Univ ersity of Massachusetts Amherst Univ ersity of Massachusetts Amherst [604392]

Univ ersity of Massachusetts Amherst Univ ersity of Massachusetts Amherst
ScholarW orks@UMass Amherst ScholarW orks@UMass Amherst
Doct oral Disser tations Disser tations and Theses
2019
Dimensions of US Global Financial P ower: Essa ys on Financial Dimensions of US Global Financial P ower: Essa ys on Financial
Sanctions, Global Imbalances, and So vereign Default Sanctions, Global Imbalances, and So vereign Default
Mariam Majd
Follow this and additional works at: https:/ /scholar works.umass.edu/disser tations_2
Part of the International E conomics Commons
Recommended Citation Recommended Citation
Majd, Mariam, "Dimensions of US Global Financial P ower: Essa ys on Financial Sanctions, Global
Imbalances, and So vereign Default" (2019). Doct oral Disser tations . 1793.
https:/ /scholar works.umass.edu/disser tations_2/1793
This Open Access Disser tation is br ought t o you for fr ee and open access b y the Disser tations and Theses at
ScholarW orks@UMass Amherst. It has been accepted for inclusion in Doct oral Disser tations b y an authoriz ed
administr ator of ScholarW orks@UMass Amherst. F or mor e information, please contact
scholar works@libr ary.umass.edu .

DIMENSIONS OF US GLOBAL FINANCIAL POWER:
ESSAYS ON FINANCIAL SANCTIONS, GLOBAL IMBALANCES, AND
SOVEREIGN DEFAULT
A Dissertation Presented
by
MARIAM MAJD
Submitted to the Graduate School of the
University of Massachusetts Amherst in partial fulllment
of the requirements for the degree of
DOCTOR OF PHILOSOPHY
September 2019
Department of Economics

c
Copyright by Mariam Majd 2019
All Rights Reserved

DIMENSIONS OF US GLOBAL FINANCIAL POWER:
ESSAYS ON FINANCIAL SANCTIONS, GLOBAL IMBALANCES, AND
SOVEREIGN DEFAULT
A Dissertation Presented
by
MARIAM MAJD
Approved as to style and content by:
Gerald Epstein, Chair
Michael Ash, Member
Douglas Cliggott, Member
Bernard Morzuch
Resource Economics, Member
L eonce Ndikumana, Department Head
Economics

DEDICATION
For Je
iv

ACKNOWLEDGEMENTS
This dissertation has beneted from much unwavering support. I am deeply grateful to
the Chair of my Committee, Gerald Epstein, for his consistent and constructive feedback,
patient mentoring, and enthusiastic interest in this dissertation project. The global nancial
arena is not easily corralled into understanding. But, through my many discussions with
Jerry and study of his work, I was able to gain a steady glimpse into a dynamic that will
undoubtedly hold my interest for some time. I am a better researcher for his ability to ask
questions that gently reveal gaps in thinking, and my future work will surely be a re
ection
of his teaching.
I am also deeply indebted to my remaining Committee members. Since the coursework
phase of my graduate studies, Michael Ash has not only consistently encouraged me to
improve my theoretical arguments and empirical methods but also patiently weathered the
storm that was (is) my incessant questioning. I am a better researcher for his ability to
explain empirical methods intuitively, and I am a better teacher for realizing through his
patience how improvement of the student: [anonimizat], so generously shared, not only regularly captured my fascination but also cultivated
my own understanding to the benet of this project. This dissertation is also much improved
owing to the generosity with which Bernard Morzuch gave of his time and expertise. Through
v

his careful critique and detailed instruction on how to improve, my methodology and writing
gained in its precision. I am exceedingly grateful to him for sharpening the tools at my
disposal as an economist.
Finally, it would be remiss of me if I did not thank, along with my Committee, my
professor of economics during my undergraduate studies, G. Reza Ghorashi, who kindly
introduced me to the topic of my third essay|nancial sanctions on Iran|and provided
helpful guidance in designing my methodology.
In addition to my Committee, I am enormously thankful to my colleagues in the Depart-
ment of Economics at UMass-Amherst who sel
essly gave not only of their friendship and
humor but also of their expertise to make my experience in graduate school more pleasant. I
am especially grateful to Amal Ahmad; Jackson Allison and his partner, Savannah; Khwlah
Almutair and her daughter, Whateen; Ben Chalmers and his wife, Sally; Raphael Gouvea
and his wife, Mariana; Osman Keshawarz; Sai Madhurika Mamunuru; Carly McCann; and
Amanda Page-Hoongrajok and her partner, Kylie. It would be dicult for me to overstate
the degree to which any success I enjoyed in graduate school is owing to their eorts.
I am also enormously grateful to my professors at what was formerly The Richard Stock-
ton College of NJ, many of whom continued to guide me throughout my graduate studies.
In particular, I thank Alan Arcuri, William Daly, Deborah Figart, Patrick Hossay, Ellen
Mutari, Judith Vogel, and Linda Wharton. I also especially thank Anne F. Pomeroy to
whom I owe an especially large intellectual debt.
I am grateful to my community in Southern New Jersey who eortlessly provided a
laugh, shrug, and pat on the back when that sequence turned out to be the best answer. I
am especially thankful to my neighbors, the McClay and Morris families.
I am indebted beyond measure to my family. My father was the rst theorist of global
politics with whom I had extended conversations, and the logic underpinning my thinking
vi

is tighter owing to our discussions. My mother rst demonstrated the joy in being curious
and nurtured it in me. My excitement to learn is owing to her example. My four siblings|
Mohammad, Ali, Ibrahim, and Jamie|trained me in resilience and continuously gave of their
expertise to strengthen my weaknesses. My niece, Amelia, repeatedly narrowed my focus
and renewed my determination. Her joyful curiosity is a consistent reminder and motivation.
Finally, this dissertation is dedicated to my husband, Je, who inspires with quiet kind-
ness and humble strength. Without him, the pursuit manifested in the following pages simply
would not exist.
vii

ABSTRACT
DIMENSIONS OF US GLOBAL FINANCIAL POWER:
ESSAYS ON FINANCIAL SANCTIONS, GLOBAL IMBALANCES, AND SOVEREIGN
DEFAULT
SEPTEMBER 2019
MARIAM MAJD
B.A., THE RICHARD STOCKTON COLLEGE OF NEW JERSEY
M.A., UNIVERSITY OF MASSACHUSETTS AMHERST
Ph.D., UNIVERSITY OF MASSACHUSETTS AMHERST
Directed by: Professor Gerald Epstein
This dissertation examines how U.S. capabilities in the global nancial arena enable it to
aect outcomes to its advantage. The rst essay presents theoretical support for the hypoth-
esis that holdings of U.S. sovereign debt collateralize public and private dollar borrowing in
developing and emerging market economies, thus enabling the U.S. to receive continued easy
nancing despite declining economic fundamentals. The consensus view|what we refer to as
theSafe Assets theory |contends that what motivates continued investment in US sovereign
debt despite its drawbacks is its status as a safe asset : an asset that is strongly demanded
during and following adverse macroeconomic shocks. Though a safe asset is primarily iden-
tied by virtue of its eect rather than by underlying determinants of safeness, the former
is understood to be motivated by perceptions of the latter. Thus, the Safe Assets theory
viii

argues that perceptions of a country's economic health buer against consequences of its
decline.
While we do not argue against the notion that perception can play an important role
in sustaining demand for US sovereign debt, we draw on the well-developed sovereign debt
literature and argue that perception is unlikely to act alone. Namely, for lending to a
sovereign to occur, the structure of the relation between lender and sovereign borrower must
be such that it facilitates endogenous enforcement of the lending contract. Indeed, the Safe
Assets theory also concedes as much and assumes that if the US were to default on its
sovereign debt obligations, it would suer a large reputational cost levied by its lenders.
We argue this assumption is unwarranted, however. To credibly threaten or levy such a
cost necessarily requires that a lender possess greater capabilities than a borrower (i.e., the
lender must be strong relative to the borrower.) It is therefore implausible to imagine that
developing and emerging market economy lenders could credibly threaten or impose a cost
onto the United States large enough to deter the latter from defaulting.
We argue that persistent demand for US Treasury securities from developing and emerg-
ing market economies is owing to its role as collateral for dollar credit. That is, public lending
enables private and public borrowing. Specically, we argue that the rapid growth in dollar
credit to developing and emerging market economies leads to unfavorable investor percep-
tions of country volatility and increases the likelihood that in
ows will reverse. Against this
prospect, developing and emerging market economies are incentivized to oer their lenders
collateral against the lending contract to secure continued access to dollar funds. US Trea-
sury securities are an ideal form of collateral to dollar lenders for three main reasons: rst,
their scale of issuance enables lenders to require as collateral an asset large enough to dis-
incentivize default; second, as a slightly less liquid form of a country's savings, US Treasury
securities enable the lender to seize the means by which a defaulted borrower would secure
ix

another lending contract; third, because custodianship of US Treasury securities largely lies
with the US, the lender is capable of costlessly retaliating against a defaulting borrower.
The second essay empirically tests the theory presented in Chapter 1 that public lend-
ing enables public and private borrowing for developing and emerging market economies.
Considering that most foreign holdings of US Treasury securities are in long-term securities
held by foreign ocial institutions, we model the decision to hold long-term US Treasury
securities as one to hold dollars in foreign exchange reserves, allowing us to draw from well-
developed models of the latter decision-making process. We use a rst-dierence estimator
to control for country-specic factors aecting both US Treasury security holdings and out-
standing dollar credit in a panel of thirteen developing and emerging market economies. To
incorporate a persistence eect of US Treasury security holdings, we introduce as a regressor
the dependent variable lagged by one period and estimate the regression equation using an
instrumental variable method.
We provide evidence that, indeed, a statistically signicant relationship exists between
a country's ocial holdings of US Treasury securities and its level of outstanding dollar
credit. Our results demonstrate that even after controlling for a persistence eect (i.e.,
inertia) in US Treasury security holdings, increases in outstanding dollar credit lead to a
statistically signicant ( p<0:01) increase in holdings of US Treasury securities. Specically,
the estimated increase in US Treasury security holdings resulting from a $1 billion increase
in outstanding dollar credit is $0.11 billion, all other factors held constant. Our result
is robust to alternative denitions of our control variables and to the removal of outliers.
Namely, our result is not driven by China's disproportionate holdings of both US Treasury
securities and outstanding dollar credit. In fact, when China is removed from the sample
and our empirical model estimated again, the eect of outstanding dollar credit on holdings
of US Treasury securities increases in statistical signicance ( p < 0:001) and magnitude.
x

Specically, when China is removed from the sample, a $1 billion increase in outstanding
dollar credit is associated with a $0.18 billion increase in US Treasury securities, all other
factors held constant.
Finally, the third essay of this dissertation examines the United States' unique and rela-
tively recent ability to wield access to global nancial networks as a distinctly eective sanc-
tioning tool. Through a review of the literature on sanctions, we highlight the similarities
between this latest form of nancial sanction and its heavy-handed forerunner, comprehen-
sive trade sanctions. We specically trace the evolution of US command over the Society for
Worldwide Interbank Financial Telecommunication (SWIFT) and its subsequent removal of
the Islamic Republic of Iran from the platform in 2012 (hereafter referred to as the SWIFT
sanction ). To evaluate the impact of this new category of sanction, we utilize quarterly data
on Iran's real GDP during the period 1988-2016 and employ a time-series forecasting tech-
nique to measure the cost of the SWIFT sanction to Iran's real GDP, where cost is measured
by the dierence between forecasted and actual real GDP.
Results generated from estimating a seasonal autoregressive integrated moving average
(ARIMA) model indicate that the impact to Iran's real GDP of the SWIFT sanction is
sizeable. Specically, the average quarterly cost of the SWIFT sanction to Iran's real GDP
is approximately $204.3 billion (PPP-adjusted, 2015 international dollars), or 14.7% and
13.8% of Iran's average quarterly actual and forecasted GDP, respectively.
xi

TABLE OF CONTENTS
Page
ACKNOWLEDGEMENTS : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : v
ABSTRACT : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : viii
LIST OF TABLES : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : xiv
LIST OF FIGURES : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : xv
CHAPTER
1. WHAT EXPLAINS PERSISTENT FOREIGN DEMAND FOR US SOVEREIGN
DEBT? A THEORETICAL ASSESSMENT : : : : : : : : : : : : : : : : : : : : : 1
1.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1
1.2 The Literature on Global Imbalances : : : : : : : : : : : : : : : : : : : 8
1.3 The Safe Assets Theory : : : : : : : : : : : : : : : : : : : : : : : : : : 14
1.3.1 A Critique of the Safe Assets Theory : : : : : : : : : : : : : : : 18
1.4 US Treasury Securities as Collateral for Private Dollar Borrowing : : : 25
1.5 Concluding Remarks : : : : : : : : : : : : : : : : : : : : : : : : : : : : 28
2. LENDING TO BORROW: US SOVEREIGN DEBT AS COLLATERAL FOR
DOLLAR CREDIT : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 30
2.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 30
2.2 Determinants of Demand for US Treasury Securities : : : : : : : : : : 38
2.3 Data and Descriptive Statistics : : : : : : : : : : : : : : : : : : : : : : 44
2.3.1 Dependent Variable: Foreign Holdings of Long-term US Treasury
Securities ( USTS it): : : : : : : : : : : : : : : : : : : : : : : : : : : 44
2.3.2 Independent Variable of Interest: Outstanding Dollar Credit
(CRED it): : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 45
2.3.3 Independent Control Variables : : : : : : : : : : : : : : : : : : 46
2.3.3.1 Relative Financial Depth of the US ( DEPTH t): : : : : 46
2.3.3.2 Stability ( EXCH t): : : : : : : : : : : : : : : : : : : : 48
2.3.3.3 Network Eects ( NETW t): : : : : : : : : : : : : : : : 49
2.3.3.4 Inertia ( USTS i;t1): : : : : : : : : : : : : : : : : : : : 50
2.3.3.5 Yield( YIELD t): : : : : : : : : : : : : : : : : : : : : : 50
2.3.4 Summary : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 51
2.4 Econometric Model : : : : : : : : : : : : : : : : : : : : : : : : : : : : 53
xii

2.5 Main Results : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 58
2.6 Robustness Checks : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 67
2.7 Concluding Remarks : : : : : : : : : : : : : : : : : : : : : : : : : : : : 71
3. THE COST OF A SWIFT KICK: ESTIMATING THE COSTS OF FINAN-
CIAL SANCTIONS ON IRAN : : : : : : : : : : : : : : : : : : : : : : : : : : : : 75
3.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 75
3.2 The Development of Financial Sanctions : : : : : : : : : : : : : : : : : 77
3.3 The Evolution of Targeted Financial Sanctions : : : : : : : : : : : : : 80
3.3.1 Targeted Financial Sanctions in Iran : : : : : : : : : : : : : : : 83
3.4 Empirical Analysis : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 88
3.4.1 Causality Test : : : : : : : : : : : : : : : : : : : : : : : : : : : 88
3.4.2 Univariate Time-series Forecasting Method : : : : : : : : : : : : 93
3.4.3 Discussion : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 100
3.5 Concluding Remarks : : : : : : : : : : : : : : : : : : : : : : : : : : : : 101
APPENDICES
A. ADDITIONAL TABLES AND FIGURES FOR CHAPTER 2 : : : : : : : : : 103
B. ADDITIONAL TABLES AND FIGURES FOR CHAPTER 3 : : : : : : : : : 109
BIBLIOGRAPHY : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 111
xiii

LIST OF TABLES
Table Page
1 Descriptive Statistics : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 52
2 Public versus Private Borrowers and Holders : : : : : : : : : : : : : : : : : : : 54
3 Panel unit root tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 56
4 Regression Results : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 59
5 Panel Granger Causality Test : : : : : : : : : : : : : : : : : : : : : : : : : : : 66
6 Regression Results – Alternative Measure of Financial Depth (M3/GDP) : : : 70
7 Descriptive Statistics : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 89
8 Robustness Time-Series Regression : : : : : : : : : : : : : : : : : : : : : : : : 92
9 ADF Test : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 97
10 Forecast Summary : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 98
11 Forecast and Cost Estimates : : : : : : : : : : : : : : : : : : : : : : : : : : : : 100
12 Data and Sources : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 104
13 Country Codes : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 105
14 Correlation Matrix : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 105
15 Variance In
ation Factors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 106
16 First Stage for
USTS i;t1: : : : : : : : : : : : : : : : : : : : : : : : : : : : : 107
17 Correlation Matrix- Alternative Measure of Financial Depth : : : : : : : : : : 108
xiv

LIST OF FIGURES
Figure Page
1 Foreign Holdings of US Long-term Treasury Securities 1984-2016 : : : : : : : : 2
2 Foreign Ocial Holdings of US Treasury Securities : : : : : : : : : : : : : : : 3
3 Federal Debt, the Real Yield and the US Dollar Index : : : : : : : : : : : : : : 5
4 Holdings of Long-term US Treasury Securities by Indicator and Country : : : 24
5 US Treasury Security Holdings and Outstanding Dollar Credit 2003-2016 : : : 34
6 Real GDP for Iran 1988-2016 : : : : : : : : : : : : : : : : : : : : : : : : : : : 84
7 GDP Comparison : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 86
8 Confounders : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 90
9 Seasonal Component of Iran's GDP : : : : : : : : : : : : : : : : : : : : : : : : 95
10 Forecast of Iran's real GDP : : : : : : : : : : : : : : : : : : : : : : : : : : : : 99
11 ACF and PACF Patterns of Transformed GDP for Iran : : : : : : : : : : : : : 110
xv

C H A P T E R 1
WHAT EXPLAINS PERSISTENT FOREIGN DEMAND FOR US
SOVEREIGN DEBT? A THEORETICAL ASSESSMENT
1.1 Introduction
The movements of global capital
ows have been the subject of much debate in large part
because of their selective and persistent
ow toward long-term US sovereign debt. Figure
1 illustrates the growth of monthly foreign holdings of long-term US Treasury securities
during the period 1984-2016. The gure demonstrates that foreign holdings of long-term US
Treasury securities has grown rapidly since 2002 and at an increasing rate just prior to the
2008 global nancial crisis.
Largely behind this in
ow are foreign ocial institutions in emerging market and devel-
oping economies who, accordingly, constitute the largest proportion of investors in long-term
US Treasury securities. Figure 2 shows foreign holdings of US Treasury securities by holder
(private or foreign ocial) and type (long or short) during the period for which this data is
available.
The gure demonstrates that foreign investors hold long-term Treasury securities in larger
proportion than short-term Treasury securities; further, foreign ocial institutions hold the
largest proportion of long-term Treasury securities held by foreigners, though foreign private
1

Figure 1: Foreign Holdings of US Long-term Treasury Securities 1984-2016
020004000
1990 2000 2010USD billion
Notes : Data on holdings of long-term US Treasury securities is adjusted for valuation eects. Data is from
Bertaut and Judson (2014).
2

Figure 2: Foreign Ocial Holdings of US Treasury Securities
0200040006000
2012 2014 2016 2018USD billion
Foreign Private Holdings
of Long−term TSecuritiesForeign Official Holdings of Long−term TSecuritiesForeign Official Holdings of Short−term TSecuritiesForeign Private Holdings of Short−term TSecurities
Source: Treasury International Capital Reporting System
Notes : Data is from the Treasury International Capital Reporting System.
3

holdings have grown since approximately 2013.
Interestingly, demand for US sovereign debt has persisted despite its naturally accompa-
nying corollary: an increasing debt to gross domestic product (GDP) ratio. Indeed, by 2006,
the US current account decit had reached an unprecedented level of 6% of GDP (Gourin-
chas and Rey, 2014). Continued funding of the US to and past this point is puzzling since
high debt to GDP ratios are conventionally understood to work against long-term prospects
for economic growth and should therefore disincentivize investment in a country's public
debt (Reinhart and Rogo, 2010; Cecchetti, Mohanty, and Zampolli, 2011; Alfaro, Kalemli-
Ozcan, & Volosovych, V., 2014; He, Krishnamurthy and Milbradt, 2016). More striking,
demand for US debt has persisted despite two additional forces that typically operate to
reduce it: declining yields and dollar depreciation.
Figure 3 shows federal debt held by foreign and international investors, the real yield on
10-year Treasury securities, and the US dollar index during the period 2003-2018.
The gure shows that during the period 2008-2012 the real yield on 10-year Treasury
securities dropped from a high of approximately 2.6% to a low of -0.76% while the level of
federal debt held by foreign ocial and international investors steadily rose. This means
that demand for US debt persisted even as investments yielded increasingly less of a real
return and actually imposed a cost to investors. Additionally, during the period 2009-2011
the real value of the dollar was falling against other major advanced economy currencies,
meaning holders could reasonably expect to be repaid in currency worth relatively less than
it had been at the time the security was purchased.
According to conventional economic theory on capital
ows and sovereign debt, none
of what has been described thus far should be happening. First, that capital should
ow
from emerging market and developing economies to advanced ones contradicts the standard
economic theory developed by Robert Solow and Trevor Swan in 1956 (hereafter referred to
4

Figure 3: Federal Debt, the Real Yield and the US Dollar Index
23456
012
808590951002005 2010 2015
2005 2010 2015
2005 2010 2015USD trillion percent Index 2003=100held by Foreign and International InvestorsFederal Debt
Real Yield
10−year Treasury Securities
US Dollar Index: Major Currencies
Notes : The US dollar index is a weighted average of the foreign exchange value of the US dollar against the
following currencies: the Euro, Canadian Dollar, Japanese yen, British pound, Swiss franc, Australian dollar,
and Swedish krona. Data on federal debt held by foreign and international investors is from the Federal
Reserve Bank of St. Louis. Data on the real yield for 10-year Treasury securities is from the Treasury
International Capital Reporting System. Data on the US Dollar Index is from the Federal Reserve Bank of
St. Louis.
5

as the Solow-Swan model ). The Solow-Swan model denes the rate of return on capital as
being equal to its marginal product minus deprecation. Assuming that the marginal product
of capital is decreasing in the level of capital stock, the model implies that the rate of return
on capital is higher in countries possessing a lower capital-to- worker ratio. Further assuming
that it is in \rich" countries where the capital-to-worker ratio is highest, then the Solow-
Swan model's central prediction is that in an open and competitive environment, capital
will
ow \downhill" in search of higher rates of return until the latter are equalized; that is,
capital will
ow from rich to poor countries until international rates of return on capital are
the same.
Standard economic theory is also challenged since increasing demand for US sovereign
debt exists simultaneously with dollar depreciation and decreasing real yields. The liter-
ature on sovereign debt predicts that in response to dollar depreciation|considered to be
synonymous with US default (albeit implicitly) on its sovereign debt obligation|demand
and yields should decrease and increase, respectively.
The conventional approach to explaining sustained demand for US sovereign debt, even
despite these declines in quality is what we will refer to as the Safe Assets theory , which
argues that US Treasury securities are a safe asset : an asset that is
ooded to during
and following adverse macroeconomic shocks. Though a safe asset is primarily dened and
identied by way of its eect rather than its characteristics, the attraction of US Treasury
securities is argued to be ultimately owing to its safeness, the latter of which is itself reliant
upon favorable investor perception of US nancial development.
Thus, persistent demand for US Treasury securities in the Safe Assets theory is owing
to the combination of risk-aversion among developing and emerging market economies and
their perception that these securities constitute a safe asset. An important implication of the
Safe Assets theory which follows from its reliance on perception is that the stability of the
6

US nancial system|and the global nancial system tethered to it|is tenuous as continued
funding of the former only hinges on a change in perception.
Despite the popularity of the Safe Assets theory, it is subject to a logical
aw. Namely,
the Safe Assets theory fails to oer a plausible explanation of what mechanism enables
sovereign lending in the rst instance. The literature has yielded strong conclusions that
sovereign lending does not occur absent of a lender's ability to impose large punitive costs
onto a defaulted borrower. That is, at least in the case of sovereign lending, the proverbial
carrot is helpless to motivate when unaccompanied by the stick. The Safe Assets theory
assumes US willingness to repay its sovereign debt obligation is motivated by a desire to
avoid incurring a large reputational cost levied by its lenders; it does not consider, however,
that the strength of lenders (here, developing and emerging market economies) relative to
the borrower makes it highly unlikely that the former can impose a large punitive cost onto
the latter.
In this paper, we ll the gap left by the Safe Assets theory and provide an explanation
of what enables lending to the US in the rst instance. Our theory is that developing and
emerging market economies hold US Treasury securities to collateralize private dollar bor-
rowing. That is, the persistent demand for US sovereign debt is at least partially motivated
by the reality that public lending enables private and public dollar borrowing.
We argue that the persistent demand for US Treasury securities from developing and
emerging market economies is a logical conclusion of the rapid growth in private dollar credit
to the latter (McCauley, McGuire and Sushko, 2015). As is well-known, capital in
ows have
a perverse eect upon its recipients: in
ows increase not only the recipient's reliance on still
more in
ows but also its debt to GDP ratio. This leads to unfavorable investor perceptions of
country volatility and increases the likelihood that in
ows will reverse, leaving crisis in their
wake. Against these prospects, developing and emerging market economies are incentivized
7

to oer their lenders collateral against the lending contract to secure continued access to
dollar funds.
US Treasury securities are an ideal form of collateral to dollar lenders for three main
reasons: rst, the scale of its issuance enables lenders to collateralize an asset large enough
to make the consequences of default larger than its benet; second, by seizing a country's
savings, the lender prevents the borrower from defaulting one debt obligation to enter into
another; nally, US Treasury securities are held in the lender's jurisdiction, making it rela-
tively costless for the lender to retaliate against a defaulting borrower.
Because it identies a motivation for investment in US Treasury securities less
eeting
than perception, one important implication of our theory is that the US ability to receive
continued and easy nancing is more stable than the Safe Assets theory assumes. The theory
is put to empirical test in Chapter 3.
The paper proceeds as follows. Section 1.2 reviews the literature on global imbalances to
set the stage for the introduction of the Safe Assets theory. Section 1.3 presents an analysis
and critique of the Safe Assets theory. Section 1.4 then presents an alternative theory to
explain persistent demand for US Treasury securities among developing and emerging market
economies. Section 1.5 concludes.
1.2 The Literature on Global Imbalances
In 1990, Robert E. Lucas, Jr. articulated what is popularly known as the Lucas Paradox :
despite the predictions of the Solow-Swan model, capital does not
ow from rich to poor
countries to the degree expected (Lucas, 1990). Using the example of capital
ows between
India and the United States, Lucas argued that return to capital should be much higher
in India than in the US, thus eliciting a
ow of capital to the former of a magnitude not
8

observed in reality. Addressing two possible explanations commonly cited to reconcile the
lack of downhill
ows with higher returns to capital|dierences in labor productivity and
political risk|Lucas also challenges each in turn.
Dierences in labor productivity can dampen the singular eect of higher worker to
capital ratios if the latter are also accompanied by a high marginal product of capital. Con-
trolling for possible dierences in labor productivity, Lucas nds that though the protability
of investment projects in poor countries is revised downwards once controlling for these dif-
ferences, capital investments in poor countries still generate a rate of return ve times that
of what would be yielded by capital investments in a rich country. That is, even controlling
for the fact that levels of productivity may be higher in rich countries, economic theory still
predicts that capital should still
ow from rich to poor countries at a rate much higher than
was observed.
Lucas further considers whether the puzzle can be solved if investments in human capital
have an exponential eect on productivity through technological innovation (that is, where
the investments in human capital are higher, the return to productivity is increasing and
not diminishing). Lucas nds that this adjustment does seemingly reconcile the puzzle
he presents, but only if unrealistic assumptions are adopted. Finally, Lucas addresses the
possibility that capital fails to
ow downhill to the extent predicted because political risk
in poor countries compromises their ability to credibly commit to repayment. Rejecting the
notion, he argues that if political risk was relevant to the
ow of capital to poor countries,
we might expect to have seen global capital
ows equalized|at least between Europe and
India|during the period when poor countries were subject to the ostensibly stabilizing force
of colonial rule. Ultimately, Lucas leaves the puzzle he presents intact (Lucas, 1990).
Little more than two decades after the Lucas puzzle was articulated, the allocation puzzle
again presented a challenge to standard economic theory. Because it is assumed that pro-
9

ductivity growth rises with technological progress, the neoclassical growth model predicts
that in a search for higher returns, capital should
ow into countries that have reached the
same level of technological progress as the rest of the world (i.e., whose total factor produc-
tivity|or Solow residual |reaches the world frontier) and out of countries whose level of the
same has receded. The allocation puzzle, however, presented the nding that capital
ows to
developing countries tend to be negatively correlated with their productivity growth. That
is, whereas Lucas articulated a dynamic in which too little funds were being received by
developing countries given their comparatively low capital-to-worker ratios, the allocation
puzzle added the mystifying detail that those developing countries experiencing higher rates
of productivity growth tend to receive less capital in
ows.
More contemporary evidence suggests that the driving force behind this inverse rela-
tion between capital in
ows and productivity growth is the public sector, as the sovereign
governments of developing and emerging market economies have largely saved the income
generated by comparatively high rates of economic growth (Gourinchas and Jeanne, 2013).
Again, standard economic theory is challenged by this phenomenon: higher volatility of
income and lower relative levels of development among developing and emerging market
economies yields the expectation that they will possess higher rates of borrowing and in-
vestment to smooth consumption intertemporally and promote development (Obstfeld and
Rogo, 1995).
A number of theories have been posited to explain why levels of national savings have
increased in developing and emerging market economies. Highlighting the importance of an
export-led growth strategy for emerging markets, the Bretton Woods II hypothesis initially
focused on the dynamic between the United States and China to argue that emerging mar-
ket economies accumulate advanced economy debt (namely, US debt) to undervalue their
own currencies, thereby allowing their exports to remain competitive in advanced economy
10

markets (Dooley, Folkerts-Landau and Garber, 2003). Because the demand for advanced
economy debt re
ects a kind of dependency of emerging market economies on advanced
economies, the earliest version of the Bretton Woods II hypothesis implies that global im-
balances are fairly sustainable. That is, so long as advanced economy funding is used to
support an export-led growth strategy in emerging market and developing economies, global
imbalances will persist uninterrupted (Dooley, Folkerts-Landau and Garber, 2003, 2004).
Research eorts have been focused not only on discovering why developing and emerging
market economies have largely become net savers but also on why their savings are stored
in advanced economy assets. The latter reality is naturally an exacerbation of the Lucas
puzzle as it implies that capital not only fails to
ow downhill in sucient quantity (as
Lucas noticed), but that it also
ows in the reverse direction. That is, capital
ows uphill
from poor to rich countries.
One explanation is that the advanced economy with the highest rate of productivity
growth relative to other advanced economies will receive a disproportionate amount of cap-
ital in
ows (Engel and Rogers, 2006). This justication does not, however, explain why
global capital
ows are found distributed among advanced economies in the rst place. In
addition, investment in advanced economies is mainly in xed-income instruments and in
equity; if productivity growth was a factor attracting investment to advanced economies,
then we might reasonably expect those investments to be in assets that earn a rate of return
proportional to the rate of productivity growth (i.e., investors would want a portion of the
return generated from the increasing rate of productivity growth.) What is seen instead,
however, is that investments in advanced economies are in assets that oer a xed rate of
return (Balakrishnan, Tulin and Bayoumi, 2007).
Another explanation highlights the function advanced economy debt serves as a reserve
asset : an asset that acts as a highly-liquid, international store of value. This view sees the
11

increased savings rates among developing and emerging market economies as a re
ection of
lessons learned from the series of domestic nancial crises occurring in the mid-late nineties,
allegedly catalyzed by poorly allocated capital in
ows. Thus, while the earliest version of
the Bretton Woods II hypothesis suggests the contemporary saving behavior of developing
and emerging market economies is opportunistic, this view understands it as precautionary:
saving re
ects a precautionary motive according to which developing and emerging market
economies save to fortify their stock of foreign reserves in case severe macroeconomic shocks
are experienced again (Bernanke, 2005; Caballero, 2006; Bussi ere, Cheng, Chinn, and Lisack,
2015). The notion that developing and emerging market economies possess a precautionary
motive to save may imply a somewhat more forgiving view of saving behavior than the
Bretton Woods II hypothesis. Importantly, however, blame for precaution is squarely placed
upon developing and emerging market economies since it is theorized to be the ineciencies
in the latter that ostensibly catalyzed the nancial crises now acting as a motivating force
(Bernanke, 2005).
Why, then, investment in advanced economy debt? The conventional view argues that
uphill capital
ows are the result of heterogeneity in securitization capacity; that is, uphill
capital
ows are the result of the ability of countries to produce nancial assets that are
both innovative and capable of acting as a store of value for a large amount of savings (i.e,
nancial depth and breadth) (Balakrishman, Tulin, and Bayoumi, 2007; Caballero, Farhi
and Gourinchas, 2008; Caballero and Krishnamurthy, 2009; Vermeulen and de Haan, 2014).
The ability of a country to generate nancial assets is understood to depend on a host of in-
stitutional, political, and economic factors that allow future output to be credibly pledged to
nanciers; these factors include strong property rights that minimize the probability of expro-
priation, strong economic fundamentals, institutions that are capable of enforcing contracts,
economic policies amenable to market liberalization, high levels of macroeconomic stability,
12

and low levels of political corruption and cronyism (Caballero, 2006; Alfaro, Kalemli-Ozcan
and Volosovych, 2014; Caballero, Farhi and Gourinchas, 2008).
When these securitization-enabling factors are absent in a country that also possesses
a high growth rate and high desired rate of precautionary savings, the conventional view
argues that the result will be a
ow of funds from asset-poor to asset-rich countries (i.e., from
countries where nancial assets are scarce to countries where nancial assets are plentiful).1
The empirical evidence on the link between nancial depth and uphill
ows has been
mixed. If a factor drawing global nancial
ows toward advanced economies is nancial
depth and breadth, then we might expect those countries with less of a capacity to generate
deep and innovative nancial assets to be more likely to invest in advanced economy nancial
assets. Yet, some evidence demonstrates the opposite is true: levels of nancial deepening
(measured by the ratio of near-money [M2] to GDP) is positively correlated with current
account balances; that is, among developing and emerging market economies, more nancial
development is associated with a greater level of national saving (Chinn and Prasad, 2003;
Chinn and Ito, 2007; Gruber and Kamin, 2009). In addition, developing and emerging
market economy savings overwhelmingly
ows to perhaps the least innovative, advanced-
economy nancial asset: sovereign debt. This oers a reason to question the appeal of
nancial breadth in drawing global capital
ows toward advanced economies.
The asset-rich country receiving a disproportionate in
ow of developing and emerging
1The formal treatment of this explanation relies on a model of global intertemporal equilibrium. Given
two, one-good, pure endowment economies and two time periods (1 and 2), the autarky rate of interest
for either country will be driven upwards when the savings rate in a country is lower and impatience (the
degree to which agents prefer to consume in period 1 and therefore prefer to borrow against future income)
is higher; accordingly, the autarky rate of interest will be driven downwards when the savings rate is higher
and impatience lower. If the model is populated by two countries, AandB, that dier in their preferred
rate of savings and level of impatience|where country Apossesses more impatience and a low savings rate
(characteristic of advanced economies) and country Bpossesses less impatience and a high savings rate
(characteristic of developing and emerging economies)|then country Awill have a higher autarky rate of
interest than country B; accordingly, in period 1, we would expect that country Awill borrow from country
B. That is, we would expect that capital would
ow uphill (Obstfeld and Rogo, 1995).
13

market economy savings for at least the past two decades is the United States. Indeed, this
reality prompted many to identify a global savings glut as the culprit of the 2008 global
nancial crisis and the subsequent ineectiveness of US monetary policy to respond. That a
country's nancial sector would act as such an in
uential determinant of uphill capital
ows
is challenged by this persistent demand for US sovereign debt, especially prior to the 2008
nancial crisis. Figure 1 illustrates that in 2008, foreign holdings of long-term US Treasury
securities grew at an increasing rate even as New York-based Bear Stearns and Lehman
Brothers collapsed.
Thus, demand for US sovereign debt|ostensibly motivated by the nancial depth and
breadth of the US|was most strong when by all reasonable measure, the US was on the
verge of nancial collapse. What is more, at this same time, the real yield on 10-year Trea-
sury securities precipitously fell and the dollar depreciated against other advanced economy
currencies (see Figure 3). Relatedly, empirical evidence suggests that large trade balance
adjustments in the US have little eect on the demand for US sovereign debt (Kamin, Reeve
and Sheets, 2007).
In response to these uniquely puzzling aspects of developing and emerging economy de-
mand for US sovereign debt, the predominant Safe Assets theory argues that US Treasury
securities are a safe asset . We now turn to that theory in more detail.
1.3 The Safe Assets Theory
The Safe Assets theory argues that uphill capital
ows and the persistence of demand for
US Treasury securities can be explained by the latter's status as a safe asset, particularly in
the foreign exchange reserves of developing and emerging market economies. What exactly
determines a safe asset is somewhat dicult to discern, however, as is readily acknowledged
14

by the literature itself (Caballero and Farhi, 2014). In its most concrete form, a safe asset
is dened as one that is
ocked to during periods of crisis, thus allowing it to retain its
value even (and especially) under such circumstances (Maggiori, 2013; Caballero and Farhi,
2014). Thus, the Safe Assets theory is in many ways a restatement of an earlier explanation
of uphill capital
ows whereby investments in advanced economy debt act as a reserve asset
(Bernanke, 2005; Caballero, 2006; Bussi ere et al., 2015). What is dierent about the Safe
Assets theory, however, it that it now adds as a quality of reserve assets a strong appeal
during periods of severe macroeconomic instability.
Thus, a safe asset is primarily detected by virtue of its eect rather than by underlying
determinants of safeness. That is, a safe asset is safe because it behaves as if it is safe,
though determinants of safeness are not subject to rigorous analysis. Curiously, however,
to the degree that safeness is re
ective of belief in the solvency of a sovereign borrower,
the perception of safeness can make the quality manifest since continued nancing allows
a borrower to meet past debt obligations. This dynamic is referred to in the literature as
strategic complementarity whereby investor behaviors are complements since more safe asset
purchases generates even more safe asset purchases. The conclusion yielded from a stylized
model developed to explain the safe asset status of US Treasury securities is illustrative: \The
safety of a safe asset depends on investor beliefs. Safety is endogenous, and when investors
believe an asset will be safe, their actions can make that asset safe" (He, Krishnamurthy,
and Milbradt, 2016, p. 523).
Though the Safe Assets theory is borne from dynamics that challenge whether a country's
nancial asset-generating capacity is a determinant of capital in
ows, it is still tethered to
this view in the following ways. First, the theory argues that what determines the choice
between safe assets is understood to be relative debt capacity. That is, if the existence of
safe assets is given (i.e., we begin from a state of the world where safe assets exist though for
15

reasons unknown), then what determines investor selection between safe assets|i.e., whether
a foreign ocial or private investor chooses to hold the safe assets of country AorB, for
instance|is how much debt a country can issue.
Debt capacity plays a large role in safe asset selection according to the Safe Assets
theory because of investor preferences and expectations: investors prefer to sustain the
largest quantity of safe assets possible and expect other investors to prefer the same; as a
result, investors expect investment
ows into countries with the largest capacity to issue
debt and will therefore also invest in the same. This behavior constitutes an endogenous
process whereby the debt of countries with the largest debt capacity actually becomes safe.
The theory argues that this herding behavior towards the safe assets of the sovereign nation
with the largest debt capacity illustrates a \nowhere else to go" principal: investors seeking
safe assets have \nowhere else to go" but the debt of the sovereign nation with the largest
debt capacity (He, Krishnamurthy, and Milbradt, 2016, p.519).
Second, though not readily apparent, a detailed analysis of the safe asset theory reveals
that debt capacity is understood to stand as a proxy for the nancial depth of a country
which is, in turn, generated by the strength of its economic fundamentals. Given that the
goal of the safe asset theory is to explain the resiliency of US sovereign debt in the absence of
strong economic fundamentals, this last point could reasonably be a source of confusion. The
argument can be made clearer, however, once thresholds are considered. The Safe Assets
theory eectively holds that absolute economic fundamentals may sustain a level of debt
issuance so large that it surpasses a certain threshold beyond which the debt enters a class
of its own: it is perceived to be safe simply by virtue of its size (i.e., the debt is safe because
it was able to reach such a large level.)
Once beyond this threshold, it becomes assumed that debt capacity stands as a proxy for
nancial depth generated from strong economic fundamentals, where the strategic comple-
16

mentarity of investor behavior serves to reinforce the perception. The assumption becomes
so embedded and the reinforcing eect of investor behavior so strong that the primary de-
terminant of safe asset selection becomes debt capacity, even to the extent that it overrides
the implications of its required result: poor economic fundamentals of a safe-asset issuing
country (He, Krishnamurthy and Milbradt, 2016). In this way, the debt capacity of a safe-
asset issuing country is both dependent on and a departure from the literature attempting to
explain the phenomenon of uphill capital
ows by way of the nancial depth and breadth of
advanced economies (Balakrishnan, Tulin, and Bayoumi, 2007; Caballero, Farhi and Gour-
inchas, 2008; Antras and Caballero, 2009; Caballero and Krishnamurthy, 2009).
One important implication of the Safe Assets theory is that because the safeness of
safe assets is ultimately perceived, its loss as an attribute is simply dependent on the loss
of this perception. Applied to the case of the United States, which is understood to have
possessed the world's largest store of safe assets for at least the past fty years, the reliance on
investor perceptions of safeness implies that the solvency of the United States is necessarily
tenuous as it only requires a change in investor condence to threaten it (Caballero and
Farhi, 2014; He, Krishnamurthy and Milbradt, 2016; Caballero, Farhi and Gourinchas 2016,
2017; Farhi and Maggiori, 2017). That is, the stability of the US nancial system (and
by extension, the global nancial system) stands on a knife-edge where the central factor
determining continued balance is investor perception of the US ability to repay its sovereign
debt. Further, because investor condence may be strengthened to an extent that sustains
declining economic fundamentals of the safe-asset issuing country, it becomes unclear at
what point a safe asset-issuing country will have gone too far in stretching its debt capacity.
The system is, therefore, fragile (Farhi and Maggiori, 2017).
17

1.3.1 A Critique of the Safe Assets Theory
In this section, we will critique the logical foundation of the Safe Assets theory, thereby
challenging the notion that perception can provide a comprehensive explanation to the puzzle
that is persistent demand for US sovereign debt. To set the stage for our critique, we begin
this section with a brief overview of the sovereign debt literature. Our main point will be
that this literature yields the strong conclusion that lending to a sovereign fails to occur
in the absence of a mechanism enforcing the debt contract. We argue that the Safe Assets
theory does not provide such a mechanism to a plausible degree, thus rendering it insucient
as a complete explanation of persistent foreign demand for US sovereign debt.
A central focus of the sovereign debt literature is the risk of lending to a sovereign in the
presence of sovereign immunity |a legal doctrine protecting the independence of sovereign
governments by ensuring their immunity to criminal or civil prosecution. The doctrine
necessarily restrains lenders from enforcing debt contracts with sovereign borrowers since
there does not exist any supra-national legal system that can facilitate enforcement. As
a result, sovereign lending belongs to a distinct group of economic exchanges where an
agency problem exists because the diculty inherent in monitoring the good or service to
be exchanged makes a contract unable to be third-party (exogenously) enforceable at the
same time that one of the parties (namely, the borrower) has the ability to benet or harm
another by repaying the loan or defaulting on it, respectively.2
Nevertheless, lending to sovereigns does indeed occur and the current consensus view
2In this context of a sovereign borrower who enters into a non-contingent debt contract|a contract that
species a set of dates and payments that must be made to the lender irrespective of (i.e., not contingent
upon) the state of the sovereign|default is a failure of a sovereign borrower to honor its debt contract,
either by a refusal to pay or by necessitating a renegotiation of the debt contract in terms less favorable to
the lender (Aguiar and Amador, 2014). If the debt is to be repaid in the borrower's currency, then the latter
form of default|repayment in terms less favorable to the lender|can occur implicitly through in
ation or
depreciation of the borrower's currency. Through implicit default, the lender may receive repayment that
fullls the debt contract in nominal terms but is worth less in real terms.
18

argues this is because there isenforcement power governing debt contracts with sovereign
borrowers; it simply is not exercised by a third-party. Rather, enforcement power is under-
stood to arise endogenously from the relation between the parties to the contract. That is,
by virtue of the relationship between lender and sovereign borrower, a structure either exists
or is created that makes default costlier to the sovereign borrower than repayment. The
methods by which repayment is endogenously secured are called endogenous enforcement
mechanisms (Bowles and Gintis, 1990, as cited in Prem, 1994).
Endogenous enforcement mechanisms generally fall within three categories: contingent
renewal, collateralization, and retaliation (Bowles and Gintis, 1990, as cited in Prem, 1994).
Contingent renewal utilizes the desire of the borrower to borrow again (i.e., to renew the debt
contract) and accordingly requires future borrowing to be contingent upon present payment.
As applied to the case of international credit markets and sovereign borrowing, contingent
renewal implies that lending to sovereign borrowers occurs because lenders are condent that
the threat of exclusion from international credit markets in the case of default operates to
endogenously enforce repayment; in such a circumstance, sovereign borrowers are said to be
incentivized to protect their reputation (where exclusion from international credit markets
is said to constitute a reputational cost .)
Collateralization as an endogenous enforcement mechanism refers to the loss of a default-
ing borrower's asset posted as collateral; importantly, for the loss to disincentivize default,
the asset must be large enough so that the borrower's cost of default is greater than the cost
of compliance. Collateralization is also linked to the endogenous enforcement mechanism
ofretaliation since it is when the lender retaliates against a defaulting borrower by seizing
collateral that collateralization is employed as an enforcement mechanism. But, retaliation
also encompasses a wider set of methods, especially in the context of international credit
markets and sovereign borrowers. These methods include seizing the borrower's assets over-
19

seas (nancial or otherwise), impeding its trade, and/or intervening militarily. Again, these
methods only operate to endogenously enforce repayment if the cost levied onto the sovereign
borrower outweighs the benet from default.
Importantly, for endogenous enforcement mechanisms to enable lending to sovereign bor-
rowers, it must necessarily be the case that lenders are strong, or better situated, than
borrowers. This is so because excluding defaulting borrowers from international credit mar-
kets requires coordination among lenders and other market participants to make the cost of
the latter's potential engagement with defaulting borrowers larger than its benet. But, en-
couraging coordination requires capital and in
uence since cooperation must extend widely
in international capital markets and imposing costs is not costless (Eaton and Fernandez,
1995; Epstein and Gintis, 1995). Similarly, military intervention or seizure of the borrower's
assets requires that the lender possess a powerful military and be powerfully situated in
trade and nancial networks, respectively (Epstein and Gintis, 1995). If the borrower pos-
sesses greater capabilities than the lender, then the outcome is straightforward: the borrower
will be capable of circumventing the lender's attempts of endogenously enforcing the debt
contract, thus rendering endogenous enforcement mechanisms ineective.
The literature on sanctions is also helpful in demonstrating the requisite relative ca-
pabilities between borrower and lender. For sanctions levied onto a sovereign nation to be
successful, researchers nd that the recipient, or target , must bear more costs than the sender
(Eaton and Engers, 1992, 1999; Davis and Engerman, 2003; Drezner, 1999; Hufbauer, Schott
and Elliott, 1990). For instance, Eaton and Engers (1992) describe the impacts of what they
calltoughness |a measure of a country's willingness to incur costs and ability to withstand
them|on sanction outcome. They nd that the likelihood of sanction success is greatest
when the sender is tough relative to the target (i.e., when the sender's ability and willing-
ness to withstand costs is high relative to a target). Indeed, reality is a testament to this
20

requisite dynamic: sanctions generally tend to be levied by stronger countries onto weaker
ones (Hanlon, 1986; Drury, 1998; Levy, 1999). As reported in Davis and Engerman (2003),
\In the 115 cases of economic sanctions deployed since 1914. . . the GNP of the sender (or
principal initiator) of sanctions was nearly always over ten times that of the target and in
the majority of cases more than 50 times greater" (p. 191).
Having brie
y reviewed some of the basic ndings of the literature on sovereign lending,
we can now return to the Safe Assets theory. The Safe Assets theory assumes that the
endogenous enforcement mechanisms enabling lending to the US are contingent renewal
and retaliation ; specically, it assumes that if the US defaults on its debt obligation, it
will incur a reputational or direct cost. As has been shown, such an assumption relies on
conceptualizing the lending dynamic as one between a strong lender and a weak borrower
(Farhi and Maggiori, 2017).
But, can the United States reasonably be considered weak relative to its developing and
emerging market economy counterparts? That is, is it reasonable to assume that standard
endogeneous enforcement mechanisms are operative because developing and emerging market
economies can coordinate to exclude the US from international capital markets or credibly
threaten to seize assets large enough to make the cost of default higher than its benet?
The evidence suggests that this is not a reasonable assumption. Research ndings
strongly suggest that the United States maintains a central position in the structure of
what can be thought of as a global nancial network.3The research in international political
economy, particularly from the vantage point of political science, is instructive in describing
3It is worth noting that we are aware of the declinist narrative predicting the demise of US hegemony
(see, for example, Wallerstein, 2006). While a description or analysis of this literature is beyond the scope
of this paper, it is useful to point out that the declinist narrative argues that, currently, US hegemony has
declined in a relative sense and not in an absolute sense. That is, while the absolute structural position of
the United States in the global arena is subject to debate, its current hegemony relative to other countries is
not. Thus, even if the declinist narrative is adopted, the point highlighted in this paper|that relative to the
United States, emerging and developing countries possess a less central position in the global arena|should
not constitute a challenge to our conclusion.
21

the central position and consequent capabilities of the United States in the international
nancial arena. For instance, using network analysis to analyze data on cross-national bank-
ing ties from the Bank of International Settlements, Wineco (2015) concludes that the
United States possesses the most cross-national banking ties of any other leading economy
such that the US can properly be described as the \world's banker" (p. 507).4Wineco
argues that this position not only failed to be diminished by the 2008 global nancial crisis
but was actually enhanced as US competitors were weakened.5
Additional research demonstrates that the Federal Reserve's access to liquidity enables
it to take on the global nancial role of international lender of last resort during crises
and especially to countries where instability would threaten US interests (McDowell, 2017).6
Further demonstrative of US capabilities in global nancial networks is its ability, post 9-
11, to monitor global nancial institutions and transactions. In 2001, Title III of the USA
Patriot Act gave the US Treasury Secretary broad discretion to subject banking and nancial
institutions (domestic and global) to increased scrutiny. Further, in 2006, the CIA, overseen
by the Treasury department, subpoenaed and won access to the nancial records database of
the Society for Worldwide Interbank Financial Telecommunications (SWIFT): a privately-
run, Brussels-based global nancial messaging company currently possessing the world's only
centralized infrastructure through which international nancial institutions interact.
Considering the centrality of the US in the global arena, the notion that lenders and
market participants from developing and emerging market economies could credibly threaten
or impose a reputational cost onto the United States that would disincentivize default is
4Per Wineco (2015), a tie is formed when banks in one country hold assets in another.
5See also Oatley, Wineco, Pennock and Bauerle Danzman (2013).
6The term \international lender of last resort" originates with Kindleberger (1986). An international
lender of last resort is dened as: \An actor that is prepared to respond to international nancial crises
by providing credit to illiquid institutions in foreign jurisdictions when no other actor is willing or able"
(McDowell, 2017, p. 4).
22

implausible. Realistically these lenders and market participants possess neither the access
nor the reach to obstruct US circumvention of their eorts.7Further, by virtue of the
US role as \world's banker," it is necessarily the case that it holds the largest variety and
amount of foreign nancial assets, reasonably warranting consideration of how, in the case of
default, developing and emerging market economy lenders might seize the requisite amount of
nancial assets that would make retaliation eective. Finally, though it is generally accepted
that military intervention is no longer employed as a response to sovereign default, it is worth
noting that this antiquated strategy is unlikely to reemerge as a tactic given US military
capabilities.
The loss of standard endogenous enforcement mechanisms leaves only perceptions of
safeness to account for the phenomenon of sovereign lending in the Safe Assets theory. While
perception likely plays some role in sovereign debt dynamics, it is implausible that it can be
the only factor enabling sovereign lending. That something more substantive than perception
motivates persistent demand for US Treasury securities is also suggested by the variation of
holdings across emerging market economy foreign ocial institutions. To illustrate, Figure 4
plots data on long-term Treasury security holdings for a select group of emerging economies
as a percentage of their gross domestic product (GDP), imports and exports during the
period 2006-2016.8
Given its reliance on investor perception, the Safe Assets theory concludes that investors
have \nowhere else to go" but the advanced economy with the largest debt capacity. The
empirical reality, however, belies such a strict constraint. Given a set of countries motivated
7It is useful to point out that even China, perhaps the most capable of emerging economy holders of US
sovereign debt, is not understood to exert substantial in
uence in nancial markets (Aizenman & Ito, 2016).
For this reason, critics of the declinist narrative argue that the latter puts too much emphasis on factors such
as GDP to argue that the US is in decline. Rather, they argue that a more comprehensive set of indicators
should be analyzed to determine prominence in the global arena (Beckley, 2012; Norlo, 2014).
8Long-term Treasury security holdings are expressed as the percentage of the country's GDP to control
for country size across selected emerging-market countries.
23

Figure 4: Holdings of Long-term US Treasury Securities by Indicator and Country
050010001500
ArgentinaBrazil
BulgariaChileChina
Colombia HungaryIndia
Indonesia MalaysiaMexicoPeru
PhillippinesRomaniaRussiaThailandTurkeyUkrainePercentHoldings of Long−term US Treasury Securities as a Percentage of Exports
0500100015002000
ArgentinaBrazil
BulgariaChileChina
Colombia HungaryIndia
Indonesia MalaysiaMexicoPeru
PhillippinesRomaniaRussiaThailandTurkeyUkrainePercent2006
2007200820092010201120122013201420152016Holdings of Long−term US Treasury Securities as a Percentage of Imports
051015
ArgentinaBrazil
BulgariaChileChina
Colombia HungaryIndia
Indonesia MalaysiaMexicoPeru
PhillippinesRomaniaRussiaThailandTurkeyUkrainePercentHoldings of Long−term US Treasury Securities as a Percentage of Real GDP
Notes : Data on long-term holdings of US Treasury securities is adjusted for valuation eects. Data on long-
term US Treasury security holdings is from Bertaut and Judson (2014). Export and import data specically
refers to merchandise exports and imports, respectively, and is from the World Trade Organization. Data
on real GDP is from the World Bank.
24

to invest a disproportionately high level of national savings (i.e, emerging market economies),
however, we would expect to see a somewhat more even distribution of US government debt
holdings if investors have \nowhere else to go." That is, a singular determinant should
produce a more uniform outcome across similarly-situated countries.9What is seen from
Figure 4, however, is that holdings of US government debt among emerging market countries
displays quite a bit of variation across country and through time.
In the next section, we introduce a more plausible alternative to the Safe Assets theory
that is consistent with standard sovereign lending dynamics and implications arising from
the relative strength of borrowers in this circumstance.
1.4 US Treasury Securities as Collateral for Private Dollar Bor-
rowing
Occurring simultaneously with the persistent demand for US Treasury securities, yet
unconnected to it in the literature thus far, is dollar credit to developing and emerging
market economies. McCauley, McGuire and Sushko (2015) report that dollar credit to non-
US residents reached approximately $7 trillion or 13% of non-US GDP by 2014. Indeed,
dollar credit to non-US residents picked up faster than to US residents after the global
nancial crisis and at rates between 10-20%. China accordingly saw non-nancial businesses
and households more than double its holdings of foreign currency credit immediately after
the crisis (Borio, McCauley and McGuire, 2011).
In one respect, capital in
ows fulll its conventional purpose of allowing developing and
emerging market economies to smooth consumption. At the same time, however, developing
9Naturally, we do not argue that all emerging-market economies are similarly situated; rather, we argue
that they are conceived to be similarly situated in their demand for US sovereign debt.
25

and emerging market economies face signicant obstacles in constraining the capital in
ows
on which they are reliant. For instance, non-nancial residents and businesses have his-
torically reacted to higher domestic interest rates by borrowing from non-domestic sources.
Further, the standard prescription for damming capital in
ows|allowing the exchange rate
to appreciate (i.e., enabling it to
oat)|can actually exacerbate international credit in
ows
since appreciation relieves the debt-to-cash-
ow ratio, thereby creating room for a greater
debt burden (Borio, McCauley, and McGuire, 2011).
Supply of capital in
ows is similarly unwieldy since the forces dictating the
ow of in-
ternational credit are beyond the purview of recipient countries. For instance, McCauley,
McGuire and Sushko (2015) nd that during tranquil times (i.e., non-crisis periods when
advanced economies' nancial institutions are healthy) easy US monetary policy drives a
surge of private dollar credit into developing and emerging economies. Similarly, Rey (2015)
famously argues that US monetary policy travels via a \global nancial cycle," the move-
ments of which are dictated by perceptions of risk and volatility manifested in the Chicago
Board Options Exchange Volatility Index (VIX).10
A principal risk of unwieldly capital in
ows is derived from the fact that leverage ratios
are directly factored into calculations of country volatility by investors (McCauley et al.,
2015; Avdjiev, Kuti and Takats, 2012). Thus, higher leverage is accompanied by the risk
that capital in
ows will be reduced|often spontaneously and rapidly|below desired levels,
inciting crisis upon departure. Indeed, a higher growth rate of international credit relative
to overall total credit has generally become cause for concern given that the trend tends
to precede crisis, with the Asian Financial Crisis serving as one particularly severe example
(Borio, McCauley and McGuire, 2011).
10The Bank for International Settlements loosely denes a nancial cycle as \self-reinforcing interactions
between perceptions of value and risk, risk-taking and nancing constraints which translate into nancial
booms and busts." (Borio, McCauley and McGuire, 2011, p.2)
26

Given the unwieldiness of capital in
ows, developing and emerging market economies
necessarily lack control over the degree to which their nancial systems are leveraged. They
can, however, control the degree to which they are vulnerable to endogenous enforcement
mechanisms. That is, developing and emerging market economy borrowers can calm investor
fears and secure access to credit by increasing the degree to which they suer a penalty in
the case of default.
Specically, the sovereign debt literature yields the conclusion that borrowers can increase
their vulnerability to lenders by oering the latter collateral against the debt contract. In-
deed, from the lender's vantage point, collateralization is the most preferred endogenous
enforcement mechanism because it allows the lender to avoid most of the diculties inherent
in other endogenous enforcement mechanisms. To illustrate, recall that seizing the bor-
rower's assets in the case of default requires that a lender be suciently strong to withstand
the costs incurred (e.g., fending o an aggrieved borrower and its allies). It also requires
the borrower's assets be in the jurisdiction of a legal system amenable to the lender's de-
mands. Indeed, the diculty in gaining the cooperation of foreign jurisdictions where a
borrower's assets are held is one that the literature highlights as a main factor preventing
the eectiveness of retaliation (Eaton and Gersovitz, 1981; Bulow and Rogo, 1989; Eaton
and Fernandez, 1995; Pitchford and Wright, 2013).
The seizure of collateral, however, obviates these diculties as the asset to be seized
is presumably already in the possession of the lender and within the jurisdiction of a legal
system amenable to the latter's interests. Additionally, because it was oered prior to the
debt contract taking eect, the lender is assured upon entering the agreement that collateral
is large enough to make the consequence of default larger than its benet.
Accordingly, we theorize that faced with an inability to control the degree to which their
nancial systems are leveraged but wanting to maintain access to much needed dollar credit,
27

developing and emerging market economies collateralize in
ows of dollar credit with US
Treasury securities. Indeed, given those qualities of collateral that make collateralization
a particularly preferred endogenous enforcement mechanism among lenders, US Treasury
securities are uniquely suited to act as collateral for dollar credit.
Because the US is unique in the scale of its debt issuance, dollar lenders are presumably
able to require borrowers to hold US Treasury securities in sucient quantity to make the
cost of default greater than its benet. Additionally, because foreign ocial holdings of US
Treasury securities are held with the Federal Reserve Bank of New York, lenders are assured
that assets to be seized reside with a custodian amenable to lender interests, making retal-
iation virtually costless.11Finally, US Treasury securities are ideally suited to collateralize
sovereign borrowing because being the form in which country savings are held, Treasury
securities are also the means by which a defaulting sovereign would service another loan.
Thus, in the case of default, dollar lenders are able to costlessly exclude the borrower from
international capital markets by disrupting the latter's future payments.
1.5 Concluding Remarks
The persistent
ow of investment toward long-term US sovereign debt by foreign ocial
investors in developing and emerging market economies is puzzling for two reasons. First,
that capital should
ow \uphill" from developing and emerging market economies to ad-
vanced ones contradicts standard economic theory assuming higher marginal rates of return
in relatively poorer economies; that is, standard economic theory predicts capital will
ow
11If retaliation is costly to the lender (perhaps because impeding the borrower's trade also signicantly
aects the lender's current account), the borrower has less reason to believe the lender will in fact retaliate
(i.e., the threat of retaliation will not disincentivize default because it will not be viewed as credible) (Eaton
and Fernandez, 1995).
28

downhill until marginal rates of return on capital are equalized. Second, holdings of US
Treasury securities among developing and emerging market economies persist even despite
dollar depreciation, decreasing real yields, and declining US economic fundamentals.
The conventional view, which we refer to as the Safe Assets theory , explains the per-
sistence of demand for US Treasury securities among developing and emerging market
economies by way of the asset's safeness, which is a function of the issuer's nancial breadth
and depth. The Safe Assets theory is logically
awed, however, in that its assumptions
run counter to sovereign lending dynamics. Specically, the Safe Assets theory assumes
endogenous enforcement mechanisms enabling sovereign lending are satised when the rela-
tive strength between borrowers and lenders in this circumstance renders such mechanisms
ineective. In the absence of an endogenous enforcement mechanism, perceptions of safe-
ness alone cannot enable sovereign lending and this suggests that another factor besides
perception motivates demand for US Treasury securities.
We present an alternative theory to explain what motivates the persistent demand for
US Treasury securities among developing and emerging market economies. We contend that
US Treasury securities act as collateral for private dollar borrowing. In the next chapter, we
put this theory to empirical test.
29

C H A P T E R 2
LENDING TO BORROW: US SOVEREIGN DEBT AS
COLLATERAL FOR DOLLAR CREDIT
2.1 Introduction
Over the past two decades, foreign investment in US sovereign debt (i.e., Treasury secu-
rities) has grown at an increasing rate, more than quintupling to approximately $5.3 trillion
dollars by 2016. Persistent foreign demand for US Treasury securities presents a number
of challenges to conventional economic theory. Because high debt to GDP ratios are con-
ventionally understood to counteract a country's long-run economic growth prospects, the
former should disincentivize investment in a country's sovereign debt (Reinhart and Rogo,
2010; Cecchetti, Mohanty and Zampolli, 2011; Alfaro, Kalemli-Ozcan and Volosovych, 2014;
He, Krishnamurthy and Milbradt, 2016). Yet, foreign demand for US Treasury securities
remains strong despite an unprecedented US current account decit of 6% of GDP (Gourin-
chas and Rey, 2014). What is more, since roughly 2000, the real yield on 10-year Treasury
securities and the foreign exchange value of the dollar against leading currencies has fallen.
That is, the US continues to receive nancing even as it oers less of a return to investors
and implicitly defaults on its current debt obligations.1
1Aguiar and Amador (2014) dene default as a failure of a sovereign borrower to honor its debt contract,
either explicitly by an outright refusal to pay (which is how borrower default is conventionally understood) or
30

Largely behind this increasing level of nancing are foreign ocial investors in developing
and emerging market economies who hold US Treasury securities in foreign exchange reserves.
This rapid accumulation gained new attention as a culprit not only of the 2008 global nancial
crisis but also of the ineectiveness of monetary policy to respond.2While subsequent
research has revealed new insights on the causes of the crisis and its aftermath, the persistent
demand for US Treasury securities from developing and emerging economies continues to fuel
debate.3
What explains these puzzling features of demand for US sovereign debt and can we expect
them to continue? One prominent theory argues that US Treasury securities are perceived
to be a \safe" asset. Specically, debt capacity beyond a certain threshold is assumed to be
generated from strong economic fundamentals, leading to the perception that those countries
with the highest debt capacity must necessarily be the safest sources of investment. Thus,
while standard determinants of asset selection are thought to inform investment decisions
between safe assets, the perception that an asset is safe increases the degree of deterioration
in terms lenders will tolerate.
Owing to the ubiquity of this perception, the latter is self-reinforcing since during periods
of crisis investors will
ood the asset they perceive to be safe. This herd behavior drives a
secular decline in real interest rates, thus rendering monetary policy ineective once nominal
interest rates hit the zero lower bound. According to this view, which we will call the Safe
Assets theory , the continued nancing of the US decit is highly tenuous since its reliance on
perception makes it highly vulnerable to a self-fullling debt crisis (Maggiori, 2013; Caballero
implicitly by repaying in terms less favorable to the lender. When the loan is to be repaid in the borrower's
currency, then one form of implicit default occurs through depreciation of the borrower's currency since
repayment only fullls the debt contract in nominal terms.
2See Ben Bernanke's well-known Sandridge Lecture on the \Global Savings Glut" (Bernanke, 2005)
3For instance, Shin (2011) argues that European global banks played a key role in generating easy credit
conditions in the US prior to the 2008 Global Financial Crisis.
31

and Farhi, 2014; He, Krishnamurthy and Milbradt, 2016).
Despite its popularity, the Safe Assets theory is
awed. If perceptions based on debt
capacity played such a large role in determining the degree of investment in US Treasury
securities among developing and emerging market economies, then holdings among the latter
should be somewhat uniformly distributed. Yet, developing and emerging market economy
holdings of US Treasury securities are heterogeneous, indicating that country-specic factors
may in
uence levels of holdings.4Additionally, it is unlikely that a loan to the US would
serve as an investment in the traditional sense since there exists no plausible enforcement
mechanism to secure it. The literature on sovereign debt has long established that sovereign
immunity makes it highly unlikely that sovereign lending will occur in the absence of a
lender's ability to reasonably secure repayment (Eaton and Gersowitz, 1981; Bulow and
Rogo, 1989; Epstein and Gintis, 1995).5This ability is exercised through an enforcement
mechanism that imposes a cost to the borrower larger than the benet from default. (Bulow
and Rogo, 1989).
But, naturally, imposing a cost and especially a large one, is itself costly; thus, we
can reasonably expect that when loans are made, lenders will be strong and borrowers
weak (Epstein and Gintis, 1995). Otherwise, lenders expecting a sovereign borrower to
circumvent any attempt to enforce the lending contract, would not lend in the rst instance.
Given the relative strength between lenders (developing and emerging market economies)
and borrowers (the US) in this circumstance, it is implausible that lenders would be capable
of threatening the borrower with a cost of default high enough to outweigh its benets.
Thus, while perception can certainly play a role in the appeal of US Treasury securities, it
4See Figure 4 in Chapter 1 of this dissertation for an illustration of heterogeneity in US Treasury security
holdings across a subset of developing and emerging market economies.
5Though a review of the sovereign debt literature is beyond the scope of this essay, a review can be found
in the Chapter 2 of this dissertation.
32

is unlikely that this perception alone can oer a comprehensive explanation of the puzzle.
In this paper, we present and provide empirical evidence supporting the hypothesis that
US Treasury securities play the role of collateral for private and public dollar credit. That is,
public lending to the US enables private and public borrowing for developing and emerging
market economies. Our argument highlights a signicant development in the global nan-
cial arena that, despite occurring simultaneously with the buildup of US Treasury security
holdings, has yet to be connected to the latter: the growth of dollar credit in developing and
emerging market economies. Indeed, dollar credit to borrowers outside of the US reached
approximately $7 trillion, or 13% of non-US GDP (McCauley, McGuire and Sushko, 2015).
But, credit is not secured on hope. As mentioned, lenders must be reasonably sure the
loan will be repaid and employ enforcement mechanisms to maximize the likelihood of this
outcome. While these mechanisms are easily availed to lenders in the domestic context
owing to the presence of courts and an abiding legal system, lenders to sovereign borrowers
must rely on endogenous enforcement mechanisms, or disciplining devices arising from the
structure of the lending relationship itself. One of the more eective mechanisms can be
provided by the borrower who oers the lender collateral to be seized in the event of a
default.
The role collateral plays in facilitating the smooth transaction of funds has gone relatively
unnoticed in the more recent literature, but it is crucial nevertheless.6US Treasury securities
are an ideal form of collateral to dollar loans not only because the borrower's nancial assets
are denominated in the lender's currency but also because custodianship enables the lender to
costlessly seize these assets in the case of default. Importantly, because retaliation is virtually
costless (i.e., all the lender must do is seize a nancial asset already in its possession), the
6For a full treatment of the role collateral plays in facilitating the smooth transition of funds, see Singh
(2016).
33

Figure 5: US Treasury Security Holdings and Outstanding Dollar Credit 2003-2016
05001000
0 200 400 600
Outstanding Dollar Credit($B)USTS Holdings ($B)Country
AR
BRCHCLIDINKRMXMYRUTRTWNZA(a) China Included
050100150200250
0 50 100 150 200 250
Outstanding Dollar Credit($B)USTS Holdings ($B)(b) China Omitted
Notes : See Table 12 in Appendix A for data sources and denitions. See Table 13 in Appendix A for country
codes. The regression illustrated by the tted line in the left panel (China included) results in a coecient
of 1.90 (standard error = 0.058), N=689, and Adjusted R-squared=0.612. The regression illustrated by the
tted line in the right panel (China Omitted) results in a coecient of 0.51 (standard error = 0.037), N=636,
and Adjusted R-squared=0.231
threat of retaliation is credible, making default less likely in the rst instance. Finally, while
collateralization is naturally the lender's most preferred enforcement mechanism because it
is virtually costless to employ, it has typically been limited in use because of the diculty in
nding an asset large enough to outweigh the benet of default to the borrower. With US
Treasury securities, however, the lender can directly supply collateral in proportion to the
loan.
Panel (a) in Figure 5 plots US Treasury security holdings against outstanding dollar credit
for the countries included in our sample and illustrates a pronounced, positive relationship
between US Treasury security holdings and outstanding dollar credit. It is clear, however,
34

that China is not only an outlier but also seems to be driving the slope of the tted line. Panel
(b) demonstrates that even when China is omitted from our sample of countries, however,
US Treasury securities and outstanding dollar credit continue to be positively related.
To empirically investigate whether US Treasury securities serve as collateral to dollar
borrowing, we focus on the stylized fact that most foreign holdings of US Treasury secu-
rities are in long-term securities held by ocial institutions (i.e., central banks) as foreign
exchange reserves (Gourinchas and Rey, 2014; Sunner, 2017).7We use this fact to model the
decision to hold long-term US Treasury securities as one to hold dollars as opposed to other
feasible currencies (the Euro, Pound Sterling, and Japanese Yen) in foreign exchange re-
serves, which allows us to draw from well-developed models of the latter dynamic.8We then
use a rst-dierence estimator to control for country-specic factors aecting US Treasury
security holdings in a panel of developing and emerging-market economies. To incorporate a
persistence eect of US Treasury security holdings, we introduce as a regressor the dependent
variable lagged by one period and estimate the regression equation using an instrumental
variable (IV) method.
Our results demonstrate that even after controlling for a persistence eect (i.e., inertia)
in US Treasury security holdings, increases in outstanding dollar credit are related to a
statistically signicant ( p<0:01) increase in holdings of US Treasury securities. Specically,
the estimated increase in US Treasury security holdings related to a $1 billion increase in
outstanding dollar credit is $0.11 billion, all other factors held constant. Our result is robust
to outliers. Namely, our results are not driven by China's disproportionate holdings of US
7See Figure 2 in Chapter 1 of this dissertation for an illustration of this stylized fact.
8Our approach follows that taken by Ito, Jongwanich, and Terada-Hagiwara (2009) and Terada-Hagiwara
(2011). Our selection of feasible alternatives to the US dollar in foreign exchange reserves is informed by the
currencies included in the International Monetary Fund's Special Drawing Rights which is composed of the
most widely used currencies in global trading and nancial systems. We do not include the Chinese Yuan
as an alternative to the dollar because our data set spans 2003Q1-2016Q4 and the Yuan was only added to
Special Drawing Rights' basket of currencies in the last quarter of 2016.
35

Treasury securities and outstanding dollar credit. In fact, when China is removed from the
sample and our empirical model estimated again, the relationship between outstanding dollar
credit on holdings of US Treasury securities increases in statistical signicance ( p<0:001)
and magnitude. Specically, when China is removed from the sample, a $1 billion increase in
outstanding dollar credit is associated with a $0.18 billion increase in US Treasury securities,
all other factors held constant.
Our results are non-trivial, especially given the magnitude of US Treasury security and
outstanding dollar credit holdings, as reported in Table 1. Generally speaking, a $1 billion
increase in outstanding dollar credit is minimal given that the mean level of outstanding
dollar credit and US Treasury security holdings in our sample of developing and emerging
economy countries is approximately $91.5 billion and $105.2 billion, respectively. In more
recent years (2010-2016), the average yearly increase in outstanding dollar credit among
countries in our sample is $14.5 billion when China is included and $9.2 billion when it is
not. Perhaps more tellingly, outstanding dollar credit in each sample country has grown
during the period 2003Q1-2016Q4 by approximately $115 billion on average.9
We also nd that with the exception of outstanding dollar credit and the relative yield
on long-term US Treasury securities, none of our explanatory variables are . statistically
signicant when China is included in our sample. Because our analysis compares the US
to other nations or regions which might reasonably also issue safe assets (the Euro Area,
United Kingdom, and Japan), this result challenges the notion presented in the Safe Assets
theory that standard determinants of asset selection, including debt capacity, are operative
in the selection between safe assets.
Finally, while our motivating question is specic to developing and emerging market
9If China is omitted from our sample, then the level of outstanding dollar credit in 2016Q4 is approxi-
mately $84.8 billion higher on average than it was in 2003Q1.
36

economy holdings of US Treasury securities, the role of these securities in foreign exchange
reserves also allows our results to provide insight on the global role of the dollar generally.
Indeed, our assumption that holdings of long-term US Treasury securities eectively repre-
sent the dollar composition of foreign exchange reserves allows us to work around the highly
condential nature of this data.10Though a detailed review of the literature on the currency
composition of foreign exchange reserves is beyond the scope of this paper, a central concept
emerging from it is that the global role of a currency is strongly dictated by its use as a
reserve currency.11Thus, the research on determinants of reserve currency choice is typi-
cally part of a larger eort to develop insights on the continued dominance of a particular
currency, usually the dollar.12
Contrary to the Safe Assets theory that envisions continued nancing of the US decit as
teetering on a knife-edge and related research in the currency composition of foreign reserves
that warns of a spontaneous demise in the international role of the dollar, our results suggest
that these fears may be less than warranted. Specically, if holdings of US Treasury securities
and the global role of the dollar are driven at least in part by outstanding dollar credit, then
naturally the former are at least partially dependent on demand for dollar credit and the
ability of the US to provide it.
The paper proceeds as follows. Section 2.2 reviews the determinants of demand for US
Treasury securities. Section 2.3 presents our variables and describes the data. Section 2.4
presents our econometric model and describes our estimation strategy. In Section 2.5, we
10A number of researchers have commented on the unavailability of country-level data on the currency
composition of foreign exchange reserves (Truman and Wong, 2006; Galati and Woolridge, 2006; Eichengreen,
Chitu and Mehl, 2016; Sunner, 2017). The closest alternative is the IMF's Currency Composition of Ocial
Foreign Exchange Reserve (COFER) data|discontinued in 2015|which disaggregates by country groupings
(advanced and developing/emerging) annually for the period 1995-1998 and quarterly for the period 1999Q1-
2015Q2.
11An older but excellent treatment can be found in Prem (1994).
12See, for instance, Krugman (1984), Prem (1994), Chinn and Frankel (2005), Eichengreen, Chitu and
Mehl (2016), Eichengreen, Mehl and Chitu (2017).
37

present our results using a rst-dierence and IV strategy for samples including and omitting
China (an outlier). Section 2.6 presents the results from a series of robustness checks. Section
2.7 concludes.
2.2 Determinants of Demand for US Treasury Securities
Owing to the role of US Treasury securities in the foreign exchange reserves of developing
and emerging market economies, demand for these nancial assets are typically modeled as
a preference to hold dollars in foreign exchange reserves. Likewise, determinants of US
Treasury security demand can properly be considered synonymous with determinants of the
currency composition of foreign exchange reserves. We exploit this fact to model the demand
for US Treasury securities as demand for dollars in foreign exchange reserves.
Empirical eorts to determine whether a variable acts as a determinant of the currency
composition of foreign exchange reserves typically introduces it as a regressor among a set of
standard determinants (Prem, 1994; Eichengreen, Mehl and Chitu, 2017). These standard
determinants, or factors conventionally believed to aect the currency composition of foreign
exchange reserves, are theorized to operationalize more general motivations among foreign
ocial institutions of reducing transaction costs and preserving a store of value. That is,
those currencies that are disproportionately held in foreign exchange reserves are generally
theorized to be those that are most stable and least costly, and the standard determinants
are those which signal these qualities in a currency. These standard determinants typically
account for the following characteristics of issuing countries: nancial depth, exchange rate,
trade relations, and inertia.
Deeper nancial markets are understood to be more resilient and better able to lower
transaction costs of a currency. More specically, deep nancial markets are theorized to
38

facilitate better absorption of adverse shocks through diversication of risk, thus enhancing
stability (Prem, 1994; Sahay et al., 2015). One particularly popular proxy for nancial
depth is the ratio of broad money to GDP (Eichengreen, Chitu and Mehl, 2016).13Though
its attractiveness as a measure is likely owing at least in part to data availability, the ratio
of broad money to GDP is also lacking as a proxy for nancial depth. Specically, the
ratio of broad money to GDP more appropriately acts as a proxy for the depth of nancial
institutions in a given country but ignores the depth of nancial markets in the same. A
more robust measure of nancial depth should also account for the depth of nancial markets
in a given country by incorporating the level of a country's domestic private credit, stock
market capitalization and bond market capitalization relative to a country's GDP (King and
Levine, 1993; Chinn and Frankel, 2005; Cihak, Demirguc-Kunt, Feyen and Levine, 2012).
A currency's exchange rate re
ects the degree to which a currency is a stable source
of value.14The intuition is straightforward: if a nancial asset is held in a currency that
depreciates, then its holders will be repaid in currency worth relatively less than it had been
at the time the asset was initially invested in; as a consequence, the nancial asset will yield
less of a return than was originally anticipated by the investor.
The most ecient way of measuring the stability of a currency is to measure its exchange
rate against the Special Drawing Rights (SDR), an international reserve asset created by the
IMF (Chinn and Frankel, 2005; Eichengreen, Chitu and Mehl, 2016). The value of the SDR
is determined by a weighted average of the currencies where the assigned weights correspond
to a currency's importance in global nance and trade. The exchange rate of a currency to
the SDR is calculated on the basis of its dollar exchange rate; that is, a given currency is
13Typically, M3 is preferred as a measure of broad money but where data on M3 is not available, M2 has
been utilized instead (King and Levine, 1993).
14Thus, later in section 2.3, we view our variable capturing the exchange rate of the dollar as operational-
izing the latter's stability.
39

rst converted to dollars and then to SDRs using the latter's value against the dollar. As a
result, this method has the advantage of comparing the value of the dollar not only relative
to any currency of notable importance in global nance and trade but also in proportion to
this importance. For instance, if the dollar depreciates against any of the currencies included
in the currency basket comprising the SDR, the dollar will depreciate against the SDR and
in proportion to the importance of the currency it has depreciated against, oering a more
robust measure of currency stability.
Though goods trade is typically associated with the private sector, the currency compo-
sition of a country's foreign exchange reserves is understood to re
ect international trade
relations. This re
ection is mostly owing to eorts by central banks in developing and
emerging market economies to manage the domestic exchange rate. Thus, central banks in
developing and emerging market economies are prompted to hold in its foreign exchange
reserves that currency which is commonly used in international trade transactions.
Recently, research on the determinants of an international currency has attributed a
greater weight to an issuing country's overall size in international trade networks than indi-
vidual trade ties. That is, countries are increasingly less likely to hold the currencies of their
trade partners simply on account of the partnership and more likely to hold the currency
of a country dominant in international trade networks. Empirical evidence of this reality
and especially in favor of the dollar is relatively easy to see: for instance, Goldberg and
Tille (2008) show that while dollar-use varies across countries, its prevalence is high even in
non-US transactions, so that the role of the dollar is larger than can be explained by the
number of US trade partners.
The use of the US dollar in non-US transactions can be explained by way of a network
eect whereby transaction costs are lowered by using the US dollar as a vehicle currency , or
as a medium of exchange between currencies (Krugman, 1979; Portes, Rey and Oh, 2001;
40

Devereux and Shi, 2013). The network eect of currencies is essentially an economies of
scale argument: as the number and scale of a country A's international trade ties grows, so
do the cost advantages to its trade partners of transacting in country A's currency. That is,
if countries BandCare weaker trade partners of each other than either are of country A,
then transacting in country A's currency allows countries BandCto avoid costs associated
with exchanging their own respective currencies for country A's currency.15
Additionally, using the currency of the largest issuer is convenient in the sense that
it obviates the need, which is sometimes costly, of acquiring information about multiple
currencies. Thus, assuming that countries prefer to use a currency that minimizes transaction
costs and that the latter is inversely proportional to the volume of transactions, then countries
will opt to use the currency of leading trade nations and this result will be self-reinforcing as
more countries follow suit (Krugman, 1979, 1984). A common proxy for a country's role in
international trade networks is its economic size, or share of world GDP (Chinn and Frankel,
2005).
Portes, Rey and Oh (2001) demonstrate the network eects of a currency by focusing
on foreign exchange markets, assuming transactions occur through a nancial intermediary,
and modeling gains as those costs that would otherwise have to be spent on additional
intermediary services if a vehicle currency was not used. She nds that a country's position
in international trade networks largely determines whether its currency serves as a vehicle
currency. Devereux and Shi (2013) conrm that Rey's results do not depend on the existence
of a nancial intermediary since vehicle currencies facilitate international trade by lowering
the average cost of currency trade between countries.
Because the benet of adopting a currency is partly owing to the network made up of
other users, it is possible for the network to outlast its original generating source because
15This is essentially a restatement of the argument made in Krugman (1984).
41

the cost-benet of ubiquity acts as a buer against loss. That is, network eects may give a
currency inertia orpersistence (Trin, 1960; Krugman, 1979, 1984; Eichengreen, Mehl and
Chitu, 2017). To illustrate the dynamic, consider a simple thought experiment: let xbe
the benet to country Bfrom holding country A's currency owing to the latter's intrinsic
characteristics and let ybe the benet from holding country A's currency simply because
it is widely used. Then, country Bwill continue to hold country A's currency even if x
declines, so long as xy.16
CountryB's willingness to continue holding country A's currency even after xdeclines
indicates that a particular currency (or, in this analysis, US Treasury securities) exhibits
inertia. Importantly, though network eects may elicit an inertial force in a currency, they
are not unique in this regard. Inertia may also be caused by habit formation or lack of better
alternatives (Eichengreen, Mehl and Chitu, 2017). We might even consider that the Safe
Assets theory has added another reason that an asset may exhibit inertia: it is perceived
to be \safe." Where some measure or share of a currency is a regressand in an empirical
exercise, its inertia is typically captured through a lagged dependent variable included as a
regressor (Chinn and Frankel, 2005; Chitu, Eichengreen and Mehl, 2014; Eichengreen, Mehl
and Chitu, 2017).
One critique of explaining currency use by way of the latter's inertia is that inertia is
exactly what is to be explained.17That is, practically speaking, including a variable rep-
resenting inertia as a regressor simply captures whatever variation in the level or share of
16The continued dominance of the pound sterling as a global currency during World War II serves as a
prime example of this phenomenon (Aliber, 1966; Krugman, 1979).
17Echoing this sentiment, Edwin M. Truman writes of Chinn and Frankel (2005), \The Chinn and Frankel
results conrm the well-known observation that there is substantial inertia in international reserve holdings.
The issue is what explains this inertia" (Chinn and Frankel, 2005, p.329). Similarly, Prem (1994) asserts, \We
take issue with the convention[al] approach to inertia, on grounds that by dismissing anomalies as remnants
of past conventions, it merely serves to justify the status quo and hardly oers any real explanation. It fails
to specify the dynamics|how long before a break in the old convention might be expected, and in particular,
what might cause this transition" (p.33).
42

currency holdings that cannot be explained by the variation in the explanatory variables.
Nevertheless, empirical studies conrm that what has been designated inertia is indeed an
important determinant of the currency composition of foreign exchange reserves. Chinn and
Frankel (2005) empirically assess determinants of reserve currencies to determine whether
the dollar would maintain its status as a key international reserve currency following the
euro's introduction. The authors utilize a logistic model and test the following determinants
of reserve currency status: inertia, income share and nancial depth of an issuing country,
the exchange rate, and network externalities (i.e., network eects). Their results suggest that
inertia exercises a statistically signicant ( p<0:10) and large eect on the currency compo-
sition of reserves; specically, inertia propels forward approximately 90% of each currency
in foreign exchange reserves from one year to the next.
It is also worth mentioning that Chinn and Frankel's (2005) ndings also support the
statistical signicance at the 10% level of an issuer's income share and a currency's exchange
rate as determinants. Interestingly, nancial depth measured through a series of nancial
market capitalization to GDP ratios is not only statistically insignicant but also possesses
an unexpected negative sign.
In sum, the demand for US Treasury securities can properly be modeled after the de-
mand for the US dollar in foreign exchange reserves. This equivalency is owing to the stylized
fact that that most foreign-held US Treasury securities are held by ocial institutions (i.e.,
central banks) as foreign exchange reserves. Generally speaking, foreign ocial institutions
seek currencies as foreign exchange reserves that are stable and minimize transaction costs;
thus, the determinants of the currency composition of foreign exchange reserves (or, equiv-
alently, foreign demand for US Treasury securities) are those that signal these qualities in a
currency. Empirical exercises testing whether a variable can be considered a determinant of
the currency composition of foreign exchange reserves typically introduces the variable as a
43

regressor among standard determinants: nancial depth of an issuing country, a currency's
exchange rate, the issuer's position in international trade networks, and inertia of currency
use.
2.3 Data and Descriptive Statistics
2.3.1 Dependent Variable: Foreign Holdings of Long-term US Treasury Secu-
rities (USTS it)
Long-term US Treasury security holdings for country iat timetis the dependent variable
in our empirical analysis and is represented by the variable USTS it. To recall, our goal is to
determine what motivates developing and emerging market economies to hold US sovereign
debt where we hypothesize the reason to be demand for dollar credit. Thus, while it is not
necessary that we focus on long-term Treasury securities since our analysis seeks to under-
stand the motivations of holding US sovereign debt generally, doing so oers an advantage.
Foreign holdings of Treasury securities are overwhelmingly in long-term securities and for-
eign ocial investors hold the largest proportion of the latter as foreign exchange reserves.
Thus, focusing exclusively on long-term Treasury securities allows us to better account for
unobserved heterogeneity across countries by controlling for the well-dened determinants
of currency composition in foreign exchange reserves.18
The link between long-term Treasury securities and foreign exchange reserves can be
made stronger still if we could focus exclusively on that portion of the former that are
held by foreign ocial institutions. Unfortunately, however, country-level data on foreign
ocial holdings of long-term Treasury securities is unavailable, likely re
ecting the highly
18Specically, we draw from the models of Chinn and Frankel (2005) and Eichengreen, Mehl and Chitu
(2017).
44

condential nature of currency compositions of foreign exchange reserves by country. Thus,
we follow Terada-Hagiwara (2011) in using long-term Treasury securities as a proxy for
foreign ocial holdings.
Data on US Treasury security holdings by country is from Bertaut and Judson (2017)
which oers monthly U.S. cross-border securities holdings for 92 countries and 6 regions.
The Bertaut and Judson data is a substantial improvement from the more conventional
source for this data, the Treasury International Capital (TIC) System. Most importantly,
the Bertaut and Judson data reports country holdings of US Treasury securities adjusted
for valuation eects, or changes in reported positions resulting from a simple change in the
market price of a security. Despite this strength, however, it is important to point out that
one weakness of the Bertaut Judson data (shared with its TIC counterpart) is that it suers
from custodial bias whereby custodians of an asset are assumed to be its owner (Bertaut and
Judson, 2014, p.10). Because country-level positions in US Treasury securities are reported
on a geographical basis (i.e., an asset is assumed to reside with its owner), a weakness of our
data is that some country holdings may be underestimated.19
2.3.2 Independent Variable of Interest: Outstanding Dollar Credit ( CRED it)
Outstanding dollar credit for country iat timetis the independent variable of interest
and is represented by the variable CRED it. Dollar credit to developing and emerging market
economies is specically measured as total dollar credit to non-bank borrowers and is taken
from the Bank for International Settlements (BIS).
Country-level data on outstanding credit is available at dierent frequencies and for
19Bertaut and Judson (2014) oer the following example to illustrate the problem of custodial bias: "If
a Russian investor chooses to hold the securities with a custodian in the United Kingdom, the liability
would be recorded against the United Kingdom rather than Russia. As a result, the total liability position
worldwide is correct, but the geographic allocation is not" (Bertaut and Judson, 2014, p.10). Further, it
is useful to point out that while some country positions may be underestimated, we do not anticipate any
being overestimated as no country included in our data set houses large custodians.
45

a number of sources including the World Bank, the BIS, and the International Financial
Statistics database published by the IMF. Data on the currency composition of this credit,
however, is only provided by the BIS and for non-bank borrowers in each country. We utilize
the most comprehensive measure on foreign-currency credit oered by BIS which combines
international debt securities, cross-border bank loans and local bank loans.
Because country-level data on local bank loans in foreign currency is dicult to procure,
BIS only reports outstanding foreign currency credit for fourteen developing and emerging
market economies, one of which (Saudi Arabia) is only included from 2013 onwards.20Thus,
data on outstanding dollar credit contributed the most to constraining the cross-sectional
units of our rst panel to thirteen countries.
2.3.3 Independent Control Variables
2.3.3.1 Relative Financial Depth of the US ( DEPTH t)
Financial depth of the US at time trelative to what we will call its currency competitors |
the Euro Area (EA), the United Kingdom (UK), and Japan (JP)|is represented by the
variableDEPTH t.21Including this variable among our regressors controls for the possibility
that the US attracts investment in its sovereign debt because it possesses the deepest nancial
markets among a set of feasible alternatives (Portes and Rey, 1998; Papaioannou and Portes,
2008; Chinn and Frankel, 2005; Eichengreen, Mehl and Chitu, 2017).
Financial depth for each currency competitor is dened and measured as the sum of a
countryj's domestic private credit ( DPRIV jt), stock market capitalization ( SCAP jt), and
bond market capitalization ( BCAP jt) relative to GDP ( GDP t) at timet.22Domestic private
20These fourteen economies are: Argentina, Brazil, Chile, China, India, Indonesia, Malaysia, Mexico,
Republic of Korea, Russia, Saudi Arabia, South Africa, Taiwan, and Turkey.
21See footnote 8 for our rationale in selecting these currencies to be the US' \currency competitors."
22To clarify, the subscript iwill be used to denote a developing and emerging market economy and the
46

credit is operationalized using all credit to the private non-nancial sector from banks.
Stock market capitalization typically refers to a specic company and is simply the prod-
uct of a company's outstanding total shares and the current market price of one share.
Broadened to the country or regional level (as it is in this analysis), stock market capitaliza-
tion refers to the sum of outstanding shares listed on a stock exchange(s) multiplied by their
respective current market prices. Stock market capitalization for the US, the UK, Japan,
and the Euro Area is calculated using the New York Stock Exchange, the London Stock
Exchange, Japan Exchange Group Inc., and Euronext, respectively.23Data on these stock
market exchanges is obtained from the World Federation of Exchanges. Quarterly data on
bond market capitalization proved dicult to secure so we operationalized this variable using
the commonly utilized proxy of outstanding domestic debt securities issued by nancial and
non-nancial corporations; data is from the BIS.
Once nancial depth for the US and each of its currency competitors is ascertained,
relative nancial depth for the US is then measured by the dierence between its measure
for nancial depth and the maximum value of nancial depth found among its currency
competitors. DEPTH tis therefore given by,
DEPTH t=FIN US;tmaxfFIN EA;t;FIN UK;t;FIN JP;tg (2.1)
where
FIN jt=DPRIV jt+SCAP jt+BCAP jt
GDP jt
The subscript jdenotes one of the US' currency competitors (EA, UK, or JP). Thus, the
subscriptjwill be used to denote a country (or region) that is a US currency competitor.
23Japan Exchange Group, Inc. was formed by the merger of the Tokyo Stock Exchange and the Osaka
Securities Exchange. Thus, prior to 2014, stock market capitalization for Japan was measured as the sum
of stock market capitalization for the Tokyo Stock Exchange and the Osaka Securities Exchange.
47

proper way to understand, say, a positive value XforDEPTH t, is as follows: in period t,
the US nancial depth to GDP ratio is Xpercentage points above the maximum nancial
depth to GDP ratio found between the Euro Area, the United Kingdom, and Japan.
Perhaps surprisingly, it should be noted that the values for DEPTH tare negative across
all time periods. That is, the US is not the most nancially developed country among its
currency competitors during the observed time period; in fact, it is the least nancially
developed of this group for much of the observed period. This result remains if we utilize
the ratio of broad money to GDP as a proxy for nancial depth instead. Though it is clear
that nancial depth of the US is not signicant as a determinant of US Treasury security
holdings in an absolute sense, we still include the variable as a control in case variations in
relative US nancial depth can help to explain variations in holdings.
2.3.3.2 Stability ( EXCH t)
Stability at time tis represented by the variable EXCH tand refers to the stability of
the currency in which a nancial asset is denominated at time t. Stability is included in our
empirical model to control for the possibility that developing and emerging market economies
simply prefer to hold nancial assets that are stable sources of value. We dene EXCH tas
the USD to SDR exchange rate at time t.
To capture the probable reality that changes in currency stability will not immediately
yield changes in holdings of nancial assets denominated in those currencies, we utilize a
5-year moving average of the USD to SDR exchange rate in our construction of this variable.
Data is from the Federal Reserve Economic Data (FRED) database published by the Federal
Reserve Bank of St. Louis.
48

2.3.3.3 Network Eects ( NETW t)
Network Eects is represented by the variable NETW tand is representative of the eco-
nomic size of the US (operationalized by the ratio of US GDP to world GDP) relative to the
economic size of its currency competitors (operationalized as the ratio of a country j's GDP
to world GDP). The introduction of NETW tas a regressor in the model controls for the
possibility that the variation in holdings of US Treasury securities in our sample of countries
is owing to changes in the relative economic size of the US.
The relative advantage of the US in terms of its network eect is given by the dierence
between its economic size and the economic size of its currency competitors. That is,
NETW t=RATIO US;tmaxfRATIO EA;t;RATIO UK;t;RATIO JP;tg (2.2)
where
RATIO jt=Countryj's GDP at time t
World GDP at time t
Quarterly data on world GDP is unavailable and so had to be constructed by aggregating
quarterly GDP at market exchange rates for individual countries, based on data availability.
Data is taken from the International Financial Statistics database published by the IMF,
the FRED database, and CEIC data. World GDP is comprised of 53 countries, 28 of which
are classied as advanced economies by the International Monetary Fund.24
24The countries used to calculate world GDP are as follows: Albania, Australia, Brazil, Bulgaria, Canada,
Chile, China, Colombia, Costa Rica, Croatia, Czech Republic, Denmark, Hungary, Iceland, India, Indonesia,
Japan, Mexico, Morocco, New Zealand, Norway, Poland, Republic of Korea, Romania, Singapore, South
Africa, Sweden, Switzerland, Taiwan, Thailand, Tunisia, Turkey, the United Kingdom, the United States,
and Euro Area countries.
49

2.3.3.4 Inertia ( USTS i;t1)
Inertia or persistence in holdings of US Treasury securities for country iis represented
by theUSTS itvariable lagged by one period (i.e, USTS i;t1). Including this variable among
our regressors controls for the possibility that current holdings of US Treasury securities are
simply re
ections of past holdings.
One critique, mentioned previously, of our inclusion of a variable representing inertia of
US Treasury securities holdings is that inertia is exactly what is to be explained. While we
agree thatUSTS i;t1is something of a catchall that is unsatisfactory for its vagueness, we
also consider it plausible that at least some of the variation in US Treasury security holdings
is owing to one if not more of the factors manifested in an inertial quality. Prudence therefore
requires that these factors be controlled for, though to reconcile the con
ict we rst estimate
a regression equation omitting USTS i;t1as a control variable.
2.3.3.5 Yield( YIELD t)
Yield at time tis represented by the variable YIELD t. Including this variable among
our regressors controls for the possibility that the variation in our dependent variable is
simply owing to US Treasury securities oering investors a higher yield than other feasible
alternatives (i.e., the yield on long-term sovereign debt of US currency competitors). Though
return considerations of foreign ocial institutions are not typically included as a determi-
nant of the currency composition of foreign exchange reserves in relevant empirical models,
it is plausible that these considerations become important in the selection between safe as-
sets. Data is obtained from FRED. To capture the yield of long-term US Treasury securities
relative to feasible alternatives, YIELD tis calculated as the dierence between the yield on
long-term US Treasury securities and the maximum yield on long-term government bonds
found among the Euro Area, United Kingdom, and Japan. YIELD tis therefore given by,
50

YIELD t=GOV US;tmaxfGOV EA;t;GOV UK;t;GOV JP;tg (2.3)
where
GOV jt= Nominal yield at time tfor country j's 10-year government bond
2.3.4 Summary
Table 3.1 contains descriptive statistics for the main variables. The sample is a balanced
panel for the period 2003-2016 at quarterly intervals. For country-variant variables ( USTS it,
USTS i;t1,CRED it), data is collected for 13 developing and emerging market economies
based on data availability.25
25The countries included are Argentina, Brazil, Chile, China, India, Indonesia, Malaysia, Mexico, Republic
of Korea, Russia, South Africa, Taiwan, and Turkey.
51

Table 1: Descriptive Statistics
Panel A (China Included) Panel B (China Omitted)
Statistic Mean St. Dev. Min Max Mean St. Dev. Min Max
USTS holdings ( USTS it) 105 :23 249 :05 0 :15 1293 :57 43:64 57:37 0:15 246:67
Inertia (USTS i;t1) 102 :92 246 :40 0 :15 1293 :57 42:59 56:57 0:15 246:67
Dollar credit outstanding ( CRED it) 91:45 102 :62 6 :31 689 :58 62:73 54:21 6:31 248:78
Exchange rate ( EXCH t) 1 :49 0 :07 1 :33 1 :56
Relative nancial depth ( DEPTH t)402:57 50 :40509:10288:26
Relative world GDP share ( NETW t) 31:99 3 :21 26 :28 39 :49
Relative yield on USTS ( YIELD t)0:45 0 :642:14 0 :83
Notes : USTS holdings ($B) is obtained from Bertaut and Judson (2017) and is adjusted for valuation eects. USTS holdings is lagged by one period to
account for inertia (or persistence) in the level of holdings across countries. Dollar debt outstanding ($B) is from the Bank for International Settlements
(BIS). The exchange rate ($/SDR) is a 5-year moving average of the USD per SDR exchange rate and data is from the Federal Reserve Bank of St. Louis
(FRED). Relative nancial depth (pp) is the percentage point dierence between the nancial depth of the US and the maximum nancial depth found
among the Euro Area, Japan, and the United Kingdom (henceforth referred to as currency competitors); data is from the BIS and the World Federation of
Exchanges. Relative world GDP Share is the percentage point dierence between the GDP share of world total GDP for the US and the maximum share
found among its currency competitors; data is from the IMF International Financial Statistics and CEIC data . Relative yield on USTS is the percentage
point dierence between 10-year government bond yields for the US and its currency competitors; data is from FRED. Summary statistics for EXCH it,
DEPTH t,NETW t, andYIELD tdo not change from Panel A to Panel B. For detailed denitions and sources, see Table 12 in Appendix A.
52

2.4 Econometric Model
The econometric model which will serve as the basis for the analysis when we do not
include inertia of US Treasury security holdings ( USTS i;t1) as an explanatory variable is
given by,
USTS it=
CRED it+x0
t+i+t+uit (2.4)
whereUSTS itis holdings of long-term US Treasury securities for country iin period
tandCRED it|our variable of interest|is total outstanding dollar credit to non-bank
borrowers for country iin periodt. The parameter
therefore measures the magnitude of
the relationship between outstanding dollar credit and holdings of US Treasury securities.
Further,x0
tis a vector comprised of our control variables listed in Section 2.3.3, iis an
unobserved eect capturing all unobserved, time-invariant factors aecting our dependent
variable, and tis a complete set of time eects accounting for secular changes in US Treasury
security holdings. If our hypothesis that US Treasury securities serve as collateral to dollar
credit is correct, then we would expect the coecient on CRED itto be statistically signicant
and positive.
Further, given our eorts to tailor foreign holdings of US Treasury securities to a model of
currency composition in foreign exchange reserves, it is perhaps useful at this point to clarify
how a positive and statistically signicant coecient on CRED itrelates to the public and
private domains. To this end, consider Table 2 illustrating the distinction between private
and public holdings of US Treasury securities and private and public dollar borrowing.
Using Table 2, we can say that our analysis practically applies to cases (A) and (B); that
is, a statistically signicant coecient on CRED itwill indicate that there is a statistically
signicant relationship between private and public dollar borrowing and public holdings of
53

Table 2: Public versus Private Borrowers and Holders
Dollar Borrowers
Public Private
Collateral ProvidersPublic A B
Private C D
Notes :Collateral providers refers to holders of US Treasury
securities. Case (A) refers to public dollar borrowers and public
collateral providers. Case (B) refers to private dollar borrowers
and public collateral providers. Case (C) refers to public dollar
borrowers and private collateral providers. Case (D) refers to
private dollar borrowers and private collateral providers.
US Treasury securities in the developing and emerging market economies included in our
sample (all other factors held constant.)26
To continue explanation of our empirical model, when our inertia variable ( USTS i;t1) is
omitted, the model is best estimated simply using a rst dierences estimator. Given that
the unit of analysis is developing and emerging market economies, we cannot reasonably
adopt the crucial assumption needed for a random eects model that we have suciently
controlled for all sources of variation in the dependent variable such that there exists no i
wherebyCov(i;xit)6= 0. For instance, perhaps the form of government in a developing or
emerging-market economy is correlated with dollar credit and the currency denomination of
nancial assets and liabilities. The main point here is that though we cannot be certain of
the omitted, country-specic factors that would bias our results, the fact that such factors
are likely to exist given such large geographical units of observation renders the use of a
26We should note that because our data on US Treasury securities does not distinguish between public and
private foreign holdings of US Treasury securities by country, our analysis technically also applies to cases (B)
and (D). But, as previously mentioned, the overwhelming majority of long-term US Treasury securities are
held by Central Banks in foreign exchange reserves, which we can exploit to restrict the practical application
of our results to cases (A) and (B).
54

random eects estimator inappropriate for our current analysis.
A consideration in choosing between the xed eects and rst dierences model is that the
number of cross-sectional units (countries) is less than the time periods per cross-sectional
unit (i.e.,N <T ), meaning that non-stationarity of the data presents a more serious chal-
lenge to the estimation process. We thusly stationarize our variables. A standard test for
stationarity of a time series, say y, is the Augmented Dickey Fuller Test (ADF) which tests
the estimated coecient ^in the model,
4yt=+t+yt1+hX
l=1
l4yt1+ut (2.5)
whereh=p1 andtare the deterministic components of the model. The goal is to
determine whether stationarity of the series has been achieved for plags ofy. While the
ADF is suitable for pooled cross-sections, panel data requires that we establish stationarity
of the time series for each cross-sectional unit N. That is, equation (2.5) becomes,
4yit=it+iyi;t1+hX
l=1
il4yi;t1+uit (2.6)
Both Levin, Lin and Chu (2002) and Im, Pesaran and Shin (2003) consider equation
(2.6) above and propose applying the ADF test to the null hypothesis H0:it= 0, or
that the series has a unit root.27Table 3 presents our results from performing the ADF,
Levin-Lin-Chu, and Im-Pesaran-Shin tests to all of our main variables in equation (2.4).
The test results presented favor adopting a rst-dierences model since our variables in
both panels (with and without China included as an observation) are integrated of order one,
or I(1), suggesting that rst-dierencing the series is likely to result in stationarity. Further,
inspection of the residuals generated by estimating the original model given in equation (2.4)
27The main dierence between the Levin-Lin-Chu and Im-Pesaran-Shin tests is that in the former, iis
assumed to be the same for all cross-sectional units whereas in the latter, iis permitted to vary.
55

Table 3: Panel unit root tests
Level Augmented Dickey Fuller Levin-Lin-Chu Im-Pesaran-Shin
1st Dierence Test Statistic Test Statistic Test Statistic
USTS it 3:246 1 :870 2 :009

USTS it 3:96017:80813:938
CRED it 3:287 8 :289 4 :278

CRED it 4:10214:55322:025
EXCH t 6:2610:7162:726

EXCH t 6:5373:19414:766
DEPTH t10:5402:1329:285

DEPTH t10:81620:91818:032
NETW t 9:4841:994 1 :068

NETW t9:72817:9814:663
YIELD t 5:8554:1862:696

YIELD t5:72921:42918:867
Notes : Lag orders are chosen to minimize the Bayesian information criterion. Statistical signif-
icance of the test statistic indicates rejection of the null hypothesis that the series follows a unit
root process.
p<0.001;p<0.01;p<0.05
56

for panel A and B indicates that uitin both cases follows a random walk process so that rst
dierencing equation (2.4) is likely to remove serial correlation from our error term.
Thus, to determine the eect of dollar credit holdings on US Treasury security holdings
when inertia of the latter is not included as an explanatory variable, we estimate the following
model:
4USTS it=
4CRED it+4×0
t+4t+4uit (2.7)
Equation (2.7) is simply the result of time-dierencing equation (2.4), which necessarily
removes the unobserved heterogeneity, i.
When inertia or persistence of our dependent variable, US Treasury security holdings
(USTS i;t1), enters into our analysis, equation (2.4) becomes,
USTS it=USTS i;t1+
CRED it+x0
t+i+t+uit (2.8)
Naturally, transforming equation (2.8) by time-dierencing prior to estimation may seem
like an appropriate strategy given that panels A and B are rst-dierence stationary and that
the residuals generated by estimating equation (2.4) follow a random walk process. While
rst-dierencing equation (2.8) is certainly needed for the preceding reasons, introduction
of a lagged dependent variable among our regressors requires that we slightly modify our
former approach. To elucidate, consider rst dierencing equation (2.8) which results in the
following model,
4USTS it=4USTS i;t1+
4CRED it+4×0
t+4t+4uit (2.9)
Notice that using OLS to estimate equation (2.9) will not result in a consistent estimation
sinceUSTS i;t1is used to generate the residuals ut1and, by extension, 4uit; thus, the
57

requisite assumption that our residuals be uncorrelated with the regressors in the model, is
violated. We therefore adopt the method proposed by Anderson and Hsiao (1982) whereby
4USTS i;t2is used to instrument for 4USTS i;t1in equation (2.9).28
2.5 Main Results
Our results are presented in Table 4 along with robust standard errors (clustered at the
country level for estimates reported in columns [2]-[4]). Each specication uses a full set of
time dummies among the explanatory variables to control for any variation in holdings of
US Treasury securities among developing and emerging-market economies that is owing to
a time-dependent eect.
28We follow the procedure outlined in Woodridge (2010) according to which we must conrm that the
proposed variable ( 4USTS i;t2) serves as a proper instrument for the variable in question ( 4USTS i;t1).
This determination can easily be made by estimating the model, 4USTS i;t1=4USTS i;t2+uitfor
i= 1;2;:::;N: Results from estimation of the preceding model for panels 1 and 2 are reported in Table 16
of Appendix A and indicate that 4USTS i;t2is indeed a good instrument for 4USTS i;t1in both cases.
58

Table 4: Regression Results
Dependent variable: USTS i;t
Panel A (China Included) Panel B (China Omitted)
Pooled First Dierences Anderson-Hsiao Pooled First Dierences Anderson-Hsiao
OLS OLS IV OLS OLS IV
(1) (2) (3) (4) (5) (6)
USTS i;t1 0:7240:636
(0:115) (0 :153)
CRED i;t 2:0650:2780:1120:442 0 :2160:180
(0:40) (0 :05) (0 :05) (0 :32) (0 :05) (0 :05)
EXCH t 23:796 74 :41677:495 13 :69922:28387:533
(421:21) (142 :34) (100 :56) (131 :35) (123 :77) (137 :90)
DEPTH t0:6380:0440:046 0 :0300:0520:027
(0:36) (0 :12) (0 :12) (0 :15) (0 :03) (0 :03)
NETW t11:692 2 :043 1 :042 1 :2370:534 0 :295
(8:499) (2 :41) (1 :80) (3 :92) (0 :54) (0 :56)
YIELD t8:3995:3006:603 5 :6093:76 4 :194
(12:943) (10 :13) (9 :84) (2 :61) (3 :07) (2 :49)
Observations 728 715 689 672 660 636
R20.640 0.184 0.580 0.257 0.205 0.319
Adjusted R20.610 0.115 0.544 0.190 0.131 0.256
Notes: Pooled OLS regression in columns 1 and 4, with robust standard errors in parentheses. First-dierences OLS regression in columns 2
and 5. Columns 3 and 6 use the instrumental variables method of Anderson and Hsiao (1982); we instrument for USTS i;t1using a two-period
lag. Year dummies are included in all regressions. Robust standard errors clustered by country in parentheses for First Dierences OLS and
Anderson-Hsiao IV. For detailed data denitions and sources, see Table 12 in Appendix A.
p<0.001;p<0.01;p<0.05
59

Column (1) of Table 4 reports the results from estimating a standard pooled OLS re-
gression including all explanatory variables except inertia ( USTS i;t1) for panel A (where
observations for China are included.) Though the pooled OLS estimates cannot technically
be correct, they can still provide a useful baseline.29The coecient on the main variable
of interest, outstanding dollar credit ( CRED it), is statistically signicant ( p <0:001) and
positive, indicating that increases in holdings of US Treasury securities is associated with
an increase in outstanding dollar credit. Specically, a $1 billion increase in outstanding
dollar credit is associated with a $2.1 billion increase in US Treasury security holdings, hold-
ing other factors xed. This result supports our hypothesis but should be viewed with a
considerable degree of skepticism for reasons outlined in Section 2.4.
In addition to the main variable of interest, the nancial depth of the US relative to its
currency competitors ( DEPTH t) is signicant at the 5% level but the coecient estimate
has the \wrong" sign. Our coecient estimate for DEPTH tindicates a negative relationship
between relative US nancial depth and developing and emerging economy holdings of US
Treasury securities; specically, a 1 percentage point increase in the relative nancial depth
of the US is associated with a $0.64 billion decrease in US Treasury security holdings among
developing and emerging economy countries. Indeed, the coecient estimates on all of
our control variables suggest counterintuitive relationships, though with the exception of
CRED itandDEPTH t, no other coecient estimates from the pooled OLS are statistically
dierent from zero.
These counterintuitive results can be owing to a number of reasons including the omission
of an important explanatory variable, multicollinearity among regressors, and measurement
error. In our analysis, it is likely that the direction of the relationship between our control
29When stationarity of the variables is either achieved or not a concern in the case of a short panel, then
pooled OLS is especially useful: when compared to rst dierences or random eects, pooled OLS estimates
provide insight into the degree of bias arising from unobserved heterogeneity.
60

variables and USTS itindicated is possibly owing to non-stationarity of the series (the conse-
quences of which are discussed in detail in Section 2.4), non-inclusion of a persistence eect
of US Treasury security holdings (represented by our variable USTS i;t1), the inclusion of
China in our rst panel which is an outlier both in terms of its holdings of US Treasury
securities and outstanding dollar credit, failure to remove unobserved heterogeneity, and/or
misguided economic theory. Before commenting further, we continue to our results from
estimating a rst-dierenced version of equation which necessarily resolves some of these
issues.
Column (2) presents our results from estimating equation (2.7) with robust standard
errors clustered by country. The coecient on CRED itindicates that there is a positive,
statistically signicant ( p<0:001) relationship between changes in a country's outstanding
dollar credit and holdings of US Treasury securities; specically, a $1 billion increase in
outstanding dollar credit is associated with a $0.28 billion increase in US Treasury security
holdings, all other factors held constant. Thus, our results from estimating equation (2.7)
conrm that the relationship between outstanding dollar credit and US Treasury securities
estimated by pooled OLS holds even after removing unobserved heterogeneity. Importantly,
compared to the results from pooled OLS, dierencing has reduced the sensitivity between
changes in holdings of outstanding dollar credit and US Treasury securities by about $0.7
billion; that is, the same increase in outstanding dollar credit is associated with less of an
increase in US Treasury security holdings after unobserved heterogeneity is removed from
the estimating equation.
The rst-dierenced equation yields no statistically signicant control variables and this
includes our formerly signicant variable DEPTH t. While the coecient estimates on most
of our control variables continue to have the \wrong" sign, the expected sign on our variable
NETW tis yielded through rst-dierencing. That is, the dierence between the ratio of
61

GDP to world GDP for the US and its currency competitors is associated with an increase
in US Treasury Security holdings among developing and emerging economies included in
our sample. One nal observation is that, though not statistically signicant, the coecient
estimate on our stability variable EXCH tis relatively large and in an unexpected direction,
indicating that an increase in the weakness of the dollar relative to the SDR is associated
with an increase in US Treasury security holdings among developing and emerging economy
countries in our sample.
Results from introducing a persistence eect of US Treasury Securities are presented in
column (3). As discussed earlier, we instrument for the representative variable USTS i;t1
with a second lag of USTS itto avoid a violation of the requisite assumption that our re-
gressors be uncorrelated with our error term. The coecient on USTS i;t1indicates that
holdings of US Treasury security holdings indeed demonstrate strong and statistically signif-
icant (p<0:001) inertia. Specically, a $1 billion increase in US Treasury security holdings
is associated with a $0.72 billion increase in the next time period, all other factors held
constant.
Importantly, our results demonstrate that even after controlling for inertia in US Treasury
Security holdings, the coecient on our main variable of interest, CRED it, continues to
be positive and statistically signicant. However, CRED itdisplays slightly less statistical
signicance, indicating that inertia has absorbed some of the variation in USTS it, formerly
attributed to outstanding dollar credit. Our variable CRED itis also reduced in magnitude
when a persistence eect of US Treasury securities is added to the model; specically, the
estimated increase in US Treasury security holdings resulting from a $1 billion increase
in outstanding dollar credit is $0.11 billion, or approximately $0.17 billion lower than the
rst-dierences estimates, all other factors held constant.
The relationship between outstanding dollar credit and US Treasury security holdings
62

in our sample of developing and emerging market economies is non-trivial. Our descriptive
statistics reported in Table 1 indicate that, on average, countries in our sample hold ap-
proximately $91.5 billion dollars in outstanding dollar credit. Even when China is omitted
from the sample, the mean level of outstanding dollar credit is still large at approximately
$62.7 billion. More telling, however, is the growth rate of outstanding dollar credit over the
sample period. From 2003Q1-2016Q4, outstanding dollar credit in our sample of developing
and emerging market economies has grown by approximately $115 billion ($84.7 billion when
China is omitted). This growth has occurred at a higher rate in the later part of the sample
period (2010-2016) with countries in the sample adding on average about $14.5 billion per
year to their stock of outstanding dollar credit ($9.16 billion when China is omitted).
In 2016, developing and emerging market economies held roughly $3 trillion in US Trea-
sury securities. Our results indicate that if outstanding dollar credit among developing and
emerging market economies in our sample continues to grow at its more recent pace, then the
collateral needed to secure this credit will be roughly $188.5 billion in US Treasury securities
over the next decade.30That is, holdings of US Treasury securities among developing and
emerging market economies will increase by roughly 6% over the next decade.
Echoing previous studies, the statistically signicant coecient on our lagged dependent
variable indicates that inertia or persistence in US Treasury security holdings is of consid-
erable magnitude: a $1 billion increase in US Treasury security holdings in the previous
time period is associated with a $0.64 billion increase in the current time period.31To
put it simply, about half of US Treasury security holdings carried forward from one time
period to the next in developing and emerging market economies are propelled by inertia.
30To get this number, we simply multiply $14.5 billion in increased outstanding dollar credit per year by
13 (for the number of countries in the sample).
31Chinn and Frankel (2005) nd that inertia explains approximately 90% of the currency composition
of reserves. Eichengreen, Mehl and Chitu (2017) estimate a considerably lower but statistically signicant
eect of inertia that is closer to our estimate, at about 60%.
63

With the exception of our variable representing inertia, our control variables continue to be
statistically insignicant, though now the coecient on our variable representing stability
(EXCH t) possesses the conventionally expected sign. Interpreted causally, our results in-
dicate that a weaker dollar (measured against the SDR) yields a decrease in US Treasury
security holdings. Our coecient estimates for the variables representing relative US -
nancial depth ( DEPTH t) and relative US yield of long-term government bonds ( YIELD t),
however, continue to indicate an unexpected negative relationship with US Treasury security
holdings.
If conventional economic theory is to be our guide, the reasons underlying our unexpected
estimated signs on DEPTH tandYIELD tare possibly owing to a mistake in our specica-
tion, empirical strategy, and/or data set. We consider those factors that are plausibly related
to the current undertaking. Inspection of the correlation coecients and variance in
ation
factors (VIF) for the variables in our sample suggest that multicollinearity is unlikely to be
the cause of our unexpected signs on DEPTH tandYIELD t.32
Simultaneous-equations bias may be another reasonable concern if YIELD tandDEPTH t,
respectively, are determined simultaneously with the dependent variable. That is, if foreign-
held US Treasury security holdings are determined by and determine YIELD tand/or
DEPTH t, then our estimates will be biased and inconsistent. That portion of our variable
DEPTH twhich might reasonably be determined by foreign-held, long-term US Treasury
securities is bond market capitalization. But, because we have proxied for this dimension of
DEPTH twith outstanding debt securities issued by the private sector, it is quite unlikely
that relative US nancial depth would be determined by our dependent variable.
With regard to our variable YIELD t, while we recognize that the yield on long-term US
32A formal presentation of our correlation matrix and variance in
ation factors can be found in Tables 14
and 15, respectively, in Appendix A.
64

Treasury securities is determined by demand for the latter (i.e., by holdings of US Treasury
Securities), recall that our variable YIELD trepresents the yield of long-term Treasury
securities relative to the yield on long-term government bonds of US currency competitors.
Additionally, to the extent that holdings of US Treasury securities may in fact determine the
yield on long-term US Treasury securities, it is plausible that it does so with a lag whereas
in our model the variables DEPTH tandYIELD tare contemporaneous with each other.
Nevertheless, as an additional precautionary measure, we employ a Granger causality
test to assess whether the relative yield of US Treasury securities is determined by foreign
holdings of the same. Generally speaking, the Granger causality statistic allows us to test
whether past values of one variable, say Xt, is a signicantly useful addition to a model using
past values of Ytto forecast Yt. IfXtis indeed a signicantly useful addition (i.e., possesses
predictive content useful in forecasting Yt) beyond past values of Yt, then we conclude that Xt
\Granger causes" Yt. More specically, consider the following time-series regression model
with one predictor, Xt:
Yt=0+ipX
i=1Yti+iqX
i=1Xti+ut (2.10)
wherepandqare the lag order of YtandXt, respectively. Then, the Granger causality
statistic is simply an F-statistic generated to test the null hypothesis that the coecients i
in equation (2.10) are all zero.
As is well-known, the Granger causality test cannot determine contemporaneous causality
between two variables; that is, it cannot reveal information on whether the value of Xtin
periodtis related to Ytin periodt. But, applied to the current circumstance, it can provide
some useful insight because if YIELD tis potentially determined by USTS t, then we would
reasonably expect past values of the latter to provide predictive content to forecast the
65

former.
Our results from performing the Granger causality test are reported in Table 5 and indi-
cate that while the relative yield on US Treasury securities Granger causes foreign holdings
of long-term US Treasury securities, the reverse is not true. Because past values of USTS it
do not provide useful predictive content to forecast YIELD t, we are unconvinced that simul-
taneity between the two is the cause of the unexpected estimated coecients on our control
variables.
Table 5: Panel Granger Causality Test
Alternative Hypothesis F-statistic Direction of Granger-causality
USTS Granger-causes YIELD
for at least one country 1 :487 USTS!YIELD
YIELD Granger-causes USTS
for at least one country 2 :297YIELD!USTS

USTS Granger-causes
YIELD
for at least one country 1 :432
USTS!
YIELD

YIELD Granger-causes
USTS
for at least one country 2 :629
YIELD!
USTS
Notes : Signicance of the F-statistic reported in the second column supports rejection of the null
hypothesis in favor of the alternative hypothesis. For detailed data denitions and sources, see Table
12 in Appendix A.p<0.001;p<0.01;p<0.05
Another reason that can possibly explain why our coecient estimates on our variables
DEPTH tandYIELD tdo not possess the expected sign is the presence of outliers. Indeed,
Figure 5 clearly illustrates that China is an outlier in our sample in both its level of out-
standing dollar credit and holdings of US Treasury securities. Moreover, Figure 5 also seems
to indicate a strong positive relationship between the level of outstanding dollar credit and
66

US Treasury securities for China; thus, we should be concerned that the presence of China
in our sample is not only driving unexpected signs in our coecient estimates but also the
statistical signicance of our main variable of interest, CRED it.
Thus, in the next section we remove China from our dataset and repeat our empirical
exercise to determine how sensitive our results are to the exclusion of this outlier. As an
additional precautionary measure, we also repeat the empirical analysis outlined in section
2.4 using an alternative denition of nancial depth: the ratio of broad money (M3) to GDP.
2.6 Robustness Checks
Our rst robustness exercise entails the removal of China as an observation in our sample
and estimating equations (2.4), (2.7), and (2.9) again following the same procedure outlined
in section 2.4. Our results are reported in columns (4)-(6) of Table 4.
A rst observation is that while the estimated coecient on our main variable of interest
CRED itis not signicant in the pooled OLS regression reported in column (4), estimating
the rst-dierenced equation results in statistical signicance being gained at the .1% level.
Thus, it seems that when China is removed from the sample, unobserved heterogeneity biases
our pooled OLS results to a considerably larger extent than when it is included. Interestingly,
comparison of the IV estimates presented in columns (3) and (6) indicate that when China is
omitted from our sample, CRED itgains in statistical signicance ( p<0:001) and magnitude
by approximately $.07 billion. Specically, when China is removed from the sample, a $1
billion increase in outstanding dollar credit is associated with a $0.18 billion increase in US
Treasury security holdings, all other factors held constant.
The gain in statistical signicance of our outstanding dollar credit variable indicates that
though China holds a disproportionately high level of US Treasury securities and outstanding
67

dollar credit, the linkbetween the two is less strong for China than it is for other countries
included in our sample (i.e., outstanding dollar credit explains less of the variation in hold-
ings of US Treasury securities for China than it does for our other cross-sectional units.)
Relatedly, because the inertia of US Treasury security holdings seems to absorb more of the
variation in our dependent variable when China is included in the sample, it is possible that
inertia of US Treasury security holdings is relatively stronger for China.
The IV estimate of the coecient on YIELD tis now not only positive but also statisti-
cally signicant at the 5% level.33Specically, our results indicate that a 1 percentage point
increase in the relative yield of long-term US Treasury securities is associated with a sizeable
$4.2 billion dollar increase in holdings of the same among developing and emerging market
economies countries included in the sample. Coupled with the potential that inertia plays
a larger role in motivating Chinese holdings of US Treasury securities, the signicance of
the coecient on YIELD tfor panel B hints that China's motivation for accumulating US
Treasury securities is potentially unique and warranting of further research.
Notwithstanding YIELD t, the IV estimates of the coecients on our other control vari-
ables, reported in column (6), are not statistically signicant. Moreover, the coecient on
DEPTH tcontinues to be negative, indicating that increases in relative US nancial depth
continue to be associated with decreases in US Treasury security holdings, even after omit-
ting outliers from the sample. The prevailing view argues that if the existence of a safe
asset is taken as given (i.e., we do not concern ourselves with what makes an asset \safe"
but, rather, simply accept that investors believe certain assets possess this characteristic),
then the selection between safe assets is explained by the nancial depth of an issuing coun-
33A negative coecient on our variable EXCH reported in column (6) indicates that increases in the
dollar to Special Drawing Rights exchange rate (i.e., dollar depreciation) is associated with a decrease in
US Treasury security holdings. A positive coecient on NETW tindicates that increases in the US network
(measured as the ratio of US GDP to world GDP) relative to the network of its currency competitors is
associated with an increase in US Treasury security holdings among countries included in Panel B.
68

try. The sign on the estimated coecient of our variable DEPTH tcoupled with its lack of
statistical signicance challenges the prevailing view.
A natural concern is whether an unexpected, negative sign on DEPTH tis perhaps owing
to the way we have dened the variable. To account for this possibility, we estimate equations
(2.4), (2.7), and (2.9) again using an alternative but commonly used proxy for nancial depth:
the ratio of broad money (M3) to GDP. Inspection of the correlation coecients indicates
that while our original measure of nancial depth (the sum of a country's domestic private
credit [DPRIV jt], stock market capitalization [ SCAP jt], and bond market capitalization
[BCAP jt] relative to GDP) is very weakly correlated with our new measure, the latter
is highly correlated with our variable EXCH it(r=-0.903).34Thus, we follow the same
procedure outlined in section (2.4) to estimate equations (2.4), (2.7), and (2.9) with and
withoutEXCH itincluded as a regressor. Our results are presented in Table 6.
34A formal presentation of our correlation matrix using an alternative denition of nancial depth can be
found in Table 17 of Appendix A.
69

Table 6: Regression Results – Alternative Measure of Financial Depth (M3/GDP)
Dependent variable: USTS i;t
Panel A (China Included) Panel B (China Omitted)
Pooled First Dierences Anderson-Hsiao Pooled First Dierences Anderson-Hsiao
OLS OLS IV OLS OLS IV
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
USTS i;t1 0:7240:7240:6360:636
(0:11) (0 :11) (0 :15) (0 :15)
CRED i;t 2:0652:0650:2780:2780:1120:1120:442 0 :442 0 :2160:2160:1790:179
(0:40) (0 :40) (0 :05) (0 :05) (0 :05) (0 :05) (0 :32) (0 :32) (0 :05) (0 :05) (0 :05) (0 :05)
EXCH t186:83 116 :224 86:901 23 :760 27 :373 162:63
(355:43) (176 :88) (121 :47) (108 :52) (120 :27) (376 :40)
DEPTH t 1:1561:4720:1150:1260:1810:2650:0560:0950:1360:1390:390:203
(0:65) (0 :67) (0 :309) (0 :32) (0 :49) (0 :59) (0 :28) (0 :29) (0 :067) (0 :07) (0 :41) (0 :10)
NETW t 3:440 2 :842 2 :940 3 :285 2 :334 1 :808 0 :514 0 :5900:533 0 :614 0 :8180:838
(3:03) (2 :59) (2 :97) (3 :33) (1 :78) (1 :25) (0 :47) (0 :30) (0 :41) (0 :24) (0 :39) (0 :40)
YIELD t 46:28463:8499:33111:90213:13213:086 2 :997 0 :763 1 :027 1 :632 10 :367 3 :260
(27:70) (26 :15) (6 :63) (8 :64) (11 :16) (11 :11) (12 :87) (13 :80) (3 :04) (1 :77) (15 :82) (2 :32)
Observations 728 728 715 715 689 689 672 672 660 660 636 636
R20.640 0.640 0.184 0.184 0.580 0.580 0.258 0.258 0.205 0.205 0.319 0.319
Adjusted R20.610 0.610 0.115 0.115 0.545 0.545 0.190 0.190 0.131 0.131 0.256 0.256
Notes : The nancial depth variable ( DEPTH t) is measured as the ratio of broad money (M3) to GDP. Pooled OLS regression in columns 1,2,7 and 8
with robust standard errors in parentheses. First-dierences OLS regression in columns 3-4 and 9-10. Columns 5-6 and 11-12 use the instrumental variables
method of Anderson and Hsiao (1982); we instrument for USTSi;t1using a two-period lag. Year dummies are included in all regressions. Robust standard
errors clustered by country in parentheses for First Dierences OLS and Anderson-Hsiao IV. For detailed data denitions and sources, see Table 12 in
Appendix A .p<0.001;p<0.01;p<0.05
70

Importantly, the coecient on our variable of interest, CRED it, is robust to an alterna-
tive measure of nancial depth. Our results also indicate that the estimated coecient for
DEPTH tcontinues to be negative when rst dierences and IV estimation strategies are
adopted and remains so when China is omitted as an observation and EXCH tis removed
as a control variable. The estimated coecient on DEPTH tis not statistically signicant
whenEXCH tis included in the model as a control variable but becomes signicant at the
1% level when EXCH tis omitted. While we favor the more comprehensive denition of
nancial depth that was utilized in the original empirical analysis (the results of which are
reported in Table 3.3), the persistence of a negative sign on the estimated coecient on
DEPTH tserves to conrm the robustness of our original result.
Additionally, while an alternative operationalization of nancial depth has not changed
the direction in which our independent variables in
uence USTS t, the magnitudes and sta-
tistical signicance of some coecients have indeed changed. Namely, YIELD tis no longer
statistically signicant and increases in magnitude when EXCH tis included as a control
variable, suggesting that perhaps those elements of a more robust denition of nancial depth
(utilized in the rst set of estimations reported in Table 3.3) serve to articially reduce the
correlation between YIELD tandUSTS t. Further, the coecient on our variable represent-
ing network eects, NETW t, is now statistically signicant, indicating that perhaps those
elements of a more robust denition of nancial depth are correlated with NETW t.
2.7 Concluding Remarks
The prevailing view is that the persistent foreign demand for US Treasury securities
(i.e., persistent lending to the US) is owing to favorable investor perception of US economic
fundamentals, namely nancial depth. In this paper, we rely on the sovereign debt literature
71

to argue that perception alone cannot motivate sovereign lending. Rather, sovereign lending
requires the existence of an endogenous enforcement mechanism to secure repayment. This
mechanism cannot plausibly be said to exist in the current lending environment given the
disparity in strength between developing and emerging market economy lenders and advanced
economy borrowers.
In this essay, we provide evidence that US Treasury securities act as collateral to dollar
credit. Additionally, while perception is impossible to capture empirically, our results indi-
cate that variation in the relative nancial depth of the US does not explain the variation
in holdings of US Treasury securities. Our results imply that the continued nancing of the
US decit and the dollar's global role are not solely dependent on investor perception and
so cannot be entirely vulnerable to the latter.
This research has generated possibly fruitful areas for future inquiry. Naturally, the
analysis could be improved dramatically with a panel dataset that includes more cross-
sectional units and time periods. This possibility, however, necessarily relies on factors
beyond the researcher's control, so we oer suggestions that can feasibly be undertaken.
Our research has shed light on the contemporaneous relationship between outstanding
dollar credit and holdings of US Treasury securities among developing and emerging market
economies. Future research may try to determine whether and how our results might change
when lags are introduced into the estimating equation (i.e., how current and past values of
outstanding dollar credit are related to holdings of US Treasury securities, given current and
past values of our control variables. It may also be interesting to test whether the relationship
between outstanding dollar credit and holdings of US Treasury securities in developing and
emerging market economies was more or less pronounced during dierent time periods.
Future research could also be devoted to investigating the determinants of inertia in a
reserve asset. Our research has demonstrated that at least part of the variation in holdings
72

of US Treasury securities that would otherwise be resigned to inertia, is actually owing to
the asset's role as collateral to dollar borrowing. While we do not claim that inertia plays
no role in the persistence of an asset in foreign exchange reserves, the empirical research
thus far (including our own current endeavor) has shown that it is quite substantial and in
the order of at least 50% (i.e., at least 50% of the level of US Treasury securities in foreign
exchange reserves from time period to the next is explained by way of the latter's inertia).
While this may indeed re
ect reality, the re
ection does not seem to have been rigorously
challenged in light of other possibilities.
Guidance in what these other possibilities or determinants of inertia might be is provided
by the key highlights and results of this essay. Sovereign immunity threatens virtually any
contract where a sovereign nation is a counterparty whose compliance is required. Thus,
it may be the case that US Treasury securities act to collateralize other contracts between
sovereign nations and future endeavors may empirically test whether this is the case.
Another avenue for future research is investigation of our unexpected negative coecient
on the variable representing the nancial depth of the US relative to its currency competitors
(DEPTH t). This negative coecient persisted after the estimating equation was transformed
to account for unobserved heterogeneity, the outlier China was removed, and an alternative
denition of nancial depth was adopted. This result also echoes Chinn and Frankel's (2005)
ndings whereby the nancial depth of a currency issuer is found to be statistically insigni-
cant and negatively related to the currency composition of foreign exchange reserves. Future
work could seek to explain the mechanisms that underlie these counterintuitive results.
Finally, our results demonstrate that though China has a disproportionate amount of
outstanding dollar credit and US Treasury securities, the relationship between these two
variables is less pronounced for China than it is for other developing and emerging market
economies in our sample. Future research could therefore be devoted to investigating what
73

seems to be a unique set of determinants of the currency composition of China's foreign
exchange reserves.
74

C H A P T E R 3
THE COST OF A SWIFT KICK: ESTIMATING THE COSTS
OF FINANCIAL SANCTIONS ON IRAN
3.1 Introduction
The reality of one country (or city-state) sanctioning another has been in existence at
least since the time of Ancient Greece.1Since that time, this reality has also generated an
accompanying discussion on sanction eectiveness; that is, whether sanctions achieve their
desired impact, what negative externalities they occasion, what costs are borne by sender
countries, and so on.2Over time, the accompanying questions have not changed as much as
the sanctions themselves. Prior to 2000, most sanctions took the form of trade sanctions:
those banning imports from or exports to a state (Kirshner, 1997). But, these sanctions came
to be viewed as not only inecient but also devastatingly heavy-handed. The response to
the bluntness and poor performance of trade sanctions was a targeted (i.e.,\smart") nancial
sanction.
In this paper, we argue that the existence of global nancial networks has enabled a unique
potential for nancial sanctions to be comprehensive in eect. To illustrate, we measure the
1Pericles levied sanctions against nearby Megara in 432 B.C. (Hufbauer, Schott and Elliott, 1990)
2In this paper, we adopt the convention of referring to the sovereign nation imposing a sanction as the
sender and that receiving the sanction as the target
75

cost on Iran's economy of a unique nancial sanction requiring its removal from the Society
for Worldwide Interbank Financial Telecommunications (the \SWIFT sanction"). In so
doing, we demonstrate how global nancial networks enable ostensibly targeted nancial
sanctions to resemble the heavy-handed tools of compliance they were meant to replace.
The SWIFT sanction presents a unique opportunity to investigate the impact of global
nancial networks wielded as a sanctioning tool. First, Iran is the rst and only nation to be
punitively excluded from SWIFT. Second, because the Iranian economy has operated under
a host of sanctions for nearly half a century, the impact of the SWIFT sanction provides
insight into how its unique form compares with its predecessors.
To evaluate the impact of the SWIFT sanction on Iran's economy, we utilize quarterly
data on Iran's real gross domestic product (in purchasing power parity [PPP] adjusted,
2015 international dollars) during the period 1988-2016. This data is obtained from the
Central Bank of Iran and the Statistical Center of Iran. We employ a time-series forecasting
technique to measure the cost of the SWIFT sanctions to Iran's real gross domestic product
(GDP), where cost is measured by the dierence between forecasted and actual real GDP.
We further perform a robustness check using a distributed lag model to control for potential
confounders. We nd that the average quarterly cost of the SWIFT sanction to Iran's
real GDP is approximately $204.3 billion (PPP-adjusted, 2015 international dollars) which
represents approximately 14.7% and 13.8% of Iran's average quarterly actual and forecasted
GDP, respectively.
The paper proceeds as follows. Section 3.2 describes the development of nancial sanc-
tions. Section 3.3 outlines how nancial sanctions have evolved with the existence of global -
nancial networks. Section 3.4 describes the empirical analysis employed and results obtained,
including a robustness check to control for potential confounders. Section 3.5 concludes.
76

3.2 The Development of Financial Sanctions
A sanction or punishment levied by one sovereign nation onto another can take many
forms: a sovereign nation and/or a coalition of countries might intervene against another
militarily, sever diplomatic ties, close its borders to citizens of a target country, and/or
restrict membership to international organizations. When sanctions specically aect the

ow of resources|whether trade or nancial |to and from a target country, then sanctions
are considered economic sanctions.
A more specic and widely-accepted denition of what constitutes an economic sanction
does not exist. That is, groups of theorists have arrived at separate conclusions and this has
long been a point of frustration for researchers (Nossal, 1989). Broad characterizations of
denition-groupings are possible, however. Generally speaking, denitions of sanctions are
led into two broad categories depending on whether the objective of sanctions is believed
to be to punish orexpress . The punishment perspective of sanctions can be further broken
down into two groups: means-ends and retributive . In the means-ends perspective, the goal
of sanctions is to elicit compliance from a target; that is, sanctions are the means to the
endof amending the target's behavior (Daoudi and Dajani, 1983; Askari, Forrer,Teegan and
Yang, 2002; Davis and Engerman, 2003; Early, 2012).
In the retributive perspective , the goal of sanctions is simply to punish with no necessary
requirement that a target comply with a sender's demands. In other words, a sanction can be
retribution for violating an international norm and can thusly be successful even if the target
does not amend objectionable behavior. Whereas the goal of sanctions in the punishment
perspective is to mete out a sentence for the violation of some law or norm, the expressive
view of sanctions holds that sanctions can be used as a signaling mechanism to demonstrate
support or disfavor (Galtung, 1967). For instance, Klotz (1995) argues that US sanctions
against apartheid in South Africa operated against US material interests but were imposed
77

to signal to the global community that the US supports racial equality.
In addition to dierent denitions of economic sanctions, there are also two generally
dierent types: trade and nancial. Though both types of sanctions aim to stop the
ow of
resources to and from a target country, their dierence lies in the nature of the
ows they aim
to stop. The goal of trade sanctions is to stop the
ow of goods , thereby hindering a target
nation from earning gains from trade. Financial sanctions, however, aim to stop the
ow of
funds to and from a target country. Thus, the methods of nancial sanctions include but are
not limited to freezing a target's assets; cutting o sources of credit, aid, and/or payment
facilitation; and imposing monetary pressures onto a target economy (Drezner, 2011). Until
recently, most sanctions have taken the form of trade sanctions.
For roughly the past two decades, however, trade sanctions have increasingly come to be
viewed as blunt and inecient tools of coercion, primarily owing to their performance in Iraq.
In August 1990, four days after Saddam Hussein's invasion of Kuwait, the United Nations
Security Council levied comprehensive trade sanctions onto Iraq.3The sanctions required
that members and non-members of the United Nations refrain from exporting to or importing
from Iraq or Kuwait. Given the likelihood of a military intervention, the sanctions were
easily seen as a more humane alternative to outright war, but some researchers argued the
consequences were just as harmful (Normand, 1996; Lopez and Cortright, 1997). Shortages
of medicine, clean water and food were widespread, risks of child mortality increased four-
fold, and infant mortality jumped 6.1 percentage points (Ali and Shah, 2000; Daponte and
Gareld, 2000; Gareld, 2002). By the time sanctions were lifted, they had levied a cost
onto Iraq's economy equaling 54% of its gross national product (Hufbauer, Schott, Elliott
and Oegg, 2007).
The sanctions levied onto Iraq in 1990 not only elicited a humanitarian crisis but were also
3United National Security Council Resolution 661
78

counterproductive. Lopez and Cortright (1997) argue that in targeting the civilian popula-
tion, the sanctions imposed onto Iraq spared government elites who possessed the means to
circumvent them. Indeed, the evidence suggests that the potential for circumvention of key
groups within a target country is at least partly exacerbated in a non-democracy. Exercising
a relatively larger degree of control over the economy, key groups in non-democracies are bet-
ter capable of diverting sanction costs, extracting rents from the citizenry, and orchestrating
smuggling eorts to circumvent sanctions (Kaempfer and Lowenberg, 1999). Additionally,
the sanctions were also counterproductive because they deprived opposition groups of cru-
cial resources needed to mount a viable campaign against the Iraqi government (Lopez and
Cortright, 1997).
Sanctions did not only disable opposition groups, however. They also enabled the Iraqi
government to play the hero to its citizenry which facilitated political integration in the target
(Lopez and Cortright, 1997). Galtung (1967) calls this phenomenon a rally around the
ag
eect : sanctions foster the target's political integration because its citizenry collectively
blames a common enemy (the sender) for their economic hardship. In non-democracies, this
eect has the perverse result that sanction costs are more easily diverted to the citizenry at
the same time that the existence of sanctions provides a scapegoat for their adverse eects.
In such a context |as characterized Iraq in 1990 |any aid given by a target government to
its citizenry is viewed by the latter sympathetically as a rescue eort.
The general response to the failures of comprehensive trade sanctions was to increasingly
employ nancial sanctions, argued to be "smarter" because they could be applied in a more
targeted fashion: individual nancial assets could be frozen, select loans could be refused for
debt rescheduling, nancial assistance for particular individuals could be rejected, etc. Thus,
targeted or "smart" nancial sanctions promised to accomplish the seemingly unachievable
79

goal of being both more humane and successful.4Specically, by targeting nancial sanctions
to a specic elite, civilian populations would be spared |thereby sidestepping humanitarian
impacts and removing the potential for target governments to facilitate a rally around the
ag
eect |and pivotal target groups would be prevented in their eort to circumvent sanctions.
Owing to their theorized advantages, nancial sanctions have become increasingly pop-
ular as a tool of compliance. Indeed, the use of sanctions surged following the events of
September 11, 2001, and targeted nancial sanctions were utilized almost exclusively. So
far, their relative eectiveness is supported by the results: the success rate of nancial
sanctions relative to trade sanctions is found to be 41% and 25%, respectively (Lopez and
Cortright, 1997).
3.3 The Evolution of Targeted Financial Sanctions
The adoption of targeted nancial sanctions also coincided with an important develop-
ment: the enhancement of US capabilities to monitor and in
uence global nancial trans-
actions. Most notably, in 2006 the CIA, overseen by the Treasury department, secretly
subpoenaed and won access to the nancial records database of SWIFT.
SWIFT is a private, Brussels-based global nancial messaging network. Because it is
simply a messaging network, the company neither maintains deposit accounts nor transfers
actual funds; rather, it simply delivers a standardized nancial message (Lichtbau and Risen,
2006). To gain a more concrete sense of how SWIFT operates, consider a simple thought
experiment: imagine three individuals| Q,R, andS|residing in the same town of the
United States. Suppose that Qwants to give $10 to Rbut that instead of walking to R's
house and handing it to her, Qopts to employ the services of S. For a fee, Swill askQ
4In this context, "success" is meant to indicate a change in the target's policies.
80

to complete a form for deposit (we might properly consider it to be a simple check) that S
created in consultation with QandR's banks (in this way, S's nancial message is convenient
to and therefore preferred by QandR's banks.) Once Qcompletes the form, Sdelivers it to
R's banking institution and upon its receipt, a process is initiated whereby $10 is removed
fromQ's banking account and deposited into R's. In this simple example, Sessentially
performs the same function as SWIFT: Ssimply delivers a nancial message (i.e., a form or
check).
Given the ease with which two individuals residing in the same town can transact, Q's
use of a messenger service is perhaps only justied by her own laziness. But in the global
nancial arena, the movement of funds is naturally quite complex, and so it is in this context
that SWIFT provides a crucial service, all the more because it is unique. That is, SWIFT
is the only common infrastructure through which global nancial institutions can transact.5
To continue our previous hypothetical, we might imagine a circumstance whereby the only
wayQandR's banks can interact is through S's form for deposit. In this case, Sbecomes
not only a messenger but also a network since she possesses a unique platform through which
people and funds are connected. In the same way, SWIFT is also a network (i.e., it is also a
platform through which nancial institutions communicate.)
The ability to monitor SWIFT transactions represented a signicant breakthrough in
the US capacity to monitor global nancial transactions not only because of the function
SWIFT serves but also because of its ubiquity. When it was founded in the 1970s, SWIFT
could claim as members 518 dierent institutions from 22 dierent countries and it processed
roughly ten million messages annually. Today, it claims as members 11,000 institutions from
over 200 countries and processes approximately 26 million messages in one day (SWIFT
5It should be noted that in response to the advent of a SWIFT sanction, some alternatives to the SWIFT
platform were initiated (e.g., Russia's System for Transfer of Financial Messages or China's Cross-Border
Interbank Payment System), though none can reasonably be said to compete with SWIFT.
81

Annual Review, 2016). It is believed that the dollar-value of SWIFT transactions in 2001
was more than $7.7 trillion annually, or more than one-third of total world exports of goods
and services in 2016 (Scott and Zachariadis, 2014).
In 2012, the US used its access to the SWIFT network strategically as a sanctioning
tool.6In March of that year, the Islamic Republic of Iran became the rst sovereign nation
to be removed from the SWIFT platform. Though new in form, the sanction was touted as
being identical to its predecessors in eect and benet. For instance, David Cohen |Under
Secretary of the Treasury for Terrorism and Financial Intelligence from 2011 to 2015 |made
a number of assurances that the SWIFT sanctions were targeted, allowed for the import of
food and medicine, and minimized damage to the Iranian people (Faucon, 2012). Cohen's
2014 statements published by the Treasury are illustrative:
We have been able to move away from clunky and heavy-handed instruments of economic
power…Sanctions that focus on bad actors within the nancial sector are far more
precise and far more eective than traditional trade sanctions. And the trade
restrictions that we continue to employ today are also smarter and more surgical,
targeting specic classes of products rather than cutting o entire economies.
(Remarks of Under Secretary, 2014)
Given the vast literature on sanctions, this sanction episode has received a great deal of
attention.7Seemingly unnoticed, however, is what a radical departure from the standard
sanction a SWIFT cuto has the potential to be. Throughout much of the nineties, the terms
targeted nancial sanctions ,smart sanctions andnancial sanctions were used interchange-
ably. And this makes sense. Some thirty years ago, there did not exist any other option but
to make nancial sanctions targeted. Sovereign-to-sovereign transactions occurred, for the
6It is worth mentioning that being a private company, SWIFT was and continues to be reluctant to be
used as a sanctioning tool. The ability to leverage SWIFT strategically was therefore the product of an
active lobbying eort, namely by United Against Nuclear Iran.
7A sanction episode is the period of time between sanctions implementation and removal.
82

most part, in a dyadic setting so that there were natural limits to the kind of nancial cuto
a single country could impose onto another. That is, the US acting unilaterally could only
freeze nancial assets in US banks or stop payments from a target to specic US businesses .
Today's global environment is quite dierent, however. The existence of SWIFT demon-
strates that countries are increasingly capable of interacting within a nancial network and
the platform's ubiquity demonstrates the importance of maintaining unimpeded access to it
(Davis and Engerman, 2003; Joshi and Mahmud, 2016, 2018). Together, what these devel-
opments imply for nancial sanctions is that the latter now have the potential to become
comprehensive in nature. That is, by depriving a sovereign nation of access to the only
platform available to receive and send payments in a globalized economy, it is likely that the
eect will re
ect a comprehensive approach; indeed, we argue that the impact of the SWIFT
sanction to its rst recipient, Iran, serves to substantiate the point.
3.3.1 Targeted Financial Sanctions in Iran
The Islamic Republic of Iran has been the target of a consistent barrage of sanctions since
the Islamic Revolution in 1979. Prior to 2007, most sanctions levied against Iran were trade-
based and aimed against Iran's energy sector. In 2007, however, the Bush Administration
imposed its rst targeted nancial sanction (i.e., "smart" sanction) against Iran's fourth
largest bank, Bank Sepah, for its alleged support of entities aliated with Iran's nuclear
program. Under Executive Order 13382, the Bank's assets under US control were frozen and
US entities were banned from engaging in business with Bank Sepah. Thus, consistent with
the general form of a targeted nancial sanction, the punitive measure was contained to the
Bank and its chairman.
Measured against the Bank Sepah sanctions, the SWIFT sanction is notably dierent in
scope and correspondingly dealt an unprecedented blow to Iran's general economy. Figure
83

Figure 6: Real GDP for Iran 1988-2016
60090012001500
1990 2000 2010
YearGDP
(PPP, 2015 Intl$, Billions)
Notes : The gure plots Iran's quarterly GDP in billions of PPP-adjusted 2015 international dollars. The
shaded vertical bar represents the duration of the SWIFT sanction. Data is from the Central Bank of Iran
and the Statistical Center of Iran.
6 plots quarterly data on Iran's real GDP in purchasing power parity (PPP) adjusted, 2015
international dollars during the period 1988Q4-2016Q4. The shaded vertical bar indicates
the SWIFT sanction episode.
The gure demonstrates that when the SWIFT sanction was rst imposed in 2012Q1,
there is a sharp decline in Iran's real GDP and when it was removed in 2016Q1, there is a
sharp uptick in the same. The gure also shows that between the SWIFT sanction's initial
implementation (2012Q1) and the period when Iran's real GDP reached its lowest level in
the sanction period (2013Q4), real GDP declined by $86.35 billion.
Comparing Iran's real GDP during this time to the real GDP of other oil-exporting and
advanced economies suggests that the sharp movements in Iran's GDP during the SWIFT
sanction episode is not owing to macroeconomic conditions aecting oil-exporters or the
global economy generally. Figure 7 plots GDP in billions of purchasing power parity (PPP)
84

adjusted, 2011 international dollars for all Organization of the Petroleum Exporting Coun-
tries (OPEC) and Group of Seven (G7) member-countries. The shaded vertical bar represents
the duration of the SWIFT sanction episode.
85

Figure 7: GDP Comparison
Notes : The gure in the left panel plots the GDP in billions of PPP-adjusted, 2011 international dollars for all fourteen member-countries of
OPEC during the period 1990-2017. The gures in the right panel illustrate GDP in billions of PPP-adjusted, 2011 international dollars for
all G7 member-countries, the European Union, and Iran during the period 1990-2017. The shaded vertical bar in all gures represents the
duration of the SWIFT sanction. Data is from the FRED database and CEIC data.
86

Figure 7 demonstrates that the countries sampled did not experience a sharp increase
and decrease at the beginning and end of the SWIFT sanction, respectively. Libya seems
to be the only OPEC member-country to experience a sharp downturn in real GDP near
the start of the SWIFT sanction episode, but this is more likely owing to the 2011 US-led
military intervention into Libya.
In this paper, we interpret the sharp decline in Iran's real GDP in 2012Q1 as being a direct
result of the SWIFT sanction imposed onto the economy during that time. Accordingly, we
argue that the magnitude of the SWIFT sanction's impact to Iran's real GDP provides
evidence to our claim that the use of global nancial networks as a sanctioning tool has
caused "targeted" nancial sanctions to behave more like comprehensive trade sanctions in
eect. We focus on the impact to Iran's real GDP because a targeted nancial sanction
is one that is contained to a nancial and governmental elite within the target country.
By denition, then, any spillover eects impacting the level of goods and services in the
target's real economy naturally constitute a failure of any targeted objective. Accordingly,
Pape (1997) comments, "Economic sanctions characteristically aim to impose costs on the
economy as a whole…Accordingly, the most important measure of the intensity of economic
sanctions aggregate gross national product (GNP) loss over time" (p.4).
Though the sharp movements in Iran's GDP correspond exactly to the application and
removal of the SWIFT sanction, care should be taken when making a causal claim. To this
end, in the next section we evaluate the impact of the SWIFT sanction on Iran's real GDP
by controlling for potential confounders. Then, we calculate the annual cost to Iran's real
GDP of the SWIFT sanction using a time-series forecasting method and discuss our results.
To the best of our knowledge, this undertaking constitutes the rst empirical analysis of the
impact to Iran's real GDP of the SWIFT sanction.
87

3.4 Empirical Analysis
3.4.1 Causality Test
To further support the notion that the sharp movements in Iran's GDP can be interpreted
as measuring a causal eect of the SWIFT sanction, we provide a time-series analysis to
control for potential confounders. Specically, we control for the following: the international
price of crude oil ( OIL) to control for the possibility that a change in Iran's real GDP during
the SWIFT sanction is due to
uctuations in the global price of crude oil; Iran's government
expenditure ( EXP ) to control for the possiblity that a change in Iran's real GDP during
the SWIFT sanction is due to changes in the level of domestic government expenditure; and
global economic conditions ( G20) to control for the possibility that any change in Iran's
real GDP during the SWIFT sanction is simply owing to more general global economic
conditions.
The frequency of the data for all of our variables is quarterly and spans the time period
1988Q4-2016Q4, giving us 116 observations for each variable. Owing to data availability
issues, data on Iran's real GDP and government expenditure is obtained from two sources.
For data prior to and including 1990Q4, GDP data on Iran's real GDP is obtained from the
Central Bank of Iran; after 1990Q4, real GDP data is obtained from the Statistical Center
of Iran. For Iran's government expenditure, data prior to and including 1991Q4 is obtained
from the Central Bank of Iran and data after 1991Q4 is obtained from the Statistical Center
of Iran. In both cases, the data has been rebased in billions of purchasing power parity
adjusted, 2015 international dollars. That is, $1 of government expenditure or real GDP
has the same purchasing power as $1 in the US in 2015. The Average Petroleum Spot Price
(APSP) Crude Oil Index is used to represent the international price of crude oil and data is
88

retrieved from the Federal Reserve Bank of St. Louis.8Finally, data on the GDP for G20
countries is obtained from OECDStat. Table 1 provides summary statistics on the variables
that serve as our potential confounders.
Table 7: Descriptive Statistics
Statistic Mean St. Dev. Min Max
GDP Iran ( GDP t) 943 :19 347 :00 402 :75 1499 :20
APSP Index ( OIL t) 84 :90 61 :16 21 :81 227 :26
GDP G20 ( G20t) 2788 :47 488 :69 1965 :91 3597 :19
Government Expenditure ( EXP t) 133:24 36 :87 66 :77 205 :57
Notes : GDP Iran is obtained from the Central Bank of Iran and the Statistical Center of Iran. APSP
Index is from the Federal Reserve Bank of St. Louis Economic Database (FRED). G20 is from the
OECD Statistics database (OECDStat). Government Expenditure for Iran is from the Central Bank
of Iran and Statistical Center of Iran
Our goal is to determine whether any of these potential confounders has a substantial
eect on Iran's GDP during the SWIFT sanction. Figure 8 compares plots of our confounder
variables during the period 1988Q4-2016Q4, where the shaded vertical line corresponds to
the SWIFT sanction (the event of interest). At rst glance, it does not seem that the
movements in our potential confounder variables would have impacted Iran's real GDP
during the SWIFT sanction in a substantive way.
The behavior of our potential confounders during the SWIFT sanction episode seems
to conform to their respective trends, making it unlikely that they would have contributed
to the movements of Iran's real GDP during this periods in a signicant way. The only
exception is perhaps the price of crude oil which, as panel (d) in Figure 8 depicts, dropped
precipitously beginning in the second quarter of 2014.
In theory, it is possible that this sharp decline in crude oil prices impacted the Iranian
economy in one of two ways, depending on whether demand for Iranian oil was the cause of
8The APSP Crude Oil Index is a simple average of three spot prices for oil: Dated Brent, West Texas
Intermediate, and Dubai Fateh.
89

Figure 8: Confounders
60090012001500
1990 2000 2010PPP, 2015 Intl$, Billions)(a) GDP for Iran
100150200
1990 2000 2010PPP, 2015 Intl$, Billions(b) Government Expenditure for Iran
2000250030003500
1990 2000 2010PPP, 2010 USD, Billions(c) GDP for G20 Countries
50100150200
1990 2000 2010Index 2005=100(d) APSP Crude Oil Index
Notes : Panel (a) plots Iran's quarterly real GDP in billions of purchasing power parity (PPP) adjusted 2015
international dollars. Data for panel (a) is obtained from the Central Bank of Iran and the Statistical Center
of Iran. Panel (b) plots government expenditure for Iran in billions of PPP-adjusted 2015 international
dollars. Data for panel (b) is from the Statistical Center of Iran. Panel (c) plots real GDP for the Group of
Twenty (G20) countries in billions of PPP-adjusted 2010 US dollars. Data is obtained from the Organisation
for Economic Co-operation and Development Statistics database (OECD.Stat). Panel (d) plots values for
the APSP Crude Oil Index where 2005=100. Data for Panel (d) is obtained from the Federal Reserve Bank
of St. Louis Economic Data (FRED) database. The shaded vertical bar in Panels (a)-(d) represents the
duration of the SWIFT sanction.
90

decline in global crude oil prices. If the precipitous decline in global crude oil prices occurred
irrespective of demand for Iranian crude oil, then we might reasonably expect the decline to
correspond with a decrease in Iran's real GDP since a lower global price of crude oil would
lessen the appeal of a cheaper alternative.9
Panel (a) in Figure 8 depicts a slight increase in Iran's real GDP beginning in roughly
the third quarter of 2015. Given the delayed and seemingly negligible change in Iran's real
GDP, it is unclear whether the sharp movements in the global price of crude oil had any
meaningful impact.
To more accurately assess the impact of our potential confounders on Iran's real GDP, we
estimate a distributed lag model which allows us to understand how current and past values
of our potential confounders aect Iran's real GDP. Specically, we use quarterly data during
the period 1988Q4-2016Q4 and estimate the following regression after rst-dierencing our
variables to obtain stationarity:
4GDP t=+pX
j=0j4OIL tj+pX
j=0j4EXP tj+pX
j=0
j4G20tj+Dt+4t(3.1)
wherepis the lag length chosen on the basis of the BIC criterion; GDP tis Iran's quarterly
real GDP at time tin billions of PPP-adjusted, 2015 international dollars; OIL tis the
global price of oil at time tbased on the APSP Crude Oil Index in US dollars per barrel;
EXP tis the level of Iran's government expenditure at time tin billions of PPP-adjusted,
2015 international dollars; G20tis the average GDP for a subset of Group of Twenty (G20)
member countries in billions of PPP-adjusted, 2010 international dollars; and Dtis a dummy
9This point is related to a strand of the sanctions literature arguing that the existence of a third-party
nation who will trade with a target (i.e., a sanctions buster ), has substantial impact on whether a sanction
will be successful. Crucially, the willingness of a third-party nation to help a target circumvent sanctions
is higher the more protable the arrangement. Applied to the present case, this argument logically implies
that the higher the global price of crude oil, the higher the likelihood that nations will be willing to incur
the risk of acting as a sanctions buster to purchase a cheaper alternative. (Hanlon, 1986; Early, 2013)
91

variable at time tfor the SWIFT sanction set to one for each quarter during the SWIFT
sanction episode and set to zero otherwise.10Table 8 shows our results from estimating
equation (3.1).
Table 8: Robustness Time-Series Regression
Dependent variable:
GDP t
(1) (2) (3) (4)
SANCTION ( Dt)14:52715:33914:32713:454
(3:601) (3 :349) (3 :352) (3 :547)

APSP t0:0190:012 0 :055
(0:067) 0 :067 (0 :052)

G20 t 0:1990:1930:190
(0:080) 0 :081 (0 :067)

EXP t0:452 0:4500:434
(0:383) (0 :383) (0 :393)
Observations 678 678 678 678
R20:558 0 :155 0 :175 0 :147
Adjusted R20:521 0 :132 0 :152 0 :123
Notes : The dependent variable in all four specications is Iran's real GDP in billions of PPP-
adjusted, 2015 international dollars. Columns (1)-(4) reports results from estimating the distributed
lag model specied in equation 3.1 when all potential confounders are included (Column [1]),
EXP
is omitted (Column [2]),
APSP tis omitted (Column [3]), and
G20tis omitted (Column[4]),
respectively. All regressions include robust standard errors.p<0.001;p<0.01;p<0.05
Given our potential confounders, the coecient estimate for our sanction dummy variable
(Dt) is statistically signicant at the .1% level and large, indicating that the quarterly change
in Iran's real GDP (in PPP-adjusted, 2015 international dollars) was approximately $14.5
billion lower when the SWIFT sanction was in eect than when it was not. The sanction
dummy variable ( Dt) remains nearly unchanged across the four specications of our model
(reported in Columns [1]-[4], respectively), suggesting that failure to control for our potential
10G20 member countries included in the calculation of this variable were selected based on data availability
and include Australia, Canada, France, Germany, Italy, Japan, Korea, South Africa, Turkey, the United
Kingdom, and the United States.
92

confounders in forecasting Iran's real GDP will have no meaningful eect on our results.11
We thus proceed to forecasting Iran's GDP in the next section.
3.4.2 Univariate Time-series Forecasting Method
In this section, a univariate time-series forecasting model is used to determine what Iran's
GDP would have been if the SWIFT sanction had not been imposed in 2012. Students of
econometrics will recall that a key assumption needed to develop estimators for a classical
regression model is that there is no covariance between disturbances of dierent observations
(i.e., the disturbances are non-autoregressive). Though violation of this assumption poses
a distinct problem for regression analysis, the correlation between data can be harnessed in
time-series analysis to make predictions. Applied to forecasting, time-series models use past
data values to predict future values and so are unlike structural macroeconomic models that
attempt to dene a trend through causal relationships between variables that is informed
by economic theory. Though unable to explain causation, time-series forecasting techniques
are particularly useful tools when the movement of values is thought to be in
uenced by a
complex set of relationships. With regard to these complex dynamics (as naturally underlies
movements in macroeconomic indicators), structural macroeconomic models are vulnerable
to misspecication, endogeneity, and omitted variable bias, thus making time-series fore-
casting techniques an especially eective alternative (Marcellino, 2008; Keck, Raubold and
Truppia, 2010; Wang, 2016).
The dynamics that underlie movements in Iran's GDP are complex, indeed. The econ-
11It should be noted, however, that the average GDP of G20 member countries (
G20) has a positive
and statistically signicant eect ( p<0:01) on Iran's real GDP given changes in the market price of oil and
Iran's government expenditure, as well as the SWIFT sanction episode. Yet, a convincing case emerges for
pursuing a simple forecast of Iran's real GDP using its own structural processes when we consider that to the
extent that failure to control for 4G20 biases our forecast estimates, our results indicate that the direction of
bias is downward. Note, for instance, that when
G20 is omitted from equation 3.1, the coecient estimate
for our sanction dummy variable (reported in Column [4]) is approximately $1.1 billion lower in absolute
value terms than when
G20 is included in equation 3.1.
93

omy has operated under a host of sanctions since 1979 that are dierent in scope, eect,
and duration. A signicant literature exists theorizing the eect of sanctions on economies,
governments, and citizenries, but there does not exist a wide-ranging consensus view on the
eect of individual sanctions, much less on the dynamics of multiple and variable sanctions
(Hufbauer, Schott, Elliott, and Oegg, 2007; Eriksson, 2016; Haidar, 2017). In short, the
sanctions regime imposed on Iran since the Revolution makes it dicult to understand the
dynamics of Iranian GDP in a way that would readily lend itself to forecasting using a struc-
tural macroeconomic model. Use of a structural macroeconomic model is also complicated
by a common problem that is only exacerbated in the case of a pariah state: data availability
(Valadkhani, 2003; Alaedini and Ashrafzadeh, 2016).
In this analysis, we circumvent these diculties by using Iran's own structural processes
to predict what real GDP would have been if the SWIFT sanctions had not been imposed.
Accordingly, we follow the Box and Jenkins methodology (1976) to t a seasonal, autoregres-
sive integrated moving average (seasonal ARIMA) model to forecast Iran's GDP during the
time-period 2012Q1-2016Q4 (the entire duration of the SWIFT sanction implementation).
We are prompted to use a seasonal ARIMA because the periodic
uctuations in Iran's GDP
prior to the SWIFT episode illustrated in Figure 6 suggest that the series possibly possesses
a seasonal component which, if present, must be accounted for in our time-series model.
That the time series variable representing Iran's GDP possesses a seasonal component
is more clearly seen in Figure 9 which presents a seasonal plot of Iran's GDP prior to the
SWIFT sanction episode in panel (a) and a seasonal subseries plot of the residuals of a linear
model relating Iran's GDP and the time period in panel (b).
By plotting each observation against each quarter (or season), panel (a) in Figure 9
suggests that the series indeed possesses a seasonal component: Iran's GDP typically peaks
in Q1, declines slowly into the second quarter, and then slowly increases.
94

Figure 9: Seasonal Component of Iran's GDP
Notes : Panel (a) plots Iran's quarterly real GDP in billions of PPP-adjusted 2015 international dollars
against each quarter. Panel (b) plots the residuals of a linear model relating Iran's GDP (in billions of
PPP-adjusted, 2015 international dollars) and the time period.
95

Panel (b) of Figure 9 also conrms the existence of a seasonal component. The seasonal
subseries plot featured plots the residuals of a linear model relating Iran's GDP and the
quarterly time period (to detrend the data prior to plotting), where the horizontal blue line
delineates the mean for each quarterly subseries. The plot demonstrates that, on average,
the series reaches a maximum in the rst quarter, a minimum in the second quarter, and
then slowly increases. Thus, an ARIMA model that accounts for a seasonal component in
our time series variable to be forecasted is necessitated.
A seasonal ARIMA model forecasts a time-series based on the assumption that it is a
multiplicative combination of its non-seasonal and seasonal components. Its general form in
backshift notation (for conciseness) is expressed as,
P(Bs)p(B)rD
srdZt=q(B)Q(Bs)at (3.2)
wherep,d, andqrepresent the order of the autoregressive (AR), integrated (I), and moving
average (MA) processes, respectively, for the non-seasonal component representing the time
series;P,D, andQrepresent the order of the AR, I, and MA processes, respectively, for
the seasonal component representing the time series; sis the period of seasonality; Bis the
backshift operator;  P(Bs) andp(B) are the seasonal AR operator of order Pand the
non-seasonal AR operator of order p, respectively;  Q(Bs) andq(B) are the seasonal MA
operator of order Qand the non-seasonal MA operator of order q, respectively;rD
sis the
seasonal backward dierence operator for Ddierences and speriods andrdis the non-
seasonal backward dierence operator for ddierences and speriods;Ztis the time-series
to be forecasted; and atis a white noise process. In simplied form, the seasonal ARIMA
model is written as ARIMA( p;d;q )(P;D;Q )s.
Fitting an ARIMA ( p;d;q )(P;D;Q )smodel requires that the underlying time series be
96

stationary. Figure 6 demonstrates that Iran's GDP exhibits a strong upward trend; thus, the
data is rst-dierenced ( d=1) and an Augmented Dickey-Fuller (ADF) test | a standard
formal test for stationarity | is used to conrm stationarity. The results from performing
an ADF test on our variable for Iran's real GDP are presented in Table 1.3 and conrm our
series is integrated of order 1.
Table 9: ADF Test
ADF
Test Statistic
GDP t 0:920

GDP t4:656
Notes: The variable GDP trepresents Iran's GDP
(PPP-adjusted, 2015 International dollars) in time pe-
riodt. Lag orders are chosen to minimize the Bayesian
information criterion. Statistical signicance of the
test statistic indicates rejection of the null hypothesis
that the series follows a unit root process.p<0.001;
p<0.01;p<0.05
To account for the seasonal component of our quarterly time series for Iran's GDP, we
apply a seasonal dierence where s, the period of seasonality, equals 4.12
The auto.arima() function in R is utilized to select an ARIMA( p;d;q )(P;D;Q )smodel
based on minimizing the Bayesian information criterion (BIC).13The result is an ARIMA
(1;1;1)(0;0;1)4model with drift. Thus, our model takes the following form where GDP t
12The resulting autocorrelation function (ACF) and partial autocorrelation function (PACF) of the trans-
formed series (i.e., the seasonal dierence of the rst dierence of Iran's GDP) is shown in Figure 13 of
Appendix B. The ACF plots the correlation between the variable being forecasted (here, Iran's real GDP)
and its past values. The PACF plots the correlation between the variable being forecasted and its past values
that is not the result of previous correlations.
13The BIC oer one solution to the challenge of choosing parameter values for an ARIMA ( p;d;q )(P;D;Q )s
model. The problem arises from the fact that if too high a value is chosen, forecasts may suer from increased
estimation error whereas if too low a value is chosen, forecasts may ignore relevant information contained in
past values. The BIC penalizes increases to the parameter values so that the true parameter values can be
determined.
97

represents Iran's GDP (in billions of PPP-adjusted, 2015 international dollars) at time tand
is a constant.
1(B)(1B)GDP t=+1(B)1(B4)at (3.3)
Table 10 displays our results from estimating equation (3.3) and the calculated Ljung-Box
statistic through 8 lags is 4.40 and statistically insignicant, indicating that the residuals of
our estimated model follow a white-noise process.
Table 10: Forecast Summary
ARIMA(1,1,1)(0,0,1) 4Estimate
with drift
Constant 11.47
(2.19)
AR(1) 0.96
(0.04)
MA(1) -0.65
(0.09)
SMA(1) -0.79
(0.09)
N 113
Log-likelihood -356.88
AIC 723.75
BIC 736.36
Notes : Standard error in parenthesis
Using our ARIMA (1 ;1;1)(0;0;1)4model with drift, we forecast values for Iran's real
GDP over the entire SWIFT sanction episode (2012Q1-2016Q1). Figure 10 illustrates our
results.
The dashed black line above the noticeable plunge in Iran's real GDP starting in 2012Q1
is the forecasted value of Iran's real GDP in billions of purchasing power adjusted, 2015
international dollars. That is, the dashed line represents what Iranian GDP would have
98

Figure 10: Forecast of Iran's real GDP
YearGDP (PPP, 2015 Intl$, Billions)
500 1000 1500
1990 1995 2000 2005 2010 2015
Notes : The dashed line in the gure represents a forecast of Iran's GDP during the period 2012Q1-2016Q4
generated from an ARIMA(1,1,1)(0,0,1) 4model. The lightly and darkly shaded areas represents the 80%
and 95% prediction intervals, respectively.
99

been if the SWIFT sanctions had not been imposed and real GDP had continued to follow
the pre-2012Q1 trend. Table 11 compares quarterly forecasted values to actual values for
Iran's real GDP.
Table 11: Forecast and Cost Estimates
Date Actual GDP Forecasted GDP Lo95 CI Hi95 CI Forecast-Actual
2012Q1 1434 :57 1458 :13 1435 :36 1480 :91 23 :56
2012Q2 1415 :68 1481 :63 1443 :80 1519 :46 65 :95
2012Q3 1395 :85 1493 :17 1439 :82 1546 :53 97 :32
2012Q4 1372 :37 1505 :57 1435 :75 1575 :39 133 :20
2013Q1 1362 :09 1519 :38 1441 :62 1597 :14 157 :29
2013Q2 1359 :12 1533 :16 1447 :61 1618 :71 174 :04
2013Q3 1354 :66 1546 :92 1453 :67 1640 :16 192 :26
2013Q4 1348 :22 1560 :64 1459 :75 1661 :54 212 :42
2014Q1 1357 :38 1574 :34 1465 :82 1682 :87 216 :96
2014Q2 1367 :37 1588 :02 1471 :87 1704 :16 220 :65
2014Q3 1373 :09 1601 :66 1477 :88 1725 :44 228 :57
2014Q4 1373 :54 1615 :28 1483 :83 1746 :73 241 :74
2015Q1 1373 :64 1628 :87 1489 :72 1768 :01 255 :23
2015Q2 1376 :33 1642 :43 1495 :54 1789 :32 266 :10
2015Q3 1376 :55 1655 :96 1501 :29 1810 :64 279 :41
2015Q4 1383 :70 1669 :47 1506 :95 1831 :99 285 :77
2016Q1 1397 :17 1682 :95 1512 :53 1853 :37 285 :78
2016Q2 1426 :65 1696 :41 1518 :03 1874 :78 269 :76
2016Q3 1456 :74 1709 :83 1523 :44 1896 :23 253 :09
2016Q4 1499 :20 1723 :23 1528 :76 1917 :71 224 :03
Notes : Figures are in billons of purchasing power adjusted 2015 international dollars. Forecasted
GDP is an average of the low and high 95% prediction interval values.
3.4.3 Discussion
The impact to Iran's real GDP of the SWIFT sanction is sizeable. We calculate that the
average quarterly cost of the SWIFT sanction to Iran's real GDP is approximately $204.3
billion (PPP-adjusted, 2015 international dollars), or 14.7% and 13.8% of Iran's average
100

quarterly actual and forecasted GDP, respectively. This impact is striking given the steady
upward trend of Iran's real GDP since roughly 1996, suggesting a resilience to the variety of
sanctions levied against it before SWIFT sanctions were imposed.
One reason why a seemingly targeted nancial sanction levied such a substantial cost
onto the Iranian economy is owing to the unique form nancial sanctions take when access
to global networks becomes a tool for compliance. By using access to a unique global -
nancial network as a sanction, targets are selectively removed from an edice upon which
international economic relations take place. In practice, therefore, ostensibly targeted nan-
cial sanctions wielding access to global nancial networks will resemble the comprehensive
trade sanctions they were intended to replace, with all the latter's attending humanitarian
impacts.
Indeed, and perhaps unsurprisingly given the shock to Iran's economy, journalistic ac-
counts began to emerge in and around 2012 reporting shortages of crucial medicines, food-
stus, and medical devices in Iran (Bozorgmehr, 2012; Mohammed, 2012; Gladstone, 2013).
Also reported were higher poverty levels owing to rising food prices, high in
ation, and in-
creasing unemployment (Warrick and Ball, 2012; Bozorgmehr, 2013; Nasseri and Motevalli,
2015). Formal empirical analyses also conrm journalistic reports; for instance, Setayesh
and Mackey (2016) nd that of the 73 drugs in shortage in Iran during the period when the
SWIFT sanction was levied, 44% are classied as essential medicines according to the World
Health Organization.
3.5 Concluding Remarks
In this paper we have argued that the development of global nancial networks has
enabled nancial sanctions|sanctions aimed at stopping the
ow of funds to and from a
101

target country|to have a more comprehensive eect. As a result, nancial sanctions are
coming to resemble the heavy-handed and inecient trade sanctions they were meant to
replace. To provide evidence in support of our argument, we focus on a unique set of
nancial sanctions levied against the Islamic Republic of Iran.
In 2012, Iran was the rst country to have been removed from the Society for World-
wide Interbank Financial Telecommunications (SWIFT), a global nancial messaging service.
Though advertised at the time to be targeted to Iran's governmental and nancial elite, the
nancial sanction delivered a sizeable negative shock to Iran's real economy. To analyze
the impact of this shock, we employ a time-series forecasting model to measure the cost of
exclusion from the SWIFT network on Iran's real GDP.
By calculating the dierence between Iran's forecasted and actual real GDP, we nd that
the cost of the SWIFT sanctions to Iran's real GDP was, on average, approximately $204.3
billion (PPP-adjusted 2015 international dollars), which represents approximately 14.7% and
13.8% of Iran's average quarterly actual and forecasted GDP, respectively. We conclude that
the SWIFT sanctions made a considerable impact to Iran's real GDP which belies the claim
that these sanctions are targeted or \smart."
How the development of global nancial networks has changed the nature of nancial
sanctions has been, to the best of our knowledge, unnoticed in the sanctions literature. This is
at least partly because the latter largely still conceptualizes the global arena through a dyadic
framework (that is, sanctions are analyzed on a sovereign-to-sovereign basis rather than on
a network-sovereign basis.)14Given the potential for nancial sanctions to demonstrate the
heavy-handedness and ineciency characteristic of comprehensive trade sanctions, however,
we conclude that more research is warranted.
14Exceptions include Joshi and Mahmud (2016, 2018).
102

A P P E N D I X A
ADDITIONAL TABLES AND FIGURES FOR CHAPTER 2
103

Table 12: Data and Sources
Variable Description Source
USTS it Foreign holdings of long-term US Treasury securities. Monthly data avail-
able for 92 countries and 6 regions during the period 1984-2017. Data
adjusted for valuation eects.Bertaut and Judson (2017)
CRED it Outstanding dollar credit. Quarterly data available for fourteen develop-
ing and emerging market economies during the period 2000Q1-2018Q4.Bank for International Settlements
EXCH t 5-year moving average of the US Dollar to Special Drawing Rights ex-
change rate. Quarterly data is available from 1975Q1-2019Q1.Federal Reserve Bank of St. Louis
(FRED Economic Data)
DEPTH tFinancial Depth of the US relative to the Euro Area, United Kingdom
and Japan. Financial depth is the sum of a country's domestic private
credit, stock market capitalization, and bond market capitalizationWorld Federation of Exchanges. Bank
for International Settlements
DEPTH 2tFinancial Depth of the US relative to the Euro Area, United Kingdom
and Japan. Financial depth is the ratio of broad money (M3) to GDP.FRED Economic Data. CEIC data.
Bank for International Settlements.
NETW t Economic size of the US relative to Euro Area, United Kingdom and
Japan. Economic size is the ratio of country GDP to world GDP.International Monetary Fund. CEIC
data
YIELD t Yield on long-term US Treasury securities relative to yield on long-term
government bonds for Euro Area, United Kingdom and Japan. Quarterly
data for all regions is available during the period 1989Q1-2019Q1.Federal Reserve Bank of St. Louis
(FRED Economic Data)
104

Table 13: Country Codes
Country Code
Argentina AR
Brazil BR
Chile CL
China CH
India IN
Indonesia ID
Malaysia MY
Mexico MX
Republic of Korea KR
Russia RU
South Africa ZA
Taiwan TWN
Turkey TR
Table 14: Correlation Matrix
USTS it CRED it EXCH tDEPTH tNETW tYIELD tUSTS i;t1
USTS it 1
CRED it 0:783 1
EXCH t 0:199 0:399 1
DEPTH t0:0360:0790:176 1
NETW t 0:004 0:099 0:0250:034 1
YIELD t0:010 0:0760:0400:196 0 :722 1
USTS i;t1 0:999 0:794 0:2050:036 0 :0090:006 1
Notes : See Table 12 in Appendix A for data sources and denitions.
105

Table 15: Variance In
ation Factors
Model: Equation 2.9
Dependent Tolerance VIF
Variable (1) (2)

USTS i;t1 0:939 1 :065

CRED it 0:945 1 :058

EXCH it 0:926 1 :080

DEPTH it 0:832 1 :202

NETW it 0:831 1 :203

YIELD it 0:898 1 :114
Notes: Columns (1) and (2) report the level of tol-
erance and VIF for each regression of the depen-
dent variable on a linear combination on the re-
maining independent variables in equation 2.9, re-
spectively.
USTS i;t2is used as an instrument
for
USTS i;t1.
106

Table 16: First Stage for
USTS i;t1
Dependent variable:
USTS i;t1
Panel A Panel B
China included China Omitted
(1) (2)

USTS i;t2 0:7100:475
(0:029) (0 :077)

CRED it 0:0890:071
(0:025) (0 :075)

EXCH it 234:998 141 :789
(433:145) (248 :297)

DEPTH it 0:017 0 :044
(0:062) (0 :049)

NETW it 1:181 0 :208
(0:979) (0 :902)

YIELD it 3:061 1 :019
(6:550) (2 :885)
Observations 689
R20.558
Adjusted R20.521
Notes: Column 1 is the rst-stage regression for Panel A
where China is included in the sample. Column 2 is the
rst-stage regression for Panel B where China is omitted
from the sample. Both regressions include time dummies and
robust standard errors clustered by country in parentheses.
For detailed data denitions and sources, see Table 12 in
Appendix A.p<0.001;p<0.01;p<0.05
107

Table 17: Correlation Matrix- Alternative Measure of Financial Depth
USTS it CRED it EXCH tDEPTH1 tDEPTH2 tNETW tYIELD tUSTS i;t1
USTS it 1
CRED it 0:782 1
EXCH t 0:174 0:272 1
DEPTH1 t0:0390:0840:098 1
DEPTH2 t0:1620:2270:915 0 :002 1
NETW t 0:003 0:096 0:1880:032 0 :242 1
YIELD t0:010 0:0760:2870:197 0 :345 0 :722 1
USTS i;t1 0:999 0:793 0:1750:0400:160 0 :0070:005 1
Notes :DEPTH1 trepresents nancial depth dened as the sum of a country's domestic private credit, stock market capitalization,
and bond market capitalization relative to GDP. DEPTH2 trepresents nancial depth dened as the ratio of broad money (M3)
to GDP. See Table 12 in this Appendix A for data sources and denitions.
108

A P P E N D I X B
ADDITIONAL TABLES AND FIGURES FOR CHAPTER 3
109

Figure 11: ACF and PACF Patterns of Transformed GDP for Iran
−0.6−0.4−0.20.00.2
5 10 15
LagACF(a) ACF
−0.6−0.4−0.20.00.2
5 10 15
LagPACF(b) PACF
Notes : Panels (a) and (b) illustrate the autocorrelation function (ACF) and partial autocorrelation function
(PACF), respectively, of the seasonal dierence of the rst dierence of Iran's GDP (in billions of PPP-
adjusted, 2015 international dollars).
110

BIBLIOGRAPHY
[1] Aguiar, M. & Amador, M. (2014). Sovereign debt. In G. Gopinath, E. Helpman, and K.
Rogo (Eds.), Handbook of International Economics (Vol. 4, pp. 647-687). Amsterdam,
The Netherlands: Elsevier.
[2] Aizenman, J., & Ito, H. (2016). East Asian economies and nancial globalization in the
post-crisis world (NBER No. 22268). Cambridge, MA: National Bureau of Economic
Research.
[3] Alaedini, P., & Ashrafzadeh, H. (2016). Iran's post-revolutionary social justice agenda
and its outcomes: evolution and determinants of income distribution and middle-class
size. In M.R. Farzanegan & P. Alaedini (Eds.), Economic Welfare and Inequality in
Iran(pp.15-45). New York, NY: Palgrave Macmillan.
[4] Alfaro, L., Kalemli-Ozcan, S., & Volosovych, V. (2014). Sovereigns, upstream capital

ows, and global imbalances. Journal of the European Economic Association , 12(5),
1240-1284.
[5] Ali, M.M., & Shah, I.H. (2000). Sanctions and childhood mortality in Iraq. The Lancet ,
355(9218),1851-1857.
[6] Aliber, R.Z. (1966). The future of the dollar as an international currency . New York,
NY: Praeger.
[7] Anderson, T.W., & Hsiao, C. (1982). Formulation and estimation of dynamic models
using panel data. Journal of Econometrics , 18(1), 47-82.
[8] Antras, P., & Caballero, R. J. (2009). Trade and capital
ows: A nancial frictions
perspective. Journal of Political Economy , 117(4), 701-744.
[9] Askari, H., Forrer, J., Teegan, H., & Yang, J. (2002). Economic sanctions and US
international business interests. Banca Nazionale del Lavoro Quarterly Review , 55(220).
[10] Avdjiev, S., Kuti, Z., & Takats, E. (2012). The euro area crisis and cross-border bank
lending to emerging markets. BIS Quarterly Review , December, pp 37-47.
[11] Babakhani, S., Kazemi, F., Junemann, M., & Jozi, A. (2016). Banking industry Iran:
current status, opportunities and threats . ILIA Corporation.
111

[12] Balakrishnan, R., Tulin, V., & Bayoumi, M. T. (2007). Globalization, Gluts, Innovation
or Irrationality: What Explains the Easy Financing of the US Current Account Decit?
(International Monetary Fund No. 7-160). International Monetary Fund.
[13] Beckley, M. (2012). China's century? Why America's edge will endure. International
Security , 36(3), 41-78.
[14] Bernanke, B. (2005). The global saving glut and the US current account decit (No. 77).
Board of Governors of the Federal Reserve System (US).
[15] Bertaut, C., & Judson, R. (2014). Estimating the U.S. cross-border securities positions:
new data and new methods. (Board of Governors International Finance Discussion Pa-
pers No. 1113). Board of Governors of the Federal Reserve System.
[16] Borio, C. E., McCauley, R. N., & McGuire, P. (2011). Global credit and domestic credit
booms. BIS Quarterly Review , September, 43-57.
[17] Bowles, S., & Gintis, H. (1990). Contested exchange: new microfoundations for the
political economy of capitalism. Politics & Society , 18(2), 165-222.
[18] Bozorgmehr, N. (2012, September 4). In Iran, sanctions take toll on the sick. The Wash-
ington Post .
[19] Bozorgmehr, N. (2013, October 26). In
ation and weak rial squeeze Iran's middle. Fi-
nancial Times , p. 7.
[20] Bulow, J., & Rogo, K. (1989). A constant recontracting model of sovereign debt.
Journal of Political Economy , 97(1), 155-178.
[21] Bussi ere, M., Cheng, G., Chinn, M.D., & Lisack, N. (2015). For a few dollars more:
reserves and growth in times of crises. Journal of International Money and Finance , 52,
127-145.
[22] Caballero, R.J. (2006). On the macroeconomics of asset shortages (NBER No. 12753).
Cambridge, MA: National Bureau of Economic Research.
[23] Caballero, R.J., & Farhi, E. (2014). On the role of safe asset shortages in secular stag-
nation. In C. Teulings & R. Baldwin (Eds.), Secular stagnation: facts, causes and cures .
London, UK: Centre for Economic Policy Research.
[24] Caballero, R.J., Farhi, E., & Gourinchas, P.O. (2008). An equilibrium model "global
imbalances" and low interest rates. American Economic Review , 98(1), 358-93.
[25] Caballero, R.J., Farhi, E., & Gourinchas, P.O. (2016). Safe asset scarcity and aggregate
demand. American Economic Review , 106(5), 513-18.
112

[26] Caballero, R.J., Farhi, E., & Gourinchas, P.O. (2017). Rents, technical change, and risk
premia accounting for secular trends in interest rates, returns on capital, earning yields,
and factor shares. American Economic Review , 107(5), 614-20.
[27] Caballero, R.J., & Krishnamurthy, A. (2009). Global imbalances and nancial fragility.
American Economic Review , 99(2), 584-88.
[28] Cecchetti, S.G., Mohanty, M.S., & Zampolli, F. (2011). The real eects of debt (Bank for
International Settlements No. 352). Switzerland: Bank for International Settlements.
[29] Chinn, M., & Frankel, J. (2005). Will the euro eventually surpass the dollar as leading
international reserve currency? (NBER Working Paper Series No. 11510). Cambridge,
MA: National Bureau of Economic Research.
[30] Chinn, M.D., & Ito, H. (2007). Current account balances, nancial development and
institutions: assaying the world "savings glut." Journal of International Money and
Finance , 26(4), 546-569.
[31] Chinn, M.D., & Prasad, E.S. (2003). Medium-term determinants of current accounts in
industrial and developing countries: an empirical exploration. Journal of International
Economics , 59(1), 47-76.
[32] Chitu, L., Eichengreen, B., & Mehl, A. (2014). Journal of Development Economics , 111,
225-245.
[33] Daoudi, M.S., & Dajani, M.S. (1983). Economic sanctions, ideals and experience . Rout-
ledge Kegan & Paul.
[34] Daponte, B.O., & Gareld, R. (2000). The eect of economic sanctions on the mortality
of Iraqi children prior to the 1991 Persian Gulf War. American Journal of Public Health ,
90(4), 546.
[35] Davis, L., & Engerman, S. (2003). Sanctions: neither war nor peace. Journal of Eco-
nomic Perspectives , 17(2), 187-197.
[36] Devereux, M.B., & Shi, S. (2013). Vehicle currency. International Economic Review
54(1), 97-133.
[37] Dooley, M.P., Folkerts-Landau, D., & Garber, P. (2003). An essay on the revived Bret-
ton Woods system (NBER No. 9971). Cambridge, MA: National Bureau of Economic
Research.
[38] Dooley, M.P., Folkerts-Landau, D., & Garber, P. (2004). The US current account decit
and economic development: collateral for a total return swap (NBER No. 10727). Cam-
bridge, MA: National Bureau of Economic Research.
113

[39] Drezner, D.W. (1999). Cambridge Studies in International Relations: Vol 65. The sanc-
tions paradox: economic statecraft and international relations. Cambridge, UK: Cam-
bridge University Press.
[40] Drezner, D.W. (2011). Sanctions sometimes smart: targeted sanctions in theory and
practice. International Studies Review , 13(1), 96-108.
[41] Drury, A.C. (1998). Revisiting economic sanctions reconsidered. Journal of Peace Re-
search (407), 35(4), 497-509.
[42] Early, B.R. (2012). Alliances and trade with sanctioned states: a study of US economic
sanctions, 1950-2000. Journal of Con
ict Resolution , 56(3), 547-572.
[43] Eaton, J., & Engers, M. (1992). Sanctions. Journal of Political Economy , 100(5), 899-
928.
[44] Eaton, J., & Engers, M. (1999). Sanctions: some simple analytics. American Economic
Review , 89(2), 409-414.
[45] Eaton, J. & Fernandez, R. (1995). Sovereign debt. In Grossman, G. & Rogo, K. (Eds.),
Handbook of International Economics (pp 2031-2077). Amsterdam, The Netherlands:
Elsevier Science.
[46] Eaton, J., & Gersovitz, M. (1981). Debt with potential repudiation: theoretical and
empirical analysis. The Review of Economic Studies , 48(2), 289-309.
[47] Eichengreen, B., Chitu, L., & Mehl, A. (2016). Stability or upheaval? The currency
composition of international reserves in the long run. IMF Economic Review , 64(2),
354-380.
[48] Eichengreen, B., Mehl, A.J., & Chitu, L. (2017). Mars or Mercury? The geopolitics
of international currency choice (NBER Working Paper No. 24145). Cambridge, MA:
National Bureau of Economic Research.
[49] Engel, C., & Rogers, J.H. (2006). The US current account decit and the expected share
of world output. Journal of Monetary Economics , 53(5), 1063-1093.
[50] Epstein, G., & Gintis, H. (1995). International capital markets and national economic
policy. Review of International Political Economy , 2(4), 693-718.
[51] Eriksson, Mikael.(2016). Targeting peace: understanding UN and EU targeted sanctions.
New York, NY: Routledge.
[52] Farhi, E., & Maggiori, M. (2017). A model of the international monetary system. The
Quarterly Journal of Economics , 133(1), 295-355.
114

[53] Faucon, B. (2012, August 2). In Iran, private sector feels squeeze of sanctions. The Wall
Street Journal .
[54] Galati, G. & Woolridge, P. (2006). The euro as a reserve currency: a challenge to the
pre-eminence of the US dollar? (BIS Working Papers No. 218). Basel, Switzerland:
Bank for International Settlements.
[55] Galtung, J. (1967). On the eects of international economic sanctions, with examples
from the case of Rhodesia. World politics , 19(3), 378-416.
[56] Gareld, R. (2002). Economic sanctions, humanitarianism, and con
ict after the Cold
War. Social Justice , 29(3), 94-107.
[57] Gladstone, R. (2013, April 2). Double-digit in
ation worsens in Iran. The New York
Times .
[58] Goldberg, L.S., & Tille, C. (2008). Vehicle currency use in international trade. Journal
of International Economics , 76(2), 177-192.
[59] Gourinchas, P.O., & Jeanne, O. (2013). Capital
ows to developing countries: the
allocation puzzle. Review of Economic Studies , 80(4), 1484-1515.
[60] Gourinchas, P.O., & Rey, H. (2014). External adjustment, global imbalances, valuation
eects. In G. Gopinath, E. Helpman, & K. Rogo (Eds.), Handbook of International
Economics (pp. 585-645). Amsterdam, The Netherlands: Elsevier.
[61] Gruber, J. & Kamin, S. (2009). Do dierences in nancial development explain the
global pattern of current account imbalances? Review of International Economics , 174,
667-688.
[62] Haidar, J.I. (2017). Sanctions and export de
ection: evidence from Iran. Economic
Policy , 32(90), 319-355.
[63] Hanlon, J. (1986). Sanctions. Capital & Class , 10(3), 38-42.
[64] He, Z., Krishnamurthy, A., & Milbradt, K. (2016). What makes US government bonds
safe assets? American Economic Review , 106(5), 519-23.
[65] Hufbauer, G.C., Schott, J.J., & Elliott, K.A. (1990). Economic sanctions reconsidered:
history and current policy (Vol. 1). Peterson Institute for International Economics.
[66] Hufbauer, G.C., Schott, J.J., Elliott, K.A., & Oegg, B. (2007). Economic sanctions
reconsidered (Vol. 3). Peterson Institute for International Economics.
[67] Im, K.S., Pesaran, M.H., & Shin, Y. (2003). Testing for unit roots in heterogeneous
panels. Journal of Econometrics , 115(1), 53-74.
115

[68] Ito, H., Jongwanich, J., & Terada-Hagiwara, A. (2009). What makes developing Asia
resilient in a nancially globalized world? (ADB Economics Working Paper Series No.
181). Phillipines: Asian Development Bank.
[69] Joshi, S., & Mahmud, A.S. (2016). Sanctions in networks: \The Most Unkindest Cut
of All." Games and Economic Behavior , 97, 44-53.
[70] Joshi, S., & Mahmud, A.S. (2018). Unilateral and multilateral sanctions: a network
approach. Journal of Economic Behavior & Organization , 145, 52-65.
[71] Kaempfer, W.H., & Lowenberg, A.D. (1988). The theory of international economic
sanctions: a public choice approach. The American Economic Review , 78(4), 786-793.
[72] Kamin, S., Reeve, T., & Sheets, N. (2007). \U.S. external adjustment: is it unique?
Will it disrupt the rest of the world?" International Finance Discussion Papers, Federal
Reserve Board, 2007-892.
[73] Keck, A., Raubold, A., & Truppia, A. (2010). Forecasting international trade. OECD
Journal: Journal of Business Cycle Measurement and Analysis , 157-176.
[74] Kindleberger, C.P. (1986). The world in depression, 1929-1939 Berkeley: University of
California Press.
[75] King, R.G., & Levine, R. (1993). Finance and growth: Schumpeter might be right. The
Quarterly Journal of Economics , 108(3), 717-737.
[76] Kirshner, J. (1997). The microfoundations of economic sanctions. Security Studies , 6(3),
32-64.
[77] Klotz, A. (1995). Norms reconstituting interests: global racial equality and US sanctions
against South Africa. International Organization , 49(3), 451-478.
[78] Krugman, P.R. (1979). Vehicle currencies and the structure of international exchange
(NBER Working Paper Series No. 333). Cambridge, MA: National Bureau of Economic
Research.
[79] Krugman, P.R. (1984). The international role of the dollar: theory and propsect. In
J.F.O. Bilson & R.C. Marston (Eds.), Exchange Rate Theory and Practice (pp. 261-
278). University of Chicago Press.
[80] Levin, A., Lin, C.F., & Chu, C.S.J. (2002). Unit root tests in panel data: asymptotic
and nite-sample properties. Journal of Econometrics , 108(1), 1-24.
[81] Levy, P.I. (1999). Sanctions on South Africa: what did they do? American Economic
Review , 89(2), 415-420.
116

[82] Lopez, G.A., & Cortright, D. (1997). Financial sanctions: the key to a 'smart' sanctions
strategy. Die Friedens-Warte , 72(4), 327-336.
[83] Lucas, R.E. (1990). Why doesn't capital
ow from rich to poor countries? American
Economic Review , 80(2), 92-96.
[84] Maggiori, M. (2013). The U.S. dollar safety premium. 2013 Meeting Papers (No. 75).
Society for Economic Dynamics.
[85] Marcellino, M. (2008). A linear benchmark for forecasting GDP growth and in
ation?
Journal of Forecasting , 27(4).
[86] McCauley, R.N., McGuire, P.M., & Sushko, V. (2015). Dollar credit to emerging market
economies. BIS Quarterly Review .
[87] McDowell, D. (2017). Brother, can you spare a billion?: the United States, the IMF,
and the international lender of last resort. New York, NY: Oxford University Press.
[88] Mohammed, A. (2012, March 20). Of diapers and drugs, Iran's trouble paying bills.
Reuters .
[89] Nasseri, L., & Motevalli, G. (2015, April 13). Iran's middle class plans for life after a
deal. Bloomberg Businessweek , pp.14-15.
[90] Norrlof, C. (2014). Dollar hegemony: a power analysis. Review of International Political
Economy , 21(5), 1042-1070.
[91] Normand, R. (1996). Iraqi sanctions, human rights and humanitarian law. Middle East
Report , 200, 40-43.
[92] Nossal, K.R. (1989). International sanctions as international punishment. International
Organization , 43(2), 301-322.
[93] Oatley, T., Wineco, W.K., Pennock, A., & Danzman, S.B. (2013). The political econ-
omy of global nance: a network model. Perspectives on Politics , 11(1), 133-153.
[94] Obstfeld, M. & Rogo, K.S. (1995). The intertemporal approach to the current account.
Handbook of International Economics , 3, 1733-1799.
[95] Papaioannou, E., & Portes, R.(2008). Costs and benets of running an international
currency. (Economic Papers No. 348). Brussels, Belgium: European Commission.
[96] Pape, R.A. (1997). Why economic sanctions do no work. International Security , 22(2),
90-136.
[97] Pitchford, R., & Wright, M.L. (2013). On the contribution of game theory to the study
of sovereign debt and default. Oxford Review of Economic Policy , 29(4), 649-667.
117

[98] Portes, R., Rey, H., & Oh, Y. (2001). Information and capital
ows: the determinants
of transactions in nancial assets. European Economic Review , 45(4-6), 783-796.
[99] Prem, R. (1994). International currencies and endogenous enforcement (PhD thesis).
Retrieved from ProQuest Dissertations and Theses database. (UMI No. 9510528).
[100] Reinhart, C.M. & Rogo, K.S. (2010). Growth in a time of debt. American Economic
Review , 100(2), 573-78.
[101] \Remarks of Under Secretary for Terrorism and Financial Intelligence David S. Cohen"
at The Practicing Law Institute's \Coping with U.S. Export Controls and Sanctions"
Seminar, \The Evolution of U.S. Financial Power." (2014). U.S. Department of the
Treasury.
[102] Rey, H. (2015). Dilemma not trilemma: the global nancial cycle and monetary pol-
icy independence (NBER No. 21162). Cambridge, MA: National Bureau of Economic
Research.
[103] Sahay, R., Cihak, M., N'Diaye, P., Barajas, R.B., Ayala, D., Gao, Y.,…Youse, S.R.,
(2015). Rethinking nancial deepening: stability and growth in emerging markets . (IMF
Sta Discussion Note). International Monetary Fund.
[104] Scott, S.V., & Zachariadis, M. (2014). The society for worldwide interbank nancial
telecommunication (SWIFT): cooperative governance for network innovation, standards
and community. New York, NY: Routledge.
[105] Setayesh, S., & Mackey, T.K. (2016). Addressing the impact of economic sanctions on
Iranian drug shortages in the joint comprehensive plan of action: promoting access to
medicines and health diplomacy. Globalization and health , 12(1).
[106] Shin, H.S. (2012). Global banking glut and loan risk premium. IMF Economic Review ,
60(2), 155-192.
[107] Singh, M. (2016). Collateral
ows and balance sheet(s) space. Journal of Financial
Market Infrastructures , 5(1), 65-82.
[108] Sunner, D. (2017). Trends in global foreign currency reserves. RBA Bulletin , 75-84.
[109] SWIFT Annual Review 2016. (2016). Society for Worldwide Interbank Financial
Telecommunication, www.swift.com/about-us/nancials.
[110] Terada-Hagiwara, A. (2011). Asian holding of US Treasury securities: trade integration
as a threshold. Journal of the Japanese and International Economies , 25(3), 321-335.
[111] Trin, R. (1960). Gold and the dollar crisis: the future of convertibility . New Haven,
CT: Yale University Press.
118

[112] Truman, E.M., & Wong, A. (2006). The case for an international diversication stan-
dard. (IIE Working Paper No. 06-2). Washington, D.C.:Institute for International Eco-
nomics.
[113] Valadkhani, Abbas. (2003). The Causes of Unemployment in Iran. International Jour-
nal of Applied Business and Economic Research , 1(1), 21-33.
[114] Vermeulen, R., & de Haan, J. (2014). Net foreign asset(com) position: does nancial
development matter? Journal of International Money and Finance , 43, 88-106.
[115] Wallerstein, I. (2006). The curve of American power. New Left Review , 40, 77-94.
[116] Wang, T. (2016). Forecast of economic growth by time series and scenario planning
method. Modern Economy , 7, 212-222
[117] Warrick, J., & Ball, J. (2012). Food prices, in
ation rise sharply in Iran. The Wash-
ington Post .
[118] Wineco, W.K. (2015). Structural power and the global nancial crisis: a network
analytical approach. Business and Politics , 17(3), 495-525.
119