UC Santa Cruz

UC Santa Cruz Electronic Theses and Dissertations

Title
Essays in International Finance

Permalink
https://escholarship.org/uc/item/2648m23t

Author
Chertman, Fernando

Publication Date
2020

Peer reviewed|Thesis/dissertation

eScholarship.org Powered by the California Digital Library

University of California
 UNIVERSITY OF CALIFORNIA
SANTA CRUZ

ESSAYS IN INTERNATIONAL FINANCE

A dissertation submitted in partial satisfaction of the
requirements for the degree of

DOCTOR OF PHILOSOPHY

in

ECONOMICS

by

Fernando Chertman

March 2020

The Dissertation of Fernando Chertman

is approved:

Professor Michael M. Hutchison, Chair

Professor Kenneth Kletzer

Professor Chenyue Hu

Dean Quentin Williams

Acting Vice Provost and Dean of Graduate Studies
 Copyright c by

Fernando Chertman

2020
Table of Contents

List of Figures v

List of Tables vi

Dedication ix

Acknowledgments x

1 Facing the quadrilemma: Taylor Rules, Intervention Policy and Capital

Controls in Large Emerging Markets 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Macroeconomic Management in Large Emerging Market Economies . . . . . 7
1.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Data and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5.1 Baseline and Extended Full Sample Results . . . . . . . . . . . . . . 19
1.5.2 Policy Shifts and the Global Financial Crisis . . . . . . . . . . . . . 21
1.5.3 Transmission of U.S. Interest Rates and Capital Controls . . . . . . 23
1.5.4 Linkage across policies . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.6 Robustness: Extensions to China and Chile . . . . . . . . . . . . . . . . . . 27
1.6.1 Chile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.6.2 China . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.9 Appendix - Variables description, Tables and Figures . . . . . . . . . . . . . 39

2 Regional Trade Agreements and A Gravity Model of Consumption Risk

Sharing 58
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

iii
 2.2.1 Regional Trade Agreements . . . . . . . . . . . . . . . . . . . . . . . 64
2.2.2 GDP, consumption, and risk sharing . . . . . . . . . . . . . . . . . . 67
2.2.3 Geographic Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.3 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.3.1 Cross-country Risk Sharing and RTAs . . . . . . . . . . . . . . . . . 71
2.3.2 A Gravity Model of Risk-sharing . . . . . . . . . . . . . . . . . . . . 74
2.3.3 Gravity Model and RTA . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.6 Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

3 Deviations from Covered Interest Rate Parity: evaluating drivers from

changes 101
3.1 Deviations from CIP and Global Financial Crisis . . . . . . . . . . . . . . . 101
3.2 CIP analysis through time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.2.1 Theoretical evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.2.2 Empirical evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.3 Methodology and Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.3.1 The variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.4.1 Shifts and Regressions . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.4.2 Some extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3.5 Final Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
3.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
3.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

iv
List of Figures

1.1 Output Gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

1.2 Inflation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.3 Money Market Interest Rates . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.4 Exchange Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
1.5 Reserves, Reserve Adequacy and Foreign Exchange Market Intervention . . 55
1.6 Reserve Gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
1.7 Capital Openness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2.1 Current RTAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

2.2 Historical RTAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.3 Bilateral Risk Sharing before and after RTAs . . . . . . . . . . . . . . . . . 70

3.3.1 Cross Currency Swap and Basis . . . . . . . . . . . . . . . . . . . . . . . . . 112

3.3.2 Source: Elaborated by the author. . . . . . . . . . . . . . . . . . . . . . . . 115
3.3.3 Source: Elaborated by the author. . . . . . . . . . . . . . . . . . . . . . . . 115

v
List of Tables

1.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

1.2 Baseline Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
1.3 Pre and Post Global Financial Crisis . . . . . . . . . . . . . . . . . . . . . 46
1.4 Capital Account Liberalization (Openness) . . . . . . . . . . . . . . . . . . 47
1.5 Residual Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.6 Chile Policy Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
1.7 China Policy Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.1 Summary Statistics of Risk-sharing Coefficients . . . . . . . . . . . . . . . 69

2.2 Bilateral Risk Sharing and RTA . . . . . . . . . . . . . . . . . . . . . . . . 74
2.3 A Gravity Model of Risk-sharing . . . . . . . . . . . . . . . . . . . . . . . . 78
2.4 Gravity model - Robustness check . . . . . . . . . . . . . . . . . . . . . . . 81
2.5 Gravity Model with RTA . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.6.1 List of Countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
2.6.2 Bilateral Risk Sharing and RTA — Robustness . . . . . . . . . . . . . . . . 98
2.6.3 A Gravity Model of Risk-sharing — Robustness with Alternative Distance 99
2.6.4 A Gravity Model of Risk-sharing — Robustness with Projected Trade . . . 100

3.3.1 OTC foreign exchange turnover by currency pair . . . . . . . . . . . . . . . 109

3.3.2 Basis evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.3.3 Summary Statistics for Cross-Currency Swap Basis Points . . . . . . . . . 114
3.4.1 T-tests for mean differences pre x post Lehman . . . . . . . . . . . . . . . 118
3.4.2 Direction hypothesis for the dependent variables . . . . . . . . . . . . . . . 121
3.4.3 Australian dollar Cross-Currency Swaps regressions . . . . . . . . . . . . . 126
3.4.4 Canadian dollar Cross-Currency Swaps regressions . . . . . . . . . . . . . . 127
3.4.5 Swiss Franc Cross-Currency Swaps regressions . . . . . . . . . . . . . . . . 128
3.4.6 British Pound Cross-Currency Swaps regressions . . . . . . . . . . . . . . . 129
3.4.7 Singaporean dollar Cross-Currency Swaps regressions . . . . . . . . . . . . 130
3.4.8 Cross-currency Swaps panel regressions . . . . . . . . . . . . . . . . . . . . 131
3.4.9 T-tests for mean differences pre x post QE1 . . . . . . . . . . . . . . . . . 132
3.4.10 T-tests for mean differences pre x post QE2 . . . . . . . . . . . . . . . . . 132
3.7.1 G-Sibs defined by Financial Stability Board . . . . . . . . . . . . . . . . . 136

vi
 Abstract

Essays in International Finance

by

Fernando Chertman

This dissertation studies topics of international finance, such as the use of inter-

national reserves as a tool of monetary policies by Emerging Markets (EM), the impact

of trade costs in cross-country risk sharing, and the new observed deviations from covered

interest rate parity. Each chapter of the dissertation approach one of these three topics.

The first chapter investigates extended Taylor rules and foreign exchange inter-

vention functions in large Emerging Markets (EM), measuring the extent to which policies

are designed to stabilize output, inflation, exchange rates and accumulate international re-

serves. We focus on two large emerging markets - India and Brazil. We also consider the

impact of greater capital account openness and which rules dominate when policy conflicts

arise. We find that output stabilization is a dominant characteristic of interest rate policy

in India, as is inflation targeting in Brazil. Both countries actively use intervention policy

to achieve exchange rate stabilization and, at times, stabilizing reserves around a target

level tied to observable economic fundamentals. Large unpredicted intervention purchases

(sales) accommodate low (high) interest rates, suggesting that external operations are sub-

vii
ordinate to domestic policy objectives. We extend the work to Chile and China for purposes

of comparison. Chile’s policy functions are similar to Brazil, while China pursues policies

that substantially diverge from other EMs.

The second chapter empirically examines whether trade costs impede cross-country

consumption risk sharing. Using the data for a large panel of countries over the period

1970-2014, we document that bilateral risk sharing improves once a pair of countries be-

come partners under a regional trade agreement. Moreover, we establish a gravity model

of consumption risk sharing by finding that bilateral risk sharing decreases in geographi-

cal distance between countries. The effect is more pronounced in the absence of regional

trade agreements. These empirical findings support the argument that lifting trade barriers

promotes risk sharing across countries.

The third and last chapter examines the deviation from covered interest rate parity.

Being a new phenomena observed especially after the 2008 Great Financial Crisis (GFC),

we measured the deviations in a cross-country setup, and explored possible causality chan-

nels for this behavior. The empirical findings in this chapter show a big heterogeneity of

causalities per country, limiting the possibilities of a unified theoretical model to explain it.

viii
To Thais, Rebeca, and Naomi,

my biggest accomplishment.

ix
 Acknowledgments

It is hard to express in few pages how thankful I am to a lot of people that helped me

to conclude this dissertation. I thank my advisor Michael Hutchison for the continuous

support throughout the PhD study and research. His experience, knowledge and his careful

and practical approach have taught me invaluable lessons about how to do research. I

would like to thank my dissertation committee, composed by professors Kenneth Kletzer

and Chenyue Hu for their generous support. Professor Kletzer and professor Hu have always

been so encouraging and thoughtful and it has been a great privilege to share their wisdom

and experience, and work side by side with them these years.

Besides my advisors, I also would like to thank professor Eric Aldrich and professor

Amilcar Menichini for their help and support through my Qualifying Exam. Their insights

and suggestions were very important for the development of this dissertation as well. I

am also grateful to many other professors from the UCSC Economics Department: Natalia

Lazzati, Carlos Dobkins, Nirvikar Singh, Carl Walsh, Alan Spearot, Justin Marion, Grace

Gu, and Daniel Friedman.

Beyond research, the program also offered some great opportunities to develop

my teaching skills. I would like to thank professors Julie Gonzalez and Kai Pommerenke

for all the advising and help on supervising me as a Teaching Assistant. The Economics

Department Staff was always so helpful these years, and I can’t thank enough all the support

from Sandra Reebie, Susan Leach, Lisa Morgan, Inga Tromba, and Leigh Faulk.

There is quote that says “A man is nothing without friends”. This journey would

not be possible if done alone. Special thanks to my UCSC friends: Luka Kocic, Andrew

x
Barber, David Zink, Zijing Zhu, Jiayi Xu, Evan Miao, Yifei Sheng, Rachel Zhang, Kelsey

Pilch, Jae Hoon Choi, Akatsuki Sukeda, Alan Ledesma, Dongwan Choo.

Finally, my deep gratitude and love to my family. My parents, Moacyr and Beti,

and my brothers Marcio and Carla, Telma and Ranieri. My nephews and nieces, DanDan,

Lele, Dudu, and Hannah. I am nothing without you. I would also want to make a special

thank you to my best friend, girlfriend, and wife Thais Nagata. All that happened these

years would not happen without her help and support. And for my beatiful daughters,

Rebeca and Naomi, for their smile.

xi
Chapter 1

Facing the quadrilemma: Taylor

Rules, Intervention Policy and

Capital Controls in Large

Emerging Markets

1.1 Introduction

The traditional “trilemma” set of policy constraints, where a country needs to bal-

ance tradeoffs between degrees of monetary independence, exchange rate stability and con-

trolled capital account openness, has in the recent literature been extended to a “quadrilemma”

with a fourth policy goal of financial stability (Aizenman,2017). The later consideration for

emerging markets is frequently focused on stability from international financial shocks in

1
the form of sharp movements in capital flows, exchange rate instability and U.S. interest

rate fluctuations. Emerging markets have always looked beyond the domestic objectives

of inflation and output gaps, emphasized in large advanced economies and embodied in

interest rate Taylor Rules, toward external objectives.

In attempting to achieve these external objectives, emerging markets frequently

complement policy interest rates with foreign exchange market intervention and capital

controls as additional policy instruments. Given that four policy objectives are combined

with only three policy instruments (policy interest rate, intervention and capital controls),

the “Tinbergen Principle” doesn’t hold (i.e. equal instruments and objectives) and policy

makers may at times face tradeoffs in achieving all their goals. In this context, the IMF

(2012) finds that the number of countries actively managing their exchange rates has in-

creased substantially since the Global Financial Crisis and that Brazil, Chile, Colombia,

Turkey, and other emerging markets with announced inflation targeting regimes have in-

creased both the frequency and the size of their interventions. Changes in capital controls

are also a powerful macroeconomic management tool in some emerging markets (Fernandez

et al., 2015), but are generally used infrequently.

Theoretical work has investigated the tradeoffs associated with domestic and ex-

ternal policy objectives, and where intervention and capital controls may contribute to

macroeconomic and financial stability (e.g. Gonçalves (2008), Cavallino (2019), Farhi and

Werning (2012), Jeanne (2012)). For example, the theoretical framework of Gonçalves

(2008) argues that official accumulation of foreign reserves may be perceived as interven-

tions to influence the exchange rate, undermining the credibility of floating exchange rates

2
and inflation targets. He develops a theoretical framework to study the interaction between

reserve accumulation and monetary policy, and highlights the trade-off between the speed

of reserve accumulation and anti-inflationary credibility.

In related work, Cavallino (2019) develops a New Keynesian small open economy

model that characterizes the optimal use of foreign exchange intervention in response to ex-

change rate fluctuations driven by capital flows. In his model, an increase in foreign demand

for domestic assets appreciates the domestic currency and generates a boom-bust cycle in

the economy. In response to such a shock, the optimal foreign exchange intervention in his

model is to lean against the wind and stabilize the path of the exchange rate. By lean-

ing against the wind, the central bank reduces the real appreciation (and the consumption

boom triggered by the inflow of capital) and reduces the output gap. It is not optimal for

the central bank to fully stabilize the exchange rate in this framework since it reduces some

of the benefits of portfolio capital flows.

Most empirical work on macroeconomic policy functions, especially for advanced

economies, emphasize policy interest rates as reflected in Taylor rules. Taylor rules for

emerging markets often recognize external considerations by including an exchange rate

stabilization objective, e.g. Aizenman et al. (2011). We extend previous work investigating

modified Taylor rules by considering a second policy rule linking foreign exchange market

intervention to exchange rate stability and an objective to accumulate reserves to a tar-

get level. Specifically, we explore how large emerging-market economies have in practice

managed to accumulate substantial reserve levels over time (for precautionary purposes,

reducing the likelihood of financial instability), despite substantial cyclical variation, while

3
at the same time following monetary policy rules designed to stabilize inflation, output and

the exchange rate.

We focus on two policy instruments, interest rates and foreign exchange market

intervention, and four policy objectives—inflation, output, exchange rates and foreign re-

serve target. Against this background, we also investigate (1) the impact of changes in the

intensity of capital controls, though this instrument is only infrequently cyclically applied in

most EMs, and the impact of the transmission of U.S. interest rates; and (2) cases of very

large discretionary (unpredicted) intervention operations and interest rate changes, eval-

uating whether the interest rate instrument (internal balance) or intervention operations

(external balance) dominate when policy conflicts arise. Although not able to capture all

aspects of the quadrilemma with our analysis, we are able to shed light on practical policy

considerations for internal and external balance in the use of the two major tools - monetary

policy and intervention policy.

Our primary interest is in two large emerging market economies, Brazil and India,

with a comparative analysis of the largest EM, China, and one small open economy, Chile.

Most theoretical and empirical work in this area focuses on small open economies (SOEs)

and attempts to measure where each country lies on a spectrum of policy tradeoffs. However,

large emerging markets should display somewhat different characteristics than SOEs in the

reserves-exchange rate-monetary policy nexus. In particular, large EM interest rates should

not in principle be completely determined by the “center country” (some inherent monetary

independence compared with the SOEs) and potential foreign capital inflows are not infinite

(as in the SOE model).

4
 Brazil and India use capital controls extensively as a macroeconomic management

tool. Although India has been gradually reducing capital controls over the past two decades,

it continues to have quite strict international capital controls. Brazil is much more open

financially but continues with fairly extensive controls. According to the Fernandez et al.

(2016; updated online June 2019) data set on capital control restrictiveness using the IMF

Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER) as the

underlying data source, India and Brazil placed 0.93 and 0.65, respectively in 2017. (The

range is from 0 with no restrictions to 1 as completely closed). The authors characterize

India with “walls” to external financial flows and Brazil with a “gate.” Net liberalization

has occurred over the past two decades as corresponding values for India and Brazil in

2000 were 1.0 and 0.85, respectively1 . (The U.S. had a restrictiveness index of 0.16 in 2017

and 0.13 in 2000 using this methodology). This allows us to explore whether variations

in this instrument has impacted the effectiveness of other instrument of macroeconomic

management.

These emerging markets have also experienced very large reserve accumulations,

motivated at least in part by the desire to reduce the likelihood or severity of financial crises.

This fact, in combination with active foreign exchange policies, is an important element of

macroeconomic and macro-prudential management. However, their stated macroeconomic

policies and monetary regimes are very different. In particular, the Central Bank of Brazil

has had an explicit inflation targeting regime since 2001 while the Reserve Bank of India is
1
China is also characterized by Fernandez et al. (2016) as having “walls” with a capital account re-

strictiveness measure of 0.85 in 2017 and 1.0 in 2000. Chile is more much more open, with a restrictiveness

measure of 0.45 in 2017 (and 0.88 in 2000).

5
characterized by substantial discretion in policy actions2 .

We empirically evaluate the significance of these regime differences on Taylor rules

as well as intervention policy functions, and whether capital controls influence policy ac-

tions and the transmission of U.S. interest rate changes to policy rates. We also consider

whether interest rate policy (internal balance) dominates or is subordinate to intervention

policy (external balance) when policy conflicts arise. We use time-series methods for our

methodology and employ quarterly data. Additional features of our analysis are the incor-

poration of a measure of “adequate” reserves, calculated by the IMF, into our intervention

equation, and a measure of capital account openness, based on the work of Pasricha et al.

(2015) and Pasricha (2017), into the interest rate rule (Taylor rule) and intervention rule

equations.

We include China in our study as a counterpoint to the other large EMs. As

China’s institutions are quite different, it is an interesting comparison case. And, as a

counterpart to our analysis of large emerging markets, we also consider a small commodity-

based emerging market - Chile. Chile is a small open economy, largely commodity-based

and with very open capital markets. We investigate whether the revealed policy choices for

large emerging markets carry over to small emerging markets like Chile. The remainder of

the paper is organized as follows. Section 2 presents some background on macroeconomic

management and external considerations in Brazil and India. Section 3 presents the basic

model. Section 4 presents data and methodology. Section 5 presents the empirical results

for Brazil and India. Section 6 extends the analysis to China and Chile. Section 7 concludes.
2
Chile also has an inflation targeting regime, while the People’s Bank of China monetary policy demon-

strates substantial discretion.

6
1.2 Macroeconomic Management in Large Emerging Market

Economies

Our focus emerging markets - India and Brazil- have experienced challenges to

macroeconomic and financial stability similar to other emerging markets and advanced

economies. Managing domestic output and inflation objectives in tandem with exchange

rate and balance of payments stability has frequently been a balancing act between multi-

ple targets and limited policy instruments. Neither of these countries explicitly state that

they follow a Taylor rule in setting interest rates, but in monetary policy statements note

that inflation is a priority and usually point to the state of the economy as a considera-

tion in setting policy. Our objective is to quantify the relative importance of these factors.

Similarly, authorities rarely provide an explicit intervention policy guide but ex post pol-

icy statements often refer to “disorderly” exchange market conditions, reserve and current

account developments, and so forth in explaining their actions. Again, our objective is to

quantity, if possible, the relative weight that these various considerations play in system-

atically influencing intervention operations. Previous research and policy statements help

guide us in our empirical specifications.

In particular, the Reserve Bank of India formally states that its primary objec-

tive is to maintain price stability, while “. . . keeping in mind the objective of growth” and

announced recently a “flexible inflation targeting” regime3 . Empirical work has found that
3
The Reserve Bank of India (July 2019) states that the goals of monetary policy are:

“The primary objective of monetary policy is to maintain price stability while keeping in mind the objective

of growth. Price stability is a necessary precondition to sustainable growth.” Moreover, in May 2016, the

Reserve Bank of India (RBI) Act, 1934 was amended to provide a statutory basis for the implementation of

7
India alternates between an emphasis on output and inflation in pursuing domestic macroe-

conomic stability (Hutchison et al. 2013; Gupta and Sengupta, 2014; Kaur, 2016), and

maintaining orderly conditions in the foreign exchange markets as an official objective of

the Reserve Bank of India (RBI) (Hutchison and Pasricha, 2016). RBI is the manager of

the foreign exchange regulation act (FEMA, 2004), which also gives it the power to im-

pose capital controls. In practice, this objective has meant very active management of

controls on international capital movements and frequent foreign exchange market inter-

vention operations, as well as at least one episode (in 2013) of interest rate defense of the

exchange rate. These considerations make understanding the linkages between monetary

policy, capital controls and foreign exchange market intervention operations central to a

study of macroeconomic management in India.

Hutchison and Pasricha (2016) find that India has followed active foreign exchange

market intervention and capital control policies. They argue that intervention policy is

mainly directed toward limiting exchange rate appreciation, during which times dollar pur-

chases were generally large, and not directed toward limiting depreciation. This policy may

have allowed relative stability in the real exchange rate, hence maintaining India export

competitiveness, as the exchange rate depreciated over longer-periods to offset relative high

inflation in India. Intervention policy and exchange rate depreciation also allowed greater

monetary autonomy, especially during a period associated with increased financial liberal-

ization of the international capital account. Moreover, reserve accumulation—through USD

purchases on the foreign exchange market—is a desirable objective to the extent that it pro-

vides a stock of precautionary reserves in the event of a balance of payments/currency crisis

the flexible inflation targeting framework.

8
or sudden stop in private capital inflows that generally finance persistent current account

deficits in India. On the other hand, the exchange rate has not been a “nominal anchor”

for monetary policy in India, and as a consequence high inflation is a recurring problem.

Control of international financial capital movements is another policy instrument

that has been frequently employed to influence financial flows in and out of India and the

exchange rate (Hutchison et al. (2012); Patnaik and Shah, 2012; Hutchison and Pasricha,

2016). Although the overall trend was towards financial liberalization of the capital account,

capital control actions (i.e. tightening and easing of restrictions on capital flows) have been

actively used as an instrument to “lean against the wind” of exchange rate pressures in

both directions. Whether or not capital controls policies have been effective is evaluated by

Patnaik and Shah (2012).

Similar, tradeoffs between domestic and external objectives have also confronted

the Central Bank of Brazil. The country is the largest emerging market to adopt an inflation

targeting regime (IT), starting in July 1999 and formally continuing to date. Cortes and

Paiva (2017) argue that the Central Bank of Brazil (BCB) succeeded in anchoring inflation

expectations and gaining credibility until 2011, when a new discretionary-based policy was

adopted despite a formal IT rule. However, it is evident from numerous policy statements

that output stabilization is also an important element in setting interest rate policy in Brazil.

Minutes from a recent monetary policy report from the Central Bank of Brazil (2019), for

example, note that: ”The Copom members assessed that economic conditions with anchored

inflation expectations, underlying inflation measures at appropriate levels, 2020 inflation

projected around or slightly below target, and high level of slack in the economy prescribe

9
stimulative monetary policy, i.e., interest rates below the structural interest rate level. The

structural interest rate is a reference for the conduct of monetary policy”4 . Hence, in this

case it is also of interest to measure the weights the central bank places on the inflation

target as opposed to output stabilization and other factors in setting interest rates. Other

factors may include the exchange rate. For example, Aizenman et al. (2011) find that

commodity-based emerging markets with an IT regime such as Brazil are still very likely

to smooth exchange rates as part of their Taylor Rule interest rate setting policy.

The Central Bank of Brazil also intervenes in the foreign exchange market to

smooth excessive exchange rate volatility and to manage the level of international reserves

(Gnabo et al., 2010). Although intervention activity varies over time, waning in recent years,

spot-market interventions and the sale of exchange swaps are predominantly against the

wind in terms of USD. In terms of the effectiveness of intervention, several studies find that

FX intervention, including through swaps, can affect the exchange rate, e.g. Kohlscheen and

Andrade (2014), Barroso (2014), Chamon et al. (2017), and Oliveira and Novaesk (2007).

Oliveira and Novaesk (2007), for example, find that in periods of relative tranquility the

level of the exchange rate is affected more strongly by interventions (in both the spot and

the derivatives markets) than the stance of monetary policy, while interventions appear

ineffective during episodes of high exchange rate volatility.

4
Minutes of the 223rd Meeting of the Monetary Policy Committee (Copom) Banco Central do Brasil,

June 18-19, 2019. Italics in the quote are our own.

10
1.3 Model

The basic analytical framework consists of two policy rules: a modified Taylor rule

and a foreign exchange intervention policy function. Policy is directed toward achieving

two domestic objectives, output and inflation stabilization, and two international macroe-

conomic objectives, exchange rate stabilization and a target level of international reserves

to reduce the risk of capital stops and financial instability. Two instruments are associated

with policy functions, and one instrument, fluctuations in capital controls, is taken as a

pre-determined variable. In addition to the two policy reaction functions, foreign exchange

market is directly linked to changes in international reserves through an accounting identity.

The Taylor rule is modified to capture the central bank’s objective of reducing

output variations around trend, inflation variations from target, and stabilize the nominal

exchange rate. Given hysteresis found in policy actions we include a lagged interest rate as

is standard in most studies. The modification of the Taylor rule to include an exchange rate

target is standard in the emerging markets literature (e.g. Aizenman et al., 2011). This

formulation takes the form:

it = α1 + α2 (yt − y ∗ ) + α3 (πt − π ∗ ) + α4 (et − et−1 ) + α5 it−1 + εt (1.1)

where it is the central bank interest rate operating instrument, (yt − y ∗ ) is (log) output

less (log) output trend (i.e. percentage deviation from trend output), (πt − π ∗ ) is inflation

deviation from target, (et − et−1 ) is the (log) nominal exchange rate change, and εt is the

error term. Stabilizing objectives (“leaning against the wind”) of output, inflation and the

exchange rate suggests that α2 > 0, α3 > 0, and α4 > 0.

The foreign exchange management fund is postulated to intervene in the foreign

11
exchange market (foreign exchange purchases are positive values) to stabilize the exchange

rate and to manage foreign reserves around the target level. Hence, there are potentially

two instruments focused on exchange rate management. In addition, the target level may

itself vary over time as suggested by the very rapid buildup of international reserves by

emerging market economies during the period prior to the Global Financial Crisis (GFC).

The intervention equation takes the form:

It = β1 + β2 (et − et−1 ) + β3 (Rt − Rt∗ ) + µt (1.2)

where It is foreign exchange market intervention (USD purchases (purchases of foreign

exchange are positive values and sales are negative values, as a percent of last quarter’s

stock of international reserves), (R − R∗ ) is the (log) stock of international reserves less

the (log) of the target reserve level (i.e. percentage deviation from target reserves) and µt

is the error term. Foreign exchange sales intervention to slow exchange rate depreciation

(et − et−1 > 0) suggests β2 < 0. A rise in the stock of reserves above the target value also

suggests foreign exchange sales intervention, β3 < 0.

Intervention is linked to international reserves through an accounting identify, i.e.

the rise (fall) in international reserves equals foreign exchange intervention purchases (sales)

plus interest earnings on foreign reserves and valuation changes:

Rt − Rt−1 = It−1 + i∗t−1 Rt−1 + VALt−1 (1.3)

where i∗t−1 is the interest rate on foreign exchange reserves and VALt−1 is valuation changes

on international reserve holdings. Hence, intervention is directly linked to the target for

international reserves. Our assumption is that i∗t−1 and VALt−1 are exogenous variables.

12
 As extensions of the basic models represented by equations (1) and (2), we also

include the terms-of-trade and the current account in both equations. A rise in either the

terms-of-trade or the current account have wealth and liquidity effects on the economy

and could elicit a monetary response. Similarly, a terms-of-trade change could impact

the foreign exchange market (increasing foreign exchange receipts), as could a rise in the

current account by increasing liquidity in the market. Both of these variables also have

proved important in other studies of macroeconomic policy in EMs (e.g. Aizenman et al.

2011).

We also investigate the extent to which U.S. interest rates (i∗t ) and capital ac-

count openness (opennesst ) constrain domestic interest rate policy (Taylor rule) and, for

(opennesst ), enters into decisions to intervene in the foreign exchange market. We would

expect U.S. interest rates to enter directly into interest rate policy decisions, in addition to

the indirect channel via the exchange rate, especially in the post-GFC period when greater

movement of international capital was generally allowed in both Brazil and India. The effect

of greater capital market openness (liberalization) on both interest rate and intervention

policies would depend on the directional response of net private capital flows, which in turn

on market conditions and whether institutional measures liberalized controls on inflows or

outflows.

13
1.4 Data and Methodology

1.4.1 Data

We employ quarterly data over the period 1999q1-2018q4 in our analysis. The

exact sample period varies slightly between regression specifications due to data availability.

Descriptions of each variable and the date range over which they are available are explained

in the appendix5 .

Macroeconomic developments for both countries are detailed in the summary

statistics of Table 1 and Figures 1-7. Panel A of Table 1 shows the full sample period,

Panel B shows the pre-GFC crisis sample period and Panel C shows the post-GFC crisis

period. India generally has a much more stable macro-economy than Brazil, with lower

interest rates, lower inflation and more stable (lower standard deviation) exchange rates,

intervention and reserves (relative to “adequate” reserves)6 . Figure 1 shows the output gap;

Figure 2 inflation (and, for Brazil, evolution of the inflation target); Figure 3 money market

interest rates; Figure 4 exchange rates (left panel, level of the domestic currency per USD;

right panel, percent change); Figure 5, left column, is the level of international reserves and

the “adequate reserves” level (estimated by the IMF) and the right column is the net spot
5
Two appendices - sources of data and detailed variable definitions - are omitted from the text for brevity

but are available from the authors upon request.

6
It is an intriguing question as to why Brazil has had a much more volatile economy than India, with

prime candidates more restrictive capital controls in India and, hence, less volatile capital movements; more

volatile external shocks in Brazil associated with dependence on commodities and terms-of-trade fluctuations;

and so on. Our focus is not in addressing this issue but to compare monetary and intervention policies in the

two countries. Differences in policies, however, may play an important role in explaining relative volatility

of these economies.

14
foreign exchange market intervention; Figure 6 is the reserve gap (difference between actual

reserves and adequate reserves as a percent of adequate reserves; Figure 7 is the measure

of cumulative step of external capital account openness (cumulative net changes).

We use a standard measure of the output gap given by the cyclical deviation of

industrial production from its trend. We seasonally adjusted both series using the U.S.

Census Bureau X-13 procedure. HP filter estimates of the logged series are employed

to obtain trend and cyclical output measures. The cyclical portion is multiplied by 100,

yielding an output gap measure that can be interpreted as the percent deviation of industrial

production from its trend level. The output gap measures are shown in Figure 1. This series

has been employed in other studies investigating monetary policy in both Brazil and India.

(Kaur, 2016; Gupta and Sengupta, 2014; De Almeida, 2003). It is evident from the figure

that output gap volatility has been much larger in Brazil than India.

As noted, Brazil has had an inflation target since 1999. This target has changed

several times over the sample period, shown in Figure 2, but for most of the sample the

midpoint target was 4.5%. India does not have an announced inflation target. For purposes

of econometric estimation, we assume the target is constant and therefore subsumed in the

constant term of the estimated Taylor rule for India. We follow other studies (e.g. Gupta

and Sengupta, 2014; Modenesi et al., 2013) and use the WPI index to construct the inflation

rate in India and the IPCA index for Brazil. Inflation averaged 4.7% in India and 5.2% in

Brazil over the sample period, with similar volatility, shown in Table 1. Brazil has been

slightly above its inflation target over the sample period (0.4% above).

Money market interest rates are employed in both studies, shown in Figure 3.

15
Despite similar inflation rates, Brazil has almost double the nominal (and real) interest rates

than India. This may reflect both real growth equilibrium factors (determining equilibrium

real interest rates), risk premium differences, institutional features of the two economies,

and that Brazil is more financially open. The stance of monetary policy is measured with

the money market interest rate. For India, this is the 3-month interbank lending rate. For

Brazil, we use the SELIC rate, which is the overnight interbank lending rate. The nominal

exchange rate employed in the study, shown in Figure 4, is the value of local currency against

the USD. Brazil has experienced higher average depreciation (1.0% quarterly average) over

the sample than India (0.7% quarterly average), shown in Table 1, and much higher exchange

rate volatility.

Foreign exchange market intervention is defined as foreign currency purchases (do-

mestic currency sales) in the foreign exchange market, valued in millions USD, shown in the

right panels of Figure 5. This data is obtained from the Central Banks of Brazil and India,

respectively. Negative values represent foreign currency sales (domestic currency purchases)

in the foreign exchange market. The advantage of this measure is that it only reports active

intervention in the foreign exchange market and excludes interest earnings and valuation

effects on reserves. (Many studies proxy intervention by changes in reserves). Both coun-

tries actively intervened in the foreign exchange market during most of the sample period,

though Brazil ceased its intervention activity in recent years.

Reserves are defined as international reserves less gold but including SDRs, shown

in the left panels of Figure 5. Reserve data for Brazil and India are obtained from the central

bank of each country. No reserve targets are announced in either country. As a proxy, we use

16
the IMF series on reserve adequacy for both Brazil and India. The IMF defines international

reserve adequacy (RA) for emerging market economies with floating exchange rates as RA =

(5%×Exports)+(5%×Broad Money)+(30%×Short Term Debt)+(15%×Other Liabilities).

The IMF measure of reserve adequacy is only available at the annual level. An approximate

quarterly series is estimated using a cubic spline interpolation. The resulting quarterly series

are also plotted in the left panels of Figure 5. It is apparent that both countries grew reserves

very substantially since the early 2000s, pausing at the time of the GFC. After that period,

reserve growth in reserves continued in India and flattened out in Brazil.

The reserve “gap,” measured by the difference between actual reserves and reserve

adequacy (as a percentage of reserve adequacy), is shown in Figure 6. This figure shows

that India exceeded its “reserve adequacy” metric from around 2002, peaking at almost

100% just before the GFC. Since that time, the reserve gap declined before stabilizing at

about 30%. Brazil’s reserve gap was negative until 2007 but has been consistently positive

since 2010, fluctuating around 50% from 2014 until 2018.

Capital Openness Index, shown in Figure 7, is taken by accumulating net capital

account liberalization or restrictiveness changes based on the Pasricha et al. (2015) dataset,

updated in Pasricha (2017). This is a dataset of capital control actions for 16 emerging

market economies, where country-level measures of capital control changes are based on a

weighted sum of the capital account changes for a given year, where the weights are given by

the share of the country’s international investment position that are affected by the policy

change. We take the cumulative sum of these changes so that they can be interpreted as

the level of capital openness for a given country, albeit not comparable across countries in

17
level form. The resulting time series for Brazil and India is shown in Figure 7. This index

has been used in Pasricha et al. (2015), Pasricha (2017), and Aizenman and Binici (2016).

Some of the advantages of this series are that it results in a measure of capital openness

that varies more regularly than several measures such as the Chinn-Ito index (Chinn and

Ito, 2006) or Fernandez et al. (2016). This is because it presumably takes into account

all regulatory changes for a given country and weights them according to their estimated

impact on capital flows.

1.4.2 Methodology

Turning to methodology, our baseline time series models for Brazil and India are

estimated over the 1999q1-2018q4 period. We allow for sample shifts before (1999q1-2008q4)

and after the Global Financial Crisis (2009q1-2018q4), as the external environment changed

markedly at this time, likely impacting policy behavior. We employ a methodology that

considers the endogeneity of the reserve gap. The contemporaneous reserve gap is influenced

by the scope of intervention operations. Consequently, we treat the reserve gap variable as

endogenous and instrument for it with its lagged value. Exchange rate fluctuations are likely

to suffer from a two-way causality issue as well. However, we do not employ instrumental

variables for the exchange rate. There are two reasons for this decision. First, exchange

rates are notoriously difficult to predict and thus finding a strong instrument is a daunting

task. Weak instruments lead to results that perform poorer than OLS estimates (Stock,

Wright, and Yogo 2002), and it isn’t clear that instrumenting for the exchange rate leads to

improved estimates. The second reason is that the bias of the exchange rate coefficient works

against our hypothesis. This is because lower interest rates and foreign currency purchases

18
lead to exchange rate depreciation, whereas we expect depreciation to cause higher interest

rates and purchases of domestic currency. Our results for the exchange rate can therefore

be interpreted as a lower bound on the true effect of exchange rates on interest rate and

intervention policy. Both inflation and the output gap are assumed to respond to interest

rate changes only with a lag and are treated as pre-determined variables. We estimate HAC

Newey-West standard errors to account for potential autocorrelation and heteroscedasticity

in the error term.

1.5 Results

1.5.1 Baseline and Extended Full Sample Results

Table 2 shows the full-sample baseline results for Brazil and India (column 1),

together with the extended model including the terms-of-trade and the current account

(column 2). Panel A reports the extended Taylor rule model estimations and Panel B

the intervention functions. Spot intervention operations are employed in the intervention

function estimates reported in Panel B7 .

The results shown in Panel A indicate very different monetary policies pursued

by India and Brazil over the full sample period. India has systemically pursued output

stabilization, raising domestic interest rates on average by 11 basis points in response to a

one percentage point rise in the output gap. We find no evidence that the Reserve Bank of

India systematically responds to inflation or exchange rates in setting money market rates
7
We also considered a measure of intervention aggregating spot and forward transactions. The results

were unchanged, omitted for brevity, and are available from the authors upon request.

19
over the full sample period. Brazil, on the other hand, responds strongly to deviations from

its inflation target, confirming the central bank’s commitment to an IT regime, increasing

the interest rate by 60 basis points for every 1 percentage point above the inflation target.

The extended results also suggest that the Central Bank of Brazil responds to exchange rate

depreciation by raising interest rates. In sharp contrast with India, no output stabilization

by Brazil’s central bank is indicated over the full sample.

The additional variables (terms-of-trade and current account) of the extended

model do not appear significant for India, but the terms-of-trade does enter significantly for

Brazil. An improvement in the terms-of-trade in Brazil is associated with a (statistically

significant) decline in interest rates. Interest rate policy is highly persistent in both coun-

tries, especially in India (lagged dependent variable coefficient equals 0.81-0.82 in India and

0.65-0.66 in Brazil).

Although following quite different Taylor rules, India and Brazil are similar in

foreign exchange market intervention policy responses to exchange rate changes, shown

in Panel B of Table 2. Both countries respond strongly to exchange rate movements in

“leaning against the wind” intervention operations, selling (buying) about 0.17-0.22% in

Brazil and 0.30-0.48% in India, of the stock of international reserves in response to a one

percent depreciation (appreciation) of the domestic currency against the USD.

Only India appears to systematically target reserves around a level associated

with observable economic fundamentals. A rise (fall) in actual reserves above (below) the

target induces a significant sale (purchase) in foreign exchange (as a percent of last period’s

total reserves)8 . Differences also emerge between the two countries in terms of responses
8
This result is statistically significant in the baseline model at the 1% level, but not statistically significant

20
to terms-of-trade fluctuations and the current account. A terms-of-trade improvement in

Brazil reduces U.S. dollar intervention purchases - most likely attributable to higher foreign

exchange earnings for Brazilian exports. No intervention response is noted to changes

in the current account in Brazil. By contrast, the current account is estimated to be

highly significant for intervention policy in India, with a rise in the surplus (as a percent of

GDP) leading to a significant increase in U.S. Dollar purchases, perhaps absorbing excess

liquidity generated by the surplus in the foreign exchange market in the face of fairly

restrictive capital controls. Although the exchange rate response remains significant in

Indian intervention policy, albeit weaker than in the basic equation, targeting of reserves

is no longer statistically significant (although the coefficient estimate is very similar, it is

measured with less precision).

It is noteworthy that both India and Brazil built very substantial foreign exchange

reserve positions during the sample period. This is reflected in the empirical model by the

significant positive constant terms in the intervention regressions, indicating substantial

average foreign exchange purchases (as a percentage of existing reserves).

1.5.2 Policy Shifts and the Global Financial Crisis

We address whether policy shifts occurred at the time of the GFC in Table 3,

comparing the pre-GFC 1999Q1-2008Q4 period with the post-GFC 2009Q1-2018Q4 period.

We present both the baseline model and the extended model in Table 3, but focus our

discussion on the extended model results.

The full sample results on output and inflation carry over to the sub-sample re-
in the extended model.

21
sults—during both sub-samples India focused on output stabilization and Brazil focused

on inflation targeting. Nonetheless, we find some evidence that India began responding to

inflation deviations in the post-crisis period9 and also to terms-of-trade changes in both

pre- and post-crisis samples. The current account is only statistically significant for India

in the pre-crisis sample.

As stated, inflation targeting dominated the Central Bank of Brazil’s interest rate

policy in both sub-periods, as it did in the full sample period, but the estimated response is

weaker in the post-GFC period10 . This finding sheds some light on the concern that Brazil

is adhering less to inflation targeting in recent years (Cortes and Paiva, 2017). However,

no output response is estimated in Brazil in either sub-period, nor is there evidence of

systematic responses to exchange rates, terms-of-trade or current account movements.

Exchange rate stabilization is a dominant feature of intervention policy for India

in the pre and post-GFC, with quite similar responses, as for the full sample period. All

the coefficient estimates are significant at the 5% level or better. By contrast, the estimates

for the two sub-samples in Brazil are not statistically significant (unlike the full sample).

Stronger responses are suggested in the management of foreign exchange reserves

in India from the pre to the post-GFC11 , and the response in the latter period - selling

foreign exchange when reserves are above target - is consistent with a stabilizing role. The

response for the reserve gap is significantly negative in Brazil both periods, with policy
9
The coefficient is 0.04 (not statistically significant) for the early period and 0.03 (statistically significant)

for the later period. The difference in coefficient values is not statistically significant.
10
However, this difference in coefficient estimates is not statistically significant at conventional levels

(z-statistic 0.96).
11
The z-statistic measuring differences in coefficient estimates is 2.53 (significant at the 5% level).

22
targeting a desired reserve level, and the coefficient estimates are similar. The terms-of-

trade played a role in intervention policy for both countries in the pre-GFC period, but not

in post-GFC period. A rise in the current account surplus induced USD purchases in both

periods for India, probably to absorb surplus liquidity in the foreign exchange market and

limit pressure on the Rupee to appreciate in the face of capital controls. Surprisingly, the

opposite result is obtained (negative and statistically significant) for Brazil in the post-GFC

period.

1.5.3 Transmission of U.S. Interest Rates and Capital Controls

In this section we explore the extent to which policy interest rates in India and

Brazil are directly tied to U.S. interest rates in addition to the indirect link via the exchange

rate. We also consider the impact of external financial account openness on policy interest

rates and foreign exchange market intervention policy.

The results are reported in Table 4. U.S. interest rates did not move enough

during the post-GFC, encompassing the zero-lower-bound period, to warrant inclusion in

the sample so only the pre-GFC period is presented in our Taylor rule equation estimates.

Column (1) in Panel A for India and Brazil include the U.S. interest rate in the baseline

Taylor rule regression, while column (2) reports estimates with the U.S. interest rate and

openness. The estimates indicate that domestic money market rates move about 18-27

(Brazil) to 24-25 (India) basis points for a 1 percentage point move in U.S. interest rates,

though only the estimates for India are statistically significant.

The results in Table 4 suggest quite different policy responses to capital account

liberalization in India and Brazil. For India, in the pre-GFC period, an increase in openness

23
led to lower money market interest rates (8 basis points, Panel A) and sales of foreign

exchange (0.97 percent of reserves) by the central bank (Panel B). No significant impact on

intervention policy from greater openness is seen in the post-GFC. In Brazil, steps toward

greater openness (restrictiveness) also is associated with lower (higher) domestic interest

rates (61 basis points), but prompted the purchase of foreign currency by the central bank

in the pre-GFC (6.17 percent of reserves) and sales of foreign currency in the post-GFC

(1.5 percent of reserves).

These differences may be explained in part by how the pattern of financial market

liberalization/openness and market conditions affected net capital flows in the two peri-

ods and across the two countries, leading to varying policy responses. Shown in Figure

7, India—though much more financially closed generally than Brazil—set out on a gradual

process of external financial liberalization over the sample period. The number of liberaliza-

tion measures (positive steps in the figure) far exceeded the number of restrictive measures

(negative steps in the figure), so that over 50 net liberalization steps were taken between

2001 and the end of 2015. Brazil, on the other hand, used capital control more as a cyclical

policy instrument, at times loosening and at times tightening controls. The number of net

liberalization steps (positive) only slightly outnumbered the number of restrictive (negative)

steps over course of the full sample.

For India, it appears that a rise in openness led to net capital outflows in the

pre-GFC, perhaps because of a tendency to liberalize outflows more than inflows, indirectly

creating incipient pressure for currency depreciation, and in turn prompting the central

bank to “absorb” the impact on the foreign exchange market by selling foreign exchange

24
(an official capital inflow). Less private capital inflow may also have adversely impacted

domestic investment, leading the Reserve Bank of India to respond by lowering the policy

rate. The effect of liberalization of inflows and outflows may have been more balanced

post-GFC as no impact on intervention operations is found.

The results for Brazil, on the other hand, suggest that an increase in openness led

to a surge in net private capital inflows during the pre-GFC period, leading the central bank

to offset the impact on the foreign exchange market by making large USD purchases. The

capital inflow associated with greater openness during pre-GFC was also associated with

lower money market rates, suggesting that the central bank allowed private capital inflows

to loosen domestic financial market conditions. The contrasts with post-GFC, where a net

increase in openness was associated with net capital outflows and official sales of foreign

exchange reserves. Liberalization in this period may have been more directed to relaxation

of controls on outflows than inflows or attributable to adverse market conditions.

1.5.4 Linkage across policies

Tradeoffs between interest rate and intervention policies are not explicitly ad-

dressed using our basic methodology. It is possible that “errors” in one policy function, i.e.

deviations from predicted values, are discretionary policy actions connected to the second

policy function. For example, unexpectedly low interest rates (intervention) may be linked

to unexpectedly low intervention (interest rates) as authorities are attempting to manage

the exchange rate via the Taylor rule rather than direct intervention operations. In other

words, there may be tradeoffs and substitutions between the internal and external policy

functions that are manifested in the error terms.

25
 We address this issue in two ways. Our first approach is to estimate the two

equations using a Three Stage Least Squares (3SLS) systems estimator12 . This method

takes into account systemic linkages among the errors of the two policy equations while

also accounting for the endogeneity of the reserve gap in the intervention equation. The

estimates, not reported for brevity are virtually identical to the extended model results

reported in Table 2, column 2 for both India and Brazil’s interest rate (Panel A) and

intervention (Panel B) policy equations13 . This indicates that the error terms in the two

equations are not significantly correlated in a simple way. This is confirmed by the simple

error correlations across the two equations– statistically insignificant correlation coefficients

of -0.16 (standard error 0.11) for India and 0.02 (standard error 0.11) for Brazil.

We also explore possible linkages between large policy errors in the two equations

since policy tradeoffs or conflicts may only be manifested during particular episodes. For

example, a country may not respond to substantial pressure on the exchange rate in the

Taylor rule if domestic conditions are clearly not warranting an interest rate change, placing

greater emphasis on intervention policy. We identify the intervention policy errors (interest

rate policy errors) that are equal to or larger than the 90th percentile in absolute value and

regress these on the associated interest rate policy (intervention policy) function errors in

Table 5. These results indicate that the equations are related in a highly non-linear way.

In particular, large intervention policy errors in both India and Brazil are negatively and
12
Greene (2012) shows that the seemingly unrelated regressions model, estimated equation by equation, is

inefficient compared with an estimator that makes use of the cross-equation correlations of the disturbances.

Following Greene (2012), we estimate both equations jointly with a three-stage least squares estimator (the

IV estimator is simply equation-by-equation 2SLS). This procedure is asymptotically efficient.

13
These results are available from the authors upon request.

26
significantly correlated with corresponding interest rate policy errors. That is, unexpectedly

large USD purchases (sales) by the foreign exchange fund are associated with lower (higher)

than predicted interest rates. This suggests that episodes of especially large unexpected

intervention purchases/sales may be designed to limit the need for interest rate changes in

macro policy management. Interestingly, we do not find that large interest rate errors are

correlated with associated intervention errors14 . Discretionary intervention policy actions

appear to serve as a “pressure valve” when policy conflicts arise, subordinate to interest

rate policy.

1.6 Robustness: Extensions to China and Chile

In this section we contrast our results for Brazil and India with two other emerging

markets, Chile and China. The contrasts between Chile and China are stark. Chile is a small

open economy with inflation targeting, high dependence on commodity exports, flexible

exchange rates and a very open capital account. China, on the other hand, is the largest

emerging market—the second largest economy in the world after the United States—with

discretionary monetary policy, dominance of manufacturing exports, rigid exchange rate

and largely closed to (non-FDI) external capital flows. China is also characterized by heavy

government involvement in the financial sector, government majority ownership in large

banks, and regulated interest rates.

Chile is included to check the robustness of the results to a small open market-

oriented EM with high dependence on commodity exports and a policy commitment to

14
Not reported for brevity but available from the authors upon request.

27
inflation targeting. China, of course, is the obvious choice to include in our study due

simply to its importance to the world economy, rapid growth and buildup of international

reserves. It is not a country of emphasis in this study, but rather an extension of our work,

because China’s macroeconomic institutions differ so markedly from other large EMs.

1.6.1 Chile

Chile was the second country in the world to adopt inflation targeting (IT), setting

its first annual target in September 1990, and IT was used as a device to bring inflation

gradually down to a stationary 3% level (Schmidt-Hebbel and Tapia, 2002). As noted in

its 2019 monetary policy report:15 ”The main objective of the Central Bank of Chile’s

monetary policy is to keep inflation low, stable, and sustainable over time. Its explicit

commitment is to keep annual CPI inflation at around 3% most of the time, within a range

of plus or minus one percentage point.”16 Although the main objective of policy is focused

on inflation, it does not preclude secondary objectives and several articles suggest that both

internal and external factors may play a role in determining domestic interest rates (e.g.

Edwards, 2015). Navdon and Vial (2016), for example, emphasize the impact of commodity

prices and the exchange rate on inflation in Chile. Nonetheless, monetary policy statements

from the central bank generally do not refer to output stabilization as a reason for policy

changes.
15
Monetary Policy Report, June 2019, Central Bank of Chile.
16
This quote continues to state that output stabilization is a derivative of achieving stable inflation, but

not an explicit objective of policy: “Low, stable inflation promotes economic activity and growth while

preventing the erosion of personal income. Moreover, focusing monetary policy on achieving the inflation

target helps to moderate fluctuations in national employment and output.”

28
 Table 6 shows the empirical estimates results for Chile. Panel A indicates that

over the full sample period interest rate policy responded significantly in the expected ways

to both inflation and the output gap. But the estimates suggest that greater focus in

Chile was on inflation targeting in the pre-GFC period and on output targeting during

the post-GFC period17 . In the pre-GFC period, improvements in the terms-of-trade (and

associated wealth gains and improving economy) were associated with interest rate hikes.

Rising current account surpluses, in tandem with increased financial market liquidity, led

to nominal interest rate declines. No statistically significant responses to either the terms-

of-trade nor the current account were found in the post-GFC period, reflecting in part a

low and largely unchanged policy interest rate during this period18 .

Panel B of Table 5 indicates that Chile’s intervention policy targeted the reserve

gap and was also impacted by the current account (with official purchases of USD declining
17
These differences are statistically significant. The z-statistic measuring the significance of the difference

in the output gap (inflation target) is -2.60 (1.74), significant at the 1% (5%) level.
18
These differences are statistically significant. The z-statistic for the difference in coefficients on the

terms-of-trade (current account) between the two periods is 2.83 (-2.40), significant at the 1% (5%) level.

29
with a rise in the surplus) during the post-GFC19,20 . There is no systemic evidence of

intervention policy directed towards exchange rate management in the full sample period

or either sub-sample.

1.6.2 China

Analyzing monetary policy in China is not straightforward as the People’s Bank

of China (PBoC) uses more than one instrument for monetary policy and these instruments

have evolved over time (Chen et al., 2017). The PBoC currently uses seven instruments

of implementation of monetary policy, including the rediscount rate on loans to banks and

other benchmark interest rates.21 Moreover, stronger emphasis has been placed on targeting

interest rates as the major monetary policy instrument in recent years (He and Jia, 2019).

Given China’s extensive use of capital controls and direct involvement in the banking sec-
19
We do not have central bank data on intervention for Chile and China (as we do for India and Brazil).

We proxy for intervention by the change in international reserves, adjusted for interest earnings and valuation

effects (as in equation 3). We estimate interest earnings as the U.S. interest rate multiplied by lagged level

of reserves. This adjusted series is divided by the lag level of reserves and regressed on the U.S. interest

rate, as a proxy for valuation effects. The estimated coefficient on the U.S. interest rate is multiplied by the

observed U.S. interest rate in each quarter to extract valuation effects from our intervention measure. As

a robustness test of this approach, we made the same calculation of adjusted reserves for Brazil and India,

and correlated our estimated intervention with actual intervention data. The correlations are 0.71 and 0.62,

respectively, for Brazil and India. This suggests that our “adjusted reserve change” proxy for intervention

is a reasonable estimate of actual intervention.

20
However, only the shift in the reserve gap coefficient between the two periods is statistically significant

(z-statistic of 3.90, significant at the 1% level).

21
Other instruments noted on the PBoC website in 2018 were open market operations, reserve requirement

ratios, standing lending facility, medium-term lending facility, and pledged supplementary lending facility.

30
tor and foreign exchange market, we modify the intervention equation in two ways beyond

the models estimated for the other three countries investigated. First, we extend the in-

tervention equation by including the broad money supply as an explanatory variable (M2,

measured in USD as 100*log(M2) divided by the log lag of nominal GDP)22 . In addition we

treat the current account as an endogenous variable23 . This methodological adjustment is

taken because tight capital controls on the financial account in China could lead to either

current account surpluses or FDI inflows automatically increasing international reserves.24

Table 7 shows the empirical estimates for China. Panel A shows the Taylor rule

estimates and panel B the intervention rule estimates. It is apparent that the central bank

in China raises the policy rate in respond to an uptick in inflation, a very robust link that

holds across sample periods and model specifications. Policy rates are also linked to the

output gap, but with unexpected and significant negative sign, indicating that interest rates

are reduced the larger is the output gap. Since GDP is only available for China on an annual

basis, this result could be associated with the interpolation methodology. However, when

employing industrial production rather than GDP as the output measure25 , the significant
22
This follows Schroder (2017) who finds both M2 and portfolio equity liabilities as significant determi-

nants of reserve demand. The latter variable is not available past 2011 and not employed in our study. (It

from the Lane and Milesi-Ferretti database, updated online through 2011 only).
23
We instrument the contemporaneous current account in China with three lags of itself.
24
This is related to the discussion of what constitutes intervention, “passive” increases in reserves that

may be caused by interest earnings or valuation effects or “active” purchases and sales in the foreign exchange

market. This is further complicated in the Chinese case by extensive capital controls.
25
GDP data in China is only available at an annual level. Quarterly estimates of GDP are obtained by

implementing a cubic spline interpolation. As a result, it is not possible to decompose the approximate

quarterly series into the trend and cyclical components that would be needed to calculate the output gap. A

31
positive coefficient is also obtained, and stands in contrast to estimates from the other EMs

in the sample. There is also evidence that large current account surpluses in the pre-GFC

period were associated with substantial liquidity in the Chinese financial system, leading

the central bank to reduce interest rates. No estimated linkage with the terms-of-trade is

statistically significant.

On the external side, we find no evidence that intervention policy systemically

responds to (albeit small) variations in the nominal exchange rate or to the broad money

supply (M2). However, we find a strong and robust intervention response to deviations in

the reserve gap—the central bank systemically reduces its USD purchases when the reserve

gap increases. This result holds across sub-samples and model specifications. This robust

result is obtained despite the massive buildup of reserves by China, far exceeding “adequate”

levels. Moreover, there is evidence that higher current account surpluses also led to more

USD purchases prior to the GFC period, as the foreign exchange fund moved to absorb

liquidity in the foreign exchange market, but not afterwards26 .

In summary, applying our methodology to Chile, our small EM extension, is in

simple quarter over quarter growth rate is calculated from the interpolated series and used as an alternative

measure of the output gap in China. A potential concern with this methodology is that the variation in the

interpolated series is being driven by statistical noise rather than actual output fluctuations in China. To

alleviate this concern, the baseline Taylor rule in China is re-estimated using both the official annual measure

of industrial production, interpolated to a quarterly series, and a quarterly measure of industrial production

growth from the OECD. These two alternative measure of the output gap leaves the results qualitatively

unchanged. Most noteworthy is that negative and statistically significant coefficients on the output gap are

robust to using industrial production. Results omitted for brevity but are available from the authors upon

request.
26
This difference is statistically significant at the 1% level (z-statistic equals 2.65)

32
line with our previous results. On the other hand, the results for China are at odds with the

estimates for the other EMs. Estimation of the output gap in the Taylor rule is particularly

problematic due to the lack of reliable quarterly output data in China. Nonetheless, we

find a strong and robust inflation response in the Taylor rule and an intervention function

consistent with targeting international reserve levels.

1.7 Conclusion

Large emerging markets follow quite different policy configurations in attempting

to achieve internal and external balance. India has quite stringent capital controls, and

follows a Taylor rule dominated by an output stabilization objective. Inflation has played

a much smaller part in influencing interest rates in India, mostly evident in recent years,

and the terms-of-trade occasionally plays a role. Brazil, by contrast, has a much more

financially open economy and follows an inflation target regime that generally dominates

other considerations. Though exchange rate and terms-of-trade fluctuations occasionally

influence interest rates in Brazil, we find no evidence that the central bank attempts to

stabilize output fluctuations directly.

External policies are more similar in Brazil and India despite differences in capital

control regimes. Intervention policies in both countries focus on exchange rate stabilization,

i.e. stabilizing the exchange rate with “leaning against the wind” foreign change purchases

and sales. In terms of an external financial stability objective, India uses intervention

operations to target reserves at a level justified by economic factors. Brazil, on the other

hand, started targeting a specific level of reserves only after the Global Financial Crisis

33
(GFC). Controlling for the exchange rate and the international reserves gap, both countries

still made large net quarterly purchases of foreign exchange on average over the sample

period.

The impact of the liberalization of international capital controls on policy is com-

plex, depending on market conditions and the specific actions taken to lift restrictions on

capital inflows or outflows. We find that greater financial openness affected India and Brazil

differently, depending on the particular sequence of administrative measures. This led to

varying private capital movements and intervention policy responses. We also find that

conflicts in internal and external policy occur occasionally and, for both countries, very

large discretionary intervention operations appear negatively linked to discretionary inter-

est rate changes. That is, large unpredicted intervention purchases (sales) accommodate

low (high) interest rates, suggesting that external operations are subordinate to domestic

policy objectives.

The results for Chile, the extension of our study to a small open economy, suggests

the central bank follows a true Taylor rule in balancing output and inflation targets but with

more emphasis on inflation prior to the GFC and on output after the GFC. The exchange

rate does not appear as a factor either in setting interest rates or intervention operations,

and targeting a particular level of reserves only appears after the GFC. China has a more

complicated institutional framework for macroeconomic policy than the other three EMs,

and quality of output data is also a concern. Nonetheless, we find that Chinese interest rate

policy responds strongly to inflation and intervention responds to an international reserves

target.

34
 In conclusion, each country has its own idiosyncratic policies, varying over time,

but commonalities emerge. Policy interest rates always respond to either inflation or output

gaps, frequently both, with varying intensities, and intervention is directed toward managing

targeted international reserve levels and usually to exchange rate stabilization. Terms-of-

trade and current account fluctuations also occasionally influence intervention operations.

In conflicts between interest rate and intervention policies, the former — focused on internal

balance — appear to dominate policy.

1.8 References

[1] Aizenman, J. (2017). “International Reserves, Exchange Rates and Monetary Policy:

From the Trilemma to the Quadrilemma”, September, mimeo.

[2] Aizenman, J., Binici, M. (2016). “Exchange market pressure in OECD and emerging

economies: Domestic vs. external factors and capital flows in the old and new normal”,

Journal of International Money and Finance, (66) p. 65-87.

[3] Aizenman, J., Hutchison, M., Noy, I. (2011). “Inflation Targeting and Real Exchange

Rates in Emerging Markets”, World Development, May (Vol. 39, No. 5).

[4] Barroso, J.B. (2014). “Realized volatility as an Instrument to Official Intervention”.

Working Paper 363, Central Bank of Brazil.

[5] Cavallino, P. (2019). “Capital Flows and Foreign Exchange Intervention”, American

Economic Journal: Macroeconomics 11(2): 127-170.

35
 [6] Central Bank of Brazil (2019). “Minutes of the 223rd Meeting of the Monetary Policy

Committee (Copom) Banco Central do Brasil”, June 18-19.

[7] Chamon, M., Garcia, M., Souza,L. (2017). “FX Interventions in Brazil: A synthetic

control approach”, Journal of International Economics 108 (2017), 157-188.

[8] Chen, H., Chow, K., Tilmann, P. (2017). “The Effectiveness of Monetary Policy in

China: Evidence from a Qual VAR”, China Economic Review 43 (April), pp. 216-231.

[9] Chinn, Menzie D., Ito, Hiro (2006). “What Matters for Financial Development? Cap-

ital Controls, Institutions, and Interactions”, Journal of Development Economics, Vol-

ume 81, Issue 1, Pages 163-192 (October).

[10] Cortes, G.S., Paiva, C. (2017). “Deconstructing credibility: The breaking of monetary

policy rules in Brazil”. Journal of International Money and Finance 74 (2017) 31–52.

[11] De Almeida, C.L., Peres, M.A., Souza, G., Tabak, B. (2003). “Optimal monetary

rules: the case of Brazil”, Applied Economics Letters, 10:5, 299-302.

[12] Edwards, S. (2015). “Monetary Policy Independence under Flexible Exchange Rates:

An Illusion?”. The World Economy Volume 38, Issue 5 Pages: 773-898 May 2015.

[13] Farhi, E., Werning, I. (2012). “Dealing with the Trilemma: Optimal Capital Controls

with Fixed Exchange Rates”. NBER Working Paper No. 18199. 95p.

[14] Fernandez, A., Klein, M., Rebucci, A., Schindler, M., Uribe, M. (2016). “Capital

Control Measures: A New Dataset”. IMF Economic Review, Vol. 64 (3): 548-574,

2016.

36
[15] Gnabo, J., Mello, L., Moccero, D. (2010). “Interdependencies between Monetary

Policy and Foreign Exchange Interventions under Inflation Targeting: The Case of

Brazil and the Czech Republic”, International Finance 13:2, 195–221.

[16] Goncalves, F. (2008). “Accumulating Foreign Reserves Under Floating Exchange

Rates”, IMF Working Paper WP/08/96.

[17] Greene, W.H. (2012) Econometric Analysis. Prentice Hall Publishers, 7th Edition.

[18] Gupta, A.S., Sengupta, R. (2014). “Is India Ready for Flexible Monetary Targeting?”,

Indira Gandhi Institute of Development Research (Mumbai) Working Paper WP-2014-

019 (June).

[19] He, Z., Jia, G. (2019), “An institutional analysis of China’s reform of their monetary

policy framework”, Levy Economics Institute of Bard College, Working Paper No. 925

(April).

[20] Hutchison, M., Pasricha, G. (2016). “Exchange Rate Trends and Management in

India”, in C. Ghate and K.M. Kletzer (eds.), Monetary Policy in India: A Modern

Macroeconomic Perspective, Spring, 2016.

[21] Hutchison, M., Sengupta, R., Singh, N. (2013). “Dove or Hawk: Characterizing Mon-

etary Policy Regime Switches in India”, Emerging Markets Review 16, 183-202.

[22] Hutchison, M., Pasricha, G., Singh, N. (2012). “Effectiveness of capital controls in

India: evidence from offshore NDF market”. IMF Economic Review, 60(3), 395–438.

[23] International Monetary Fund (2012). Annual Report on Exchange Rate Arrangements

37
 and Restrictions. Washington, DC: International Monetary Fund.

[24] Jeanne, O. (2012). “Capital Account Policies and the Real Exchange Rate”. NBER

Working Paper No. 18404. 42p.

[25] Kaur, K. (2016), “Monetary Policy Behavior in India”, Journal of Business Thought

Vol. 6 April 2015-March 2016.

[26] Kohlscheen, E., Andrade, S. (2014). “Official FX Interventions through derivatives”.

Journal of International Money and Finance 47, 202-216.

[27] Modenesi, A.M., Martins, N.M., Modenesi, R.L. (2013). “A modified Taylor rule

for the Brazilian economy: convention and conservatism in eleven years of inflation

targeting (2000-2010)”, Journal of Post Keynesian Economics, 35:3, 463-482.

[28] Navdon, A., Vial, J. (2016), “The evolution of inflation in Chile since 2000”, BIS

Papers No. 89, November.

[29] Oliveira, F.N., Novaes, W. (2007), “The demand of foreign exchange derivatives in

brazil. hedge or speculation?”. Working Paper 152, Central Bank of Brazil. Mimeo.

[30] Pasricha, G.K., Falagiarda, M., Bijsterbosch, M., Aizenman, J. (2015). “Domestic and

Multilateral Effects of Capital Controls in Emerging Markets”, NBER Working Paper

No. 20822

[31] Pasricha, G.K. (2017). “Policy Rules for Capital Controls”, BIS Working Paper No.

670, November 2017.

38
[32] Patnaik, I., Shah, A. (2012). “Did the Indian Capital Controls Work as a Tool of

Macroeconomic Policy?”, Policy Corner, IMF Economic Review, Vol. 60, No. 3 (2012),

pp. 439-464.

[33] Schmidt-Hebbel, K., Tapia, M. (2002). “Inflation Targeting in Chile”, North American

Journal of Economics and Finance 13(2) August, pp. 125-146.

[34] Schroder, M. (2017). “Mercantilism and China’s hunger for international reserves”,

China Economic Review 42, 15-33.

[35] Stock, J., Wright, J., Yogo, M. (2002). “A Survey of Weak Instruments and Weak

Identification in Generalized Method of Moments”. Journal of Business and Economic

Statistics 20, 518 – 529.

1.9 Appendix - Variables description, Tables and Figures

• ∆e: Percent change in nominal exchange rate, closing price reported by the

Central Bank of Brazil and Reserve Bank of India. Quotations denominated in local cur-

rency per unit of US dollar. For quarterly data, exchange rate is for March 31st, June 30th,

September 30th, and December 31st (or the closest date available). We applied the log

changes and presented as percentage, ∆e = 100 × (ln(et ) − ln(et − 1).

• Ŷ : India output measured by Industrial Production. Brazil output is quar-

terly GDP series reported by the Central Bank of Brazil. Log of output series filtered by

Hodrick-Prescott (HP) tecnhique. Output gap is the cyclical component of the HP-filtered

39
log(GDP) series.

• π: Inflation calculated as the anualized log change over local price index. India

is the wholesale price index, Brazil is the IPCA (National Index of Consumer Prices, elab-

orated by the Brazilian Institute of Geography and Statistics). Percent Annualized change,

π = 100 × (ln(CP It ) − ln(CP It−4 )).

• π ∗ : India does not publish inflation target. We assume the implicit target con-

stant through the whole period. For Brazil, IT is officially defined by the National Monetary

Council and the Central Bank is required by law to pursue it, with some allowed deviations

(tolerance bands). The IT changes through time. For 2019, it is defined as 4.25% with a

tolerance band of 2% (meaning an accepted interval of [2.25%, 6.25%]).

• (π − π ∗ ): The inflation gap is measured as the deviation from the target, i.e.

[100 × (ln(CP It ) − ln(CP It−4 )) − inflation target] = [100 × (ln(CP It ) − ln(CP It−4 )) − π ∗ ].

• i: Money market rate defined and controlled by the Central Bank of Brazil and

Reserve Bank of India, respectively. For Brazil we have used the “SELIC” rate, and for

India we’ve used 3 months money market defined by RBI & India: 1999Q1-2018Q4; Brazil:

2000Q1-2018Q4;

• i∗ : The US interest rate is the 3-Month Treasury Bill Rate, published by the

40
Federal Reserve Economic Data (FRED).

• openness: This variable is from Pasricha et al.(2015). The author provided a

detailed dataset for the period 2001-2015 with quarterly frequency. Each data series counts

the number of capital flow measures (for example, number of easings of inflow controls or

tightenings of outflow controls) undertaken by each country. The variables used from the

dataset weighted each policy action by the share of the country’s international assets or li-

abilities that the measure was designed to influence. The policy actions in the dataset were

counted by effective dates and included changes for which the announcement and effective

dates are different. From the dataset we explored two specific series: “wgt nettighteningin”,

and “wgt neteasingout”, that correspond to number of net inflow tightenings, weighted, and

number of net outflow easenings, weighted, respectively. As we are insterested to under-

stand the degree of openness of the countries studied, we have transformed the first series

“net inflow tightenings” to “net inflow easing” by inverting its sign (a positive tightening

means a negative easing and a negative tightening means a positive easing). With the

quarterly values of easing inflow and easing outflow we chose to work with the cumula-

tive measures of both easing inflow and outflow combined. As this variable was intended

to measure openness, we need to measure the easing policies, regardless of inflow or outflow.

• R: Level of Foreign Reserves in USD reported by the Central Bank, includes

SDRs and excludes Gold holdings.

41
 • R∗ : The Reserve Target values are from IMF “Assessing Reserve Adequacy”.

The institution’s work compares the reserve holdings and alternative metrics of reserve ad-

equacy. This reserves adequacy measure was initially developed in the IMF Board Paper

”Assessing Reserve Adequacy” - RAM1 (February 15, 2011), and adjusted in the latest

IMF Board Paper ”Assessing Reserve Adequacy- Specific Proposals” (December 19, 2014),

in order to reflect the outflows during the Global Financial Crisis which were not addressed

in RAM1. The IMF Reserve Adequacy estimates adequate volume of reserves for a specific

country taking into account exports, imports, broad money, and other liabilities.

• (R − R∗ ):The Reserve Gap is calculated by the difference of the level of reserves

and the adequate level proposed by the IMF (R∗ ). Log-transformation and percentage pre-

sentation is also applied: 100 × (ln(R) − lnR∗ )

• Appreciation: Dummy variable that assumes value equals to 1 if the local cur-

rency appreciates versus US dollar, i.e., ∆e < 0 and value equals 0 otherwise (∆e ≥ 0).

• Spot Intervention: Amount of USD bought and sold in the spot market relative

to the level of Reserves.

• Forward Intervention: Amount of USD bought and sold in the forward market

relative to the level of Reserves.

42
 • Terms of Trade: Ratio of exports over imports. We have used the following

monthly series elaborated by the IMF: Commodity Export Price Index, Individual Com-

modites Weighted by Ratio of Exports to Total Commodity Exports, Commodity Import

Price Index, and Individual Commodites Weighted by Ratio of Imports to Total Commod-

ity Imports. All for the 1999-2018 period.

• Current account: Quarterly data on the net current account balance is obtained

from the IMF. The series is normalized by dividing the current account balance by the first

lag of nominal GDP and multiplying by 100.

43
 Table 1.1: Descriptive Statistics
Panel A: Entire Sample, 1999Q1 - 2018Q4
India Brazil

Statistic N Mean St. Dev. N Mean St. Dev.

i 84 6.98 1.62 76 13.447 4.579
Ŷ 84 0.00 2.24 76 −0.207 9.554
π 80 4.56 3.19 76 5.242 3.385
π − π∗ 80 4.56 3.19 76 0.419 1.023
∆e 83 0.73 3.04 76 1.019 8.498
R − R∗ 84 33.12 27.68 76 1.244 49.978
Ispot 84 1.56 3.89 76 2.63 6.769
Itotal 84 0.01 11.64 76 2.581 7.12
openness 60 20.76 15.84 60 1.802 1.193
t.o.t. 76 107.27 11.33 80 95.15 14.37
curr. acc. 83 -1.37 2.01 88 -1.89 2.19
Panel B: Pre Crisis, 1999Q1 - 2008Q4
India Brazil

Statistic N Mean St. Dev. N Mean St. Dev.

i 44 6.93 1.63 36 16.931 3.775
Ŷ 44 0.25 2.61 36 −0.624 10.049
π 40 4.56 3.19 36 6.268 3.870
π − π∗ 40 4.56 3.19 36 0.546 1.254
∆e 43 0.50 2.87 36 −0.148 8.109
R − R∗ 44 29.68 36.78 36 −42.709 33.72
Ispot 44 2.32 4.79 36 4.263 9.358
Itotal 44 0.14 11.37 36 3.988 9.801
openness 32 8.07 5.67 32 1.409 1.346
t.o.t. 36 106.15 15.36 40 95.58 17.78
curr. acc. 43 -0.78 1.92 47 -1.43 2.61
Panel C: Post Crisis, 2009Q1 - 2018Q4
India Brazil

Statistic N Mean St. Dev. N Mean St. Dev.

i 40 7.04 1.63 40 10.312 2.5
Ŷ 40 −0.27 1.74 40 0.168 9.198
π 40 3.97 4.05 40 5.057 2.908
π − π∗ 40 3.97 4.05 40 0.305 0.755
∆e 40 0.98 3.24 40 2.069 8.801
R − R∗ 40 36.91 10.51 40 40.802 19.869
Ispot 40 0.72 2.34 40 1.161 2.199
Itotal 40 −0.16 12.08 40 1.315 2.794
openness 28 35.27 10.11 28 2.252 0.798
t.o.t. 40 103.77 2.97 40 97.82 9.34
curr. acc. 40 -2.31 1.61 41 -2.42 1.43

44
Panel A: Interest Rate Policy Dependent Variable: it
India Brazil
(1) (2) (1) (2)
c 1.13∗∗∗ 1.16∗∗ 3.51∗∗∗ 5.87∗∗∗
(0.39) (0.56) (1.31) (1.38)
Ŷ 0.11∗∗∗ 0.11 ∗∗∗ 0.03 0.03
(0.03) (0.03) (0.03) (0.04)
π − π∗ 0.02 0.02 0.60∗∗∗ 0.60∗∗∗
(0.02) (0.02) (0.22) (0.16)
∆e 0.03 0.03 0.02 0.03∗
(0.06) (0.06) (0.02) (0.015)
it−1 0.82∗∗∗ 0.81∗∗∗ 0.65∗∗∗ 0.66∗∗∗
(0.05) (0.05) (0.11) (0.04)
t.o.t. −0.00 −0.013∗∗
(0.00) (0.012)
current account −0.04 0.11
(0.03) (0.17)
R2 0.83 0.82 0.85 0.86
Num. obs. 80 76 79 79

Panel B: Intervention Policy Dependent Variable: It

India Brazil
(1) (2) (1) (2)
c 3.23 ∗∗∗ 3.12 3.12 ∗ 25.17∗∗∗
(0.71) (6.21) (1.70) (9.36)
∆e −0.48 ∗∗∗ −0.30 ∗∗ −0.22 ∗∗ −0.17∗∗
(0.15) (0.11) (0.09) (0.074)
R − R∗ −0.04∗∗∗ −0.04 −0.04 −0.03
(0.01) (0.03) (0.04) (0.02)
t.o.t. 0.01 −0.23∗∗
(0.05) (0.09)
current account 0.91∗∗∗ 0.18
(0.24) (0.39)
R2 0.13 0.45 0.11 0.32
Num. obs. 83 76 75 75
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1

Table 1.2: Baseline Results

45
 Panel A: Interest Rate Policy Dependent Variable: it
India Brazil
Pre-Crisis Post-Crisis Pre-Crisis Post-Crisis
(1) (2) (3) (4) (1) (2) (3) (4)
c 1.48∗∗∗ 0.39 0.86 −6.32∗∗ 8.66∗∗∗ 5.13 1.90∗∗ −0.15
(0.50) (0.85) (0.53) (2.68) (1.55) (3, 30) (0.76) (1.27)
Ŷ 0.12∗ 0.18∗∗∗ 0.16∗∗∗ 0.15∗∗∗ 0.01 0.03 0.02 0.01
(0.07) (0.06) (0.03) (0.04) (0.03) (0.05) (0.02) (0.01)
π − π∗ −0.02 0.04 0.04∗∗ 0.03∗∗ 0.60∗∗∗ 0.61∗∗∗ 0.50∗∗∗ 0.43∗∗∗
(0.05) (0.07) (0.01) (0.02) (0.16) (0.17) (0.09) (0.08)
∆e −0.02 −0.03 0.08 0.10 0.01 −0.02 0.01 0.01
(0.03) (0.02) (0.08) (0.09) (0.03) (0.03) (0.01) (0.01)
it−1 0.79∗∗∗ 0.57∗∗∗ 0.86∗∗∗ 0.85∗∗∗ 0.41∗∗∗ 0.30∗∗∗ 0.74∗∗∗ 0.72∗∗∗
(0.05) (0.13) (0.08) (0.09) (0.08) (0.07) (0.07) (0.10)
t.o.t. 0.02∗ 0.07∗∗ 0.06 0.02
(0.01) (0.03) (0.04) (0.02)
cur. acc. −0.07∗∗ −0.07 0.13 −0.15
(0.03) (0.06) (0.22) (0.10)
R2 0.85 0.84 0.86 0.87 0.78 0.80 0.93 0.94
Num. obs. 40 36 40 40 39 39 40 40

Panel B: Intervention Policy Dependent Variable: It

India Brazil
Pre-Crisis Post-Crisis Pre-Crisis Post-Crisis
(1) (2) (3) (4) (1) (2) (3) (4)
c 3.57∗∗∗ −23.78∗ 4.63∗∗ 7.92 3.64∗ 46.71∗∗∗ 5.06∗∗∗ 3.65∗∗
(1.15) (13.46) (1.82) (8.26) (2.05) (8.65) (1.08) (1.42)
∆e −0.66∗∗ −0.37∗∗ −0.35∗∗ −0.35∗∗ −0.37∗∗∗ −0.11 0.04∗∗ 0.02
(0.30) (0.14) (0.15) (0.15) (0.10) (0.14) (0.02) (0.02)
R − R∗ −0.03∗ 0.07 −0.10∗∗ −0.10∗∗∗ −0.03 −0.13∗∗∗ −0.09∗∗∗ −0.09∗∗∗
(0.02) (0.06) (0.04) (0.03) (0.04) (0.03) (0.02) (0.01)
t.o.t. 0.21∗∗ −0.02 −0.53∗∗∗ 0.00
(0.10) (0.08) (0.10) (0.01)
cur. acc. 1.13∗∗∗ 0.51∗∗ −0.25 −0.38∗∗
(0.30) (0.24) (0.39) (0.18)
R2 0.15 0.63 0.14 0.26 0.11 0.37 0.29 0.36
Num. obs. 43 36 40 40 35 35 40 40
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Pre-Crisis corresponds to periods before 2009Q1

Table 1.3: Pre and Post Global Financial Crisis

46
Panel A: Interest Rate Policy - Pre GFC
Dependent Variable: it
India Brazil
(1) (2) (1) (2)
c 1.987∗∗∗ 3.2289∗∗∗ 6.4176∗ 8.8692∗∗
(0.3249) (0.8176) (3.4913) (4.1772)
Ŷ 0.1277∗∗ 0.2475∗∗∗ −0.0176 0.0041
(0.0691) (0.0578) (0.0390) (0.0416)
π − π∗ −0.0276 .0909 0.5248 0.5183
(0.0489) (.0849) (0.3105) (0.3798)
∆e 0.0323 0.0590 0.0089 0.0006
(0.0336) (0.0373) (0.0294) (0.0279)
it−1 0.5994∗∗∗ 0.4054∗∗∗ 0.5103∗ 0.4080
(0.0455) (0.1175) (0.2598) (0.3249)
iU S 0.2474∗∗∗ 0.236∗∗∗ 0.1872 0.2717
(0.0511) (0.0473) (0.2306) (0.3268)
openness −0.0809∗∗∗ −0.6089∗
(0.0284) (0.3550)
R2 0.8908 0.8766 0.8198 0.8369
Num. obs. 40 32 32 32

Panel B: Spot Intervention

Dependent Variable: It
India Brazil
Pre-Crisis Post-Crisis Pre-Crisis Post-Crisis
c 4.78∗∗∗ −2.09 −9.39 ∗∗∗ 8.06∗∗∗
(1.35) (4.51) (2.14) (1.77)
∆e −0.26∗∗ −0.27∗ −0.27 −0.00
(0.11) (0.16) (0.20) (0.02)
R − R∗ 0.12 −0.02 −0.14∗∗∗ −0.08∗∗∗
(0.10) (0.05) (0.03) (0.02)
openness −0.97∗∗ 0.11 6.17∗∗∗ −1.50∗∗∗
(0.42) (0.09) (0.81) (0.51)
R2 0.66 0.30 0.49 0.41
Num. obs. 32 28 32 28
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Pre-Crisis corresponds to periods before 2009Q1

Table 1.4: Capital Account Liberalization (Openness)

47
Dependent Variable ǫtaylor India Brazil
c 0.14 0.87∗∗∗
(0.12) (0.10)
ǫintervention |(|ǫintervention |) > p90 −0.21∗ −0.09∗
(0.11) (0.04)
R2 0.17 0.05
Num. obs. 16 16

∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1

Table 1.5: Residual Analysis

48
Panel A: Interest Rate Policy Dependent Variable: it
Full Sample Pre-Crisis Post-Crisis
(1) (2) (3) (4) (5) (6)
c 0.76∗∗∗ 1.60∗∗∗ 0.40 −2.98∗∗∗ 1.79∗∗∗ 2.11
(0.25) (0.54) (0.35) (0.95) (0.52) (1.43)
Ŷ 0.07 ∗∗∗ 0.10 ∗∗∗ 0.02 −0.03 0.21∗∗∗ 0.18∗∗
(0.03) (0.02) (0.03) (0.04) (0.06) (0.07)
π−π ∗ 0.20 ∗∗∗ 0.18 ∗∗∗ 0.22∗∗∗ 0.28 ∗∗∗ 0.05 0.07
(0.06) (0.05) (0.04) (0.05) (0.12) (0.11)
∆e 0.02 0.01 0.01 0.01 0.00 −0.01
(0.02) (0.02) (0.01) (0.01) (0.01) (0.01)
it−1 0.64∗∗∗ 0.58∗∗∗ 0.74∗∗∗ 0.65∗∗∗ 0.44∗∗∗ 0.40∗∗∗
(0.08) (0.09) (0.08) (0.09) (0.05) (0.04)
t.o.t. −0.01 0.04∗∗∗ −0.00
(0.00) (0.01) (0.01)
current account −0.07 −0.11 ∗∗ −0.06
(0.04) (0.05) (0.05)
R2 0.81 0.82 0.84 0.88 0.84 0.85
Num. obs. 80 76 40 36 40 40

Panel B: Intervention Policy Dependent Variable: It

Full Sample Pre-Crisis Post-Crisis
(1) (2) (3) (4) (5) (6)
c 1.63 ∗∗ 2.80 1.41 ∗∗ 3.67 13.62 ∗∗∗ 19.08∗∗∗
(0.64) (3.83) (0.60) (8.25) (4.94) (5.62)
∆e 0.08 0.07 0.08 0.04 0.26 0.24
(0.08) (0.08) (0.11) (0.06) (0.19) (0.18)
R − R∗ −0.00 −0.00 0.01 0.02 −0.14∗∗∗ −0.23∗∗∗
(0.01) (0.01) (0.03) (0.03) (0.05) (0.05)
t.o.t. −0.01 −0.02 0.01
(0.04) (0.08) (0.03)
current account −0.04 −0.14 −0.80∗∗
(0.17) (0.25) (0.37)
R 2 0.00 0.01 0.03 0.05 0.05 0.05
Num. obs. 75 75 35 35 40 40
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Pre-Crisis corresponds to periods before 2009Q1

Table 1.6: Chile Policy Rules

49
Panel A: Interest Rate Policy Dependent Variable: it
Full Sample Pre-Crisis Post-Crisis
(1) (2) (3) (4) (5) (6)
c 2.17 ∗∗∗ 1.77 ∗ 1.57∗∗∗ 2.13 ∗∗∗ 2.86∗∗∗ 4.81
(0.66) (0.90) (0.33) (0.41) (0.92) (3.20)
Y −0.39∗∗ −0.41∗ −0.28∗ −0.39∗∗∗ −0.76∗∗ −0.86∗∗
(0.18) (0.24) (0.15) (0.07) (0.35) (0.40)
π − π∗ 0.10∗∗∗ 0.16∗∗∗ 0.04∗ 0.12∗∗∗ 0.22∗∗∗ 0.17∗∗
(0.04) (0.04) (0.02) (0.03) (0.06) (0.07)
∆e 0.00 −0.02 0.01 0.01 −0.07 −0.05
(0.02) (0.03) (0.03) (0.02) (0.05) (0.05)
it−1 0.28 ∗∗ 0.17 0.52 ∗∗∗ 0.36 ∗∗∗ 0.08 0.05
(0.13) (0.12) (0.13) (0.08) (0.12) (0.15)
t.o.t. 0.01 0.00 −0.01
(0.01) (0.00) (0.02)
current account −0.07∗∗∗ −0.07∗∗∗ −0.04
(0.02) (0.01) (0.12)
R2 0.31 0.37 0.33 0.47 0.35 0.36
Num. obs. 65 64 37 36 28 28

Panel B: Intervention Policy Dependent Variable: It

Full Sample Pre-Crisis Post-Crisis
(1) (2) (3) (4) (5) (6)
c 26.25∗∗∗ 37.79∗∗∗ 29.23∗∗∗ −33.44 33.07∗∗∗ 66.64∗∗
(2.65) (12.85) (5.87) (39.84) (5.99) (24.54)
∆e −0.17 −0.19 −0.15 −0.08 −0.21 −0.24
(0.10) (0.17) (0.14) (0.24) (0.14) (0.26)
R − R∗ −0.06∗∗∗ −0.06∗∗∗ −0.07∗∗∗ −0.08 −0.07∗∗∗ −0.11∗
(0.01) (0.01) (0.02) (0.05) (0.01) (0.06)
t.o.t. −0.03 0.25 0.03
(0.04) (0.15) (0.05)
current account −0.28 1.32∗ −1.69
(0.34) (0.69) (1.39)
M2 −0.02 0.15 −0.08
(0.08) (0.10) (0.07)
R2 0.63 0.62 0.27 0.51 0.40 0.33
Num. obs. 59 59 19 19 40 40
∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Pre-Crisis corresponds to periods before 2009Q1

Table 1.7: China Policy Rules

50
Figure 1.1: Output Gap

51
Figure 1.2: Inflation

52
Figure 1.3: Money Market Interest Rates

53
Figure 1.4: Exchange Rates

54
Figure 1.5: Reserves, Reserve Adequacy and Foreign Exchange Market Intervention

55
Figure 1.6: Reserve Gap

56
Figure 1.7: Capital Openness

57
Chapter 2

Regional Trade Agreements and A

Gravity Model of Consumption

Risk Sharing

2.1 Introduction

Classic economic theories identify frictions in the goods market as an explanation for the

lack of consumption risk sharing among countries. For instance, Obstfeld and Rogoff (2001)

argue that trade costs make it costly for countries to share risk through the exchange of

goods and can therefore account for the low cross-country consumption correlations ob-

served in the data.

We test the theory empirically by exploiting the variation in trade costs amongst

country pairs. We find that regional trade agreements (RTA hereafter) facilitate bilateral

58
risk sharing between trade partners for a panel of 178 countries over the 1970-2014 pe-

riod. This finding based on policy shifts supports the viewpoint that reducing trade costs

promotes consumption risk sharing. In addition, we provide cross-sectional evidence by

establishing a gravity model of consumption risk sharing. As trade costs increase in geo-

graphical distance, we hypothesize and then confirm that bilateral risk sharing is weaker

for countries which are more distant from each other. Lastly, we find that geographical

distance becomes less relevant for consumption risk sharing once countries are RTA mem-

bers. All the evidence points towards the importance of trade costs for explaining imperfect

cross-country consumption risk sharing.

Following the literature, including Sorensen and Yosha (1998) and Kose et al.

(2009), we measure a country’s consumption risk sharing as the response of its relative

consumption growth to its relative output growth. A greater response suggests a lower

degree of consumption risk sharing. Consider the extreme case where two countries that face

output risk cannot trade assets or ship goods across borders, each country’s consumption

is equal to its own output. There is no risk sharing between the two countries since the

difference in their consumption growth equals that in output growth. In contrast, when

risk sharing is perfect the level of a country’s consumption does not fluctuate with its

own current output but that of the aggregate economy. As a result, the output difference

between countries does not influence their relative consumption to each other’s.

In this paper we focus on bilateral risk sharing which has received little attention

in the literature. In a classical model with complete markets, competitive equilibrium

coincides with the allocation of a social planner who makes centralized decisions regardless

59
of bilateral economic exchanges. Nevertheless, in the real world there exist frictions of

different magnitudes across country pairs that segment complete markets and make bilateral

risk-sharing relationship important for analyzing consumption patterns.

To examine the influence of trade costs on consumption risk sharing, we exploit

both cross-sectional and time-series variations in trade costs across country pairs. As dis-

cussed earlier, our empirical analysis consists of three parts. To start with, we examine

whether RTAs promote bilateral risk sharing. An RTA is a treaty between two or more

countries that aims to foster trade partnership. By regulating tariffs and other forms of

trade barriers, RTAs reduce the trade costs among member countries. Therefore, we ex-

amine consumption patterns around RTA events to uncover the relationship between trade

costs and risk sharing. We conduct this analysis for a panel of 178 countries who consti-

tute 31684 country pairs over the 1970-2014 period. We interact a dummy variable that

equals 1 when a pair of countries both participate in an RTA and 0 otherwise with the two

countries’ difference in output growth. With the difference in their consumption growth as

the dependent variable, the coefficient of the interaction term reveals the influence of RTAs

on bilateral risk sharing. After controlling for time fixed effects, we find that participating

in an RTA lowers the response of relative consumption to output growth by about 0.11

(equivalent to 0.9 standard deviations). The result is robust when we employ both pooled

regressions and panel analysis with country-pair fixed effect models.

In addition to exploiting policy changes, we provide cross-sectional evidence that

demonstrates the impact of trade costs on consumption risk sharing. Geographical distance

is acknowledged to be a vital determinant for trade costs. The more distant countries

60
are from one another, the higher trade costs it incurs to ship goods between them. If

consumption risk sharing is hampered by trade costs, we should expect that country pairs

with greater geographical distance in between exhibit weaker risk sharing. We conduct a

two-step analysis to test this hypothesis. In the first step we calculate the bilateral risk-

sharing coefficients using the real GDP and consumption data of the 178 countries over

the 1970-2014 period in our sample. In the second step we confirm that the risk-sharing

coefficients are negatively correlated with geographic distance and positively correlated with

the product of GDPs for country pairs. We call this finding a gravity model of consumption

risk sharing. The gravity model has emerged as a workhorse in the literature due to its

empirical success in predicting bilateral trade flows. More recently, it has been applied

in a range of areas to document the importance of geographical variables for explaining

economic linkages across countries.1 This paper contributes to this stand of literature by

establishing a gravity model of consumption risk sharing. Based on the regression results,

a 1% increase in geographic distance lowers the response of relative consumption to output

growth for a country pair by 0.11 ( or 0.36 standard deviations). The result remains robust

when controlling for other common gravity variables including GDP per capita, population,

common language, and common legal system.

Last but not least, we bring the previous analyses together to build the causal

link between trade ties and the gravity model. Trade may not be the only channel through

which geographic distance influences risk sharing. Specifically, countries can share risks
1
For instance, Portes and Rey (2005) show that a gravity model explains international transactions in

financial assets. Ramos and Surinach (2017) use a gravity model to analyze bilateral migration in Europe.

Lustig and Richmond (2019) study the gravity effect in the factor structure of exchange rates.

61
through financial exchanges and labor mobility. Since the literature has acknowledged the

importance of geographic distance for migration and financial flows, additional evidence is

needed to attribute the gravity model of risk sharing to the trade channel. To this end,

we incorporate RTAs and geographic distance in a single regression. If the trade channel

contributes to risk sharing across countries, we should expect that geographic distance

becomes less relevant for risk sharing in the presence of RTAs. We confirm the hypothesis

in the data by documenting a negative correlation between relative consumption growth

and an interaction term of the RTA dummy, distance and output growth. As a result, we

conclude that trade costs can at least partially explain why risk sharing deteriorates as the

geographical distance between countries increases. This finding echoes the main argument

of the paper that trade costs impede consumption risk sharing across countries. Therefore,

lifting trade barriers will yield welfare gains by strengthening countries’ ability to share

risks and smooth consumption.

This paper speaks to a substantial body of literature in international economics.

First and foremost, imperfect consumption risk sharing remains to be one of the major

puzzles in international macroeconomics (Obstfeld and Rogoff (2001)). On the theoretical

front, papers including Obstfeld and Rogoff (2001), Dumas and Uppal (2001), and Backus

and Smith (1993) study the role of trade costs in the goods market when explaining the

lack of international risk sharing. Our paper provides empirical evidence for their theories

directly by exploiting cross-sectional as well as time-series variations in trade costs amongst

country pairs. Besides trade costs, financial frictions that prohibit countries from trading

state-contingent assets have been acknowledged to impede cross-country risk sharing (e.g.

62
Lewis (1996) and Kollmann (1995)). In an empirical paper that also exploits institutional

changes like ours, Kose et al. (2009) examine whether financial liberalizations facilitate risk

sharing and find little evidence. In our paper we control for country-pairs’ financial liber-

alization status when studying RTA events that do not coincide with financial integration

in order to isolate the effects of the trade channel on risk sharing.

Furthermore, this paper is related to several influential studies that investigate

the patterns and consequences of cross-country risk sharing. For instance, Kalemli-Ozcan

et al. (2003) find that countries or regions with better risk sharing exhibit higher indus-

trial specialization. We follow their two-step approach in our paper when constructing the

measure of risk sharing first and then exploring its correlation with variables of interest.

In Section 3.2 we establish the gravity model by finding that risk sharing increases with

country-pairs’ GDP but decreases with geographic distance. Moreover, Fitzgerald (2012)

builds a structural model to disentangle the effects of financial frictions and trade costs on

the lack of consumption risk sharing. Our paper focuses on providing empirical evidence in

the trade channel. In addition, Callen et al. (2015) evaluate the degree of risk sharing that

can be achieved by small sets of countries given that pooling worldwide risk is costly. In a

similar spirit, we examine pairwise risk sharing acknowledging the difficulty of sharing risks

among all the countries in the world.

This paper also contributes to the extensive empirical literature on the gravity

model. Since being introduced by Isard (1954) and Tinbergen (1962), the model has

merged as a classic framework in the trade literature due to its success in matching bi-

lateral trade flows. More recently, seminal works including Anderson and van Wincoop

63
(2003) and Eaton and Kortum (2002) refine the theoretical foundations of the framework

that rationalize empirical regularities of bilateral trade. In addition to trade, the gravity

model has recently been applied to a wide range of topics including financial assets (e.g.

Portes and Rey (2005), Martin and Rey (2004), and Okawa and van Wincoop (2012) and

labor migration (e.g. Lewer and Van den Berg (2008) and Ramos and Surinach (2017). Our

paper contributes to this literature by exploring the role of geographic distance in shaping

consumption allocations.

The remainder of the paper proceeds as follows: Section 2 describes the data and

methods of constructing risk-sharing coefficients. Section 3 presents empirical results as to

how trade costs influence consumption risk sharing. Section 4 concludes.

2.2 Data

To examine the influence of trade ties on consumption risk sharing we combine

data on regional trade agreements, GDP, consumption, and geographical distance among

countries. In this section we describe how we collect and analyze the data.

2.2.1 Regional Trade Agreements

We obtain the information on regional trade agreements from the World Trade Or-

ganization (WTO) and the Centre d’Études Prospectives et d’Informations Internationales

(CEPII). The dummy for regional trade agreements (RTA) is 1 for the period where a pair

of countries both participate in a specific RTA. The WTO classifies RTAs into four groups:

customs unions, economic integration agreements, free trade agreements, and partial scope

64
agreements. We do not consider the last group as RTAs in our analysis since they only cover

specific goods and services. Meanwhile we exclude the events where economic integration

agreements coincide with policies that promote financial integration to isolate the effect of

trade ties on consumption risk sharing.

Figure 2.1 displays the global map of RTAs as of July 2019. There are close to

300 RTAs signed bilaterally or multilaterally by groups of countries. Figure 2.2 tracks the

historical occurrence of RTAs. It illustrates that the coverage of RTAs has been remarkably

expanded over the decades. Table 2.6.1 provides the list of countries in our sample. For the

countries that have ever joined in any RTA from 1950 to 2014, we also list the number of

RTAs they have been a member of, number of countries that have ever been their partners

in any RTA, and the average duration (in years) of RTAs they have participated in.

Figure 2.1: Current RTAs

Source: WTO

65
Figure 2.2: Historical RTAs

Source: WTO and CEPII

66
2.2.2 GDP, consumption, and risk sharing

We collect the real GDP, real consumption, and population data from the Penn

World Table (PWT) version 9.0. Our sample covers 178 countries over the 1970-2014 period.

Following the literature including Sorensen and Yosha (1998) and Kose et al.

(2009), we measure a country’s consumption risk sharing as the response of its relative

consumption growth to its relative output growth. Specifically, we are interested in bilat-

eral risk sharing so that we can exploit pair-specific factors including RTAs and geographic

distance in order to provide a more robust understanding of the factors that shape risk-

sharing patterns. We evaluate risk sharing between country i and j from

∆log cit − ∆log cjt = αij + βijt (∆log yit − ∆log yjt ) + ǫijt , (2.1)

where ∆log cit (∆log cjt ) denotes the growth of log real per-capita consumption of country

i(j) at time t, and ∆log yit (∆log yjt ) denotes the growth of log real per-capita output.

A higher coefficient βijt suggests a lower degree of consumption risk sharing. In the

case with perfect risk sharing, relative consumption growth should not vary with relative

output growth, which yields a coefficient of 0. In the opposite case where there is no risk

sharing, a country’s consumption is solely determined by its own output. In this scenario

relative consumption growth should equal relative output growth across countries such that

βijt = 1. Therefore, the better a country is able to share its risks with another, the smaller

will be the influence of its relative output on consumption (measured by a lower value for

βijt ). For simplicity, we define the bilateral risk-sharing coefficient as RSijt ≡ 1 − βijt . A

higher RSijt stands for better risk sharing.

Table 2.1 shows the summary statistics of RSijt estimated with the annual data

67
from 1970 to 2014 for all the country pairs in our sample. Each cell reports the mean

value and the standard error is in parenthesis. Column (1) reports the coefficients for the

years when two countries are RTA partners, column (2) reports the coefficients for the years

when they are not bound by any RTA, and column (3) reports the difference between the

two coefficients. All the estimates across the three columns are significantly different from

zero at the 1% level. From the table when country-pairs are regional trade partners, the

mean value of risk-sharing coefficients is 0.572, which is much higher than the value 0.416

when countries are not partners under RTAs. If we split countries into different groups, we

find the RTAs benefit risk-sharing between industrial and developing countries to a greater

extent compared to risk-sharing between countries in the same income group. Across all

types of country pairs, there is a robust pattern that risk sharing improves under RTAs.

To exemplify the pattern, we estimate the risk-sharing coefficients RSijt over six-

year rolling windows and show them graphically for a group of European countries. As is

illustrated in Figure 2.3, bilateral risk sharing remarkably improves after the Single Market

was established in the mid 1990’s. 2

2
Austria, Sweden, and Finland became the new member states of the treaty in 1995. The Switzerland

was not an official member, but it signed a separate treaty with the members under EFTA.

68
 Table 2.1: Summary Statistics of Risk-sharing Coefficients

(1) (2) (3)

w/ RTA w/o RTA Difference

All types of countries 0.572 (0.009) 0.416 (0.009) 0.156 (0.012)

Industrial and industrial 0.426 (0.009) 0.344 (0.024) 0.082 (0.024)

Industrial and developing 0.708 (0.025) 0.400 (0.012) 0.308 (0.027)

Developing and developing 0.477 (0.012) 0.434 (0.018) 0.043 (0.020)

This table reports bilateral risk sharing coefficients RSijt ≡ 1−βijt , where βijt is estimated

from equation 2.1. Column (1) reports the coefficients for the years when two countries

are RTA partners, and column (2) reports the coefficients for the years when they are not

bound by an RTA. Each cell reports the average value in the relevant subsample and the

standard error is in parenthesis. The designation of “industrial” and “developing” coun-

tries is based on the Statistics Division of the United Nations.

69
 Figure 2.3: Bilateral Risk Sharing before and after RTAs

Evolution of risk sharing measured as RSijt = 1−βijt for selected pairs of countries. Vertical

lines indicate the implementation dates of regional trade agreements.

2.2.3 Geographic Distance

We add spatial features to our analysis by examining how geographic distances

influence bilateral risk sharing. The benchmark measure of geographic distance between

two countries comes from the CEPII, which calculates population-weighted distance between

the biggest cities of those two countries. For robustness, we also consider simple distance

calculated with the geographical coordinates (latitudes and longitudes) of the capital cities.

70
2.3 Empirical Analysis

In this section we employ econometric analysis to examine the influence of trade

ties on risk sharing. First we test whether regional trade agreements promote bilateral risk

sharing. Second we empirically establish a gravity model of risk sharing. Last we combine

the two pieces and find that geographical distance is less of an obstacle for risk sharing in

the presence of RTAs.

2.3.1 Cross-country Risk Sharing and RTAs

In this section we study consumption patterns around RTA events to provide evi-

dence for the influence of trade costs on consumption risk sharing. We take two approaches

to evaluate the impact of RTAs: pooled panel regressions and fixed effects models. The for-

mer approach allows us to exploit both cross-sectional and time-series variations in country

pairs’ exposure to RTAs. The second approach focuses on within-country-pair variations

over time.

We use annual data for a panel of 178 countries who constitute 31684 country pairs

over the 1970-2014 period. Our pooled panel regression has the following specification

∆log cit − ∆log cjt = α + β1 (∆log yit − ∆log yjt ) + β2 RT Aijt

(2.2)
+β3 RT Aijt × (∆log yit − ∆log yjt ) + ηt + ηi + ηj + ǫijt ,

where ∆cit (cjt ) denotes the change in real consumption per capita of country i(j) at time t

and ∆yit (yjt ) denotes that of the real output per capita. As discussed earlier, the response of

the relative consumption growth to the relative output growth measures the two countries’

ability to share risks. Moreover, RT Aijt is a dummy variable that equals 1 for the periods

71
where the country pair participates in a regional trade agreement and 0 otherwise. A nega-

tive β3 suggests that bilateral risk sharing improves in the presence of RTAs. ηt represents

time fixed effects, which captures the world aggregate output shock at time t. ηi , ηj repre-

sent country fixed effects that capture time-invariant country-specific characteristics. The

standard errors ǫijt are clustered at country pairs to control for potential heteroskedasticity

and autocorrelation. In addition to the baseline specification, we consider other variables

that could potentially influence bilateral consumption risk sharing as controls, including the

product of the two countries’ population and GDP per capita in logs at time t, as well as

the two countries’ product of GDP volatility over the sample period.

Table 2.2 reports the estimation results. Panel A presents the results for the full

sample of country pairs formed by 178 countries. The coefficient estimate for the relative

output growth is around 0.3 in all the regressions. The fact that it is between 0 and 1 in

value suggests imperfect risk sharing. More importantly, the coefficient of the interaction

term with RTA and relative output growth is significantly negative, which implies that

participating a regional trade agreement facilitates a country pair’s bilateral risk sharing.

Based on the estimates, being RTA partners lowers the response of a country pair’s relative

consumption growth to output growth by 0.11 (or 0.9 standard deviations). The result holds

when we control for population, GDP per capita, and GDP volatility of the country pair.

These variables do not appear to exhibit correlations with relative consumption growth.

We then focus on the sub-sample of country pairs who have ever participated in

the same RTAs over the sample period. As is shown in Panel B, the absolute value of the

coefficient estimate for the interaction term increases, which implies that RTAs play a more

72
vital role in consumption risk sharing for countries that have a history of regional trade

partnership.

Next we employ the panel approach with a fixed effects model to quantify the im-

pact of RTAs. By including country-pair fixed effects, this approach controls for unobserved

systematic differences across country pairs around RTA events. Table 2.2 Panel C reports

the results. It demonstrates that the response of relative consumption growth to output

growth decreases by 0.112 once a country pair joins an RTA. The coefficient estimate in

this fixed effects model is similar in magnitude to that in the pooled regression for the full

sample of country pairs.

In addition to these baseline findings, we conduct a robustness check. Since having

access to broader goods and capital markets may change bilateral risk-sharing patterns, we

control for country-pairs’ ties with the rest of the world. To this end, we introduce the

number of the GATT/WTO members from CEPII and financially-liberalized economies

based on Bekaert et al. (2004) in the country-pair as as regressors. As is shown in Table

2.6.2, the GATT/WTO membership and financial liberalization do not appear to have

a significant effect. Furthermore, the coefficient estimate for the interaction term with

RTA and relative Output growth stays significant. The fact that our finding is robust to

controlling for countries’ financial liberalization status indicate that barriers in the trade

channel remain to impede consumption risk sharing even if frictions in the asset market are

taken into consideration.

To sum up, the coefficient estimate for the interaction term of RTA and relative

output growth remains statistically and economically significant across alternative specifica-

73
tions. The finding supports the theory that reducing trade barriers promotes cross-country

risk sharing.

Table 2.2: Bilateral Risk Sharing and RTA

Dep Var: Pooled Regression Panel Approach
∆ Consumption A. Full Sample B. RTA Sample C. FE Model
(1) (2) (3) (4) (5) (6) (7) (8)
∆ Output 0.302*** 0.308*** 0.308*** 0.327*** 0.455*** 0.455*** 0.302*** 0.307***
(0.005) (0.005) (0.005) (0.012) (0.013) (0.013) (0.005) (0.005)
RTA 9.16e-17 8.02e-17 6.81e-18 2.17e-17 1.11e-16
(0.000) (0.000) (0.001) (0.001) (0.001)
RTA × ∆ Output -0.111*** -0.111*** -0.259*** -0.259*** -0.112***
(0.014) (0.014) (0.018) (0.018) (0.014)
GDP 1.11e-15 7.12e-15
(0.000) (0.001)
Population -2.52e-15 -1.68e-14
(0.001) (0.003)
GDP volatility -6.59e-16 1.03e-15
(0.000) (0.001)
Country Pair FE Y Y
Country FE Y Y Y Y Y Y
Time FE Y Y Y Y Y Y Y Y
Observations 1,420,421 1,419,887 1,419,887 217,616 217,616 217,616 1,420,421 1,419,887
R-squared 0.208 0.209 0.209 0.224 0.255 0.255 0.183 0.185

The dependent variable is country i’s relative consumption growth to that of country j. ∆ Output is country i’s relative

output growth to that of country j. RTA is a dummy variable which is 1 when country i and j both participate in a regional

trade agreement at t. Population is the product of the country pair’s population at t in logs. GDP is the product of the

country pair’s GDP per capita at t in logs. GDP volatility is the product of the standard deviation of the two countries’

per-capita GDP over time. The regressions include time fixed effects. In addition, pooled regressions include country fixed

effects and the panel approach includes country-pair fixed effects. Clustered standard errors reported in parentheses. ***,

**, and * indicate significance at the 1%, 5%, and 10% level.

2.3.2 A Gravity Model of Risk-sharing

After establishing the importance of trade costs for risk sharing by exploiting policy

shifts, we derive a cross-sectional prediction for cross-country consumption allocations. In

particular, we explore the implications of geographic distance for bilateral risk sharing.

74
 The international economics literature has a long tradition of empirically studying

how geographical distance influences economic linkages across countries. For instance, since

being developed by Isard (1954) and Tinbergen (1962), the gravity model in international

trade remains to be a workhorse due to its empirical success in predicting bilateral trade

patterns. More recently, the gravity model has been applied to a growing range of areas to

document that economic ties between two countries — including financial and migration

flows — are inversely proportional to the geographic distance between them (e.g. Portes

and Rey (2005) and Ramos and Surinach (2017)). Nevertheless, little is known about the

impact of distance on macro fundamentals. Our paper fills the gap in the literature by

focusing on consumption patterns.

The economic reasoning behind our hypothesis is straightforward. Trade costs

increase with geographic distance: the farther away countries are located from one another,

the higher trade costs it incurs to ship goods between them. If trade costs impede risk

sharing, we should expect that country pairs with greater geographical distance in between

exhibit weaker consumption risk sharing. Therefore, we hypothesize that there is a gravity

model of consumption risk-sharing.

We test this hypothesis using a two-stage regression. In the first stage we compute

the bilateral risk-sharing coefficients for all the country pairs using annual data over the

sample period by estimating the equation:

∆log cit − ∆log cjt = αij + βij (∆log yit − ∆log yjt ) + ǫijt . (2.3)

In the second stage we regress the estimated βij on geographic distance distij :

βij = α + γ (ln distij ) + ǫij . (2.4)

75
 We will confirm the hypothesis if γ is positive, which implies that countries which

are more distant from each other tend to exhibit a lower degree of consumption risk sharing.

In addition to the baseline specification with distance only, we augment the analysis with

standard gravity regressors including dummies for contiguity, common language, common

legal system, and time-averaged product of population in logs and GDP per capita in logs.

The values of these variables are sourced from the CEPII gravity database.

Table 2.3 reports the results of the second-stage regression. The coefficients for

geographic distance are significantly positive across all the specifications. The estimates

indicate that bilateral risk sharing decreases by about 0.01 (or 0.36 s.d.) for a 1% increase

in geographic distance. The results obtain when other gravity variables are controlled

for. Moreover, we find that bilateral risk sharing improves as a country-pair’s economic size

increases. From Column (4), a 1% increase in the product of GDP per capita raises bilateral

risk sharing by 0.051. This result indicates that more economically developed countries are

more likely to share risks with each other. Meanwhile, bilateral risk sharing decreases by

0.034 for a 1% increase in the product of population. One potential explanation is that,

there is a higher level of intra-national risk sharing in a more populous economy which

dampens the need for inter-national risk sharing. In terms of other gravity variables in

Table 2.3, we find that sharing a common language promotes bilateral risk sharing, while

having a common legal system yields less consistent results. When we control for country

sizes, similarity in legal systems does appear to facilitate risk sharing as shown in column

(4). In the same column the coefficient estimate for contiguity is positive, which contradicts

our expectation that country pairs that share borders should exhibit stronger risk sharing.

76
However, contiguity does promote risk sharing when geographic distance is controlled for,

as suggested by the sign of the interaction term in column (4). To sum up the main

findings in Table 2.3, we confirm that the signs of distance and GDP in this gravity model

of consumption risk sharing are the same as those in other gravity models including trade,

finance, and migration.

77
 Table 2.3: A Gravity Model of Risk-sharing
Dep Var: βij (1) (2) (3) (4)

Distance 0.011*** 0.014*** 0.009*** 0.007***

(0.002) (0.003) (0.002) (0.002)
Contiguity 0.142*** 0.033***
(0.012) (0.012)
Language -0.063*** -0.016***
(0.005) (0.005)
Legal 0.008* -0.033***
(0.004) (0.004)
GDP -0.050*** -0.051***
(0.001) (0.001)
Population 0.035*** 0.034***
(0.001) (0.001)
Constant 0.481*** 0.458*** 0.319*** 0.384***
(0.021) (0.022) (0.032) (0.033)

Observations 31,684 31,659 31,684 31,659

R-squared 0.001 0.008 0.224 0.226

The dependent variable is the estimated coefficient β from the

first stage regression. Higher β suggests weaker consumption

risk sharing. Independent variables include the log of geo-

graphic distance between two countries in kms, dummies for

common language, legal system, contiguity, and time-averaged

product of population in logs, and GDP per capita in logs.

Standard errors reported in parentheses. ***, **, and * indi-

cate significance at the 1%, 5%, and 10% level.

78
 In the next step we conduct two sensitivity analyses to verify the robustness of

the gravity model. Specifically we consider an alternative measure of distance and a more

robust measure of risk sharing.

The benchmark measure of geographic distance between two countries comes from

the CEPII, which calculates population-weighted distance between the biggest cities of

those two countries. For robustness, we also consider simple distance calculated with the

geographical coordinates of the capital cities. Results reported in Table 2.6.3 suggest that

the results remain unchanged.

Furthermore, we address a potential concern with our measure of risk sharing. In

Equation 2.3 where we define and estimate the risk-sharing coefficients, we use the difference

in output growth between a pair of countries (denoted as ∆log yit − ∆log yjt ) to reflect

the countries’ idiosyncratic risks. By doing so, we implicitly assume that the two countries

have the same degree of exposure to the global shocks. In other words, when loadings of

aggregate shocks (denoted as βi , βj ) are the same, the difference in idiosyncratic risks can

be written as

(∆log yit − βi ∆log ywt ) − (∆log yjt − βj ∆log ywt ) = ∆log yit − ∆log yjt , (2.5)

where ywt is the world output per capita. However, this assumption is not valid in some

cases so that the difference in output growth is also driven by the countries’ distinct degrees

of exposure to world aggregate risks. To address this concern, we conduct a robustness

check where we adjust for countries’ different loadings of aggregate risks. First we estimate

βi , βj from

∆log yit = αi + βi ∆log ywt + ǫit , ∆log yjt = αj + βj ∆log ywt + ǫjt . (2.6)

79
Second we calculate bilateral risk-sharing coefficients from the response of consumption to

obtain a more robust measure of idiosyncratic output shocks:

∆log cit −∆log cjt = αij +βij [(∆log yit −βi ∆log ywt )−(∆log yjt −βj ∆log ywt )]+ǫijt . (2.7)

Lastly we regress the estimated βij on geographic distance.

βij = α + γ (ln distij ) + ǫij . (2.8)

Table 2.4 presents the result for this robustness check. Compared to Table 2.3, the coeffi-

cient estimates have identical signs and similar values. The magnitude of the coefficient for

distance is greater by about 0.003, indicating that geographic distance plays a more crucial

role in shaping risk sharing patterns when we control for countries’ different exposure to

world aggregate risks. The gravity model of risk sharing remains robust.

80
 Table 2.4: Gravity model - Robustness check
Dep Var: βij (1) (2) (3) (4)

Distance 0.014*** 0.019*** 0.012*** 0.011***

(0.002) (0.002) (0.002) (0.002)
Contiguity 0.145*** 0.039***
(0.012) (0.012)
Language -0.059*** -0.014***
(0.005) (0.005)
Legal 0.020*** -0.021***
(0.004) (0.004)
GDP -0.051*** -0.052***
(0.001) (0.001)
Population 0.033*** 0.033***
(0.001) (0.001)
Constant 0.453*** 0.415*** 0.369*** 0.411***
(0.020) (0.022) (0.031) (0.033)

Observations 31,684 31,659 31,684 31,659

R-squared 0.001 0.008 0.225 0.226

The dependent variable is the estimated coefficient β from the

first stage regression. Independent variables include the log of

geographic distance between two countries in kms, dummies

for colony, common language, legal system, and time-averaged

product of population in logs, GDP per capita in logs, and GDP

p.c. volatility. Standard errors reported in parentheses. ***,

**, and * indicate significance at the 1%, 5%, and 10% level.

81
2.3.3 Gravity Model and RTA

As the last part of the empirical analysis, we bring all the previous pieces together

and study the relationship between the gravity model of risk-sharing and regional trade

agreements (RTA). The finding will allow us to examine the impact of policy obstacles in

the trade channel on the lack of efficient risk sharing across countries.

Theoretically in a frictionless world, bilateral risk sharing should not be correlated

with geographical distance among countries. All the countries share risks perfectly regard-

less of the physical distance among them. Nevertheless, there exist frictions that positively

comove with distance in the channels of risk sharing. For example, shipping costs in trade,

informational asymmetries in finance, migration cost in labor mobility are factors that pro-

hibit the channels from working efficiently to ensure perfect risk sharing. These frictions

rise with geographic distance, making risk sharing across country pairs that are physically

distant from each other increasingly difficult. These frictions can justify the gravity model

established in the previous section.

This paper focuses on trade in the goods market as a channel for risk sharing,

but frictions increase with geographic distance in various channels. Therefore, we need

additional empirical evidence to build the causal link between trade and the gravity model.

To this end, we exploit variations in RTAs as in Section 3.1 in order to attribute the gravity

model of risk sharing to the trade channel.

Besides the lower shipping costs due to the shorter traveling distance, countries

that are physically closer to each other obtain better risk sharing through trade since they

typically face fewer trade policy distortions under RTAs. RTAs are usually signed to reduce

82
trade barriers including tariffs and quotas in order to protect the common economic interest

of a geographic region. If the trade channel contributes to risk sharing across countries,

we should expect that geographic distance poses a smaller obstacle for risk sharing in the

presence of RTAs.

To test this hypothesis we estimate the following specification:

∆log cit − ∆log cjt = α + β1 (∆log yit − ∆log yjt ) + β2 (ln distij )

+β3 (RT Aijt ) + β4 (RT Aijt × ln distij )

(2.9)
+β5 [RT Aijt × ln distij × (∆log yit − ∆log yjt )]

+ηt + ηi + ηj + ǫijt .

In this specification we are particularly interested in β5 , the coefficient for RT A × log dist ×

∆y. A negative β5 implies that geographic distance impedes risk sharing to a less extent

for a pair of countries when they participate in a regional trade agreement.

The results presented in Table 2.5 confirm this hypothesis. The coefficients for

the three-way interaction term are significantly negative across all the regression specifi-

cations. Based on the coefficient estimates, a 1% increase in geographic distance lowers

consumption risk sharing by 0.016 (or 0.13 s.d.) more in the absence of RTAs. The inter-

pretation of the find is that, if geographic distance is a proxy for barriers to risk sharing,

RTAs overcome these barriers regardless of distance. This finding remains robust when I

add dummies for contiguity, common language, common legal system, and time-averaged

product of population in logs, GDP in logs, and GDP volatility in the regressions. These

standard gravity controls do not show significant correlations with cross-country relative

consumption growth. The only variable that has a significant coefficient other than the

three-way interaction term is the relative output growth, implying that the failure of con-

83
sumption risk sharing is not fully explained by the listed variables. This could be driven by

the fact that there exist frictions in other channels of consumption risk sharing.

Moreover, we conduct an exercise with two stage least squares (2SLS) in order to

further identify the mechanism through which distance and RTA affect risk sharing. To

implement 2SLS, we project trade with all the gravity variables including distance and

RTA in the first step. As a second step we include the projected trade as a control variable

when testing regression 2.9. As reported in Table 2.6.4, the coefficient for the interaction

term with RTA, distance and relative output growth is no longer significant. This finding

establishes the causality that RTA and distance affect consumption patterns through their

influence on trade.

Based on these results, we confirm our hypothesis that one important channel

through which we justify the gravity model established earlier is trade in goods. Geographic

distance matters for risk sharing because they covary with trade costs. Hence once trade-

promoting policies are considered, distance becomes less relevant in shaping consumption

risk sharing patterns. Furthermore, distance and RTA no longer matter for consumption

once we control for their impact on trade.

These findings support the main argument of the paper that trade costs impede

consumption risk sharing across countries. Therefore, efforts to lift trade barriers will

strengthen countries’ ability to share risks and smooth consumption.

84
 Table 2.5: Gravity Model with RTA
Dep Var: ∆ Consumption Pooled Regression Panel Approach
(1) (2) (3) (4)
∆ Output 0.309*** 0.310*** 0.310*** 0.308***
(0.005) (0.005) (0.005) (0.005)
RTA -1.90e-11 -2.11e-11 -2.12e-11 -3.28e-11
(0.002) (0.002) (0.002) (0.005)
RTA × Distance 2.61e-12 2.86e-12 2.86e-12 4.33e-12
(0.000) (0.000) (0.000) (0.001)
RTA × Distance × ∆ Output -0.016*** -0.016*** -0.016*** -0.016***
(0.001) (0.001) (0.001) (0.002)
Contiguity 1.66e-12 1.67e-12
(0.000) (0.000)
Language 5.12e-14 5.28e-14
(0.000) (0.000)
Legal 1.25e-13 1.32e-13
(0.000) (0.000)
GDP -3.90e-13
(0.000)
Population 4.24e-13
(0.001)
GDP volatility 9.04e-14
(0.000)
Constant -7.20e-13 -8.57e-13 -4.16e-12 -1.68e-13
(0.001) (0.001) (0.013) (0.001)
Country Pair FE Y
Country FE Y Y Y
Time FE Y Y Y Y
Observations 1,419,887 1,418,802 1,418,802 1,419,887
R-squared 0.210 0.211 0.211 0.195

The dependent variable is country i’s relative consumption growth to that of country j. ∆

Output is country i’s relative output growth to that of country j. Independent variables

include the log of geographic distance between two countries in kms, a dummy for RTA

which is 1 when country i and j both participate in a regional trade agreement at t, dum-

mies for contiguity, common language, legal system, and time-averaged product of popu-

lation in logs, GDP p.c. (per capita) in logs, and GDP p.c. volatility. The regressions

include time fixed effects. In addition, pooled regressions include country fixed effects and

the panel approach includes country-pair fixed effects. Clustered standard errors reported

in parentheses. ***, **, and * indicate significance at the 1%, 5%, and 10% level.

85
2.4 Conclusion

By exploiting cross-sectional and time-series variations in trade costs amongst

country pairs, this paper empirically evaluates the role of trade costs in explaining the lack

of international consumption risk sharing. We obtain three major findings from a large panel

of countries over the period 1970-2014. First, bilateral risk sharing improves once a pair

of countries become partners under a regional trade agreement. Moreover, a gravity model

of consumption risk sharing obtains since bilateral risk sharing decreases in geographical

distance between countries. In addition, this effect is more pronounced in the absence of

regional trade agreements. All the evidence supports the argument that trade costs impede

cross-country risk sharing.

This paper contributes to the growing literature that extends the gravity model

of trade to other topics including migration, financial flows, and exchange rate determina-

tion among others. Since these cross-country economic linkages also play an essential role

in international risk sharing, disentangling the influence of each channel can help us better

understand the global consumption pattern. Fitzgerald (2012) sets a nice example by quan-

tifying financial frictions and trade costs in explaining the lack of cross-country risk sharing.

Future work should incorporate other channels to account for the growing interdependence

across countries. Counterfactual analysis based on such structural frameworks will allow

us to measure the contribution of each channel to cross-country risk sharing. In terms of

policy implications, these papers call for the need of policies that aim to reduce the frictions

in the channels of risk sharing. Doing so will allow the global community to yield welfare

gains by reducing the overall consumption volatility.

86
2.5 References

[1] Anderson, J. E. and van Wincoop, E. (2003). Gravity with gravitas: A solution to the

border puzzle.The American Economic Review, 93(1):170–192.

[2] Backus, D. K. and Smith, G. W. (1993). Consumption and real exchange ratesin

dynamic economies with non-traded goods. Technical report, Queen’s Economics De-

partment Working Paper.

[3] Bekaert, G., Harvey, C. R., and Lundblad, C. (2004). Does financial liberalizationspur

growth?Journal of Financial Economics, 77:3–55.

[4] Callen, M., Imbs, J., and Mauro, P. (2015). Pooling risk among countries.Journalof

International Economics, 96(1):88–99.

[5] Dumas, B. and Uppal, R. (2001). Global diversification, growth, and welfare withim-

perfectly integrated markets for goods.The Review of Financial Studies,14(1):277–305.

[6] Eaton, J. and Kortum, S. (2002). Technology, geography, and trade.Econometrica,

70(5):1741–1779.

[7] Fitzgerald, D. (2012). Trade costs, asset market frictions, and risk sharing.Amer-ican

Economic Review, 102(6):2700–2733.

[8] Isard, W. (1954). Location theory and trade theory: short-run analysis.TheQuarterly

Journal of Economics, pages 305–320.

[9] Kalemli-Ozcan, S., Sorensen, B. E., and Yosha, O. (2003). Risk sharing andindustrial

specialization: Regional and international evidence.The AmericanEconomic Review,

87
 93(3):903–918.27

[10] Kollmann, R. (1995). Consumption, real exchange rates and the structure of inter-

national asset markets.Journal of International Money and Finance, 14(2):191–211.

[11] Kose, M. A., Prasad, E. S., and Terrones, M. E. (2009). Does financial globalization-

promote risk sharing?Journal of Development Economics, 89(2):258–270.

[12] Lewer, J. J. and Van den Berg, H. (2008). A gravity model of immigration.Economics

letters, 99(1):164–167.

[13] Lewis, K. K. (1996). What can explain the apparent lack of international con-sumption

risk sharing?Journal of Political Economy, 104(2):267–297.

[14] Lustig, H. and Richmond, R. J. (2019). Gravity in the exchange rate factor struc-

ture.The Review of Financial Studies.

[15] Martin, P. and Rey, H. (2004). Financial super-markets: size matters for asset-

trade.Journal of International Economics, 64(2):335–361.

[16] Obstfeld, M. and Rogoff, K. (2001). The six major puzzles in international macroe-

conomics: is there a common cause?NBER macroeconomics annual, 15:339–390.

[17] Okawa, Y. and Van Wincoop, E. (2012). Gravity in international finance.Journalof

International Economics, 87(2):205–215.

[18] Portes, R. and Rey, H. (2005). The determinants of cross-border equity flows.Journal

of International Economics, 65(2):269–296.

88
[19] Ramos, R. and Surinach, J. (2017). A gravity model of migration between the encand

the eu.Tijdschrift voor economische en sociale geografie, 108(1):21–35.28

[20] Sorensen, B. E. and Yosha, O. (1998). International risk sharing and europeanmone-

tary unification.Journal of International Economics, 45(2):211–238.

[21] Tinbergen, J. (1962). Shaping the world economy; suggestions for an international

economic policy.

2.6 Appendices

Table 2.6.1: List of Countries

Country Num of RTAs Num of partners

Albania 4 39

Algeria 2 69

Angola 2 21

Anguilla 1 28

Antigua n Barbuda 2 42

Argentina 7 44

Armenia 7 9

Aruba 1 28

Australia 10 27

Austria 41 105

Azerbaijan 5 6

89
Bahamas 2 42

Bahrain 4 17

Bangladesh 7 52

Barbados 2 42

Belarus 4 7

Belgium 44 105

Belize 2 42

Benin 3 52

Bermuda 1 28

Bhutan 4 7

Bolivia 5 43

Bosnia 3 35

Botswana 3 16

Brazil 8 46

Brunei Darussalam 8 16

Bulgaria 39 105

Burkina Faso 2 14

Burundi 4 16

Cambodia 6 15

Cameroon 3 74

Canada 10 14

Cape Verde 1 14

90
Cayman Islands 1 28

Central African 1 5

Chad 1 5

Chile 27 91

China 13 23

Colombia 11 80

Comoros 2 16

Congo 2 6

Congo, D.R. 0 0

Costa Rica 13 44

Côte d’Ivoire 2 14

Croatia 38 105

Cyprus 40 105

Czech Republic 40 105

Denmark 44 105

Djibouti 1 1

Dominica 2 42

Dominican Republic 4 35

Ecuador 6 71

Egypt 9 105

El Salvador 11 39

Equatorial Guinea 1 5

91
Estonia 40 105

Ethiopia 2 16

Fiji 4 42

Finland 41 105

France 44 105

Gabon 1 5

Gambia 1 14

Georgia 10 37

Germany 44 105

Ghana 2 52

Greece 43 105

Grenada 2 42

Guatemala 10 38

Guinea 2 52

Guinea-Bissau 1 14

Haiti 1 14

Honduras 11 39

Hong Kong 4 7

Hungary 40 105

Iceland 29 61

India 18 52

Indonesia 8 48

92
Iran 2 42

Iraq 2 51

Ireland 44 105

Israel 8 49

Italy 44 105

Jamaica 2 42

Japan 13 15

Jordan 8 51

Kazakhstan 7 8

Kenya 4 16

Kuwait 3 16

Kyrgyzstan 6 8

Lao 9 17

Latvia 40 105

Lebanon 3 47

Lesotho 4 25

Liberia 1 14

Lithuania 40 105

Luxembourg 44 105

Macao 1 1

Macedonia 5 40

Madagascar 2 29

93
Malawi 3 21

Malaysia 13 48

Maldives 3 7

Mali 2 14

Malta 40 105

Mauritius 6 51

Mexico 20 85

Moldova 7 43

Mongolia 0 0

Montserrat 2 42

Morocco 8 85

Mozambique 2 51

Myanmar 7 48

Namibia 3 16

Nepal 4 7

Netherlands 44 105

New Zealand 10 28

Nicaragua 10 76

Niger 2 14

Nigeria 2 52

Norway 28 60

Oman 4 17

94
Pakistan 10 52

Palestine 3 33

Panama 16 44

Paraguay 7 21

Peru 17 84

Philippines 9 53

Poland 40 105

Portugal 42 105

Qatar 3 16

Romania 39 105

Russia 8 12

Rwanda 4 16

Saint Lucia 2 42

Sao Tome 0 0

Saudi Arabia 3 16

Senegal 2 14

Seychelles 3 41

Sierra Leone 1 14

Singapore 22 63

Slovakia 40 105

Slovenia 40 105

South Africa 4 44

95
South Korea 14 84

Spain 42 105

Sri Lanka 8 47

St. Kitts 2 42

St. Vincent n Grenadines 2 42

Sudan 4 64

Suriname 2 42

Swaziland 5 25

Sweden 41 105

Switzerland 29 62

Syria 3 44

Taiwan 6 7

Tajikistan 2 7

Tanzania 5 58

Thailand 12 48

Togo 2 14

Trinidad n Tobago 3 82

Tunisia 7 88

Turkey 20 57

Turkmenistan 5 6

Turks n Caicos 1 28

U.A.E. 4 16

96
 Uganda 4 16

Ukraine 17 45

United Kingdom 44 105

United States 14 15

Uruguay 7 21

Uzbekistan 4 6

Venezuela 4 43

Viet Nam 9 48

Virgin Islands 1 28

Yemen 1 15

Zambia 3 21

Zimbabwe 5 86

This table reports the list of countries in our sample. For the countries that participated in any

regional trade agreement (RTA) from 1950 to 2014, we list the number of RTAs they were a

member of, the number of countries that were ever their partners in any RTA, and the average

duration (in years) of RTAs they participated in. Source: WTO.

97
 Table 2.6.2: Bilateral Risk Sharing and RTA — Robustness
Dep Var: Pooled Regression Panel Approach
∆ Consumption A. Full Sample B. RTA Sample C. FE Model
(1) (2) (3)

∆ Output 0.316 *** 0.464 *** 0.307 ***

( 0.003 ) ( 0.009 ) ( 0.005 )
RTA 5.48E-17 1.38E-17 9.64E-18
( 0.000 ) ( 0.001 ) ( 0.001 )
RTA × ∆ output -0.117 *** -0.266 *** -0.112 ***
( 0.009 ) ( 0.012 ) ( 0.014 )
WTO 5.09E-17 -1.44E-18 -7.61E-17
( 0.000 ) ( 0.000 ) ( 0.000 )
BHL -4.86E-17 -2.18E-17 -7.30E-17
( 0.000 ) ( 0.000 ) ( 0.000 )
Year FE Y Y Y
Observations 1,419,887 217,616 1,419,887
R-squared 0.194 0.246 0.194

The dependent variable is country i’s relative consumption growth to that of

country j. ∆ Output is country i’s relative output growth to that of country

j. RTA is a dummy variable which is 1 when country i and j both partici-

pate in a regional trade agreement at t. WTO and BHL denote the number

of the GATT/WTO members and financially-liberalized economies based on

Bekaert et al. (2004) in the country-pair. The regressions include time fixed

effects. Clustered standard errors reported in parentheses. ***, **, and *

indicate significance at the 1%, 5%, and 10% level.

98
Table 2.6.3: A Gravity Model of Risk-sharing — Robustness with Alternative Distance
Dep Var: βij (1) (2) (3) (4)

Distance 0.012*** 0.009*** 0.015*** 0.008***

( 0.002 ) ( 0.002 ) ( 0.003 ) ( 0.002 )
GDP -0.050*** -0.051***
( 0.001 ) ( 0.001 )
Population 0.035*** 0.034***
( 0.001 ) ( 0.001 )
Language -0.063*** -0.016***
( 0.005 ) ( 0.005 )
Legal 0.008* -0.033***
( 0.004 ) ( 0.004 )
Contiguity 0.114 0.487***
( 0.106 ) ( 0.115 )
Contg × Dist 0.005 -0.067***
( 0.016 ) ( 0.017 )
Constant 0.475*** 0.320*** 0.451*** 0.371***
( 0.021 ) ( 0.031 ) ( 0.022 ) ( 0.033 )

Observations 31,684 31,684 31,659 31,659

R-squared 0.001 0.224 0.008 0.227

The dependent variable is the estimated coefficient β from the

first stage regression. Higher β suggests weaker consumption risk

sharing. Independent variables include the log of geographic dis-

tance between two countries in kms, dummies for common lan-

guage, legal system, contiguity, and time-averaged product of

population in logs, and GDP per capita in logs. Standard errors

reported in parentheses. ***, **, and * indicate significance at

the 1%, 5%, and 10% level.

99
Table 2.6.4: A Gravity Model of Risk-sharing — Robustness with Projected Trade
Dep Var: ∆ Consumption Pooled Regression Panel
(1) (2) (3) (4)
∆ Output 0.138*** 0.138*** 0.138*** 0.135***
(0.005) (0.005) (0.005) (0.005)
RTA -0.001 -4.37E-04 -0.001 2.23e-4
(0.003) (0.003) (0.003) (0.008)
RTA × Distance 1.07E-04 1.33E-05 3.28E-05 -1.15e-4
(0.000) (0.000) (0.000) (0.001)
RTA × Distance × ∆ Output 0.002 0.002 0.002 0.002
(0.001) (0.001) (0.001) (0.001)
T\
rade 2.35E-04 3.21E-04 3.35E-04 5.19e-4
(0.000) (0.000) (0.000) (0.000)
Contiguity -7.55E-04 -7.96E-04
(0.001) (0.001)
Language -2.56E-04 -2.52E-04
(0.000) (0.000)
Legal -1.13E-04 -8.16E-05
(0.000) (0.000)
GDP -9.45E-04
(0.002)
Population 0.002
(0.003)
GDP volatility 4.09E-04
(0.001)
Constant -0.004 -0.006 -0.05 -0.01
(0.005) (0.005) (0.055) (0.007)
Country FE Y Y Y Y
Year FE Y Y Y Y
Observations 431,132 431,132 431,132 431,132
R-squared 0.112 0.112 0.112 0.0839

The dependent variable is country i’s relative consumption growth to that of coun-

try j. ∆ Output is country i’s relative output growth to that of country j. Indepen-

dent variables include the log of geographic distance between two countries in kms,

a dummy for RTA which is 1 when country i and j both participate in a regional

trade agreement at t, log of trade projected by gravity variables (denoted as T\

rade),

dummies for contiguity, common language, legal system, and time-averaged prod-

uct of population in logs, GDP p.c. (per capita) in logs, and GDP p.c. volatility.

The regressions include time fixed effects. Clustered standard errors reported in

parentheses. ***, **, and * indicate significance at the 1%, 5%, and 10% level.

100
Chapter 3

Deviations from Covered Interest

Rate Parity: evaluating drivers

from changes

3.1 Deviations from CIP and Global Financial Crisis

The foreign exchange (FX) market is a significant part of the financial markets.

It allows one country’s money exchange for another, defining the exchange rate between

them. Derived from comparing exchange rates and interest rate differentials, the theoretical

concept of covered interest rate parity (CIP) holds when there are no arbitrage opportunities

among these financial instruments. Despite this theoretical assumption, and its use in open

economy and monetary models, empirical research has shown that CIP does not necessarily

hold in practice.

101
 Before the Global Financial Crisis (GFC), CIP was observed for most advanced

economies. Small deviations from it were arbitraged in the short run. However, after the

GFC, large and persistent deviations were observed for the most liquid currencies in the FX

market. This research empirically studies the deviation from CIP and the possible drivers

for this phenomenon over a set of advanced market economies.

Some factors that can cause the deviation (“basis”) and the incapacity of being

arbitraged away have been identified in the literature. They are usually explained by:

transactions costs, counter-party credit risk, lack of liquidity in secondary markets, and d)

lack of funding due to systemic withdrawal by short-term lenders in a currency. Different

papers have tried to explain the deviations by using models and/or empirical analysis to

explain and quantify it. Funding shortage and counterparty risk are present in most of the

post-GFC literature as important factors.

Between 2001 and 2007, the world economy experienced a period of stability and

growth, followed by a global financial crisis and a period of dollar squeeze in 2008-2012. As

the most liquid currency, several international banks increased their holdings of US dollar

assets. This dollar funding was raised, primarily, through market operations: the bank

raises domestic currency through deposits and lends them against US dollars. In normal

times, it is done through the interbank market, operations with central banks, and FX swaps

to convert domestic currency funding into dollars. With the financial crisis and the Lehman

Brothers bankruptcy, lenders became risk averse, increasing the difficulty of keeping these

operations active.

Several research papers present in their theoretical models and empirical analysis

102
an important role for intermediaries. Despite the theoretical assumption of costless arbi-

trage, the actual no arbitrage condition requires a lot of resources, and regulatory require-

ments also raise the its cost. Before the crisis, the collateral and margining requirements

for arbitraging interest rates and exchange rate differentials were much less prohibitive in

balance sheet requirements.

Given the scenario of positive basis in some currencies, the understanding of possi-

ble causes is an interesting and current research question. I begin this paper with a selected

and critical literature review of theoretical models, empirical analysis and stylized facts de-

veloped to explain currency basis. With this map, I am able to identify the literature gaps

and establish my contributions. My main contribution is to analyze post GFC period in a

cross-country setup. Most of the work has been done for specific currencies like US dollar,

Euro and Japanese Yen, and this broader approach aims to compare deeply the drivers for

the deviation. Additionally, the channels in the literature have been evaluated in single

fashion, i.e., added in some theoretical model alone. I have claimed to see how relevant the

channels still remain when evaluated together.

Based on data from Bloomberg, the World Bank, and the Bank of International

Settlement (BIS), I have tested how several variables might get potential explanations for

the deviations in a different set of countries. The following sections start with the literature

review - divided before and after the GFC (and this last period grouped by theoretical

and empirical evidences). After that, I present the methodology and data, as well as

justifications for the choices. Then the results are presented, and the paper concluded.

103
3.2 CIP analysis through time

The CIP literature can be divided in two periods: before and after the 2008 GFC.

In the first period, CIP was empirically confirmed to hold in practice. This means that

small deviations were arbitraged away in short durations. After the 2008 crisis, some papers

identified substantially deviations, and these basis have been showing persistent behavior,

raising questions about the CIP concept. A large literature tests the CIP condition before

the global financial crisis and documents large CIP deviations during the crisis. This work

focuses on the post 2008 period.

3.2.1 Theoretical evidence

A considerable part of the literature, besides identifying the deviations on CIP,

proposed models to explain them. Several mechanisms serve as motivation for the modeling

exercise: currency as a scarce good, banking lending behavior, application to monetary

policy and zero lower bound, and corporate funding cost arbitrage (Bottazzi et al. (2013);

Ivashina et al. (2015); Amador et al. (2016); Liao (2016)). Some models approached the

problem by using a dynamic general-equilibrium model with margin constraints (Garleanu

& Pedersen, 2011). The CIP deviation may also be some component of a model with a

distinct primary goal, such as exchange rate determination (Gabaix and Maggiori, 2015).

Below, I add more details of the theoretical development.

As the channels of liquidity during the crisis were scarcer, a model proposed by

Bottazzi et al. (2013) proposed the cross-currency basis (which captures the deviations from

CIP) as the relative value of the scarcer currency. This hypothesis was able to match the

104
data, by checking collaterals as funding constraints. In a crisis, banks are more reluctant

to lend a scarcer currency, and it is priced into the cross-currency basis.

Ivashina et al. (2015) approached the problem by the credit quality of the banks,

that ultimately work as intermediaries for these operations. Using the financial friction as

a way to sharp bank behavior, this channel also matches the data.

The zero-lower bound (ZLB) limitation faced in the crisis was also explored as a

channel for explaining the deviation from CIP. The constraint on nominal interest rates

works as a source of limitation to arbitrage (Amador et al. (2016)). Liao (2016) examined

the issue through the lens of corporate fund cost. To explain it, the author developed a

model of market segmentation, in which post-crisis regulations and intermediary frictions

hampered arbitrage.

Garleanu and Pedersen (2011) had a similar approach as Liao (2016), which was

aimed for explaining deviations from LOOP. In their reasoning, a funding-liquidity crisis

raises the price gaps between securities with identical cash-flows but different margins.

Gabaix and Maggiori (2015) explored the topic in a model with moral hazard and imperfect

financial markets.

The banking regulation evidence is developed in models of intermediary-based

asset pricing (He and Krishnamurty (2012); Brunnermeier and Sannikov (2014)). Addi-

tionally, Gromb and Vayanos (2010) survey offers useful information on limits to arbitrage,

Brunnermeier and Pedersen (2009) on funding liquidity, Vayanos and Vila (2009) and Green-

wood and Vayanos (2014) on preferred habitat.

105
3.2.2 Empirical evidence

The literature on deviations from CIP also contains papers with a purely empirical

approach, without focusing on developing new models. In this section, I present some of

these papers and their methodology. This section will be important for discussing proxies,

identification of the basis, and measurement analysis.

Baba et al. (2008) analyzed the deviation in money markets in the second half of

2007. They identified the use of swap markets to circumvent US dollar funding shortages

and linked it with deviations from CIP. Their analysis contemplated a small window of 2007

and 2008.

Coffey et al. (2009) explored margin conditions and the cost of capital as drivers

of CIP deviations, especially during the crisis period. With increasing uncertainty about

counterparty risk and scarcer swap lines, a breakdown of arbitrage transactions in the

international capital markets was evaluated.

Adding more emphasis in the post-crisis period, Du et al. (2017) identified the CIP

deviations as a combination of cost of financial intermediation and international imbalances

in investment demand and funding supply across currencies. Costly financial intermediation

can explain why the basis is not arbitraged away post crisis.

Rime et al. (2016) also focused on the role of money market segmentation on

CIP deviations. With funding liquidity differences, it becomes impossible for FX swap

intermediaries to supply the markets without eliminating arbitrage conditions.

Sushko et al. (2016) linked the estimated dollar hedging demand (quantities) to

the variation in CIP deviations (prices). The authors argue that the degree to which CIP

106
holds depends more the relationship between the forward and spot price than the interest

rate differential, by showing that the CIP deviations rely mostly to hedge the USD forward.

This is explained by the cost associated to this hedge over regulatory aspects: it causes

some allocation on the balance sheet. With limits to arbitrage, CIP arbitrageurs charge

a premium in the forward markets for taking the other side of FX hedgers’ demand in

proportion to their balance sheet exposure. This will allow us to proxy the USD funding

needs in FX swap markets by banks through the financial system net liabilities.

3.3 Methodology and Data

By mapping all the different approaches and proxies discussed in the literature, it

is interesting to check how this proxies perform together and across a larger set of countries,

instead of a particular one. Being able to identify possible the “top drivers” to the phe-

nomena has immediate applications to policy makers, like monetary authorities. Potential

candidates for drivers are justified theoretically from the economic and financial literature.

The argument above are explained in the following two bullets:

• Are the variables used as possible drivers in the literature extensible to the new

set of countries?

• What are the commonalities and differences in the deviations from CIP among

countries?

107
 Some common explaining factors can be identified in the literature about deviation

for CIP, regardless of whether they have a theoretical or empirical approach. According to

literature review, liquidity and counterparty risk play a big role in driving these deviations,

but other factors might help explain it (demand for US dollars, risk from global banks,

financial variables).

Traditionally, this deviations were around zero in developed countries, and it seems

reasonable the recent literature focuses over the most relevant currencies in the FX market,

like US dollar, Euro, and Japanese Yen. Nevertheless, according to the Bank of International

Settlement (BIS) Foreign Exchange Survey realized in 2019 (BIS, 2019), the group composed

by Australian dollar (AUD), Canadian dollar (CAD), Swiss franc (CHF), Pound Sterling /

British Pound (GBP), and Singapore dollar (SGD), respond for approximately 25% of the

daily turnover on the foreign exchange market (table 1 below). Given this relevance and

the trend of participation, it is interesting to check how deviations from CIP behaved for

this group.

108
 Table 3.3.1: OTC foreign exchange turnover by currency pair

2007 2010 2013 2016 2019

Currency pair
Amount % Amount % Amount % Amount % Amount %

USD / EUR 892 26.8 1099 27.7 1292 24.1 1172 23.1 1584 24.0

USD / JPY 438 13.2 567 14.3 980 18.3 901 17.8 871 13.2

USD / GBP 384 11.6 360 9.1 473 8.8 470 9.3 630 9.6

USD / AUD 185 5.6 248 6.3 364 6.8 262 5.2 358 5.4

USD / CAD 126 3.8 182 4.6 200 3.7 218 4.3 287 4.4

USD / CHF 151 4.5 166 4.2 184 3.4 180 3.6 228 3.5

USD / SGD - - - - 65 1.2 81 1.6 110 1.7

Source: Bank of International Settlements -Triennial Central Bank Survey on Foreign Exchange

and OTC. Net-net basis, daily averages in April, in billions of US dollars and percentages,

adjusted for local and cross-border inter-dealer double-counting. I have omitted the pairs US-

D/CNY (Chinese Yuan), USD/HKD (South Korean Won) and USD/INR (Indian Rupee) with

4.1%, 3.3% and 1.7% of participation in 2019 due to lack of information for the calculations executed.

It is a relevant research question to investigate how the deviations from CIP be-

haved for these currencies, and to investigate if the same factors found on the literature can

be applied to this new set. In the scenario where similar evidence is found, it is possible to

extend the reach of the models that absorb these stylized facts to a wider range of coun-

tries. If the empirical behavior for this new set of currencies shows a different outcome,

then a natural consequence of the research would be to explore other possible factors that

might explain such behavior, and how the existing models can be extended to absorb such

behavior in that group.

109
 As the analysis was conducted with linear regressions, the possible limitations of

heteroskedasticity was contemplated by robustness on the errors. Despite the fact that

these variables could be evaluated as time series, these are mostly financial instruments and

concepts constructions (like bid-ask spread, terms of trade), or institutional statistic (bank

concentration). To enhance the results interpretation, I have added a lagged term for the

cross-currency basis. The high frequency of the data adds much noise to the evaluation.

Even considering it, it’s safe to assume that the simple model proposed captures significant

part of the variance – with R-squared close to 1 for the countries sample).

Finally, due to some differences in the time frame for the variables – some were

available in daily basis, while others were available as year basis 1 - the analysis was con-

ducted over levels, and not changes. This is the reason why the discussion was concentrated

over the signs and not the dimension of the coefficients.

3.3.1 The variables2

In this section, I will present, discuss, and justify the variables selected for my

analysis. The research question of this paper is to evaluate possible drivers for deviation for
1
The banking concentration, for example, was given as a year number. I have conducted a cubic spline

interpolation for quarterly frequency to add some disturbance. It wouldn’t make economic sense to transform

it to daily frequency, as the bank concentration doesn’t change that fast. This fact limited the change analysis

of the variables, so I have concentrated only on the interpretation of the coefficient signs.
2
Following the literature, I have also considered as proxies: volume of currency pairs(liquidity), demand

for USD (risk), implied volatility and risk-reversal-25-Delta (financial). All these variables have a high

correlation with the covariates used and haven’t added explanation power to the results. For space limitation,

I have maintained them out of the results showed, but they are available upon request.

110
CIP for the selected countries. The first step is to measure the basis over these currencies.

For this, I will use the approach of Du et al. (2017) on considering the basis as the spread

between cross-currency swap and USD Libor.

Basis and Cross-Currency Swaps

Following Nakisa (2011), we establish an example to understand the banking role

in the FX swap market.

“Let’s take the scenario of European banks with liabilities in dollars. As the euro

falls against the dollar, the cost of these payments increases. The situation is worsened by

US investors fear to lend to any European firms and banks. Through a cross-currency swap,

banks can raise funding in Europe in euro and transform this into dollars at a fixed currency

exchange rate that is agreed up front. The basis swap will allow the bank to transform their

dollar liability into a euro liability they can fund more easily. The cross-currency basis swap

will convert the lump sum that the bank borrowed in euros into a lump sum in dollars. The

counterparty in the cross-currency basis swap will actually pay the bank a little less than

the euro rate and pocket the difference between the euro rate and the rate on the swap. If

banks are desperate for dollar funding, they will be willing to receive less interest on the euro

interest on the swap. Cross-currency basis swaps are quoted as this difference in interest

received. Turning this around, it is extremely cheap for US banks to convert euro liabilities

into dollars. Then, the cross-currency basis swap rate measures deviations from the CIP

condition where interest rates are Libor interest rate swap rates.”

111
 Figure 3.3.1: Cross Currency Swap and Basis

Source: Nakisa (2011).

Table 3.3.2: Basis evaluation

Dollar Demand vs. Euro Basis Swap Rate Swap EUR → USD Swap USD → EUR

High Decrease (more negative) More expensive Less expensive

Low Increase (more positive) Less expensive More expensive

Source: Nakisa (2011).

As demand for dollar funding has increased the euro dollar basis swap rate has

fallen sharply and has become strongly negative. The data collected from Bloomberg for

the currencies selected was compiled in the tables below for one year of maturity.

112
 The literature review argued an increase on the deviations from CIP after the

GFC, with some persistent behavior. Following Coffey et al. (2009) and Du et al. (2017),

the period break proposed to analyze the deviation is composed by two intervals: the first

part goes from January 1st, 2000 to September 15th, 2008, the official bankruptcy date of

the Lehman Brothers. The second period goes from September 16th, 2008 until December

31st, 2018. The averages and standard deviation are presented on table 3 and illustrated

on figures 2 and 3.

113
 Table 3.3.3: Summary Statistics for Cross-Currency Swap Basis Points

Panel A: Whole period (01.01.2000 - 12.31.2018)

Whole period (01.01.2000 - 12.31.2018)

Cross-currency swap Basis Points
Mean SD Min Max N

Australian dollar 9.33 6.91 (71.00) 40.00 4,936

Canadian dollar (2.11) 11.98 (37.50) 37.00 4,746

Swiss Franc (16.70) 17.13 (82.12) 3.25 4,666

British Pound (7.72) 12.48 (116.60) 11.60 4,725

Singaporean dollar (3.23) 3.61 (35.00) 9.73 4,330

Panel B1: Period I - Before Lehman Brothers Bankruptcy (01.01.2000 - 09.15.2008)

Per. I - Before Lehman Brothers Bankruptcy

Cross-currency swap Basis Points
Mean SD Min Max N

Australian dollar 5.19 2.01 (8.00) 10.70 2,230

Canadian dollar 7.10 3.22 (1.00) 21.00 2,041

Swiss Franc (0.67) 4.10 (24.00) 3.25 1,959

British Pound (1.70) 4.76 (32.40) 5.00 2,021

Singaporean dollar (2.46) 3.07 (25.00) 4.00 1,634

Panel B2: Period II - After Lehman Brothers Bankruptcy (09.16.2008 - 12.31.2018)

Per. II - After Lehman Brothers Bankruptcy

Cross-currency swap Basis Points
Mean SD Min Max N

Australian dollar 12.77 7.62 (71.00) 40.00 2,706

Canadian dollar (9.08) 11.48 (37.50) 37.00 2,705

Swiss Franc (28.29) 13.19 (82.13) (6.00) 2,707

British Pound (12.22) 14.42 (116.60) 11.60 2,704

Singaporean dollar (3.70) 3.84 (35.00) 9.73 2,696

Source: Calculated by the author.

114
 Figure 3.3.2: Source: Elaborated by the author.

Figure 3.3.3: Source: Elaborated by the author.

From table 3 we can see a significant change in the deviation from the covered

interest rate parity for these currencies. The increase is observed not only in the mean

115
basis, but also in their volatilities. The only exception is seen over the Singaporean dollar

(SGD), which has a somehow stable standard deviation on the basis (but still with a rise

in the mean).

The additional variables presented in the literature review and which will compose

the regression analysis, are explained below.

Liquidity proxies

The liquidity proxies used are the spread over the exchange rate spot and the

spread over the exchange rate future contract. The first is the difference between the prices

quoted for an immediate sale (offer) and an immediate purchase (bid) for the spot exchange

rate. The latter is the difference between the prices quoted for an immediate sale (offer)

and an immediate purchase (bid) for future exchange-rate contracts. The data for both

these variables was collected on Bloomberg.

Risk proxy

The risk proxy used in this paper and based in the literature is the average CDS

(Credit Default Swap) of the G-Sibs (Global Sistemically important banks). I have selected

the list of thirty banks defined by the Financial Stability Board (FSB) and have calculated

their average premium. Ideally, it should obtain a measure of risk for these institutions

that are responsible for intermediating the swap trades. The banks defined as G-Sibs by

the FSB are available at table A1.

116
Financial Market and Macro Variables

Regarding financial market and macroeconomic variables, I have used Terms of

Trade, Bank concentration and VIX. The terms of trade is a widely used measure in the

literature. It is defined as the ratio of export prices to import prices, and it can be defined

as the amount of import goods an economy can purchase per unit of export goods. It is

used in this paper as a context for demand for USD, and its information was obtained on

Bloomberg. The Bank Concentration is calculated by the World Bank. It measures the

weight of the five largest banks in the country (by assets).

The Volatility Index (VIX) measures of the stock market’s expectation of volatility

implied by S&P 500 index options, calculated and published by the Chicago Board Options

Exchange (CBOE). It is a widely used measure of global risk and its data was also obtained

on Bloomberg.

3.4 Results

3.4.1 Shifts and Regressions

A primary and simple check regarding the impact of the Lehman Brothers bankruptcy

over the deviations from CIP was done through a t-test and mean comparison between the

two periods. Despite a illustrative representation of the different behaviors through figures

2 and 3, a formal test is conducted, with its results available on table 5.

117
 Table 3.4.1: T-tests for mean differences pre x post Lehman

Difference Std. Error N (pre-Lehman) N (post-Lehman)

Australian dollar -7.5843*** 0.1658 2230 2706

Canadian dollar 16.1801*** 0.2615 2041 2705

Swiss Franc 27.6028*** 0.3084 1959 2707

British Pound 10.5057*** 0.3336 2021 2704

Singaporean dollar 1.2399*** 0.1119 1634 2696

Source: Elaborated by the author.

Tables 6-10 compile the regressions for all currencies having the basis as the de-

pendent variable and the statistical significance of each independent variable described on

sections 3.1.2 - 3.1.4, added accordingly to its economic meaning and relation to the litera-

ture.

Additionally, I added a dummy for post crisis period (September 15th, 2008, the

Lehman Bank bankruptcy). This dummy represents a fixed effect for pre and post period,

allowing me to control for unobservable differences in CIP before and after crisis. There

was a concern regarding the scenario where CIP was just at a different level after the crisis

for reasons not captured by the regressors, indicating the necessity of time fixed effect

control for it. As the data has daily frequency not only on the dependent variable, but

also in the regressors, this was ruled out. As I am looking for the potential changes in

the relationship between the explanatory variables and CIP during the crisis, then I will

look for the interaction among dummies and regressors for pre and post crisis periods. For

illustration purposes, I will also add the regressions without the dummies. I will start with

118
a baseline model (regressions 1 and 3), which contemplates most used liquidity variables

– spread for spot and future contracts and terms of trade. Additionally, I will evaluate

an extended model, which will embrace also other variables used as possible drivers for

deviation on CIP, like Bank concentration, average CDS premia for global systemically

important banks, and global volatility index (regressions 2 and 4). Both models were added

on a lagged term for the dependent variable.

The regressions assume the form:

CCSt = β0 + β1 × Spott + β2 × F uturet + β3 × T OTt + β4 × CCSt−1 (3.4.1)

CCSt = β0 + β1 × Spott + β2 × F uturet + β3 × T OTt + β4 × BCt +

β5 × CDSt + β6 × V IXt + β7 × CCSt−1 (3.4.2)

CCSt = β0 + β1 × Spott + β2 × F uturet + β3 × T OTt + β4 × D × Spott +

β5 × D × F uturet + β6 × D × T OTt + β7 × D + β8 × CCSt−1 (3.4.3)

CCSt = β0 + β1 × Spott + β2 × F uturet + β3 × T OTt + β4 × D × Spott +

β5 × D × F uturet + β6 × D × T OTt + β7 × BCt + β8 × CDSt +

β9 × V IXt + β10 × D × BCt + β11 × D × CDSt + β12 × D × V IXt +

β13 × D + β14 × CCSt−1 (3.4.4)

Where:

• CCS – Cross-currency swap basis points

• Spot – Spread spot (section 3.1.2)

119
• Future – Spread future (section 3.1.2)

• TOT – Terms of trade (section 3.1.4)

• D – dummy for the period related to the Lehman Brothers bankruptcy (D = 0 before the

event, and D = 1 after the event).

• BC – Bank concentration (section 3.1.4)

• CDS – Average of GSIBs CDS (section 3.1.3)

• VIX – Volatility index (section 3.1.4)

• CCSt−1 – lagged cross-currency swap basis points

The regression results are presented on tables 6-10. The economic intuition of an

increase on the cross-currency swap basis points (taken as the proxy for the deviation of

covered interest rate parity) implies that its related to a decrease in market liquidity. This

was expected to be seen or correlated with an increase in the spread of both spot and future

contracts. The term of trade variable is defined as the ratio of exports over imports. This

means that an increase in it implies a smaller necessity of US dollars, and an improvement

on the liquidity. That said, it is expected that terms of trade would be negatively related

with the cross-currency basis.

The group of variables added on regressions (2) and (4) – bank concentration,

average CDS for G-Sibs and VIX – is expected to be related to the basis in the following

fashion.

As the trades for cross currency swaps are mostly over-the-counter trades, they

are conducted through banks, and not by the brokerage firms directly on exchange houses

120
(like NYSE, CBOE, for example). If a local firm is willing to do a cross-currency swap

and its local financial system is more concentrated, i.e., offer less options for this firm, it is

expected that the basis for the operation would be higher. The hypothesis for this variable

is that it is positively related with the CIP deviation.

As the CDS is a security against the default for any specific asset issuer, the G-

Sibs average CDS evaluates how the market is pricing the risk of these financial institutions

to not honor their operations. The higher the CDS premium for this group pf banks, less

confident the clients will be to conduct transactions with, implying a positive relation with

CIP deviation. On the other hand, there is an argument that this lack of confidence on

these banks would contribute for their lower rates, increasing the number of operations.

We see that there are arguments on both directions, and the regressions will help to clarify

which one (if any) prevails.

VIX is the S&P 500 Volatility Index – a largely used measure of stocks’ market

volatility. Its increase is expected to widen the basis of cross-currency swaps.

The table below summarizes the hypothesis for the variable directions.

Table 3.4.2: Direction hypothesis for the dependent variables

Dep. Variables Independent Variables

Terms of Bank Average

Cuurency Basis Spot Future Trade Concentration G-Sibs CDS VIX

(direct quotation) ↓ ↓ ↓ ↑ ↓ Undefined ↓

Source: Elaborated by the author.

It is important to note that the Australian dollar goes in the opposite direction of

121
all the other currencies due to the form it is quoted in the market. To circumvent that, I

have transformed the AUD/USD quotation, i.e., the amount of US dollar required to buy

one Australian dollar, to USD/AUD, the amount of Australian dollars to buy one US dollar.

Baseline - Regressions (1) and (2)

The baseline cases, regressions (1) and (2) contemplates the whole period without

a dummy variable for the period after Global Financial Crisis. The addition of the second

group of control variables (regression 2) doesn’t’ change neither the significance not the

signs of the coefficients for the first regression (1). The only exception is for the Canadian

dollar, which start having the terms of trade as significant on regression 2, in contrast to

regression 1.

Regarding the expected signs, the results are mixed. The spot spread has the

hypothesized sign for all currencies, besides GBP, although none of it are significant. The

future spread has the hypothesized sign only for AUD dollar and CHF, with only CAD

and CHF significant. This means that these variables are weakly explanatory in this set-

up. For the terms of trade, the coefficients are significant for all currencies, despite AUD.

Nevertheless, the only expected sign is for the GBP.

Extended Regressions (3) and (4)

The comparison between regressions (3) and (4) is similar as the one done above

– (1) and (2), with the shift for the post-GFC considered. When the second group of inde-

pendent variables are considered, the interaction term with future spread loses significance

for the AUD and gain significance for the SGD. All other coefficients for the first block are

122
the same in terms of significance.

Regarding the expected signs hypothesized on table 5, the results are also far for

conclusion. While AUD matches the expected signs, all other currencies show deviances

from it (CAD shows divergence is spot and future spread, as well as terms of trade; CHF

shows divergence on future spread, terms of trade, and VIX; GBP shows divergence on

VIX; and SGD shows divergence on future spread, terms of trade, bank concentration and

VIX).

The most complete set-up – Regression (4)

This section focuses mostly on the most complete set-up, i.e., regression (4). For

the five countries evaluated (tables 6-10), only UK and Singapore (tables 9 and 10) show

a significant result for the second group of interactions - bank concentration, average CDS

of G-Sibs and VIX. The bank concentration coefficient shows opposite signs for these two

countries, weakening its interpretation. In a set of five countries, three don’t show signif-

icance, and the two other present conflicted intuition. It definitely suggests some deeper

analysis and treatment to understand how it is related to the deviations from CIP, and also

puts a yellow sign before using it as a channel for a theoretical model.

Regarding the average CDS of G-Sibs, it was discussed above how both signs

for the coefficient could have an economic interpretation. The empirical results for the

regression over UK and Singapore show consistency in the positive sign, suggesting that the

increase in the risk perception for these banks widens the basis.

For the VIX, the results shown for both UK and Singapore are also consistent

with the hypothesis: a larger volatility implies a higher risk perception, which correlates

123
with an increase in the basis / deviation from CIP. Even though the results are satisfactory

for two out of three variables for UK and Singapore, it is interesting to explore a deeper

institutional country analysis in order to identify why these variables are not significant for

Australia, Canada, and Switzerland. In some degree, we can say that the results suggest

the financial and macro variables as highly correlated with this new scenario of deviation

from CIP. The cause for the Singapore result might be caused by its particular degree of

freedom over its financial market, which caused less risk aversion over its currency.

Regarding the spot and future spread, the results are mixed. Australian dollar

and Canadian dollar show no correlation, while swiss franc show significance only in the

future spread, the British Pound show significance (at 10% level) only for the spot spread,

and the Singaporean dollar show significance in both spot and future spread.

The significance on terms of trade (Australian dollar and Canadian dollar) can be

an indication of the real export channel getting higher weight through this period.

The lack of significance of the bid-ask spread for spot and/or future markets for

some currencies is intriguing. The increase of CIP post GFC has as one of main hypothesis

the liquidity constraint. Also, one of the most practiced proxies for liquidity concerns the

bid-ask spread. Nevertheless, we see that the significance over the interaction with the post-

Lehman dummy is present over independent variables with institutional characteristics, like

terms of trade and bank concentration, and global factors, like VIX and the CDS premium

of G-Sibs (UK and Singapore). This fact raises questions about the proper intervention

policies from the regulators.

For example, one of the first responses from the Fed to the crisis was the incentive

124
to some mergers in order to absorb the more problematic banks. It’s noteworthy that the

main idea was to “stop the bleeding” and avoid bank-runs through all the system, but this

result try to shed light about the cost of a higher bank concentration as a narrower set of

institutions to channel the liquidity to the real economy.

Regarding the VIX and the G-Sibs CDS (British Pound and Singaporean dollar),

the main lesson provided by this result was well explored: the lack of regulation and criteria

from regulatory agencies, supposed to supervise the financial institutions, contributed to

the increase on leverage and risk profile for all the major banks.

125
 Table 3.4.3: Australian dollar Cross-Currency Swaps regressions
(1) (2) (3) (4)
Variables AUD CCS AUD CCS AUD CCS AUD CCS
Spot spread 5.911 6.554 -0.516 -0.636
(3.954) (4.241) (1.182) (1.42)
Future spread 0.0257 0.0702 -0.005 -0.00161
(0.0325) (0.0705) (0.00626) (0.00678)
Terms of trade -0.000124 -0.00284 -0.00171 0.000953
(0.00145) (0.00213) (0.00133) (0.00219)
D * (Spot spread) 9.203 9.323
(6.109) (6.236)
D * (Future spread) 0.383* 0.252
(0.233) (0.276)
D * (Terms of trade) -0.00560* -0.00717*
(0.003) (0.00417)
Bank Concentration 0.00723 0.00495
(0.00604) (0.00494)
Average CDS G-Sibs 0.00189 -0.00407
(0.00126) (0.0031)
VIX -0.0107 0.00606
(0.00736) (0.00603)
D * (Bank Concentration) 0.0673
(0.0558)
D * (Average CDS G-Sibs) 0.00569
(0.00365)
D * (VIX) -0.00886
(0.0144)
Lag – CCS 0.981*** 0.973*** 0.968*** 0.955***
(0.0151) (0.0185) (0.0238) (0.0304)
D 0.402 -5.83
(0.258) (5.16)
Constant 0.182 -0.281 0.175 -0.242
(0.138) (0.506) (0.131) (0.389)

Observations 4757 4335 4757 4335

R-squared 0.963 0.962 0.963 0.962
Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

126
 Table 3.4.4: Canadian dollar Cross-Currency Swaps regressions
(1) (2) (3) (4)
Variables CAD CCS CAD CCS CAD CCS CAD CCS
Spot spread 67.23 49.94 92.72 33.04
(98.63) (133.1) (146.5) (156.9)
Future spread -0.000768*** -0.000721*** -0.000600*** -0.000420***
(0.000211) (0.000209) (0.000155) (0.000159)
Terms of trade 0.00735 0.0176** 0.00165 -0.0122
(0.00502) (0.00861) (0.0091) (0.0121)
D * (Spot spread) -19.13 -36.78
(181.4) (214.1)
D * (Future spread) -0.00425 -0.000337
(0.0202) (0.0187)
D * (Terms of trade) 0.0291** 0.0586***
(0.0135) (0.02)
Bank Concentration 0.00758 0.0249***
(0.00587) (0.00916)
Average CDS G-Sibs -0.00119* 0.00169
(0.000688) (0.00406)
VIX 0.0124 0.00445
(0.00879) (0.00986)
D * (Bank Concentration) 0.00346
(0.0121)
D * (Average CDS G-Sibs) 0.00244
(0.00424)
D * (VIX) 0.00543
(0.0156)
Lag – CCS 0.990*** 0.984*** 0.978*** 0.955***
(0.00272) (0.00458) (0.00559) (0.0102)
D -0.554*** -1.817*
(0.174) (1.01)
Constant -0.0888** -0.988* 0.128 -1.955***
(0.0431) (0.594) (0.0838) (0.744)

Observations 4570 4298 4570 4298

R-squared 0.981 0.981 0.982 0.981
Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

127
 Table 3.4.5: Swiss Franc Cross-Currency Swaps regressions
(1) (2) (3) (4)
Variables CHF CCS CHF CCS CHF CCS CHF CCS
Spot spread 72.38 109.5 -18.39 50.98
(57.89) (68.47) (79.45) (97.75)
Future spread -0.00229** -0.00257*** -0.00222 -0.00233**
(0.00104) (0.000527) (0.00151) (0.00094)
Terms of trade 0.0167** 0.0351*** 0.0193*** 0.0263***
(0.00657) (0.0126) (0.00733) (0.0092)
D * (Spot spread) 114.1 67.36
(99.2) (120.3)
D * (Future spread) -2.134** -3.254***
(0.866) (0.907)
D * (Terms of trade) -0.012 0.0594
(0.045) (0.0847)
Bank Concentration 0.0734*** 0.0235
(0.0191) (0.0195)
Average CDS G-Sibs -0.00298** -0.00291
(0.00133) (0.00205)
VIX -0.00959 0.000695
(0.00833) (0.00543)
D * (Bank Concentration) 0.0518
(0.0356)
D * (Average CDS G-Sibs) 0.000355
(0.00247)
D * (VIX) -0.0228
(0.0169)
Lag – CCS 0.993*** 0.976*** 0.983*** 0.972***
(0.00316) (0.00649) (0.00662) (0.00753)
D -0.693 -4.376
(0.472) (2.706)
Constant -0.0207 -6.453*** 0.0836** -2.008
(0.0413) (1.742) (0.0405) (1.83)

Observations 4470 4193 4470 4193

R-squared 0.988 0.988 0.988 0.988
Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

128
 Table 3.4.6: British Pound Cross-Currency Swaps regressions
(1) (2) (3) (4)
Variables GBP CCS GBP CCS GBP CCS GBP CCS
Spot spread -7.495 -38.14 -5.415 -5.805
(54.8) (33.69) (4.411) (4.382)
Future spread 0.303 0.0687 -0.233 -0.226
(0.456) (0.177) (0.146) (0.141)
Terms of trade -1.795*** -1.988*** -0.0239 -0.0392**
(0.0718) (0.101) (0.0146) (0.0177)
D * (Spot spread) 27.90* 26.73*
(14.38) (14.31)
D * (Future spread) 0.247 0.244
(0.155) (0.15)
D * (Terms of trade) 0.130** -0.0283
(0.056) (0.0895)
Bank Concentration -0.0814*** 0.000711
(0.0155) (0.00211)
Average CDS G-Sibs -0.0413*** -0.00735**
(0.00485) (0.00294)
VIX -0.788*** 0.00753
(0.0451) (0.00662)
D * (Bank Concentration) 0.0574***
(0.0216)
D * (Average CDS G-Sibs) 0.00632**
(0.00315)
D * (VIX) -0.0577***
(0.0193)
Lag – CCS 0.303*** 0.164*** 0.985*** 0.966***
(0.0494) (0.0318) (0.00874) (0.00971)
D -1.033** -3.201
(0.438) (1.95)
Constant -0.147 26.61*** 0.0595 0.121
(0.247) (1.628) (0.0475) (0.193)

Observations 4594 4248 4495 4197

R-squared 0.096 0.508 0.978 0.978
Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

129
 Table 3.4.7: Singaporean dollar Cross-Currency Swaps regressions
(1) (2) (3) (4)
Variables SGD CCS SGD CCS SGD CCS SGD CCS
Spot spread 0.26 0.862 -261.5*** -363.9***
(19.13) (22.3) (94.98) (138.1)
Future spread -0.000353 -0.000505 0.0285 0.0531**
(0.000816) (0.000695) (0.0187) (0.0232)
Terms of trade 0.0320*** 0.0478*** 0.0236** 0.0179
(0.00748) (0.00987) (0.01) (0.0186)
D * (Spot spread) 267.7*** 372.8***
(95.65) (138.8)
D * (Future spread) -0.0288 -0.0537**
(0.0187) (0.0232)
D * (Terms of trade) 0.0159 0.0164
(0.0122) (0.0202)
Bank Concentration -0.0600*** 0.0292
(0.0126) (0.039)
Average CDS G-Sibs -0.00430*** -0.0192***
(0.000854) (0.00394)
VIX -0.0237*** -0.0033
(0.00638) (0.00778)
D * (Bank Concentration) -0.140***
(0.0506)
D * (Average CDS G-Sibs) 0.0146***
(0.0038)
D * (VIX) -0.0184*
(0.0103)
Lag – CCS 0.916*** 0.826*** 0.914*** 0.807***
(0.0189) (0.0284) (0.0193) (0.0305)
D 0.0125 13.57***
(0.0835) (4.964)
Constant -0.0336 6.384*** -0.00146 -2.495
(0.033) (1.267) (0.0615) (3.933)

Observations 4058 3895 4058 3895

R-squared 0.875 0.882 0.875 0.883
Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

130
3.4.2 Some extensions

To improve the power of the results, I have also run a panel analysis for the data to

observe how would that differ from the cross countries results (without the lagged variables).

They are available on table 11 and show similar results from the cross-country analysis.

Table 3.4.8: Cross-currency Swaps panel regressions

(1) (2)
Variables CCS CCS
Spot spread -9.519 -15.13**
(8.113) (6.92)
Future spread 0.0371 0.0154
(0.0528) (0.0254)
Terms of trade 0.116*** 0.186***
(0.00559) (0.00483)
D * (Spot spread) -7.103 -28.63
(27.32) (23.55)
D * (Future spread) -0.029 0.000059
(0.054) (0.0274)
D * (Terms of trade) 0.491*** 0.465***
(0.0107) (0.0101)
Bank Concentration 0.0851***
(0.00283)
Average CDS G-Sibs -0.151***
(0.00563)
VIX 0.328***
(0.0116)
D * (Bank Concentration) 0.425***
(0.0104)
D * (Average CDS G-Sibs) 0.140***
(0.00629)
D * (VIX) -0.487***
(0.023)
D -11.64*** -41.26***
(0.128) (0.866)
Constant 1.980*** -7.783***
(0.0514) (0.305)

Observations 27350 25231

R-squared 0.311 0.398
Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

As a frequent argument in the literature about the deviation from CIP resides

131
on the lack of liquidity for the markets, I have checked how the Cross-country swap basis

points behaved for the same six pairs of currencies for the period around the Quantitative

Easing. I have run the same t-test for the means, but now comparing the QE1 and QE2

announcements dates, and also for a smaller window – two weeks before and two weeks

after the announcement (tables 12 and 13).

Table 3.4.9: T-tests for mean differences pre x post QE1

Difference Std. Error N (pre-QE1) N (post-QE1)

Australian dollar -6.8956*** 1.0737 15 17

Canadian dollar 21.8926*** 6.1024 15 17

Swiss Franc 11.6559*** 3.6944 15 17

British Pound 0.6838 9.3846 15 17

Singaporean dollar -6.6382*** 1.2344 15 17

Source: Elaborated by the author. The QE1 date was Nov 25, 2008.

Table 3.4.10: T-tests for mean differences pre x post QE2

Difference Std. Error N (pre-QE2) N (post-QE2)

Australian dollar 1.3836** 0.5211 22 23

Canadian dollar -0.6337 0.6828 22 23

Swiss Franc 2.3772*** 0.5815 22 23

British Pound 0.224 0.2979 22 23

Singaporean dollar 0.3549 0.3927 22 23

Source: Elaborated by the author. The QE2 date was Aug 10, 2010.

The weaker results, with some currencies showing no difference in their means

132
between the periods, is consistent with the idea that market liquidity is not the main

driver for the deviation of CIP. The period of analysis for the change due to specific event

(the QE announcements) was chosen to be small in an attempt to isolate its effect over the

deviation from CIP. If extended to a larger period, the effect would be probably be impacted

from other market variables as well. That said, the regressions for these window were not

reproduced due to the small dataset and limited explanation power. This limitation raises

a possibility for a future study with intraday data around the QE announcements.

3.5 Final Remarks

In this paper, I have empirically conducted an evaluation over the deviation from

covered interest rate parity. Through an extensive literature review, I have mapped possible

drivers for explaining the failure of the no arbitrage condition in the foreign exchange market

after the great financial crisis. I have also extended the analysis for a set of countries

not explored in the literature despite having a significant weight in the FX market. I

have obtained results that claim for a larger reason for the deviation than only liquidity

constraints, which open channels for further empirical analysis as well as different channels

for theoretical proposals on monetary policies models. The lack of consistency of some

drivers discussed in the literature when evaluated in a cross-country set-up raises a question

of how important institutional characteristics of a country are when evaluating its deviation

from CIP. Despite the rise in the deviation from the CIP being a real fact obtained through

the empirical observations, its mechanisms are still not clear enough. This is an important

contribution for the theoretical developments aiming to absorb the failure to no arbitrage

133
condition in their models.

3.6 References

[1] Amador, M., Bianchi, J., Bocola, L., and Perri, F. (2016). Exchange rate policies at the

zero lower bound. National Bureau of Economic Research, Working Paper 23266:1 38.

[2] Avdjiev, S., Du, W., Koch, C., and Shin, H. S. (2016). The dollar, bank leverage and

the deviation from covered interest parity. BIS Working Papers, WP 592:1 37.

[3] Baba, N., Packer, F., and Nagano, T. (2008). The spillover of money market turbulence

to fx swap and cross-currency swap markets. BIS Quarterly Review, March:1 14.

[4] Bank for International Settlements (2019). Triennial central bank survey.

[5] Bottazzi, J.-M., Luque, J., Pascoa, M. R., and Sundaresan, S. (2013). Dollar short-

age, central bank actions, and the cross currency basis. Working Paper, Columbia

University, pages 1 49.

[6] Brunnermeier, M. K. and Pedersen, L. H. (2009). Market liquidity and funding liq-

uidity. Review of Financial Studies, pages 1 38.

[7] Brunnermeier, M. K. and Sannikov, Y. (2014). A macroeconomic model with a finan-

cial sector. American Economic Review, 104(2):379 421.

[8] Coffey, N., Hrung, W. B., and Sarkar, A. (2009). Capital constraints, counterparty

risk, and deviations from covered interest rate parity. Federal Reserve Bank of New

York Staff Reports, pages 1 46.

134
 [9] Du, W., Tepper, A., and Verdelhan, A. (2017). Deviations from covered interest rate

parity. National Bureau of Economic Research, Working Paper 23170:1 84.

[10] Gabaix, X. and Maggiori, M. (2015). International liquidity and exchange rate dy-

namics. Quarterly Journal of Economics, pages 1369 1420.

[11] Garleanu, N. and Pedersen, L. H. (2011). Margin-based asset pricing and deviations

from the law of one price. Review of Financial Studies, 24 (6):1980 2022.

[12] Greenwood, R. and Vayanos, D. (2014). Bond supply and excess bond returns. Review

of Financial Studies, 27(3):663 713.

[13] Gromb, D. and Vayanos, D. (2010). Limits of arbitrage. Annual Review of Financial

Economics, 2 (1):251 275.

[14] He, Z. and Krishnamurthy, A. (2012). A model of capital and crises. Review of

Economic Studies, 79:735 777.

[15] Ivashina, V., Scharfstein, D. S., and Stein, J. C. (2015). Dollar funding and the lending

behavior of global banks. Quarterly Journal of Economics, pages 1241 1281.

[16] Liao, G. (2016). Credit migration and covered interest rate parity. Harvard Business

School - Project on Behavioral Finance and Financial Stability, Working Paper 2016-

07:1 58.

[17] Nakisa, R. C. (2011). A financial bestiary : introducing equity, fixed income, credit,

FX, forwards, futures, options and derivatives. Amersham: Chesham Bois.

135
[18] Rime, D., Schrimpf, A., and Syrstad, O. (2017). Segmented money markets and

covered interest parity arbitrage. BIS Working Paper, 651:1 77.

[19] Sushko, V., Borio, C., McCauley, R., and McGuire, P. (2016). The failure of covered

interest parity: Fx hedging demand and costly balance sheets. BIS Working Paper,

590:1 67.

[20] Vayanos, D. and Vila, J.-L. (2009). A preferred-habitat model of the term structure

of interest rates. National Bureau of Economic Research, Working Paper 15487:1 59.

3.7 Appendix

Table 3.7.1: G-Sibs defined by Financial Stability Board

JP Morgan Chase Bank of China Ind. and Comm. Bank of China Lim.

Bank of America Barclays Mitsubishi UFJ FG

Citigroup BNP Paribas Wells Fargo

Deutsche Bank China Construction Bank Agricultural Bank of China

HSBC Goldman Sachs Bank of New York Mellon

Nordea Standard Chartered Credit Suisse

Royal Bank of Canada State Street Groupe Crédit Agricole

Royal Bank of Scotland Sumitomo Mitsui FG ING Bank

Santander UBS Mizuho FG

Société Générale Morgan Stanley

Source: Financial Stability Board

136