI and bond. In each figure, the upper panel shows the
dynamics for developed country while the lower panel shows the dynamics for emerging
176
Figure 4.3 Dynamics under full financial integration 1
economy. Dashed lines refer to financial autarky levels while plain lines refer to financial
integration level.
Developed country lends to emerging economy to accelerate capital accumulation. But
unlike Coeurdacier et al. (2015), in our model, developed country doesn’t cut consumption
since developed country benefits from FDI a lot, which is large enough to offset this cut.
Meanwhile, it’s very interesting to find that capital in developed country decreases first
and then increases. This is because developed country wants to increase FDI and has to
suffer a loss in capital for a while while developed country also needs to cut capital to
buy bond. What’s more, when emerging economy is capital scarce and far away from
its autarky steady state, its growth rate is higher after financial integration. However, in
Case 1, FDI scarcity is another kind of capital scarcity for both countries. In such case,
developed country suffers FDI scarcity while emerging economy suffers capital scarcity
177
Figure 4.4 Dynamics under full financial integration 2
and FDI scarcity. What’s more, both countries are affected by two effects: capital scarcity
effect and risk sharing effect, which will adjust FDI and bond holding.
Emerging economy is affected by three effects: FDI scarcity effect, capital scarcity effect,
and risk sharing effect. On the one hand, capital scarcity effect can lead to faster growth
under financial integration. On the other hand, country E wants to share risk via FDI and
bonds, which will decrease the growth rate. Meanwhile, FDI scarcity effect can accelerate
growth while it can also take part of capital investment to invest on FDI, which will also
decrease the growth rate. Thus, it’s important to know which effect dominate in each time
period and it depends on the initial value of capital. At the beginning, FDI scarcity effect
dominates since FDI market is just open and the marginal productivity is super high. And
it’s easy for us to find that the growth rate of FDI is super large. At the same time, we
should also notice that capital scarcity effect is also very big. Later, capital scarcity effect
178
dominates after FDI scarcity effect gets weakened. This is because at the beginning FDI is
zero while at this time FDI has increased to a high value while capital is still scarce. Finally,
FDI and capital are both relatively abundant and risk sharing effect dominates, which can
be seen from the large amount of debt. And thanks to the contribution of FDI, country
E benefits a lot and its total output and consumption are much higher under financial
integration than financial autarky.
Developed country is affected by two effects: FDI scarcity effect and risk sharing effect.
At the beginning, FDI scarcity effect dominates, which leads to the temporary decrease of
capital. This is because FDI scarcity effect is strong while capital is relatively abundant.
In such case, FDI scarcity effect takes part of the capital investment to invest on FDI. But
after a time, risk sharing effect dominates since FDI has increased to a relatively high level.
As for the bond, in line with the bond reversal observed in Coeurdacier et al. (2015),
in our model we also observe such a phenomenon. And the logic is the same: at the
beginning, emerging economy has to sell bond to get capital for development; but after
that, as emerging economy is getting more developed, risk sharing effect dominates and
emerging economy begins to buy bond to share risk while developed country is able to
tolerate this risk. However, compared with Coeurdacier et al. (2015), our results indicate
that bond reversal happens very early that is 11 years after financial integration, which is
closer to the reality while in Coeurdacier et al. (2015) this number is about 80 years.
Case 2: A risky world with high risk aversion.
In Case 2, we follow the setting of Case 1 while we adjust the degree of relative risk
aversion to be 40. In such case, both countries will be more risk averse.
Figure 4.5 shows the dynamics of consumption, capital, and total output while Figure
4.6 shows the dynamics of FDI and bond. In each figure, the upper panel shows the
dynamics for developed country while the lower panel shows the dynamics for emerging
economy. Dashed lines refer to financial autarky levels while plain lines refer to financial
179
Figure 4.5 Dynamics under full financial integration 1 (high risk aversion)
integration level. The intuitions are the same and we focus more on the comparison of the
results in Case 1 and Case 2.
For both countries, they are more risk aversion so that risk sharing effect is much
stronger. In this case, FDI scarcity effect is still strong at the beginning but risk sharing
effect is also very strong. Meanwhile, capital scarcity effect is not that strong relative to
risk sharing effect. What’s more, both countries prefer to buy or sell more bond rather than
invest on capital. Despite FDI is also a nice tool to share risk, they still prefer bond that is
much safer. In such case, the values of consumption, capital, output, and FDI decrease
while the value of bond increases. Meanwhile, they both decrease the investment on both
capital and FDI, and focus more on bond to offset risk. However, this situation leads to
the lower growth rate of capital, FDI, and output, which makes these variables slower to
converge to the steady state.
180
Figure 4.6 Dynamics under full financial integration 2 (high risk aversion)
4.6.4 The Effect of Frictions
Here we will use Case 3 to clarify the effect of frictions. Now the world is not frictionless
and there exists frictions, which is closer to the real world. In such a world with frictions,
the effect of frictions will weaken the effect of capital scarcity and risk sharing. To the best
of our knowledge, we are the first one to explore such a situation.
Case 3: A risky world with frictions.
In Case 3, there is a world with frictions that are capital controls and financial frictions.
We still follow the setting of Case 1 while we also add the frictions. In Case 3, we assume
capital controls will decrease to 1% of its initial level after 60 years, which can be treated
as no capital controls. The reason why we adopt this setting is that capital controls still
181
exist now and we are not optimistic towards the disappearance of capital controls. As for
financial frictions, we will assume the borrowing limit will increase slowly.
Figure 4.7 Dynamics under limited financial integration 1
Figure 4.7 shows the dynamics of consumption of consumption, capital, and total
output while Figure 4.8 shows the dynamics of FDI and bond. In each figure, the upper
panel shows the dynamics for developed country while the lower panel shows the dynamics
for emerging economy. Dashed lines refer to financial autarky levels while plain lines refer
to financial integration level. Intuitions are the same and here we focus more on the effect
of frictions.
For both countries, capital controls force them to focus more on domestic market and
thus they tend to buy or sell less bond and invest more on capital. Meanwhile, capital
controls make them invest less on FDI. In such case, at the beginning, the growth rate of all
variables except capital is smaller compared with Case 1. However, as the effect of capital
182
Figure 4.8 Dynamics under limited financial integration 2
controls is weaker and weaker, all variables begin to grow fast. But it’s interesting to find
that after decades the effect of capital controls doesn’t have a large impact on FDI but
affect bond very much, which is due to capital scarcity effect. Finally, we find that capital
controls decrease the consumption of emerging economy and increase the consumption
of developed country. This means capital controls don’t help emerging economy directly
but can protect emerging economy from financial crisis. As for the financial frictions,
though it’s easy to see that financial frictions can have a significant impact in Case 1, in
Case 3 financial frictions don’t play a significant role. This is because capital controls force
both countries trade less bond so that their debt holdings don’t reach the borrowing limit.
Finally, even with frictions, our results show that after 18 years of financial integration,
bond reversal happens, which is still much smaller than the 80 years in Coeurdacier et al.
183
(2015). This means our results are still closer to the reality, which makes our model more
reliable.
In a word, capital controls do have a significant effect where it slows the economic
growth in both countries while financial frictions don’t work since capital controls are too
strong.
4.7 Welfare Analysis
In this section, we will present our welfare analysis and offer quantitative estimates of the
welfare gains of financial integration. As the previous section mentions, the steady state
values of most variables under financial integration are larger than under financial autarky
except capital. Meanwhile, capital scarcity effect and risk sharing effect are conflicting
while FDI scarcity effect has a huge impact on both countries that outweigh other two
effects. What’s more, frictions have a significant effect on both countries though FDI
scarcity effect is much stronger. In such case, we can find that the welfare gains of financial
integration will be large rather than small where FDI scarcity effect contributes a lot.
4.7.1 Definition of Welfare Gains
Here we define the welfare gains as equivalent increase in permanent consumption com-
pared with relative level. First, we define the permanent certainty equivalent consumption
c
i
j
where i denotes countries and j denotes different environments. Hence, we can have
the following definition formula forc
i
j
:
1
X
t=1
t
U(c
i
j
) =E
0
1
X
t=1
t
U(c
j
i;t
) (4.45)
184
Then we can define the welfare gains of financial integration as the increase of perma-
nent consumption:
W
i
=
c
i
FI
c
i
R
c
i
R
(4.46)
From the theory, relative level should be values under financial autarky. However,
in reality we care more about the real feeling rather than impossible financial autarky
scenario. Hence, we adopt two relative levels: financial autarky level and initial level.
As for the initial level, we choose the consumption in period 1 for both countries to be
the relative level of consumption. Then we can understand the welfare gains of financial
integration in theory and reality separately.
Of course, the ideal way from the theory to compare the welfare gains of financial
integration is to compare two different situations that are financial autarky and financial
integration. And there is no doubt that this idea is absolutely correct. However, since now
we assume there will be goods or intermediate goods in the extended model produced
abroad to be used to produce final goods, the introduction of the new goods will lead to a
significant increase on the final output since under financial autarky the foreign goods
don’t exist. In such case, if we still use financial autarky level, then the value of welfare
gains will be very large that can’t be avoided. Despite from the theory, this method is
reasonable while in the reality, this result is not reasonable. So here we try to introduce a
new level that is the initial level so that we can partially avoid the large increase due to
the introduction of new goods.
4.7.2 Welfare Gains
In this section, we will provide the specific values of welfare gains of financial integration.
Table 4.6 summarizes the welfare gains of financial integration in different cases
with the relative level of financial autarky. Case 1 is a risky world; Case 2 is similar
185
Table 4.6 Welfare gains of financial integration 1
Country D Country E
Baseline (Case 1) 11.69% 21.42%
High risk aversion (Case 2) 6.64% 26.76%
Frictions (Case 3) 13.84% 19.08%
No capital scarcity 13.35% 18.52%
Symmetric 13.35% 18.52%
FDI only 15.92% 16.76%
Bond only -3.61% 3.95%
Note: the relative level is financial autarky level.
to Case 1 while there is a higher risk aversion; Case 3 is a risky world with frictions
where capital controls and financial frictions exist; ”no capital scarcity” case is the case
where emerging economy starts with the equal level of capital to developed country while
risk is asymmetric; ”symmetric” case is the case where risk is equal for each country
(
E
=
D
= 2:5%) and both countries start with the same initial capital that is the steady
state under financial autarky; ”FDI only” case is the case where there is only FDI under
financial integration; ”bond only” case is the case where there is only bond under financial
integration.
First, it’s very clear that the welfare gains of financial integration is super large if both
countries start from financial autarky to financial integration. Meanwhile, the welfare
gains are mostly from FDI rather than bond, which explains why we don’t find large
welfare gains of financial integration in theory since we don’t consider FDI in the model
before. The result of ”FDI only” case is in line with Albuquerque (2003) while the result
of ”bond only” case for emerging economy is a little bit higher than previous results like
Coeurdacier et al. (2015). What’s more, we can find that high risk aversion leads to smaller
welfare gains for developed country but larger welfare gains for emerging economy. This
is because if countries are high risk aversion, then they will tend to invest safely. In this
situation, capital will flow to safer country that is developed country via bond. But at the
same time, developed country has to pay more interest and reduce its benefit. Moreover,
186
frictions have an opposite effect with high risk aversion. This lose is the cost to achieve
the goal of capital controls that is to protect emerging economy. However, this cost is
not very large, which is similar to the results in Bai and Zhang (2012). And we can find
that capital scarcity effect has a relatively large impact on developed country while the
impact on emerging economy is larger. This means capital scarcity effect contributes to the
welfare gains for developed country a lot but makes emerging economy lose a large part of
welfare gains, which is very close to the results in Gourinchas and Jeanne (2006). From the
comparison of ”no capital scarcity” and ”symmetric” case, we can find that asymmetric
risk doesn’t have a significant effect, which may be due to the relatively small risk. Finally,
one interesting thing we can find from ”FDI only” and ”bond only” case is that bond
makes developed country worse but makes emerging economy better. This is because
developed country has to pay interest for its debt. However, this reminds us of whether
developed country should take the risk to borrow money from emerging economy since
the cost is large rather than small enough to be ignored.
Table 4.7 Welfare gains of financial integration 2
Country D Country E
Baseline (Case 1) 8.02% 9.69%
High risk aversion (Case 2) 1.48% 1.60%
Frictions (Case 3) 7.42% 8.37%
No capital scarcity 5.08% 6.00%
Symmetric 5.08% 6.00%
FDI only 5.64% 12.31%
Bond only 6.43% 8.16%
Note: the relative level is initial level. Since ”no capital scarcity”, ”symmet-
ric” and ”bond only” are very different from baseline (Case 1), we adopt the
consumption in period 1 under each case to be their own initial levels.
Table 4.7 summarizes the welfare gains of financial integration in different cases with
the relative level of reality. The definition of seven cases is same as in Table 4.6. And the
results of Table 4.7 is much closer to the reality.
187
We can still find that the welfare gains of financial integration are still large that are
around 9%. And from the huge difference of Table 4.6 and Table 4.7, we can know that
most of the welfare gains of financial integration, especially for emerging economy, are
due to the fast growth of FDI in the early stage. In history, we do find there exists such a
fast growth period. But after this period, the growth rate is slower. Another interesting
thing is that if countries are high risk aversion, the welfare gains will suddenly drops to
about 1.5%. This is because the initial level for emerging economy has increased a lot and
developed country has to pay more for risk sharing. Meanwhile, frictions can weaken the
welfare gains a lot and the loss will be about 1%, which is also a possible explanation for
the insignificant welfare gains in empirical researches that is in line with Bai and Zhang
(2012). The logic is that frictions make growth rate smaller like high risk aversion and
make people harder to feel the large welfare gains. And from ”no capital scarcity” case, we
can find that capital scarcity effect also matters a lot that is 2.94% for developed country
and 3.69% for emerging economy. And again, asymmetric risk doesn’t have a significant
effect. Moreover, we can find that the effect of FDI is significant and bond also has a
significant effect, which can be found in ”FDI only” case where we can see the welfare
gains in ”FDI only” case is not that close to Baseline (Case 1). And we can see that welfare
gains in ”bond only” case seems to be large, which is because the method of initial level is
not suitable for ”bond only” case.
4.7.3 Summary
It’s clear that the welfare gains of financial integration are large for both countries. How-
ever, the welfare gains are mainly from FDI while bond also has a significant effect. But
previously we always consider bond only, which may lead to the opposite result. As for
the effects, for emerging economy, FDI scarcity effect is the main part of welfare gains and
capital scarcity effect is relatively small while risk sharing effect is not ideal even if we
have added FDI in the model. But for developed country, FDI scarcity is also large while
188
risk sharing effect is significant. Meanwhile, frictions can highly weaken the welfare gains
especially the growth rate of capital and consumption that may be another reason of the
small welfare gains observed. Frictions may not change the aggregate welfare gains a lot
while frictions can put off the welfare gains and then affect our feeling of welfare gains
especially in the early stage of financial integration. And if we consider the long process
of frictions being removed and the short time period between the beginning of financial
integration and today, we may need more time to find the large welfare gains of financial
integration. In a word, the welfare gains of financial integration are large while ignorance
of FDI and frictions may weaken the welfare gains to be insignificant. Developed country
benefits from risk sharing effect and FDI scarcity effect while emerging economy mainly
benefits from FDI scarcity effect.
However, there are still some problems existing. First, the choice of initial level may
not be perfect that may cause a smaller welfare gains. Second, there should be some
frictions for FDI so that FDI can’t increase to a high level, which is what we don’t capture
before and there is no enough literature in this aspect. In such case, the research of FDI
friction should be paid more attention to in the future’s work. Third, frictions may cause
the quantity controls and the price controls together while we just use price controls in our
model. If we can add quantity controls in our model, the negative effect of frictions will be
larger, which may lead to a insignificant welfare gains of financial integration that fits the
past empirical results. Finally, there are many reasons to cause the low productivity of FDI
while these problems might be solved in the future and after that, FDI might contribute
much more to the welfare gains of financial integration.
Meanwhile, in line with Coeurdacier et al. (2015), we also show the heterogeneous
effects of financial integration across countries and times. The welfare gains of financial
integration may vary with heterogeneous levels of capital scarcity, risk aversion, frictions
and available assets that can be traded. According to Kose et al. (2006), empirical literature
tried to find an average effect of financial integration while the results vary significantly.
189
Compared with Coeurdacier et al. (2015), our results may offer a more reasonable explana-
tion for heterogeneity in empirical estimates of welfare gains of financial integration. This
is because our models consider more elements including FDI and frictions that enriches
our explanation.
4.8 Sensitivity Analysis
In this section, we will perform a wide range of sensitivity analysis including the role
of country size, capital scarcity, FDI productivity, speed of frictions being removed, and
observation time. Our main results still hold: financial integration does bring large welfare
gains for both developed countries and emerging economies. However, the role of country
size, FDI productivity, and observation time do have a significant effect on the welfare
gains of financial integration, which may explain why we don’t find clear evidence of large
welfare gains from empirical work.
4.8.1 The Role of Country Size
In previous analysis, our results are based on the assumption that two countries are of
equal sizes. However, some emerging economies may not integrate and other emerging
economies may integrate but are not counted in the analysis. In such case, the size of
emerging economy is not equal to developed country. Meanwhile, it’s important to explore
the impact of country size on the welfare gains of financial integration. Here we present
the results of welfare gains of financial integration given the different country sizes, trying
to see if our results are robust.
One important thing is that we can’t adjust the country size of emerging economy to
be small to test if small open economy can benefit more from financial integration that is
similar to the method in Coeurdacier et al. (2015). This is because we can’t ignore the effect
of other emerging economies that have integrated into world financial market. In such
190
case, it’s not appropriate to just adjust the country size of emerging economy. Meanwhile,
even if we want to consider this situation, we should adjust the initial capital for such a
small open economy rather than keep the initial capital same as before. In a word, this
method is not appropriate to investigate the welfare gains of financial integration for small
open economy.
Table 4.8 Welfare gains of financial integration for different country sizes
Baseline (Case 1)
A
E;0
=A
D;0
0.9 1 1.1
Country D E D E D E
Financial autarky level 9.57% 24.77% 11.69% 21.42% 13.91% 18.78%
Initial level 7.76% 9.53% 8.02% 9.69% 8.26% 9.85%
Frictions (Case 3)
A
E;0
=A
D;0
0.9 1 1.1
Country D E D E D E
Financial autarky level 11.68% 22.11% 13.84% 19.08% 16.04% 16.68%
Initial level 7.17% 8.17% 7.42% 8.37% 7.64% 8.55%
Table 4.8 shows the welfare gains of financial integration for different country sizes.
The first part is under full financial integration (Case 1) and the second part is under
limited financial integration (Case 3). Here we adjust country size to be 0.9 and 1.1
trying to investigate the impact of different country sizes on the welfare gains of financial
integration.
From Table 4.8, we can find that the welfare gains of financial integration vary a lot
under financial autarky level but stay robust under initial level. This means country size
does matter and smaller country can benefit more from financial integration. However,
smaller country will feel less benefit since if country size is smaller, capital will grow
slower and then consumption will grow slower, which will also cause FDI from developed
country and then consumption of developed country grow slower. However, our results
still stay robust since initial level can better reflect the reality situation. Meanwhile, even
191
if country size drops to 0.9, the welfare gains of financial integration are still large enough
for both countries, which means our main results are robust.
4.8.2 The Role of Capital Scarcity
In this section, we try to explore how welfare gains of financial integration depend on the
initial capital ratio or the initial degree of capital scarcity.
Table 4.9 Welfare gains of financial integration for different degrees of capital scarcity
Baseline (Case 1)
k
E;0
=k
D;0
0.3 0.5 0.7
Country D E D E D E
Financial autarky level 10.99% 22.88% 11.69% 21.42% 12.36% 20.14%
Initial level 9.24% 11.27% 8.02% 9.69% 6.83% 8.17%
Frictions (Case 3)
k
E;0
=k
D;0
0.3 0.5 0.7
Country D E D E D E
Financial autarky level 13.43% 20.20% 13.84% 19.08% 14.24% 18.13%
Initial level 8.64% 9.90% 7.42% 8.37% 6.21% 6.91%
Table 4.9 shows the welfare gains of financial integration for different degrees of capital
scarcity. The first part is under full financial integration (Case 1) and the second part is
under limited financial integration (Case 3). Here we adjust the initial capital ratio to be
0.3 and 0.5 trying to investigate the impact of different degrees of capital scarcity on the
welfare gains of financial integration.
From Table 4.9, we can find that in general the welfare gains of financial integration
doesn’t change a lot, which means our results are robust. However, we can find that welfare
gains of developing countries is super stable while welfare gains of emerging economies is
not that stable. This is because every time when we adjust the initial capital of emerging
economies, this will directly affect its welfare gains. Another interesting thing is that for
developed countries, in both cases, welfare gains under financial autarky level increase
192
with higher initial capital ratio while welfare gains under initial level decrease with higher
initial capital level. This means though the absolute benefit is larger, the benefit we feel is
smaller. But we can still find that the welfare gains of financial integration is large enough
that can’t be ignored. Meanwhile, in both cases and under both methods, for emerging
economies, welfare gains decrease with higher initial capital ratio. This means emerging
economies benefit less from higher initial capital level all the time. And again we prove
that capital scarcity effect is not significant.
In a word, our results stay robust with different degrees of capital scarcity under full
financial integration (Case 1) and limited financial integration (Case 3), and financial
integration does bring sizable benefits.
4.8.3 The Role of FDI Productivity
Here we want to discuss the role of FDI productivity on welfare gains of financial integra-
tion. Albuquerque (2003) believes FDI productivity should be same to capital productivity
or even bigger because of tax subsidy, which will cause consumption almost double after fi-
nancial integration. However, due to many reasons such as home bias and localization, FDI
productivity is not as high as we expect. According to the fact that the ratio of FDI stock
and GDP is about 30%-40%, we calibrateA
f
to be 0.3. But this number may change in the
future. Under this circumstance, we try to find the impact of different FDI productivity on
welfare gains of financial integration.
Table 4.10 shows the welfare gains of financial integration for different degrees of
capital scarcity. The first part is under full financial integration (Case 1) and the second
part is under limited financial integration (Case 3). Here we adjust FDI productivity to be
0.2 and 0.4 trying to investigate the impact of different FDI productivity on the welfare
gains of financial integration.
From Table 4.10, it’s obvious that FDI productivity matters a lot. Under both cases, the
higher the FDI productivity, the more both countries can benefit. Meanwhile, the difference
193
Table 4.10 Welfare gains of financial integration for different degrees of capital scarcity
Baseline (Case 1)
A
f
0.2 0.3 0.4
Country D E D E D E
Financial autarky level 4.97% 13.77% 11.69% 21.42% 19.48% 30.22%
Initial level 7.38% 9.09% 8.02% 9.69% 8.68% 10.30%
Frictions (Case 3)
A
f
0.2 0.3 0.4
Country D E D E D E
Financial autarky level 6.91% 11.66% 13.84% 19.08% 21.83% 27.66%
Initial level 7.16% 8.16% 7.42% 8.37% 7.67% 8.57%
is super large under financial autarky level while under initial level the difference is not
that large. This means the initial fast growth of FDI is very important while after this
period, the growth may not be that significant. What’s more, even if FDI productivity
is higher, we may not be able to feel the difference very much. But still, even if FDI
productivity drops to 0.2, the welfare gains of financial integration are still large enough
for both countries, which means our main results are robust.
4.8.4 The Role of Speed of Frictions Being Removed
In our model, we assume that the world is a world with frictions and frictions are being
removed slowly. But how long will frictions be removed totally? What’s the impact of
fading frictions? These questions still remain important. In the predicted future, frictions
will still exist and it’s hard to give a accurate time for frictions being removed totally.
In such case, we give a estimation of this period that is 60 years while it’s necessary
to investigate the role of different speed of frictions being removed on welfare gains of
financial integration.
Table 4.11 shows the welfare gains of financial integration for different speed of
frictions being removed. The first part is under full financial integration (Case 1) and
194
Table 4.11 Welfare gains of financial integration for different speed of frictions being
removed
Frictions (Case 3)
Period (years) 50 60 100
Country D E D E D E
Financial autarky level 13.82% 19.10% 13.84% 19.08% 13.83% 19.02%
Initial level 7.50% 8.46% 7.42% 8.37% 7.21% 8.15%
the second part is under limited financial integration (Case 3). Here we take the time
for frictions to decrease to 1% level of their initial level to be the time when frictions are
removed totally since 1% is small enough to be ignored. And we adjust this time to be 50
and 100 years trying to investigate the impact of speed of frictions being removed on the
welfare gains of financial integration.
From Table 4.11, we can find that our results are robust for different speed of frictions
being removed since given different speed of speed of frictions being removed, the welfare
gains almost doesn’t change. This is because though frictions still exist, the frictions
are not strong enough to have a significant impact, which is in line with Bai and Zhang
(2012). However, though the difference here is not large, according to our previous results,
frictions still have a large impact on welfare gains of financial integration.
4.8.5 The Role of Observation Time
In this section, we try to explore the impact of observation time on welfare gains of financial
integration. In our previous analysis and other theoretical literature, we always calculate
the welfare gains in a long period. For example, in our previous analysis, we consider the
change in 150 years. However, the history of financial integration is not very long and
even until now, the length is just less than 30 years. Meanwhile, it’s obvious that at the
early stage of financial integration, the frictions are strong and there are other frictions
existing to weaken the welfare gains of financial integration. And there is financial crisis
195
in 2008-2009 that may have an impact to slow the progress of financial integration. In
such case, we adjust the observation time trying to see if the results will be different.
Table 4.12 Welfare gains of financial integration for different observation time
Baseline (Case 1)
observation time (years) 10 20 150
Country D E D E D E
Financial autarky level 8.77% 22.42% 9.87% 21.99% 11.69% 21.42%
Initial level 2.80% 3.41% 4.74% 5.75% 8.02% 9.69%
Frictions (Case 3)
observation time (years) 10 20 150
Country D E D E D E
Financial autarky level 11.10% 20.78% 12.04% 19.98% 13.84% 19.08%
Initial level 2.44% 2.77% 4.21% 4.76% 7.42% 8.37%
Table 4.12 shows the welfare gains of financial integration for different observation
time. The first part is under full financial integration (Case 1) and the second part is under
limited financial integration (Case 3). Here we adjust observation time to be 10 years and
20 years trying to investigate the impact of different FDI productivity on the welfare gains
of financial integration. Since now the observed period is smaller, we choose consumption
in 10 years and 20 years under financial autarky to calculate the equivalent consumption
in these periods and use these values to be the financial autarky level for comparison.
From Table 4.12, we can find that financial integration does bring sizable benefits to
both countries even if we control observation time to be 10 or 20 years. However, under
initial level, the welfare gains of financial integration will be much smaller. And the
larger observation time is, the larger the welfare gains are. We believe that the results here
explain why at early stage researchers don’t find a very significant benefit while recently
we may find a more significant benefit. Meanwhile, the existence of frictions does weaken
the welfare gains of both countries very much. Under this circumstance, frictions make us
harder to find the sizable benefit from financial integration. If we consider the existence
196
of other frictions and the impact of financial crisis, the welfare gains might be weakened
more, which may cause us harder to find a significant benefit.
In a word, observation time matters a lot to welfare gains of financial integration. Since
even until now, the history of financial integration is not long enough, the short observation
time may cause us harder to find a significant benefit from financial integration.
4.9 Extended Model
In this section, we introduce our extended model for the welfare analysis of financial
integration. In line with our benchmark model, here we mainly consider two different
scenarios: financial autarky and full financial integration with a non-state contingent bond
and FDI while we also consider limited financial integration. The extension we make here
is mainly to solve three problems: (1) there is only one goods in the benchmark model; (2)
due to the existence of global supply chains, countries can diversify FDI investment to
different countries to provide hedge against country-specific risk, which is also in line with
the FDI trends in recent years; (3) we just cover two countries in the benchmark model
that can be extended to multiple countries.
To solve these three problems, we will first introduce intermediate goods and final
goods with their production setting to our benchmark model. Then we will extend the
model from two countries to three countries to improve the setting of FDI risk sharing
effect. Finally, we will extend the model from three countries to N countries. To better
clarify our extended model, here we will provide the full details of model setting.
The logic of FDI risk sharing effect via global supply chains is related to the trade-
investment-services-IP nexus proposed by Baldwin (2012), which is developing very fast
in the academia in recent years. Different from 20th century, 21st century trade is much
more complex since global supply chains internationalize the complex two-way flows
that was solved within factories. In such case, the rising global supply chains are not
197
traditional trade anymore. Instead, global supply chains mean the integration of trade in
goods, especially parts and components; international investment in production facilities,
technology, and business relationships; infrastructure to coordinate production; and cross-
border flows of knowledge including intellectual property and managerial skills. This
kind of interconnection is called trade-investment-services-IP nexus.
Facing this complex trade-investment-services-IP nexus, multinational corporations
will use FDI to directly control part of the property and participate in the management and
operation of the foreign companies. In such case, multinational corporations can manage
their global supply chains to diversify the risk in one specific country and ensure stable
goods supply so that even if the production in one country gets impacted, the production
in other countries can offset this shortage and ensure smooth output. Meanwhile, the
choice of FDI becomes a part of their global supply chains management. As a result,
considering the special role of FDI in global supply chains management and its special
effect of risk sharing, we add this setting to our extended model trying to see if properly
designed global supply chains have a significant impact on our main findings.
4.9.1 Model Extension
Now the world is still divided into two countries: D and E where D represents the de-
veloped country and E represents the emerging economy. Each country consists of a
benevolent government and a continuum of identical households. There is one numeraire
final good in this world, which is used for investment and consumption. And the final con-
sumption good is actually a composite of intermediate goods. There are two intermediate
goods that can be used together to produce final goods. This setting is different from our
benchmark model where there is only one numeraire good and no intermediate goods or
final goods. In other word, there is only one stage of production while here we assume
production is two-stage. Each country starts with a given initial capital stockk
i;0
.
198
4.9.1.1 Production
There are two different methods of production to produce two different intermediate goods
in each country. The first method is the standard method that is to use capital and labor
with a Cobb-Douglas production function to produce the domestic intermediate good:
y
1
i;t
=A
i;t
(k
it
)
(l
i;t
)
1
(4.47)
whereA
i;t
is TFP, which can be decomposed into two components: a transitory country-
specific componenta
i;t
and a persistent world componenta
W;t
. In such case, we can have
the following formula forA
i;t
:
A
i;t
=a
W;t
a
i;t
(4.48)
The short-run transitory country-specific componenta
i;t
is a standard setting to present
the fact that each country is hit by transitory country-specific shocks in TFP growth. And
the long-run persistent world componenta
W;t
presents the fact that the world is hit by
persistent shocks in TFP growth. Bothlog(a
W;t
) andlog(a
i;t
) follow AR(1) process:
log(a
W;t+1
) = (1
W
)log(a
W;0
) +
W
log(a
W;t
) +
W;t
(4.49)
log(a
i;t+1
) = (1)log(a
i;0
) +log(a
i;t
) +
i;t
(4.50)
where
W
is the persistence of the growth rate of the long-run persistent world compo-
nenta
W;t
and is the persistence of the short-run transitory country-specific component
a
i;t
.
W;t
is an i.i.d normal distributed process with volatility
W
. And
i;t
= (
D;t
;
E;t
)
0
is
an i.i.d normal distributed process with covariance matrix =
2
D
D
E
D
E
2
E
. a
i;0
is the
initial value of total factor productivity in each country.
W;t
and
i;t
are independent.
199
The second method is to use FDI or in alienable capital only to produce another foreign
intermediate good:
y
2
i;t
=A
i;t
A
f
(k
f
i;t
)
f
(4.51)
wherek
f
it
is the level of FDI or in alienable capital andA
f
is the relative scale factor.
Both methods of production are impacted by the same shocks so that the stochastic TFP is
the same. And what needs to be paid special attention to is that the intermediate goods
produced via FDI fully belong to the domestic country. This means domestic country can
get full return from FDI production.
The law of motion of capital is same as the law of motion of FDI in each country:
k
i;t+1
= (1)k
i;t
+k
i;t
(
i
i;t
k
i;t
) (4.52)
k
f
i;t+1
= (1)k
f
i;t
+k
f
i;t
(
i
f
i;t
k
f
i;t
) (4.53)
where 0<< 1 is the depreciation rate of capital and FDI.i
it
andi
f
it
are the gross
investment in capital and FDI.(x) is a standard quadratic adjustment cost function:
(x) =x
2
(x)
2
(4.54)
where is the degree of adjustment costs.
Labor market and capital markets of intermediate goods are perfectly competitive and
the returns for domestic capital are their marginal productivity while for FDI, marginal
200
productivity is not equal to the return. The reason why we still adopt this notation is to
simplify our equations later. Thus, we can have the following formulas:
w
i;t
l
i;t
= (1)y
1
i;t
pp
1
i;t
(4.55)
r
i;t
k
i;t
=y
1
i;t
pp
1
i;t
(4.56)
r
f
i;t
k
f
i;t
=
f
y
2
i;t
pp
2
i;t
(4.57)
wherepp
1
i;t
andpp
2
i;t
represent price of the intermediate goods, domestic intermediate
goods and foreign intermediate goods, respectively. For simplicity, we can normalize
population in domestic capital sector to be unity in each country. Then country size is
homogeneous to the total factor productivityA
it
.
Then the final goods are produced using intermediate goods from two sectors (domestic
and foreign sector):
Y
i;t
= (n
1
i;t
(y
1
i;t
)
m
1
m
+n
2
i;t
(y
2
i
0
;t
)
m
1
m
)
m
m
1
(4.58)
whereY
i;t
denotes final good output,y
1
i;t
andy
2
i
0
;t
denote intermediate inputs in domes-
tic and foreign sector.
m
measures the elasticity of substitution between intermediate
goods, andn
1
i;t
andn
2
i;t
measure the share of two intermediate goods.
To minimize the cost of final good producer, we can have:
y
1
i;t
= (pp
1
i;t
)
m
(n
1
i;t
)
m
Y
i;t
(4.59)
y
2
i
0
;t
= (pp
2
i
0
;t
)
m
(n
2
i;t
)
m
Y
i;t
(4.60)
The zero-profit condition in the final good sector implies:
1 = (n
1
i;t
)
m
(pp
1
i;t
)
1
m
+ (n
2
i;t
)
m
(pp
2
i
0
;t
)
1
m
(4.61)
201
4.9.1.2 Demand
The benevolent government in each country chooses consumption, investment, borrowing
or lending to maximize utility of the domestic consumers given by
U
i
=E
0
1
X
t=1
t
U(c
i;t
) =E
0
1
X
t=1
t
c
1
i;t
(4.62)
where is the degree of relative risk aversion and the utility function is a CRRA utility
function.
Budget constraints and market clear conditions depend on the different cases where
the availability of different assets is different.
And in all cases, the stochastic discount factor (SDF) in country i is:
M
i;t+1
=(
c
i;t+1
c
i;t
)
(4.63)
Under financial autarky, the only available asset is domestic capital and there is no
foreign debt or FDI available. Then we can treatk
f
i;t
to be 0 in each country all the time
and the only choice for the household made by the benevolent government is to consume
or invest in domestic capital with the incomes. Then we can have the following budget
constraint for household:
c
i;t
+i
i;t
=w
i;t
+r
i;t
k
i;t
(4.64)
We can also have the following Euler equations:
E
t
fM
i;t+1
[r
i;t+1
0
i;t
+
0
i;t
0
i;t+1
(1 +
i;t+1
i
i;t+1
k
i;t+1
0
i;t+1
)]g = 1 (4.65)
where
i;t
=(
i
i;t
k
i;t
) and
0
i;t
is the first derivative of the function(x).
202
Since there is no foreign intermediate good, the goods market clear conditions in
country i are:
c
i;t
+i
i;t
=A
i;t
(k
i;t
)
(n
1
i;t
)
m
m
1
(4.66)
Under full financial integration, there are two assets under financial integration case.
First, there is a riskless international bond that will pay one unit of good back in the next
period and the price at time t isp
t
. Second, there is FDI that can be used to produce
intermediate goods back to the home country to produce final goods. And one thing needs
to be clear: the whole return from the production of intermediate goods will be back to
the home country.
Then the budget constraint in country i becomes:
c
i;t
+i
i;t
+i
f
i
0
;t
=w
i;t
+r
i;t
k
i;t
+y
2
i
0
;t
pp
2
i
0
;t
+b
i;t1
b
i;t
p
t
(4.67)
where i’ represents the other country andb
i;t
represents the new bond country i buys
in time t.
Meanwhile, we have a non-Ponzi condition:
b
i;t
>b
i
(4.68)
whereb
i
is the debt limit. In our real simulation, we set it to be a very small value.
Then we can have the same Euler equation for domestic capital as before. Meanwhile,
we can have the following Euler equations for FDI:
E
t
fM
i;t+1
[r
f
i
0
;t+1
0
i
0
;t
+
0
i
0
;t
0
i
0
;t+1
(1 +
i
0
;t+1
i
f
i
0
;t+1
k
f
i
0
;t+1
0
i
0
;t+1
)]g = 1 (4.69)
where
i
0
;t
=(
i
f
i
0
;t
k
f
i
0
;t
) and
0
i
0
;t
is the first derivative of the function(x).
203
What’s more, we can have the following Euler equations for bond:
p
t
=E
t
(M
i;t+1
) +
i;t
(4.70)
where
i;t
is the Lagrange multiplier for the non-Ponzi condition.
Finally, we can have the following market clear conditions for goods market and bonds
market:
c
D;t
+i
D;t
+i
f
E;t
+c
E;t
+i
E;t
+i
f
D;t
=Y
D;t
+Y
E;t
(4.71)
b
D;t
+b
E;t
= 0 (4.72)
4.9.1.3 Equilibrium
Now, we can define an equilibrium under financial autarky. An equilibrium in country i is
a sequence of consumption and capitalfc
i;t
,k
i;t
g such that the Euler equation for capital
(4.65) and goods market clear condition (4.66) are both satisfied.
Meanwhile we can define an equilibrium under full financial integration. An equilib-
rium is a sequence of consumption, capital, FDI and bond in both countriesfc
i;t
,k
i;t
,k
f
i
0
;t
,
b
i;t
g fori =D;E and a sequence of bond pricesp
t
and intermediate goods pricepp
1
i;t
and
pp
2
i;t
such that the Euler equations for capital, FDI, and bond (4.65), (4.69), and (4.70) are
satisfied while goods market and bond market clear conditions (4.71) and (4.72) are also
satisfied.
4.9.2 Three-country Extended Model
Now the world is divided into three countries: D, E1, and E2 where D represents the
developed country and E represents the emerging economy. For simplicity, we denote
these three countries to be country 1, 2, and 3. Since most multinational corporations
204
are owned by developed country, which is in the upper stream of global supply chains
that has advantages in technology and other aspects, developed country has advantages
in allocating global supply chains while emerging economy doesn’t have much room to
choose. In such case, we assume developed country can invest on emerging economies via
FDI while emerging economy can also invest on developed country via FDI but can’t invest
on other emerging economies. Other assumptions are in line with our assumptions before.
Another change is that since now we have three countries rather than two countries, our
notation will change in order to better represent variables.
4.9.2.1 Production
There are two different methods of production to produce two different intermediate goods
in each country. The first method is the standard method that is to use capital and labor
with a Cobb-Douglas production function to produce the domestic intermediate good:
y
i;i;t
=A
i;t
(k
i;i;t
)
(l
i;t
)
1
(4.73)
wherey
i;i;t
andk
i;i;t
mean the production and capital owned by country i and located
in country i; A
i;t
is TFP, which can be decomposed into two components: a transitory
country-specific componenta
i;t
and a persistent world componenta
W;t
. In such case, we
can have the following formula forA
i;t
:
A
i;t
=a
W;t
a
i;t
(4.74)
The short-run transitory country-specific componenta
i;t
is a standard setting to present
the fact that each country is hit by transitory country-specific shocks in TFP growth. And
205
the long-run persistent world componenta
W;t
presents the fact that the world is hit by
persistent shocks in TFP growth. Bothlog(a
W;t
) andlog(a
i;t
) follow AR(1) process:
log(a
W;t+1
) = (1
W
)log(a
W;0
) +
W
log(a
W;t
) +
W;t
(4.75)
log(a
i;t+1
) = (1)log(a
i;0
) +log(a
i;t
) +
i;t
(4.76)
where
W
is the persistence of the growth rate of the long-run persistent world compo-
nenta
W;t
and is the persistence of the short-run transitory country-specific component
a
i;t
.
W;t
is an i.i.d normal distributed process with volatility
W
. And
i;t
= (
1;t
;
2;t
;
3;t
)
0
is an i.i.d normal distributed process with covariance matrix =
2
1
1;2
1
2
1;3
1
3
2;1
2
1
2
2
2;3
2
3
3;1
3
1
3;2
3
2
2
3
where
i;j
=
j;i
. a
i;0
is the initial value of total factor productivity in each country.
W;t
and
i;t
are independent.
The second method is to use FDI or in alienable capital only to produce another foreign
intermediate good:
y
i;j;t
=A
j;t
A
f
(k
i;j;t
)
f
;i,j (4.77)
where y
i;j;t
and k
i;j;t
are the production and FDI or in alienable capital owned by
country i and located in country j whilei,j, which means this production is happened in
foreign countries but the capital (FDI) is owned by domestic country;A
f
is the relative
scale factor. Both methods of production are impacted by the same shocks so that the
stochastic TFP is the same. And what needs to be paid special attention to is that the
intermediate goods produced via FDI fully belong to the domestic country. This means
domestic country can get full return from FDI production.
206
The law of motion of capital is same as the law of motion of FDI in each country:
k
i;i;t+1
= (1)k
i;i;t
+k
i;i;t
(
i
i;i;t
k
i;i;t
) (4.78)
k
i;j;t+1
= (1)k
i;j;t
+k
i;j;t
(
i
i;j;t
k
i;j;t
);i,j (4.79)
where 0 < < 1 is the depreciation rate of capital and FDI. i
i;i;t
and i
i;j;t
are the
gross investment in domestic capital and FDI in country j. (x) is a standard quadratic
adjustment cost function:
(x) =x
2
(x)
2
(4.80)
where is the degree of adjustment costs.
Labor market and capital markets of intermediate goods are perfectly competitive and
the returns for domestic capital are their marginal productivity while for FDI, marginal
productivity is not equal to the return. The reason why we still adopt this notation is to
simplify our equations later. Thus, we can have the following formulas:
w
i;t
l
i;t
= (1)y
i;i;t
pp
i;i;t
(4.81)
r
i;i;t
k
i;i;t
=y
i;i;t
pp
i;i;t
(4.82)
r
i;j;t
k
i;j;t
=
f
y
i;j;t
pp
i;j;t
;i,j (4.83)
wherepp
i;i;t
andpp
i;j;t
represent price of the intermediate goods, domestic intermediate
goods and foreign intermediate goods in country j, respectively. For simplicity, we can
normalize population in domestic capital sector to be unity in each country. Then country
size is homogeneous to the total factor productivityA
it
.
207
Then the final goods are produced using intermediate goods from two sectors (domestic
and foreign sector):
Y
i;t
= (n
i;i;t
(y
i;i;t
)
m
1
m
+n
i;2;t
(y
i;2;t
)
m
1
m
+n
i;3;t
(y
i;3;t
)
m
1
m
)
m
m
1
;i = 1 (4.84)
Y
i;t
= (n
i;i;t
(y
i;i;t
)
m
1
m
+n
i;1;t
(y
i;1;t
)
m
1
m
)
m
m
1
;i = 2;3 (4.85)
where Y
i;t
denotes final good output, y
i;i;t
and y
i;j;t
denote intermediate inputs in
domestic and foreign sector as mentioned before.
m
measures the elasticity of substitution
between intermediate goods, andn
i;j;t
measures the country i’s share of intermediate good
produced in country j.
To minimize the cost of final good producer, we can have:
y
i;i;t
= (pp
i;i;t
)
m
(n
1
i;t
)
m
Y
i;t
(4.86)
y
i;j;t
= (pp
i;j;t
)
m
(n
2
i;t
)
m
Y
i;t
;i,j (4.87)
The zero-profit condition in the final good sector implies:
1 = (n
i;i;t
)
m
(pp
i;i;t
)
1
m
+ (n
i;2;t
)
m
(pp
i;2;t
)
1
m
+ (n
i;3;t
)
m
(pp
i;3;t
)
1
m
;i = 1 (4.88)
1 = (n
i;i;t
)
m
(pp
i;i;t
)
1
m
+ (n
i;1;t
)
m
(pp
i;1;t
)
1
m
;i = 2;3 (4.89)
4.9.2.2 Demand
The benevolent government in each country chooses consumption, investment, borrowing
or lending to maximize utility of the domestic consumers given by
U
i
=E
0
1
X
t=1
t
U(c
i;t
) =E
0
1
X
t=1
t
c
1
i;t
(4.90)
where is the degree of relative risk aversion and the utility function is a CRRA utility
function.
208
Budget constraints and market clear conditions depend on the different cases where
the availability of different assets is different.
And in all cases, the stochastic discount factor (SDF) in country i is:
M
i;t+1
=(
c
i;t+1
c
i;t
)
(4.91)
Under financial autarky, the only available asset is domestic capital and there is no
foreign debt or FDI available. Then we can treatk
i;j;t
to be 0 in each country all the time
and the only choice for the household made by the benevolent government is to consume
or invest in domestic capital with the incomes. Then we can have the following budget
constraint for household:
c
i;t
+i
i;i;t
=w
i;t
+r
i;i;t
k
i;i;t
(4.92)
We can also have the following Euler equations:
E
t
fM
i;t+1
[r
i;i;t+1
0
i;i;t
+
0
i;i;t
0
i;i;t+1
(1 +
i;i;t+1
i
i;i;t+1
k
i;i;t+1
0
i;i;t+1
)]g = 1 (4.93)
where
i;i;t
=(
i
i;i;t
k
i;i;t
) and
0
i;i;t
is the first derivative of the function(x).
Since there is no foreign intermediate good, the goods market clear conditions in
country i are:
c
i;t
+i
i;i;t
=A
i;t
(k
i;i;t
)
(n
i;i;t
)
m
m
1
(4.94)
Under full financial integration, there are two assets under financial integration case.
First, there is a riskless international bond that will pay one unit of good back in the next
period and the price at time t isp
t
. Second, there is FDI that can be used to produce
intermediate goods back to the home country to produce final goods. And one thing needs
209
to be clear: the whole return from the production of intermediate goods will be back to
the home country.
Then the budget constraint in country i becomes:
c
i;t
+i
i;i;t
+i
i;2;t
+i
i;3;t
=w
i;t
+r
i;i;t
k
i;i;t
+y
i;2;t
pp
i;2;t
+y
i;3;t
pp
i;3;t
+b
i;t1
b
i;t
p
t
;i = 1 (4.95)
c
i;t
+i
i;i;t
+i
i;1;t
=w
i;t
+r
i;i;t
k
i;i;t
+y
i;1;t
pp
i;1;t
+b
i;t1
b
i;t
p
t
;i = 2;3 (4.96)
whereb
i;t
represents the new bond country i buys in time t.
Meanwhile, we have a non-Ponzi condition:
b
i;t
>b
i
(4.97)
whereb
i
is the debt limit. In our real simulation, we set it to be a very small value.
Then we can have the same Euler equation for domestic capital as before. Meanwhile,
we can have the following Euler equations for FDI:
E
t
fM
i;t+1
[r
i;j;t+1
0
i;j;t
+
0
i;j;t
0
i;j;t+1
(1 +
i;j;t+1
i
i;j;t+1
k
i;j;t+1
0
i;j;t+1
)]g = 1;i,j (4.98)
where
i;j;t
=(
i
i;j;t
k
i;j;t
) and
0
i;j;t
is the first derivative of the function(x).
What’s more, we can have the following Euler equations for bond:
p
t
=E
t
(M
i;t+1
) +
i;t
(4.99)
where
i;t
is the Lagrange multiplier for the non-Ponzi condition.
210
Finally, we can have the following market clear conditions for goods market and bonds
market:
c
1;t
+i
1;1;t
+i
1;2;t
+i
1;3;t
+c
2;t
+i
2;2;t
+i
2;1;t
+c
3;t
+i
3;3;t
+i
3;1;t
=Y
1;t
+Y
2;t
+Y
3;t
(4.100)
b
1;t
+b
2;t
+b
3;t
= 0 (4.101)
4.9.2.3 Equilibrium
Now, we can define an equilibrium under financial autarky. An equilibrium in country i is
a sequence of consumption and capitalfc
i;t
,k
i;i;t
g such that the Euler equation for capital
(4.93) and goods market clear condition (4.94) are both satisfied.
Meanwhile we can define an equilibrium under full financial integration. An equilib-
rium is a sequence of consumption, capital, FDI and bond in both countriesfc
i;t
,k
i;i;t
,k
i;j;t
,
b
i;t
g fori;j = 1;2;3;i ,j and a sequence of bond pricesp
t
and intermediate goods price
pp
i;i;t
andpp
i;j;t
fori;j = 1;2;3;i ,j such that the Euler equations for capital, FDI, and
bond (4.93), (4.98), and (4.99) are satisfied while goods market and bond market clear
conditions (4.100) and (4.101) are also satisfied.
4.9.3 Multi-country Extended Model
Now the world is divided into N countries of two types: D and E where D represents the
developed country and E represents the emerging economy. For simplicity, we denote
these these countries to be country 1, 2, ..., and N. Since most multinational corporations
are owned by developed country, which is in the upper stream of global supply chains
that has advantages in technology and other aspects, developed country has advantages
in allocating global supply chains while emerging economy doesn’t have much room to
choose. In such case, we assume developed country can invest on emerging economies via
FDI while emerging economy can also invest on developed country via FDI but can’t invest
on other emerging economies. Meanwhile, we assume there are one developed country and
211
N-1 emerging economies. For simplicity, we assume country 1 is the developed country
and others are emerging economies. Other assumptions are in line with our assumptions
before. Another change is that since now we have multiple countries rather than two
countries, our notation will change in order to better represent variables.
4.9.3.1 Production
There are two different methods of production to produce two different intermediate goods
in each country. The first method is the standard method that is to use capital and labor
with a Cobb-Douglas production function to produce the domestic intermediate good:
y
i;i;t
=A
i;t
(k
i;i;t
)
(l
i;t
)
1
(4.102)
wherey
i;i;t
andk
i;i;t
mean the production and capital owned by country i and located
in country i;A
i;t
is the stochastic total factor productivity (TFP), which can be decomposed
into two components: a transitory country-specific componenta
i;t
and a persistent world
componenta
W;t
. In such case, we can have the following formula forA
i;t
:
A
i;t
=a
W;t
a
i;t
(4.103)
The short-run transitory country-specific componenta
i;t
is a standard setting to present
the fact that each country is hit by transitory country-specific shocks in TFP growth. And
the long-run persistent world componenta
W;t
presents the fact that the world is hit by
persistent shocks in TFP growth. Bothlog(a
W;t
) andlog(a
i;t
) follow AR(1) process:
log(a
W;t+1
) = (1
W
)log(a
W;0
) +
W
log(a
W;t
) +
W;t
(4.104)
log(a
i;t+1
) = (1)log(a
i;0
) +log(a
i;t
) +
i;t
(4.105)
212
where
W
is the persistence of the growth rate of the long-run persistent world
component a
W;t
and is the persistence of the short-run transitory country-specific
component a
i;t
.
W;t
is an i.i.d normal distributed process with volatility
W
. And
i;t
= (
1;t
;
2;t
;:::;
N;t
)
0
is an i.i.d normal distributed process with covariance matrix
=
2
1
1;2
1
2
1;N
1
N
2;1
2
1
2
2
2;N
2
N
:
:
:
:
:
:
:
:
:
:
:
:
N;1
N
1
N;2
N
2
2
N
where
i;j
=
j;i
. a
i;0
is the initial value of total factor
productivity in each country.
W;t
and
i;t
are independent.
The second method is to use FDI (Foreign Direct Investment) or in alienable capital
only to produce another foreign intermediate good:
y
i;j;t
=A
j;t
A
f
(k
i;j;t
)
f
;i,j (4.106)
where y
i;j;t
and k
i;j;t
are the production and FDI or in alienable capital owned by
country i and located in country j whilei,j, which means this production is happened in
foreign countries but the capital (FDI) is owned by domestic country;A
f
is the relative
scale factor. Both methods of production are impacted by the same shocks so that the
stochastic TFP is the same. And what needs to be paid special attention to is that the
intermediate goods produced via FDI fully belong to the domestic country. This means
domestic country can get full return from FDI production.
In terms of production, we assume there are two different methods of production
to produce two different intermediate goods in each country. The first method is the
standard method that is to use capital and labor with a Cobb-Douglas production function
to produce the domestic intermediate good. The second method is to use FDI only to
produce another foreign intermediate good. Just as Baldwin (2012) mentions, 21st century
trade is much more complex since global supply chains internationalize the complex
two-way flows that was solved within factories. And the flows include trade, investment,
services, and IP . In such case, the production abroad is quite different from the traditional
production that relies more on the technology and knowledge. Moreover, the demand
213
of labor is not that large since the share of foreign intermediate goods in the production
of final goods is small. Under this circumstance, the usage of labor in the production of
foreign intermediate goods is also small. Meanwhile, multinational corporations can use
robots to produce goods, which makes the demand of labor decrease further and the usage
of robots will be counted in capital. In such case, the demand of labor in the production
of foreign intermediate goods is very small and we can choose to simplify labor here. On
the other hand, relatively speaking, the knowledge brought with FDI is more important.
And the usage of robot makes capital more important and reduce the importance of labor.
Under this circumstance, the assumption that we can use FDI only to produce foreign
intermediate goods fits the reality better. In summary, our assumption of using FDI only
to produce foreign intermediate goods is more realistic and reasonable.
Of course, if we pursue a more generalized model, we can add labor into the production
of foreign intermediate goods, which can be a nice point for improvement later. However,
we can also choose to keep the simplicity of the model and ensure the accuracy at the
same time. Actually this setting we adopt here is a simplified version of modelling
the production of foreign intermediate goods that is used in previous literature such as
Albuquerque (2003). And one thing needs to be clear is that even if we use the more
generalized model, the welfare gains won’t change very much. This is because even if
we add the labor input setting into the model, since the production share of foreign
intermediate goods is not large, the labor income is not large either. Meanwhile, this kind
of the production of foreign intermediate goods is always two-way, which means if we
assume we can use FDI and labor together to produce foreign intermediate goods, then
this new setting will have an impact on both sides that includes domestic and foreign
country. Compared with the baseline assumption, of course domestic country can have
more labor income domestically and the added labor income comes from the production
of foreign intermediate goods domestically. But on the other hand, compared with the
baseline assumption, domestic country will also lose FDI income abroad since now the
214
production of foreign intermediate goods need to pay wage to workers abroad. To sum
up, the aggregate effect of these two mechanisms is very small. Under this circumstance,
even if we add labor input setting into the production of foreign intermediate goods, the
total income of households in all countries won’t change very much and then the welfare
gains of financial integration won’t change very much either. This means our results are
robust and the need to use more generalized model is limited. Actually, we tried to add
labor input setting in the benchmark model before and found the difference of welfare
gains is very small that can be ignored.
The law of motion of capital is same as the law of motion of FDI in each country:
k
i;i;t+1
= (1)k
i;i;t
+k
i;i;t
(
i
i;i;t
k
i;i;t
) (4.107)
k
i;j;t+1
= (1)k
i;j;t
+k
i;j;t
(
i
i;j;t
k
i;j;t
);i,j (4.108)
where 0 < < 1 is the depreciation rate of capital and FDI. i
i;i;t
and i
i;j;t
are the
gross investment in domestic capital and FDI in country j. (x) is a standard quadratic
adjustment cost function:
(x) =x
2
(x)
2
(4.109)
where is the degree of adjustment costs.
Labor market and capital markets of intermediate goods are perfectly competitive and
the returns for domestic capital are their marginal productivity while for FDI, marginal
215
productivity is not equal to the return. The reason why we still adopt this notation is to
simplify our equations later. Thus, we can have the following formulas:
w
i;t
l
i;t
= (1)y
i;i;t
pp
i;i;t
(4.110)
r
i;i;t
k
i;i;t
=y
i;i;t
pp
i;i;t
(4.111)
r
i;j;t
k
i;j;t
=
f
y
i;j;t
pp
i;j;t
;i,j (4.112)
wherepp
i;i;t
andpp
i;j;t
represent price of the intermediate goods, domestic intermediate
goods and foreign intermediate goods in country j, respectively. For simplicity, we can
normalize population in domestic capital sector to be unity in each country. Then country
size is homogeneous to the total factor productivityA
it
.
Then the final goods are produced using intermediate goods from two sectors (domestic
and foreign sector):
Y
i;t
= (n
i;i;t
(y
i;i;t
)
m
1
m
+
N
X
k=2
n
i;k;t
(y
i;k;t
)
m
1
m
)
m
m
1
;i = 1 (4.113)
Y
i;t
= (n
i;i;t
(y
i;i;t
)
m
1
m
+n
i;1;t
(y
i;1;t
)
m
1
m
)
m
m
1
;i = 2;3;:::;N (4.114)
where Y
i;t
denotes final good output, y
i;i;t
and y
i;j;t
denote intermediate inputs in
domestic and foreign sector as mentioned before.
m
measures the elasticity of substitution
between intermediate goods, andn
i;j;t
measures the country i’s share of intermediate good
produced in country j.
To minimize the cost of final good producer, we can have:
y
i;i;t
= (pp
i;i;t
)
m
(n
1
i;t
)
m
Y
i;t
(4.115)
y
i;j;t
= (pp
i;j;t
)
m
(n
2
i;t
)
m
Y
i;t
;i,j (4.116)
216
The zero-profit condition in the final good sector implies:
1 = (n
i;i;t
)
m
(pp
i;i;t
)
1
m
+
N
X
k=2
(n
i;k;t
)
m
(pp
i;k;t
)
1
m
;i = 1 (4.117)
1 = (n
i;i;t
)
m
(pp
i;i;t
)
1
m
+ (n
i;1;t
)
m
(pp
i;1;t
)
1
m
;i = 2;3;:::;N (4.118)
4.9.3.2 Demand
The benevolent government in each country chooses consumption, investment, borrowing
or lending to maximize utility of the domestic consumers given by
U
i
=E
0
1
X
t=1
t
U(c
i;t
) =E
0
1
X
t=1
t
c
1
i;t
(4.119)
where is the degree of relative risk aversion and the utility function is a CRRA utility
function.
Budget constraints and market clear conditions depend on the different cases where
the availability of different assets is different.
And in all cases, the stochastic discount factor (SDF) in country i is:
M
i;t+1
=(
c
i;t+1
c
i;t
)
(4.120)
Under financial autarky, the only available asset is domestic capital and there is no
foreign debt or FDI available. Then we can treatk
i;j;t
to be 0 in each country all the time
and the only choice for the household made by the benevolent government is to consume
or invest in domestic capital with the incomes. Then we can have the following budget
constraint for household:
c
i;t
+i
i;i;t
=w
i;t
+r
i;i;t
k
i;i;t
(4.121)
217
We can also have the following Euler equations:
E
t
fM
i;t+1
[r
i;i;t+1
0
i;i;t
+
0
i;i;t
0
i;i;t+1
(1 +
i;i;t+1
i
i;i;t+1
k
i;i;t+1
0
i;i;t+1
)]g = 1 (4.122)
where
i;i;t
=(
i
i;i;t
k
i;i;t
) and
0
i;i;t
is the first derivative of the function(x).
Since there is no foreign intermediate good, the goods market clear conditions in
country i are:
c
i;t
+i
i;i;t
=A
i;t
(k
i;i;t
)
(n
i;i;t
)
m
m
1
(4.123)
Under full financial integration, there are two assets under financial integration case.
First, there is a riskless international bond that will pay one unit of good back in the next
period and the price at time t isp
t
. Second, there is FDI that can be used to produce
intermediate goods back to the home country to produce final goods. And one thing needs
to be clear: the whole return from the production of intermediate goods will be back to
the home country.
Then the budget constraint in country i becomes:
c
i;t
+i
i;i;t
+
N
X
k=2
i
i;k;t
=w
i;t
+r
i;i;t
k
i;i;t
+
N
X
k=2
y
i;k;t
pp
i;k;t
+b
i;t1
b
i;t
p
t
;i = 1 (4.124)
c
i;t
+i
i;i;t
+i
i;1;t
=w
i;t
+r
i;i;t
k
i;i;t
+y
i;1;t
pp
i;1;t
+b
i;t1
b
i;t
p
t
;i = 2;3;:::;N (4.125)
whereb
i;t
represents the new bond country i buys in time t.
Meanwhile, we have a non-Ponzi condition:
b
i;t
>b
i
(4.126)
whereb
i
is the debt limit. In our real simulation, we set it to be a very small value.
218
Then we can have the same Euler equation for domestic capital as before. Meanwhile,
we can have the following Euler equations for FDI:
E
t
fM
i;t+1
[r
i;j;t+1
0
i;j;t
+
0
i;j;t
0
i;j;t+1
(1 +
i;j;t+1
i
i;j;t+1
k
i;j;t+1
0
i;j;t+1
)]g = 1;i,j (4.127)
where
i;j;t
=(
i
i;j;t
k
i;j;t
) and
0
i;j;t
is the first derivative of the function(x).
What’s more, we can have the following Euler equations for bond:
p
t
=E
t
(M
i;t+1
) +
i;t
(4.128)
where
i;t
is the Lagrange multiplier for the non-Ponzi condition.
Finally, we can have the following market clear conditions for goods market and bonds
market:
N
X
i=1
c
i;t
+
N
X
k=1
i
1;k;t
+
N
X
i=2
(i
i;i;t
+i
i;1;t
) =
N
X
i=1
Y
i;t
(4.129)
N
X
i=1
b
i;t
= 0 (4.130)
4.9.3.3 Equilibrium
Now, we can define an equilibrium under financial autarky. An equilibrium in country i is
a sequence of consumption and capitalfc
i;t
,k
i;i;t
g such that the Euler equation for capital
(4.122) and goods market clear condition (4.123) are both satisfied.
Meanwhile we can define an equilibrium under full financial integration. An equilib-
rium is a sequence of consumption, capital, FDI and bond in both countriesfc
i;t
,k
i;i;t
,k
i;j;t
,
b
i;t
g fori;j = 1;2;:::;N;i,j and a sequence of bond pricesp
t
and intermediate goods price
pp
i;i;t
andpp
i;j;t
fori;j = 1;2;:::;N;i,j such that the Euler equations for capital, FDI, and
219
bond (4.122), (4.127), and (4.128) are satisfied while goods market and bond market clear
conditions (4.129) and (4.130) are also satisfied.
4.9.4 Welfare Analysis
In this section, we will present our welfare analysis and offer quantitative estimates of
the welfare gains of financial integration under our extended model. Of course, as the
extension of our benchmark model, the results of our extended model are similar to
benchmark model. But since now there are intermediate goods and final goods with at
least three countries existing, there are also some changes in the results. In specific, benefit
from FDI risk sharing effect is larger than before, which is because the global supply chains
can help developed country better allocate resources to reduce shock. However, in the
benchmark model where they are only one emerging economy rather than two or more
in the extended model, this allocation is not that ideal. In such case, the welfare gains in
extended model are larger than in benchmark model.
We start with the three-country extended model where there are one developed country
1 and two emerging economies 2 and 3 of same country size.
Figure 4.9 shows the dynamics of consumption, capital, and total output while Figure
4.10 shows the dynamics of FDI and bond. In each figure, the upper panel shows the
dynamics for developed country while the lower panel shows the dynamics for emerging
economy. Dashed lines refer to financial autarky level while plain lines refer to financial
integration level.
The dynamics of variables are similar to the dynamics under benchmark model, which
is because the extended model we use now is the extension of the benchmark model. Devel-
oped country lends to emerging economies to accelerate capital accumulation. Meanwhile,
we still observe that capital in developed country decreases first and then increases. This
is because developed country wants to increase FDI and has to suffer a loss in capital for a
while while developed country also needs to cut capital to buy bond. What’s more, when
220
Figure 4.9 Dynamics under extended model 1
emerging economies are capital scarce and far away from their autarky steady state, their
growth rate is higher after financial integration. However, FDI scarcity is another kind of
capital scarcity for all countries. In such case, developed country suffers FDI scarcity while
emerging economies suffer capital scarcity and FDI scarcity. What’s more, all countries are
affected by two effects: capital scarcity effect and risk sharing effect, which will adjust FDI
and bond holding.
As for the bond, in line with the bond reversal in the benchmark model, we still observe
such a phenomenon. And the logic is the same: at the beginning, emerging economies
have to sell bond to get capital for development; but after that, as emerging economies are
getting more developed, risk sharing effect dominates and emerging economies begin to
buy bond to share risk while developed country is able to tolerate this risk. Our results
indicate that bond reversal happens very early that is 6 years after financial integration,
221
Figure 4.10 Dynamics under extended model 2
which is closer to the reality compared with the 11 years in our benchmark model. This
means our extended model is better than benchmark model in this area.
In summary, the dynamics under extended model are very close to under benchmark
model and the main difference is the bond reversal while in this area extended model is
closer to the reality.
From the theory, we expect that the welfare gains of financial integration under ex-
tended model is larger and the benefit from risk sharing effect is also larger. This is
because now there are three countries existing and developed country can adopt a better
global supply chains plan to allocate its FDI investment between two countries rather
than just one country. In such case, it can better hedge the risk from one specific country
and then the benefit from risk sharing effect should be larger. However, since in the
two-country model that is our benchmark model, developed country can also share risk
222
with one country rather than can’t share risk, this extra benefit from benchmark model
to extended model won’t be that large but should still be significant. And the following
analysis supports our expectation from the theory.
Table 4.13 Welfare gains of financial integration in extended model 1
Country D Country E1
Baseline (Case 1) 15.41% 29.97%
High risk aversion (Case 2) 8.58% 35.45%
Frictions (Case 3) 18.29% 27.62%
No capital scarcity 17.58% 26.68%
FDI only 21.09% 25.27%
Bond only -4.74% 3.89%
Note: the relative level is financial autarky level.
Table 4.13 summarizes the welfare gains of financial integration in different cases with
the relative level of financial autarky in extended model. Case 1 is a risky world; Case 2 is
similar to Case 1 while there is a higher risk aversion; Case 3 is a risky world with frictions
where capital controls and financial frictions exist; ”no capital scarcity” case is the case
where emerging economies start with the equal level of capital in steady state while risk is
asymmetric; ”FDI only” case is the case where there is only FDI under financial integration;
”bond only” case is the case where there is only bond under financial integration.
First, we can still find that the welfare gains of financial integration is super large if
all countries start from financial autarky to financial integration. This supports our main
findings again. Meanwhile, we find the welfare gains under extended model is larger than
under benchmark model. This is because now there are three countries and developed
country can use global supply chains to allocate resources to two countries rather than
one to better offset risk, which increase the welfare gains of developed country. In specific,
the welfare gains of developed country increase by 3.72% and about half of this increase
is due to the increase of risk sharing effect. In specific, from the difference of Case 1 and
Case 2, we can find that the risk sharing effect contributes 6.83%, which is significantly
larger than benchmark model by 1.78%. All these prove the significant effect of global
223
supply chains on risk sharing. What’s more, according to ”FDI only” and ”bond only” case,
the welfare gains are mostly from FDI rather than bond, which explains why we don’t find
large welfare gains of financial integration in theory since we don’t consider FDI in the
model before. And the contribution of FDI risk sharing effect is also reflected. In specific,
due to the contribution of the extra risk sharing effect via FDI and global supply chains,
the difference of ”FDI only” case and Case 1 has increased by 1.45%. Moreover, frictions
have an opposite effect and now this cost is larger than benchmark model, which fits the
expectation of other literature better and proves the advantages of extended model again.
As for capital scarcity effect, again this effect is relatively small for developed country and
emerging economies, which is consistent with our benchmark model. Finally, the effect of
bond is also similar to the benchmark model, and developed country has to pay interest
for its debt so that the welfare gains are negative.
Table 4.14 Welfare gains of financial integration in extended model 2
Country D Country E1
Baseline (Case 1) 8.78% 10.58%
High risk aversion (Case 2) 1.61% 1.71%
Frictions (Case 3) 7.91% 8.94%
No capital scarcity 5.61% 6.65%
FDI only 6.19% 12.71%
Bond only 6.71% 8.67%
Note: the relative level is initial level. We adopt the consumption in period 1
under each case to be their own initial levels.
Table 4.14 summarizes the welfare gains of financial integration in different cases with
the relative level of reality in extended model. The definition of six cases is same as in
Table 4.13. And the results of Table 4.14 is much closer to the reality.
We can still find that the welfare gains of financial integration are large that are a little
bit larger than benchmark model. And other statistics are similar to the benchmark model
so that we don’t discuss that much here. One thing needs to be paid special attention to is
that the difference of Case 1 and Case 2 are larger by about 2%. This means the effect of
224
global supply chains does contribute to the risk sharing of developed country in the initial
level.
In summary, we do find the welfare gains of developed country have increased and a
large part of this increase is due to the risk sharing effect via FDI and global supply chains.
This is in line with our expectation from the theory, which also highlights the benefit of
global supply chains to hedge risk from one country.
The welfare gains of financial integration are about 20% under financial autarky level
and we do admit this value is relatively high. There are three different effects in theory that
account for this large number: capital scarcity effect, risk sharing effect, and FDI scarcity
effect. And one of the advantages of our model is that by using this model, we can analyze
simultaneously the welfare gains from these three effects together with how they interact.
Under this circumstance, we can better estimate the welfare gains of financial integration.
In specific, capital scarcity effect is the effect where emerging economies that are lack of
capital, can benefit from capital inflows due to the relatively higher price of capital. And
the contribution of capital scarcity effect is about 3%, which is small. Risk sharing effect is
the effect where countries can benefit by sharing risk with other countries. This effect is
very important that contributes a lot to the final welfare gains. Since now we have added
the setting of global supply chains, the contribution of risk sharing effect to the welfare
gains is large, which is about 7%. And FDI scarcity effect is similar to capital scarcity effect
while now the capital is FDI. This effect is of course very important since before financial
integration there is no FDI existing while after financial integration, with the existence of
FDI, different countries can invest on other countries via FDI. On the one hand, they can
produce a new intermediate goods that can be used to produce final goods. On the other
hand, they can share risk with other countries via FDI and global supply chains. Moreover,
what needs to pay special attention to is that the existence of FDI contributes to both risk
sharing effect and FDI scarcity effect. In such case, we find the contribution of FDI scarcity
effect is about 10% that is very significant.
225
Another reason why the number of 20% is high is related to the choice of the relative
level, and we do admit there is limitation on the relative level that may make this number
much larger. The relative level we choose is the financial autarky level. From the theory,
the ideal way to compare the welfare gains of financial integration is to compare two
different situations that are financial autarky and financial integration. Of course, this
idea is absolutely correct. However, since now we assume there will be intermediate goods
produced abroad to be used to produce final goods, the introduction of the new goods will
lead to a significant increase on the final output since under financial autarky the foreign
intermediate goods don’t exist. In such case, if we still use financial autarky level, then
this number will be very large that can’t be avoided. Despite from the theory, this method
is reasonable while in the reality, this result is not reasonable. But since we are among
first few to consider the importance of FDI and global supply chains, the number of our
estimated welfare gains of financial integration will be larger than other estimations in
the past literature. One method to solve this problem is to use another more reasonable
relative level and actually we do introduce a new level that is the initial level where we
choose the consumption in period 1 to be the relative level of consumption. In such case,
we can partially avoid the large increase due to the introduction of new goods. We believe
this limitation is realistic that is very hard to avoid and the usage of the initial level is at
least a method to solve this issue. And by using the initial level, we find that the welfare
gains of financial integration are smaller to be about 10%, which is not that high compared
with other literature. However, since the choice of the initial level is not perfect, we should
be careful to interpret these numbers. Of course we are thinking about a better relative
level to better capture the welfare gains of financial integration, which may be discussed
in the future draft. But so far the best choice of the relative level we have now might be
the initial level.
To ensure our results are robust, here we also consider the four-country extended model
where there are one developed country 1 and three emerging economies 2, 3, and 4 of
226
same country size. And our results are that the welfare gains when there are four countries
are very close to when there are three countries, which proves that our main findings are
robust. The reason why the extra welfare gains are small is that the channel of global
supply chains has been created in three-country model and the change from three-country
to four-country doesn’t create a new channel but amplify this channel. In such case, the
increase is not very large while we can still capture this increase. However, we can find
that the risk sharing effect has increased, which is mainly from the global supply chains.
Of course, one can argue that if the relative country size among emerging economies
are not same, the welfare gains might be different. However, since the aggregate size of
emerging economies is equal to developed country, this difference is not large enough to
have a significant effect. To prove this point, here we assume in our three-country model,
the relative country size of two emerging economies 2 and 3 is 3:2 (60%:40%). Then we
can compare the difference of two situations to prove our results are robust. Our results
are that the welfare gains when the relative country size of two emerging economies is
different are similar to when the country size is equal. And the difference is very small,
which proves that our main findings are robust. The reason why the difference is small
is that the aggregate country size of emerging economies is same and internal difference
of emerging economies doesn’t have a large impact. However, the situation where two
emerging economies are of equal size is better for developed country to share risk via
FDI and global supply chains. In such case, the welfare gains have decreased and the risk
sharing effect has also decreased while such cut is small. And this logic can be understood
by simple math so that we don’t explain in detail. In summary, the change of the relative
country size of emerging economies is not significant to affect the welfare gains so that we
don’t need to worry about the impact of this factor.
We also conduct sensitive analysis in terms of the elasticity of substitution among
intermediate inputs
m
. Elasticity of substitution indicates the substitutability between
two different intermediate inputs within the aggregate production. And
m
measures the
227
substitutability between domestic and foreign intermediate inputs. We find that despite
we set
m
to be different, the difference is still very small. In such case, we are confident
that different value of elasticity of substitution among intermediate inputs
m
doesn’t have
a large impact on welfare gains and our main results are robust.
As for the sensitive analysis of other parameters, we have clarified these in the sensitive
analysis section mentioned before and can adopt our previous results since our extended
model is the extension of our benchmark model. In a word, we are confident that our main
results are robust.
4.10 Conclusion
The welfare gains of financial integration is always an important problem in international
finance. We understand the benefits of financial integration well enough in theory and
there are a lot of empirical researches on this topic. But from the literature, the results
are not convincing to show that financial integration is ideal as we expect and financial
integration does bring sizable benefits to emerging economies.
In this chapter, we investigate the welfare gains of financial integration. First we con-
duct empirical analysis to investigate the effect of FDI, equity, and debt market integration
on economic growth. We believe the reason why we can’t find strong evidence on the wel-
fare gains of financial integration might be the elusive relationship and poor econometric
methodology used in the previous literature. Under this circumstance, considering the
effect might be nonlinear and super complicated, an advanced econometric methodology:
the Panel Smooth Threshold Regression (PSTR) model is used. Two interpretations of PSTR
model are possible. On one hand, PSTR model can be thought of as a regime-switching
model that allows for a small number of extreme regimes associated with the extreme value
of a transition function. Meanwhile, different from the indicator function, the transition
from one regime to another is smooth. On the other hand, PSTR model can be thought to
228
allow for a continuum of regimes, each one being characterized by a different value of the
transition function. The logic is then similar to that developed in the standard univariate
time series STAR (Smooth Transition Autoregressive) model.
PSTR model allows parameters to vary across countries and times. In such case,
it provides a parametric approach of the cross-country heterogeneity and of the time
instability of the effects we want to investigate, since these parameters change smoothly
as a function of a threshold variable. In such case, PSTR model can overcome two major
critics on panel data models: non stationary and cross-section heterogeneity.
Our results indicate that there is a strong nonlinear relationship existing, which means
we can’t use linear regression model to estimate the effect. We also find strong threshold
effects in the relationship between FDI, equity, and debt market integration, and economic
growth. Under this circumstance, the aggregate effect of FDI, equity, and debt market
integration on economic growth is highly impacted by the threshold effects, which means
the effect is very complicated. And since the threshold effects rely on KAOPEN that
measures capital account openness, which is affected by multiple factors, strong threshold
effects mean that the aggregate effect of FDI, equity, and debt market integration on
economic growth is heterogeneous across countries and times. But generally speaking, we
are confident that the increase in FDI market integration can promote economic growth.
What’s more, the improvement of equity market integration will first promote and then
inhibit economic growth. Moreover, the increase in debt market integration can promote
economic growth but may face some negative effect. In such case, financial integration can
bring sizable benefits that may be mainly from FDI market integration. Finally, since the
slope parameter is not very large, the transition function is not very sharp. This means
that our regression equation cannot be reduced to the sum of a limited number of regimes
and we must use PSTR model instead of PTR model.
Then we construct a theoretical model trying to explain our empirical findings. From
the theoretical aspect, model in Coeurdacier et al. (2015) offers an important framework
229
to analyze the welfare gains from capital scarcity effect and risk sharing effect together
with how they interact. Considering the importance of FDI and frictions being removed,
we improve their model by adding FDI and frictions being removed to enrich the welfare
analysis of financial integration. In such case, welfare gains of financial integration can
be from capital scarcity effect, risk sharing effect, and FDI scarcity effect. Our results
indicate that the welfare gains of financial integration are large rather than small, which
are mainly from FDI rather than bond. This finding may explain why we can’t find
significant welfare gains of financial integration in theoretical literature, which is also
in line with our empirical findings. Meanwhile, the large welfare gains may be hidden
because of many possible reasons such as frictions and short observed period, which can
be explained as the threshold effects in our empirical results.
A key finding in this chapter is that FDI scarcity effect is super important for both
developed countries and emerging economies. This is because FDI gives both countries a
chance to invest directly on other country, which is a new channel to accumulate capital.
In such case, we should pay more attention to FDI in the literature of financial integration.
Meanwhile, we find that developed countries can benefit a lot from risk sharing effect but
emerging economies doesn’t benefit from this a lot. Finally, capital scarcity effect is small
for both countries.
Another key finding is that frictions can highly weaken the welfare gains of financial
integration especially the growth rate of capital and consumption. Frictions may not
change the aggregate welfare gains a lot while frictions can put off the welfare gains and
then affect our feeling of welfare gains especially in the early stage of financial integration.
Considering the short history of financial integration, frictions may be too strong now to
hide the large welfare gains of financial integration and we should wait for more evidence.
We also find an interesting thing that observation time matters a lot to the welfare
gains of financial integration. If we restrict our observation to a short period, then we
may find a much smaller benefit from financial integration. However, even until now, the
230
history of financial integration is not long enough. In such case, the short observation
time may cause us harder to find a significant benefit from financial integration especially
when we consider the effects of frictions. This means we should be more patient to witness
the large welfare gains.
Finally, we find a heterogeneous effect of financial integration across countries and
times that is more complicated than Coeurdacier et al. (2015), which is also confirmed
by our empirical results. The welfare gains of financial integration may vary with het-
erogeneous levels of capital scarcity, risk aversion, frictions, and available assets that
can be traded. Meanwhile, country size, FDI productivity, and observation time can also
have a significant effect. These can better explain at least partially why the empirical
literature has difficulty to identify a robust average benefit across different countries and
time periods. What’s more, these can broad the door opened by Coeurdacier et al. (2015)
for new empirical research on the growth effect of financial integration.
231
Chapter 5
Conclusion
The unprecedented global financial crisis and COVID-19 have caused huge damage to
the global economy. In order to support the economy, global central banks use both
conventional and unconventional monetary policy while the monetary policy room of
global central banks is getting smaller and smaller. Meanwhile, global central banks are
developing new policies such as macroprudential policy to respond to the next economic
recession. What’s more, other problems such as the damage of rapid international capital
flows arise during the two crises, making people wonder the future of globalization. In
specific, there are three problems that need us to pay special attention to. First, the global
benchmark interest rate reform from the old London Interbank Offered Rate (LIBOR)
to the new risk-free rates (RFRs) makes both the academia and the industry worried
whether this reform will affect monetary policy transmission. Second, facing the declining
monetary policy space, it is necessary to develop macroprudential policy while the effect
of macroprudential policy is not clear. Third, considering the huge impact of cross-border
capital flows during the crisis, the impact of financial integration on the economy still
needs to be clarified. Therefore, in this paper, we conduct research on monetary policy,
macroprudential policy, and financial integration, trying to analyze their effects and
provide references for their future development.
In chapter 2, we try to learn from China’s mature experience in developing benchmark
interest rates based on actual transactions. In the global financial market, the most widely
232
used benchmark interest rate is LIBOR. As LIBOR is about to exit, major developed
economies have identified the new benchmark interest rates: RFRs, which are generated
based on actual transactions and are different from the old benchmark interest rate
LIBOR. Our results show that overall, the transmission efficiency of OMO interest rate to
the interbank market is higher than that of the exchange market, and the transmission
efficiency to overnight products is significantly higher than that of 7-day products. From
the comparison of the two products of DR and SHIBOR, the transmission efficiency of
DR001 is slightly higher than that of SHIBOR ON and the transmission efficiency of
SHIBOR 1W is slightly higher than that of DR007 while there is not much difference
between them. But different from our initial expectation, the transmission efficiency of
DR007 to the bond market is significantly higher than our expectation. In terms of the
NCD yields of various maturities, the 1-year and 10-year government bond and CDB bond
yields, the yield curve of government bonds and CDB bonds, and AAA-rated and AA-rated
enterprise bond yields of various maturities, the overall transmission efficiency of DR007
is about 1-7% weaker than SHIBOR 1W. But on the three factors of the government bond
and CDB bond yield curve, the response to DR007 is very close to that to SHIBOR 1W.
Especially in the slope and level factor of the CDB bond yield curve, the response level to
DR007 is greater than that of SHIBOR 1W by about 1 -1.5%. This shows that DR007 is
slightly better than SHIBOR 1W in the transmission of the government bond and CDB
bond yield curve. In terms of all the bonds mentioned above, the explanatory power of
DR007 is significantly greater than that of SHIBOR 1W. These show that the transmission
effect of DR007 is better than SHIBOR 1W in some respects. What’s more, according to
the results of the bank-level micro-database, the transmission efficiency of SHIBOR 1W is
slightly higher than that of DR007 in the credit market. However, the gap is very small,
which is less than 1%. Meanwhile, the transmission efficiency of both DR007 and SHIBOR
1W in the lending interest rate is significantly higher than the deposit interest rate. But the
transmission efficiency of both DR007 and SHIBOR 1W in credit market is significantly
233
lower than that in bond market, reflecting the urgency of interest rate reform in credit
market.
From China’s experience, DR007 has performed very well in monetary policy transmis-
sion. Its transmission efficiency is only slightly weaker than SHIBOR 1W, and even better
than SHIBOR 1W in some areas. Therefore, DR007 has the power to replace SHIBOR
1W as the new benchmark interest rate. Of course, considering that other countries’
experience in benchmark interest rates based on actual transactions is far behind China,
the performance of the new benchmark interest rates RFRs may be weaker than China’s
DR007. But over time, RFRs will eventually become mature so that we still have reasons
to be optimistic about the future performance of RFRs. Considering the performance of
DR007 exceeding our expectation, we also believe that the process of replacing the old
benchmark interest rate LIBOR with the new benchmark interest rates RFRS may not lead
to a decline in the monetary policy transmission efficiency. In such case, we do not need to
worry too much about the negative impact of this reform on the market and policies.
In Chapter 3, based on Chinese characteristics, we construct two measures of Chinese
monetary policy and macroprudential policy. Our results suggest that there are strong
time-varying characteristics among Chinese monetary policy, macroprudential policy
and their targets. This finding illustrates the importance of the time-varying setting
and the importance of using advanced econometric methods to analyze the effect of
Chinese policies. We also find evidence that Chinese monetary policy can promote
economic growth, and its effectiveness is growing, especially after 2015. We also construct
a model consisting of state-owned enterprises (SOEs) sector and private-owned enterprises
(POEs) sector with directed lending to explain our findings. Moreover, we argue that
accommodative macroprudential policy can cause minor damage to economic growth in
the short term and such negative effects are increasing, which implies that policy makers
need to be more cautious in adjusting macroprudential policy to avoid side effects. Finally,
our results confirm that, so far, macroprudential policy has not been an independent
234
policy, which is consistent with the PBoC’s stated position. However, the independence is
stronger, especially after 2015, which is very useful for understanding the role of Chinese
macroprudential policy. We also investigate the policy effect on inflation, housing price,
bond price and equity price, and the results meet our expectation.
Of course, China’s economy is far from fully liberalized, and quantitative monetary
instruments are likely to continue to play an important role. In such case, Chinese
monetary policy will remain different from that of other western economies for a long time.
Nevertheless, our analysis shows that the PBoC is improving its policies. In specific, the
PBoC’s two-pillar regulation framework of monetary and macroprudential policy has been
effective in achieving its targets since these two policies have different effect on different
targets, and the effectiveness of both policies has continued to increase, especially after
2015, as the PBoC accelerates the pace of interest rate liberalization and policy reform.
However, this also poses a new challenge to the PBoC, making the PBoC’s policy decisions
more cautious.
In Chapter 4, we investigate the welfare gains of financial integration. The issue on
the welfare gains of financial integration is always an important problem in international
finance. We understand the benefits of financial integration well enough in theory and
there are a lot of empirical researches on this topic. But from the literature, the results
are not convincing to show that financial integration is ideal as we expect and financial
integration does bring sizable benefits to emerging economies. In specific, first we conduct
empirical analysis to investigate the effect of FDI, equity, and debt market integration on
economic growth. Our results indicate that there is a strong nonlinear relationship existing,
which means we can’t use linear regression model to estimate the effect. We also find strong
threshold effects in the relationship between FDI, equity, and debt market integration, and
economic growth. Under this circumstance, the aggregate effect of FDI, equity, and debt
market integration on economic growth is highly impacted by the threshold effects, which
means the effect is very complicated. And since the threshold effects rely on KAOPEN
235
that measures capital account openness, which is affected by multiple factors, strong
threshold effects mean that the aggregate effect of FDI, equity, and debt market integration
on economic growth is heterogeneous across countries and time. But generally speaking,
we are confident that the increase in FDI market integration can promote economic growth.
What’s more, the improvement of equity market integration will first promote and then
inhibit economic growth. Moreover, the increase in debt market integration can promote
economic growth but may face some negative effect. In such case, financial integration can
bring sizable benefits that may be mainly from FDI market integration. Finally, since the
slope parameter is not very large, the transition function is not very sharp. This means
that our regression equation cannot be reduced to the sum of a limited number of regimes
and we must use PSTR model instead of PTR model.
Then we construct a theoretical model trying to explain our empirical findings. From
the theoretical aspect, we find that FDI scarcity effect is super important for both developed
countries and emerging economies. This is because FDI gives both countries a chance to
invest directly on other country, which is a new channel to accumulate capital. Meanwhile,
we find that developed countries can benefit a lot from risk sharing effect but emerging
economies doesn’t benefit from this a lot. Finally, capital scarcity effect is small for
both countries. What’s more, frictions can highly weaken the welfare gains of financial
integration especially the growth rate of capital and consumption. Frictions may not
change the aggregate welfare gains a lot while frictions can put off the welfare gains and
then affect our feeling of welfare gains especially in the early stage of financial integration.
Moreover, observation time matters a lot to the welfare gains of financial integration. If we
restrict our observation to a short period, then we may find a much smaller benefit from
financial integration. In such case, the short observation time may cause us harder to find
a significant benefit from financial integration especially when we consider the effects of
frictions. This means we should be more patient to witness the large welfare gains. We
also find a heterogeneous effect of financial integration across countries and times that is
236
more complicated than Coeurdacier et al. (2015), which is also confirmed by our empirical
results.
In summary, DR007 has performed very well in monetary policy transmission. Its
transmission efficiency is only slightly weaker than SHIBOR 1W, and even better than
SHIBOR 1W in some areas. In such case, the benchmark interest rate reform may not lead
to a decline in the monetary policy transmission efficiency and we do not need to worry too
much about the negative impact. Meanwhile, the PBoC’s two-pillar regulation framework
of monetary and macroprudential policy has been effective in achieving its targets, and
the effectiveness of both policies has continued to increase, especially after 2015. However,
this also poses a new challenge to the PBoC, making the PBoC’s policy decisions more
cautious. In such case, we should be confident on the effect of macroprudential policy.
What’s more, the aggregate effect of FDI, equity, and debt market integration on economic
growth is highly impacted by the threshold effects. Financial integration can bring sizeable
benefits to both developed countries and emerging economies while this is mainly via FDI
scarcity effect. However, frictions and observation time can weaken the evidence on the
large welfare gains of financial integration, and we should be more patient to get more
evidence. In such case, we should support the financial integration and accelerate the
globalization process.
237
References
Aizenman, J., Jinjarak, Y., & Park, D. (2013). Capital flows and economic growth in the
era of financial integration and crisis, 1990–2010. Open Economies Review, 24(3),
371–396.
Alam, Z., Alter, M. A., Eiseman, J., Gelos, M. R., Kang, M. H., Narita, M. M., Nier, E., &
Wang, N. (2019). Digging deeper–evidence on the effects of macroprudential policies
from a new database. International Monetary Fund.
Albuquerque, R. (2003). The composition of international capital flows: Risk sharing
through foreign direct investment. Journal of International Economics, 61(2), 353–
383.
Almaqtari, F. A., Al-Homaidi, E. A., Tabash, M. I., & Farhan, N. H. (2019). The determinants
of profitability of indian commercial banks: A panel data approach. International
Journal of Finance & Economics, 24(1), 168–185.
Aviat, A., & Coeurdacier, N. (2007). The geography of trade in goods and asset holdings.
Journal of International Economics, 71(1), 22–51.
Bai, Y., & Zhang, J. (2012). Financial integration and international risk sharing. Journal of
International Economics, 86(1), 17–32.
Baldwin, R. E. (2012). Global supply chains: Why they emerged, why they matter, and
where they are going.
Bekaert, G., Harvey, C. R., & Lumsdaine, R. L. (2002). Dating the integration of world
equity markets. Journal of Financial Economics, 65(2), 203–247.
Berkelmans, L., Kelly, G., & Sadeghian, D. (2016). Chinese monetary policy and the
banking system. Journal of Asian Economics, 46, 38–55.
Bernanke, B. S., & Gertler, M. (1995). Inside the black box: The credit channel of monetary
policy transmission. Journal of Economic perspectives, 9(4), 27–48.
Bianchi, J., & Mendoza, E. G. (2018). Optimal time-consistent macroprudential policy.
Journal of Political Economy, 126(2), 588–634.
Bordo, M. D., Meissner, C. M., & Stuckler, D. (2010). Foreign currency debt, financial crises
and economic growth: A long-run view. Journal of international Money and Finance,
29(4), 642–665.
Buckley, P . J., Clegg, L. J., Cross, A., Liu, X., Voss, H., & Zheng, P . (2010). The determinants
of chinese outward foreign direct investment. Foreign direct investment, china and
the world economy (pp. 81–118). Springer.
Burdekin, R. C., & Siklos, P . L. (2008). What has driven chinese monetary policy since
1990? investigating the people’s bank’s policy rule. Journal of International Money
and Finance, 27(5), 847–859.
238
Burger, J. D., Warnock, F. E., & Warnock, V. C. (2017). The effects of us monetary policy on
emerging market economies’ sovereign and corporate bond markets (tech. rep.). National
Bureau of Economic Research.
Bussiere, M., & Fratzscher, M. (2008). Financial openness and growth: Short-run gain,
long-run pain? Review of International Economics, 16(1), 69–95.
Cevik, S., & Teksoz, K. (2013). Lost in transmission? the effectiveness of monetary policy
transmission channels in the gcc countries. Middle East Development Journal, 5(3),
1350018–1.
Chanda, A. (2005). The influence of capital controls on long run growth: Where and how
much? Journal of Development Economics, 77(2), 441–466.
Chang, C., Liu, Z., & Spiegel, M. M. (2015). Capital controls and optimal chinese monetary
policy. Journal of Monetary Economics, 74, 1–15.
Chen, K., Ren, J., & Zha, T. (2018). The nexus of monetary policy and shadow banking in
china. American Economic Review, 108(12), 3891–3936.
Chen, Y., & Huo, Z. (2009). A conjecture of chinese monetary policy rule: Evidence from
survey data, markov regime switching, and drifting coefficients. Annals of Economics
& Finance, 10(1).
Chinn, M. D., & Ito, H. (2006). What matters for financial development? capital controls,
institutions, and interactions. Journal of development economics, 81(1), 163–192.
Choong, C.-K., Baharumshah, A. Z., Yusop, Z., & Habibullah, M. S. (2010). Private capital
flows, stock market and economic growth in developed and developing countries:
A comparative analysis. Japan and the World Economy, 22(2), 107–117.
Coeurdacier, N., Rey, H., & Winant, P . (2015). Financial integration and growth in a risky
world (tech. rep.). National Bureau of Economic Research.
Colacito, R., & Croce, M. M. (2010). The short and long run benefits of financial integration.
American Economic Review, 100(2), 527–31.
Colletaz, G., & Hurlin, C. (2006). Threshold effects of the public capital productivity: An
international panel smooth transition approach.
Contessi, S., & Weinberger, A. (2009). Foreign direct investment, productivity, and country
growth: An overview. Federal Reserve Bank of St. Louis Review, 91(2), 61–78.
Cun, W., & Li, J. (2019). Sterilized intervention, bank credit, and chinese monetary policy.
USC-INET Research Paper, (19-10).
Deng, C., & Xi, X. (2014). The time-varying characteristic of chinese monetary policy
macro regulation effect. Jilin University Journal Social Sciences Edition, 54, 75–83.
Dickinson, D., & Jia, L. (2007). The real effects of monetary policy in china: An empirical
analysis. China Economic Review, 18(1), 87–111.
Diebold, F. X., & Li, C. (2006). Forecasting the term structure of government bond yields.
Journal of econometrics, 130(2), 337–364.
Duffie, D., & Stein, J. C. (2015). Reforming libor and other financial market benchmarks.
Journal of Economic Perspectives, 29(2), 191–212.
Durham, J. B. (2004). Absorptive capacity and the effects of foreign direct investment and
equity foreign portfolio investment on economic growth. European economic review,
48(2), 285–306.
Edison, H. J., Levine, R., Ricci, L., & Sløk, T. (2002). International financial integration and
economic growth. Journal of international money and finance, 21(6), 749–776.
239
Feng, G., & Serletis, A. (2010). Efficiency, technical change, and returns to scale in large us
banks: Panel data evidence from an output distance function satisfying theoretical
regularity. Journal of Banking & Finance, 34(1), 127–138.
Fernald, J. G., Spiegel, M. M., & Swanson, E. T. (2014). Monetary policy effectiveness in
china: Evidence from a favar model. Journal of International Money and Finance, 49,
83–103.
Fok, D., Van Dijk, D., & Franses, P . H. (2005). A multi-level panel star model for us
manufacturing sectors. Journal of Applied Econometrics, 20(6), 811–827.
Fouquau, J., Hurlin, C., & Rabaud, I. (2008). The feldstein–horioka puzzle: A panel smooth
transition regression approach. Economic Modelling, 25(2), 284–299.
Gao, B., Zhu, A., & Tan, X. (2019). Research on the influence of the development of
interbank business on operating stability of commercial banks ——the empirical
analysis based on micro-data of 16 listed banks in china. Nankai Economic Research,
(5), 78–97.
Geiger, M. (2006). Monetary policy in china (1994-2004): Targets, instruments and their
effectiveness.
Georgievska, L., Kabashi, R., Manova-Trajkovska, N., Mitreska, A., & Vaskov, M. (2011).
Determinants of lending interest rates and interest rates spreads. Bank of Greece,
Special Conference Paper, 9.
Gonzalez, A., Ter¨ asvirta, T., Van Dijk, D., & Yang, Y. (2017). Panel smooth transition regres-
sion models.
Gourinchas, P .-O., & Jeanne, O. (2006). The elusive gains from international financial
integration. The Review of Economic Studies, 73(3), 715–741.
Hansen, B. E. (1999). Threshold effects in non-dynamic panels: Estimation, testing, and
inference. Journal of econometrics, 93(2), 345–368.
He, D., & Pauwels, L. L. (2008). What prompts the people’s bank of china to change its
monetary policy stance? evidence from a discrete choice model. China & World
Economy, 16(6), 1–21.
He, Q., Leung, P .-H., & Chong, T. T.-L. (2013). Factor-augmented var analysis of the
monetary policy in china. China Economic Review, 25, 88–104.
He, X. (2018). An assessment of interest rate transmission mechanism in china. Tsinghua
University National Institute OF Financial Research Research Report.
Henry, P . B. (2003). Capital-account liberalization, the cost of capital, and economic growth.
American Economic Review, 93(2), 91–96.
Holmes, T. J., McGrattan, E. R., & Prescott, E. C. (2015). Quid pro quo: Technology capital
transfers for market access in china. The Review of Economic Studies, 82(3), 1154–
1193.
Huang, K., Yao, Q., & Li, C. (2019). Impacts of financial market shock on bank asset
allocation from the perspective of financial characteristics of banks. International
Journal of Financial Studies, 7(2), 29.
Illes, A., & Lombardi, M. J. (2013). Interest rate pass-through since the financial crisis. BIS
Quarterly Review, September.
Jawadi, F., Mallick, S. K., & Sousa, R. M. (2014). Nonlinear monetary policy reaction func-
tions in large emerging economies: The case of brazil and china. Applied Economics,
46(9), 973–984.
240
Jeanne, O., & Korinek, A. (2010). Managing credit booms and busts: A pigouvian taxation
approach (tech. rep.). National Bureau of Economic Research.
Jermann, U. J. (2019). Is sofr better than libor? Available at SSRN 3361942.
Jermann, U. J. (2020). Interest received by banks during the financial crisis: Libor vs
hypothetical sofr loans. Available at SSRN 3624908.
Ji, M., Zhang, X., Niu, M., & Ma, J. (2016). Transmission of monetary policy through the
banking system. People’s Bank of China Working Paper, (2016/4).
Jia, J., Guo, J., & Wang, Z. (2015). The fiscal-monetary policy mix and exchange rate stabil-
ity: A dynamic stochastic general equilibrium model with chinese characteristics.
Emerging Markets Finance and Trade, 51(4), 729–746.
Klein, M. W., & Olivei, G. P . (2008). Capital account liberalization, financial depth, and
economic growth. Journal of international money and finance, 27(6), 861–875.
Klingelh¨ ofer, J., & Sun, R. (2018). China’s regime-switching monetary policy. Economic
Modelling, 68, 32–40.
Klingelh¨ ofer, J., & Sun, R. (2019). Macroprudential policy, central banks and financial
stability: Evidence from china. Journal of International Money and Finance, 93, 19–41.
Koivu, T., Mehrotra, A. N., & Nuutilainen, R. (2008). Mccallum rule and chinese monetary
policy.
Kose, M. A., Prasad, E., Rogoff, K. S., & Wei, S.-J. (2006). Financial globalization: A reappraisal
(tech. rep.). National Bureau of Economic Research.
Kose, M. A., Prasad, E. S., & Terrones, M. E. (2009). Does openness to international financial
flows raise productivity growth? Journal of International Money and Finance, 28(4),
554–580.
Krolzig, H.-.-M., & Sserwanja, I. (2014). Corporate bond yields in the transmission mecha-
nism of monetary policy. University of kent at canterbury, 6, 15–21.
Liu, C., Song, P ., & Huang, B. (2019). The dynamic effectiveness of monetary policy in
china: Evidence from a tvp-sv-favar model. Applied Economics Letters, 26(17), 1402–
1410.
Liu, Z., Spiegel, M. M., & Zhang, J. (2020). Optimal capital account liberalization in china.
Ma, J., & Wang, H. (2014). Theoretical model of the transmission mechanism of the central
bank policy rate in china. Journal of Financial Research, 2014, 1–22.
Ma, J., Shi, K., Wang, H., & Wang, L. (2016). Dynamic research on interest rate transmission
mechanism. Journal of Financial Research, 427(1), 31–49.
Mbao, F. Z., Kapembwa, C., Mooka, O., Rasmussen, T., & Sichalwe, J. (2014). Determinants
of bank lending rates in zambia: A balance sheet approach. Bank of Zambia.
Mishra, P ., Montiel, P ., & Sengupta, R. (2016). Monetary transmission in developing
countries: Evidence from india. Monetary policy in india (pp. 59–110). Springer.
Mizen, P ., & Hoffman, B. (2004). Interest rate pass-through and monetary transmission:
Evidence from individual financial institutions’ retail rates. Economica, 71(1), 99–
123.
Nakajima, J. (2011). Time-varying parameter var model with stochastic volatility: An overview
of methodology and empirical applications (tech. rep.). Institute for Monetary and
Economic Studies, Bank of Japan.
Nanjunga, T., Ntsosa, M. M., & Motlaleng, G. (2016). Determinants of commercial banks’
interest rate spreads in botswana. Botswana Journal of Economics, 14(1), 1–15.
241
Obstfeld, M. (2009). International finance and growth in developing countries: What have
we learned? IMF staff papers, 56(1), 63–111.
Portes, R., & Rey, H. (2005). The determinants of cross-border equity flows. Journal of
international Economics, 65(2), 269–296.
Primiceri, G. E. (2005). Time varying structural vector autoregressions and monetary
policy. The Review of Economic Studies, 72(3), 821–852.
Quinn, D. P ., & Toyoda, A. M. (2008). Does capital account liberalization lead to growth?
The Review of Financial Studies, 21(3), 1403–1449.
Sanusi, A. R. (2010). Exchange rate pass-through to consumer prices in ghana: Evidence
from structural vector auto-regression. West African Journal of Monetary and Eco-
nomic Integration, 10(1).
Schrimpf, A., & Sushko, V. (2019). Beyond libor: A primer on the new benchmark rates.
BIS Quarterly Review March.
Schwalb, R. (2017). The effects of large scale asset purchases on corporate bond yields:
Drivers & channels.
Shu, C., & Ng, B. (2010). Monetary stance and policy objectives in china: A narrative
approach. HKMA China Economic Issues, 1(10), 1–40.
Sun, G., & Duan, Z. (2017). Research on the medium-term policy interest rate trans-
mission mechanism——based on the analysis of the two-sector decision model of
commercial banks. China Economic (Quarterly), (1), 349–370.
Sun, R. (2013). Does monetary policy matter in china? a narrative approach. China Economic
Review, 26, 56–74.
Sun, R. (2015). What measures chinese monetary policy? Journal of International Money
and Finance, 59, 263–286.
Sun, R. (2018). A narrative indicator of monetary conditions in china. International Journal
of Central Banking, 14(4), 1–42.
Sun, S., Gan, C., & Hu, B. (2012). Evaluating mccallum rule as a policy guideline for china.
Journal of the Asia Pacific Economy, 17(3), 527–545.
Tai, P . N., Sek, S. K., & Har, W. M. (2012). Interest rate pass-through and monetary
transmission in asia. International Journal of Economics and Finance, 4(2), 163–174.
Tang, Y., Luo, Y., Xiong, J., Zhao, F., & Zhang, Y.-C. (2013). Impact of monetary policy
changes on the chinese monetary and stock markets. Physica A: Statistical Mechanics
and its Applications, 392(19), 4435–4449.
Taylor-Brill, S. (2020). Libor manipulation and the transition to sofr. CMC Senior Theses.
Terrones, M. M., Kose, M. M. A., & Prasad, M. E. (2007). How does financial globalization
affect risk sharing? patterns and channels. International Monetary Fund.
Wang, M. B., & Sun, T. (2013). How effective are macroprudential policies in china? Interna-
tional Monetary Fund.
Xiong, W. (2012). Measuring the monetary policy stance of the people’s bank of china: An
ordered probit analysis. China Economic Review, 23(3), 512–533.
Yang, L., van Wijnbergen, S., Qi, X., & Yi, Y. (2019). Chinese shadow banking, financial
regulation and effectiveness of monetary policy. Pacific-Basin Finance Journal, 57,
101169.
Yu, X. Z. (2018). International trade, capital flows and economic growth in the world of financial
integration (tech. rep.). Unpublished manuscript.
242
Zhang, C., & Sun, Y. (2017). Confidence in chinese monetary policy. International Review
of Economics & Finance, 52, 212–221.
Zhang, J., Chen, T., Fan, F., & Wang, S. (2018). Empirical research on time-varying charac-
teristics and efficiency of the chinese economy and monetary policy: Evidence from
the mi-tvp-var model. Applied Economics, 50(33), 3596–3613.
Zhang, W. (2009). China’s monetary policy: Quantity versus price rules. Journal of Macroe-
conomics, 31(3), 473–484.
Zheng, T., Xia, W., & Huiming, G. (2012). Estimating forward-looking rules for china’s
monetary policy: A regime-switching perspective. China Economic Review, 23(1),
47–59.
Zhou, K., & Wong, M. C. (2008). The determinants of net interest margins of commercial
banks in mainland china. Emerging Markets Finance and Trade, 44(5), 41–53.
243
Appendices
A Equations of Equilibria
The full set of equilibrium conditions under financial autarky, full financial integration,
and limited financial integration in Chapter 4 are listed below. To make the equilibrium
clearer, we add more equations with (*) than needed, which are not used in actual coding.
Financial Autarky:
y
i;t
=A
i;t
k
it
A
i;t
=a
W;t
a
i;t
log(a
W;t+1
) = (1
W
)log(a
W;0
) +
W
log(a
W;t
) +
W;t
log(a
i;t+1
) = (1)log(a
i;0
) +log(a
i;t
) +
i;t
k
i;t+1
= (1)k
i;t
+k
i;t
[ikr
i;t
2
(ikr
i;t
)
2
]
w
i;t
= (1)y
1
i;t
r
i;t
k
i;t
=y
1
i;t
c
i;t
+i
i;t
=y
i;t
ikr
i;t
=
i
i;t
k
i;t
fai
i;t
= 1(ikr
i;t
)
1 =E
t
f(
c
i;t+1
c
i;t
)
[r
i;t+1
fai
i;t
+
fai
i;t
fai
i;t+1
(1 +ikr
i;t+1
2
(ikr
i;t
)
2
ikr
i;t+1
fai
i;t+1
)]g
244
Full Financial Integration:
y
1
i;t
=A
i;t
k
it
y
2
i;t
=A
i;t
A
f
(k
f
it
)
f
y
i;t
=y
1
i;t
+y
2
i;t
A
i;t
=a
W;t
a
i;t
log(a
W;t+1
) = (1
W
)log(a
W;0
) +
W
log(a
W;t
) +
W;t
log(a
i;t+1
) = (1)log(a
i;0
) +log(a
i;t
) +
i;t
k
i;t+1
= (1)k
i;t
+k
i;t
[ikr
i;t
2
(ikr
i;t
)
2
]
k
f
i;t+1
= (1)k
f
i;t
+k
f
i;t
[ikrf
i;t
2
(ikrf
i;t
)
2
]
w
i;t
= (1)y
1
i;t
r
i;t
k
i;t
=y
1
i;t
r
f
i;t
k
f
i;t
=
f
y
2
i;t
c
i;t
+i
i;t
+i
f
i
0
;t
=w
i;t
+r
i;t
k
i;t
+y
2
i
0
;t
+b
i;t1
b
i;t
p
t
y
D;t
+y
E;t
=c
D;t
+i
D;t
+i
f
E;t
+c
E;t
+i
E;t
+i
f
D;t
()
b
D;t
+b
E;t
= 0
p
t
=E
t
[(
c
i;t+1
c
i;t
)
]
ikr
i;t
=
i
i;t
k
i;t
ikrf
i;t
=
i
f
i;t
k
f
i;t
fai
i;t
= 1(ikr
i;t
)
faif
i;t
= 1(ikrf
i;t
)
1 =E
t
f(
c
i;t+1
c
i;t
)
[r
i;t+1
fai
i;t
+
fai
i;t
fai
i;t+1
(1 +ikr
i;t+1
2
(ikr
i;t
)
2
ikr
i;t+1
fai
i;t+1
)]g
1 =E
t
f(
c
i
0
;t+1
c
i
0
;t
)
[r
f
i;t+1
faif
i;t
+
faif
i;t
faif
i;t+1
(1 +ikrf
i;t+1
2
(ikrf
i;t
)
2
ikrf
i;t+1
faif
i;t+1
)]g
245
Limited Financial Integration:
y
1
i;t
=A
i;t
k
it
y
2
i;t
=A
i;t
A
f
(k
f
it
)
f
y
i;t
=y
1
i;t
+y
2
i;t
A
i;t
=a
W;t
a
i;t
log(a
W;t+1
) = (1
W
)log(a
W;0
) +
W
log(a
W;t
) +
W;t
log(a
i;t+1
) = (1)log(a
i;0
) +log(a
i;t
) +
i;t
k
i;t+1
= (1)k
i;t
+k
i;t
[ikr
i;t
2
(ikr
i;t
)
2
]
k
f
i;t+1
= (1)k
f
i;t
+k
f
i;t
[ikrf
i;t
2
(ikrf
i;t
)
2
]
w
i;t
= (1)y
1
i;t
r
i;t
k
i;t
=y
1
i;t
w
i;t
= (1
f
)y
2
i;t
r
f
i;t
k
f
i;t
=
f
y
2
i;t
c
i;t
+i
i;t
=w
i;t
+r
i;t
k
i;t
+y
2
i
0
;t
+b
i;t1
(1 +
t
)b
i;t
p
t
(1 +
t
)i
f
i
0
;t
y
D;t
+y
E;t
=c
D;t
+i
D;t
+i
f
E;t
+c
E;t
+i
E;t
+i
f
D;t
()
b
D;t
+b
E;t
= 0
(1 +
t
)p
t
=E
t
[(
c
i;t+1
c
i;t
)
] +
c
i;t
ikr
i;t
=
i
i;t
k
i;t
ikrf
i;t
=
i
f
i;t
k
f
i;t
fai
i;t
= 1(ikr
i;t
)
faif
i;t
= 1(ikrf
i;t
)
1 =E
t
f(
c
i;t+1
c
i;t
)
[r
i;t+1
fai
i;t
+
fai
i;t
fai
i;t+1
(1 +ikr
i;t+1
2
(ikr
i;t
)
2
ikr
i;t+1
fai
i;t+1
)]g
1 +
t
=E
t
f(
c
i
0
;t+1
c
i
0
;t
)
[r
f
i;t+1
faif
i;t
+
faif
i;t
faif
i;t+1
(1 +ikrf
i;t+1
2
(ikrf
i;t
)
2
ikrf
i;t+1
faif
i;t+1
+
t
c
i;t
)]g
b
i;t+1
t
q
i;t
k
i;t
0 =
c
i;t
(b
i;t+1
+
t
q
t
k
i;t
)
q
i;t
=
1
faif
i;t
246
B Country List
B.1 Country List for Empirical Analysis
Here we list the countries in our empirical analysis in Chapter 4. The full sample of 70
countries is divided into two groups: advanced economies and developing economies:
All Countries(70)
Argentina, Australia, Austria, Bahrain, Bangladesh, Belgium, Bolivia, Brazil, Bulgaria,
Canada, Chile, China, Colombia, Costa Rica, Croatia, Cyprus, Czech Republic, Denmark,
Dominican Republic, Ecuador, Egypt, Finland, France, Germany, Greece, Guatemala,
Hong Kong SAR, Hungary, Iceland, India, Indonesia, Ireland, Israel, Italy, Japan, Jordan,
Kazakhstan, Korea, Kuwait, Latvia, Lithuania, Luxembourg, Malta, Mauritius, Mexico,
Morocco, Netherlands, New Zealand, Nigeria, Pakistan, Panama, Peru, Philippines, Poland,
Portugal, Romania, Russian, Singapore, Slovak Republic, Slovenia, South Africa, Spain,
Sweden, Switzerland, Thailand, Turkey, Ukraine, United Kingdom, United States, Uruguay
Advanced Economies(33)
Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark, Finland,
France, Germany, Greece, Hong Kong SAR, Iceland, Ireland, Israel, Italy, Japan, Korea,
Latvia, Lithuania, Luxembourg, Malta, Netherlands, New Zealand, Portugal, Singapore,
Slovak Republic, Slovenia, Spain, Sweden, Switzerland, United Kingdom, United States
Developing Economies(37)
Argentina, Bahrain, Bangladesh, Bolivia, Brazil, Bulgaria, Chile, China, Colombia,
Costa Rica, Croatia, Dominican Republic, Ecuador, Egypt, Guatemala, Hungary, India,
Indonesia, Jordan, Kazakhstan, Kuwait, Mauritius, Mexico, Morocco, Nigeria, Pakistan,
Panama, Peru, Philippines, Poland, Romania, Russian, South Africa, Thailand, Turkey,
Ukraine, Uruguay
247
B.2 Country List for Theoretical Analysis
In the theoretical part in Chapter 4, we follow the country selection in Coeurdacier et al.,
2015. Then we can use their results to decide country size and initial capital scarcity.
Sample of countries
15 always financially opened developed countries and 40 emerging economies. Emerg-
ing economies do not include countries from Central and Eastern Europe due to lack of
data before 1990.
Developed countries
Australia, Austria, Canada, Denmark, Finland, France, Germany, Ireland, Japan, Italy,
Netherlands, Sweden, Switzerland, United Kingdom, United States.
Emerging economies
Southern Europe: Greece, Israel, Malta, Portugal, Spain, Turkey.
Latin America: Argentina, Brazil, Chile, Colombia, Ecuador, Jamaica, Mexico, Peru,
Trinidad and Tobago, Venezuela.
Asia: Bangladesh, China, India, Indonesia, Malaysia, Pakistan, Philippines, South
Korea, Sri Lanka, Thailand.
Middle-East: Egypt, Jordan, Oman, Saudi Arabia.
Africa: Botswana, Ghana, Ivory Coast, Kenya, Mauritius, Morocco, Nigeria, Tunisia,
South Africa, Tunisia, Zimbabwe.
248
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Essays on monetary policy and international spillovers
PDF
Emerging market financial crises, investors and monetary policy
PDF
Early-warning systems for crisis risk
PDF
Essays on firm investment, innovation and productivity
PDF
Three essays in international macroeconomics and finance
PDF
Mercantilism and marketization? Analysis of China's reserve accumulation and changes in exchange rate regime using the monetary policy trilemma
PDF
Innovation: financial and economics considerations
PDF
Monetary policy and the term structure of interest rates
PDF
Economic policy-making in a changing environment: the politics of success in Peru vis-à-vis China
PDF
Essays on sovereign debt
PDF
Essays on Japanese macroeconomy
PDF
Two essays on financial crisis and banking
PDF
Essays on interest rate determination in open economies
PDF
Essays on econometrics analysis of panel data models
PDF
Essays on information and financial economics
PDF
Essays on business cycle volatility and global trade
PDF
Essays on health insurance programs and policies
PDF
Three essays on the identification and estimation of structural economic models
PDF
Essays on government intervention in financial crises
PDF
Essays on high-dimensional econometric models
Asset Metadata
Creator
Yu, Zhou
(author)
Core Title
Essay on monetary policy, macroprudential policy, and financial integration
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
03/25/2021
Defense Date
03/04/2021
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
financial integration,macroprudential policy,monetary policy,OAI-PMH Harvest
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Aizenman, Joshua (
committee chair
), Betts, Caroline (
committee chair
), Ferson, Wayne (
committee member
)
Creator Email
xanderyu1995@gmail.com,zyu423@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-431851
Unique identifier
UC11668422
Identifier
etd-YuZhou-9357.pdf (filename),usctheses-c89-431851 (legacy record id)
Legacy Identifier
etd-YuZhou-9357.pdf
Dmrecord
431851
Document Type
Dissertation
Rights
Yu, Zhou
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
financial integration
macroprudential policy
monetary policy