Research in International Business and Finance 49 (2019) 251–268

Contents lists available at ScienceDirect

Research in International Business and Finance

journal homepage: www.elsevier.com/locate/ribaf

Dependence of the “Fragile Five” and “Troubled Ten” emerging

T
market financial systems on US monetary policy and monetary
policy uncertainty☆
Meltem Gulenay Chadwick
Central Bank of the Republic of Turkey, Structural Economic Research Department, Ankara, Turkey

A R T IC LE I N F O ABS TRA CT

JEL classification: In this study, I measure the dependence of the financial markets of certain emerging market
C58 countries on US monetary policy and monetary policy uncertainty. To do so, I apply the time-
E52 varying copula models proposed by Patton (2006). I am particularly interested in the differences
G15 between these markets’ dependence on US monetary policy, i.e. which of the countries’ financial
Keywords: markets tend to comove in response to quantitative easing or quantitative tightening policies. The
Fragile five results are important because financial risks via contagion need monitoring, especially since the
Troubled ten 2008 subprime crisis, although emerging markets were affected less than advanced economies.
Financial vulnerability
My analysis point to a significant difference between these emerging markets regarding their
US monetary policy
dependence on US monetary policy. The correlation persistence parameters, which determine the
Time-varying copulas
evolution of time-varying dependence, show that emerging countries, especially in Latin
America, are strongly dependent on both US monetary policy and monetary policy uncertainty. It
is therefore interesting to see that the results reveal increasing dependence during the subprime
crisis but a decrease after 2009. I conclude that increasing dependence during stress periods
should be considered as a risk factor for policymakers who closely monitor financial markets in
emerging countries.

1. Introduction

“Fragile Five” is a term coined in August of 2013 by a research analyst at Morgan Stanley to represent some emerging market
economies that had become too dependent on unstable foreign investment to finance their growth prospects. The original five
members of the fragile group include Turkey, Brazil, India, South Africa and Indonesia. In 2015, these countries experienced pro-
blems as capital flowed out of emerging markets and into developed countries. After these problems affected other emerging markets
through contagion channels, the “Fragile Five” was expanded by Morgan Stanley analysts to the “Troubled Ten” in mid-2015 in
response to capital moving out of these emerging market countries. The “Troubled Ten” economies were Taiwan, Singapore, Russia,
Thailand, South Korea, Peru, South Africa, Chile, Colombia and Brazil.
Capital flows from most emerging market countries were overwhelmingly affected by the US Fed launching three periods of
quantitative easing, starting from the end of 2008. Its impact on various macroeconomic indicators, such as economic growth,
employment, inflation and especially financial indicators, has attracted considerable academic interest. Yet, little is known about the

☆
The author is grateful to the seminar participants at the IFABS conference 2017. The views expressed herein are those of the authors and do not
necessarily represent the official views ofthe Central Bank of the Republic of Turkey.
E-mail address: meltem.chadwick@tcmb.gov.tr.

https://doi.org/10.1016/j.ribaf.2019.04.002
Received 14 February 2018; Received in revised form 1 April 2019; Accepted 3 April 2019
Available online 10 April 2019
0275-5319/ © 2019 Elsevier B.V. All rights reserved.
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

impact of unconventional US monetary policy implementations on vulnerable emerging market economies and their financial
markets. Quantitative easing, which works through the international portfolio balancing channel, changes capital flows between
countries by adjusting cross-country currencies. It therefore seems reasonable to expect cross correlations between emerging market
currencies and asset markets to be affected by US monetary policy and monetary policy uncertainty. Accordingly, this study aims to
demonstrate the evolution of the dependency of emerging market currencies and asset markets on US monetary policy and monetary
policy uncertainty through particular methods.
In asset pricing theory (APT), multivariate financial returns are often assumed to be normally distributed in order to derive simple
results. In practice, however, this assumption does not hold, especially for high frequency data (weekly and daily).1 Financial returns
are instead found to be leptokurtic and skewed in univariate distributions.2 There is a considerable literature concerning the char-
acteristics of asset returns that suggests that the return generating process might be non-linear and non-normal, and that the returns
might be mutually dependent. This possible dependency does not necessarily imply market efficiency, as long as the return gen-
erating process can be represented as a martingale that satisfies the condition of efficiency. Therefore, it is necessary to find a
different approach to model the multivariate distributions of high frequency asset returns. In addition, modern risk management
requires an understanding of dependence that goes beyond simple linear correlation.3
Considering the non-normality of high frequency asset returns, in this study I will use copula methods to measure the dependency
of emerging market financial systems on US monetary policy movements, represented by Fed policy rates, and US monetary policy
uncertainty, represented by the Merrill Lynch Option Volatility Estimate (MOVE) index. The conventional Pearson correlation may be
inappropriate for measuring dependence across financial markets since it weights positive and negative returns equally, as well as
large and small realizations. It may also underestimate the risk from joint extreme events. I therefore use methods to help me measure
dependence asymmetrically. More specifically, I use daily data to quantify the dependency of the “Fragile Five” and “Troubled Ten”
emerging market countries’ exchange rates and equity markets on US monetary policy and monetary policy uncertainty, from
January 1995 to end of March 2017.
There is a vast literature examining co-movements between different financial markets, or between the same financial variables of
different countries. Although other studies have analyzed the co-movement of Fed policy and financial markets, to my knowledge,
this study is the first to combine co-movement in a tail dependence sense, i.e. using copulas to analyze co-movements between US
monetary policy and emerging financial markets. Recent studies investigating co-movement between global financial markets and US
monetary policy include Aizenman et al. (2016), who find that emerging market countries with robust fundamentals were adversely
affected by tapering news more than the fragile group. Özmen and Yılmaz (2017) report that co-movement between exchange rates
and Fed policy rates, varies in frequency and over time. Kryzanowski et al. (2017) examine the correlations between bond markets,
stock markets and currency forwards, during the three distinct quantitative easing programs launched by the US Federal Reserve.
They find the correlations differ between various asset markets.
This study therefore examines the dependence between the exchange rates and stock markets of specific emerging markets and US
monetary policy. My study differs from previous studies in its method to build up dependence through copulas as I would like to
observe how co-movement evolves over time. In addition, most studies analyzing co-movement between financial markets and US
fundamentals, conduct their evaluations after the 2008 global financial crisis. In contrast, my study covers a very long period that
includes varied US monetary policy regimes. Clearly, the global financial crisis drew attention to spillovers between financial
markets. Yet, emerging markets have always been influenced by US fundamentals through capital flows. Accordingly, my study starts
from 1995.
The first part of this paper examines how emerging market currencies comove with US monetary policy and monetary policy
uncertainty. Specifically, my study aims to reveal whether quantitative easing periods have influenced exchange rates since the 2008
global financial crisis. The second part examines the relationship between US monetary policy, monetary policy uncertainty and
emerging countries’ stock markets by analyzing whether there are significant co-movements with stock markets, which I proxy using
the Morgan Stanley Capital International (MSCI) indices.
My paper proceeds as follows: Section 2 introduces the variables, defines data sources and lists descriptive statistics related to the
dataset. Section 3 discusses the copulas used to measure and quantify the dependency of the relevant emerging countries’ financial
markets on US monetary policy and monetary policy uncertainty. Section 4 presents the results from the models described in the
previous section, discussing whether there are significant dependency patterns between the US monetary policy and emerging fi-
nancial markets. Section 5 concludes.

2. Data

I study the empirical dependence of the following emerging market countries’ financial markets on US monetary policy and
monetary policy uncertainty: Brazil, Chile, Colombia, Indonesia, India, South Korea, Peru, Russia, Singapore, Thailand, Turkey,
Taiwan and South Africa. I choose these countries as a sample because their economies are heavily dependent on capital flows.

1
See Sancetta and Satchell (2001) and Sancetta and Satchell (2004) for details.
2
Mandelbrot (1963) and Fama (1965) are the first to report fundamental divergence from the normality assumption of returns, i.e. they claim that
empirical return distributions are fat-tailed and peaked compared to the normal distribution.
3
Throughout, I use the word ‘dependence’ to represent copulas because there are numerous words in use (e.g. correlation, concordance and co-
dependency). As I do not assume that any dependence measure is ideal, I indicate the advantages and disadvantages whenever relevant.

252
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

Whenever global capital looks for a safe haven, these are the worst affected countries.4 Due to capital moving in and out, their
financial markets become dependent on the monetary policy and monetary policy uncertainty of advanced economies.
I use daily data, i.e. the MSCI stock indices and nominal exchange rates, for the emerging market countries analyzed here. The
daily Federal Reserve effective funds rate is used as a US monetary policy measure and the MOVE index as a proxy for US monetary
policy uncertainty. The latter is derived from a yield curve weighted index of the normalized implied volatility on one-month treasury
options that are weighted on two, five, 10 and 30-year contracts.
Table 1 lists the variables and the abbreviations used throughout the methodology section, the transformations applied to the data
and the definitions. Tables 2a and 2b outline the descriptive statistics of these emerging markets’ exchange rates and stock market
indices. All the nominal currencies I use are against the US dollar and an increase (decrease) means an appreciation (depreciation) in
the emerging economies’ currency. I collect data from Bloomberg from as far back as possible to observe the evolution of dependence
between these emerging markets’ financial variables and US monetary policy. Accordingly, my data starts from the January 1, 1995
and ends in March 2017.5
Close inspection of Tables 2a and 2b shows that average returns are approximately zero in all countries, whereas differences in
standard deviations indicate dispersion in volatility behavior across markets. Likewise, large differences between maximum and
minimum returns indicate that price ranges are highest for Russia and Indonesia, especially compared to other emerging markets.
Kurtosis statistics are mostly high for almost all the returns, which is consistent with the fat tail hypothesis for the return dis-
tributions. The Jarque-Bera statistics indicate that the normality assumption is rejected by any of the returns.

3. Dependence with copulas

A theorem of Sklar (1959) introduced the copula functions that have become popular in financial, macroeconomic and even
agricultural research. According to the theorem, it is possible to decompose an n-dimensional joint distribution function into its n
marginal distributions and a copula, whereby the latter completely describes the dependence amongst the variables. This type of
decomposition allows for more flexibility in constructing multivariate distributions, which explains the major role that copulas have
played in multivariate modelling lately.
Since copulas represent the dependence structure between random variables, they provide a natural way of studying and mea-
suring dependence amongst these variables. In addition, copula-based measures of dependence have the desirable property of in-
variance under strictly increasing transformations.6 In particular, tail dependence, which is considered a copula based measure,
indicates dependence in extreme values. The concept of tail dependence has become an interest to risk management practitioners,
especially after the subprime crisis.
Patton (2006) extended Sklars’ theorem to the conditional case, defining the conditional copula and rendering the dependence
parameter conditional and time-varying. Allowing for time-variation in the conditional dependence among economic time series
seems natural, since time variation in the conditional mean and variance of such series has been widely reported. For this reason, the
extension of Sklars’ theorem to the conditional case has proved very useful. Patton extended Sklars’ theorem as follows:
Theorem (Sklars’ Theorem – two dimensional conditional case) Let F1 be the conditional distribution of X1|W, F2 be the conditional
distribution of X2|W, and H be the joint conditional distributions of (X1X2)|W, where W is the conditioning variable with support Ω. For
exposition purposes, it is assumed that W has dimension 1. Then there exists a conditional copula  such that, for any (x1, x2) ∈ ℝ̄2 and each
w ∈ Ω,
H (x1, x2 |w ) =  (F1 (x1 |w ), F2 (x2 |w )) (1)
If F1 and F2 are continuous, then  is unique; otherwise, it is uniquely determined by RanF1 × RanF2. Conversely, if I let F1 be the conditional
distribution of X1|W, F2 be the conditional distribution of X2|W, and  be the conditional copula, then the function H as defined above is a
conditional bivariate distribution function with conditional marginal distributions F1 and F2.
It is clear from the above theorem that copulas are functions that bind the multivariate distributions to their marginal dis-
tributions. They contain all the information about the joint distribution that is not contained in the marginal distribution; that is, all
the information on the dependence among the variables. Thus, copulas are alternatively called “dependence functions”. According to
Patton, the opposite of Sklars’ theorem is more interesting for multivariate modelling since it implies that I may link any univariate
distribution to any copula and, through this link, define a valid multivariate distribution.
The conditional density function corresponding to the distribution function in Eq. (1) can be easily recovered, provided that F1
and F2 are differentiable and, H and  are twice differentiable:
∂2H (x 1, x2 | w )
h (x1, x2 |w ) ≡ ∂x 1 ∂x2
∂F1 (x 1 | w ) ∂F2 (x2 | w ) ∂2(F1 (x 1 | w ), F2 (x2 | w ) | w )
= ∂x 1
· ∂x · ∂u1 ∂u2
2
= f1 (x1 |w )·f2 (x2 |w )·c (u1, u2 |w ) (2)
where u1 = F1 (x1 |w ) and u2 = F2 (x2 |w ) . This result is particularly useful for the maximum likelihood estimation that I will use for the

4
That is why some of those countries are grouped as the “Fragile Five” and “Troubled Ten”.
5
I use 5535 sample points of daily data. Availability is restricted by the MSCI index of Russia.
6
See Nelsen (2007) for more details.

253
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

Table 1
Definition of variables and data sources.
Variable Transformation Definition Frequency Source

FED-DFF Difference Effective federal funds rate Daily FRED Economic Data
MOVE Log-difference Merrill Lynch Option Volatility Estimate Index Daily Bloomberg
XR-BRL Log-difference Brazilian Real/USD Exchange Rate Daily Bloomberg
XR-CLP Log-difference Chilean Peso/USD Daily Bloomberg
XR-COP Log-difference Colombian Peso Daily Bloomberg
XR-IDR Log-difference Indonesian Rupiah Daily Bloomberg
XR-INR Log-difference Indian Rupee Daily Bloomberg
XR-KRW Log-difference South Korean Won Daily Bloomberg
XR-PEN Log-difference Peruvian Sol Daily Bloomberg
XR-RUB Log-difference Russian Rouble Daily Bloomberg
XR-SGD Log-difference Singapore Dollar Daily Bloomberg
XR-THB Log-difference Thai Baht Daily Bloomberg
XR-TRL Log-difference Turkish Lira Daily Bloomberg
XR-TWD Log-difference Taiwanese Dollar Daily Bloomberg
XR-ZAR Log-difference South African Rand Daily Bloomberg
MSCI-BRL Log-difference MSCI of Brazil Daily Bloomberg
MSCI-CLP Log-difference MSCI of Chile Daily Bloomberg
MSCI-COP Log-difference MSCI of Colombia Daily Bloomberg
MSCI-IDR Log-difference MSCI of Indonesia Daily Bloomberg
MSCI-INR Log-difference MSCI of India Daily Bloomberg
MSCI-KRW Log-difference MSCI of South Korea Daily Bloomberg
MSCI-PEN Log-difference MSCI of Peru Daily Bloomberg
MSCI-RUB Log-difference MSCI of Russia Daily Bloomberg
MSCI-SGD Log-difference MSCI of Singapore Daily Bloomberg
MSCI-THB Log-difference MSCI of Thailand Daily Bloomberg
MSCI-TRL Log-difference MSCI of Turkey Daily Bloomberg
MSCI-TWD Log-difference MSCI of Taiwan Daily Bloomberg
MSCI-ZAR Log-difference MSCI of South Africa Daily Bloomberg

outcome of the copulas.

Patton (2006) allows for time variation in the conditional copula by assuming that the dependence parameter, θc, evolves through
time according to an equation that follows a kind of restricted ARMA(1,10) process, with an autoregressive component, to capture
any persistence in the dependence parameter, and a forcing variable to capture any variation in dependence. The evolution equation
of the dependence parameter can be written as:

θct = Λ(ω + βθct − 1 + αψt )

where Λ is a logistic transformation used to keep the parameter within its interval at all times and ψt is the forcing variable, defined
by Patton (2006) as the mean absolute difference between the transformed marginals u1t and u2t over the previous ten observations.
I would like to discuss several advantages and disadvantages of using copulas in finance. First, a copula is a convenient choice for
studying potentially nonlinear portfolio dependence. Second, copulas allow modelling of joint dependence in a portfolio without
specifying the distribution of its individual assets. Third, a copula is invariant under strictly increasing transformations. On the other
hand, there are two disadvantages of using copulas. First, existing financial models of asset prices are typically expressed in terms of
Pearson correlations. Therefore, if a study uses copulas, which do not have correlation as a parameter, it is difficult to relate the
results to these financial models used in the empirical literature. Second, it is not easy to determine statistically which parametric
copula best fits the data.
In this study, I use the following copulas to analyze the dependence between financial markets of the emerging markets and US
monetary policy decisions7 : the elliptical copula, i.e. normal copula, and two Archimedean copulas, namely Rotated-Gumbel and
Symmetrized Joe-Clayton (SJC) copula.8 The normal copula is the most popular one used in the financial literature due to its easy
computation of the linear correlation coefficient and the dependence structure. In addition to being easily computed, several basic
financial theories are based on linear correlations between different instruments. A drawback regarding the normal copula is that it
describes only symmetric dependence but it has been widely reported in the literature that asymmetries can be expected in financial
returns. By asymmetry in returns I mean that the dependence in the lower tail can be larger than the dependence in the upper tail and
vice versa. For this reason alone, I adopt two asymmetric copulas, i.e. the SJC copula with lower tail dependence, which is different
from upper tail dependence, and the Rotated-Gumbel copula with only lower tail dependence.
When risk distributions are moderately heavy tailed, this condition creates a potentially unbounded downside risk and upside
gain. In an environment of heavy tailed risk distributions, economists and investors often assess the diversification benefits using a
dependence measure, even though this contributes to making the entire financial system more fragile. Therefore, the measure of

7
The copula models are estimated using the Copula Toolbox provided by Andrew Patton and functions written by the authors for Matlab software.
8
Their functional forms and the evolution equations of their dependence parameters following Patton (2006) are described in Appendix A.

254
 M.G. Chadwick

Table 2a
Summary statistics of the daily log returns of exchange rates and US monetary policy variables.
FED-DFF MOVE XR-BRL XR-CLP XR-COP XR-IDR XR-INR XR-KRW XR-PEN XR-RUB XR-SGD XR-THB XR-TRL XR-TWD XR-ZAR

Mean −0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 −0.001 0.000 0.000 −0.001 0.000 0.000
Median 0.000 −0.002 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Maximum 2.800 0.337 0.103 0.042 0.076 0.231 0.035 0.203 0.036 0.278 0.037 0.086 0.083 0.032 0.116
Minimum −2.510 −0.355 −0.087 −0.045 −0.068 −0.200 −0.033 −0.153 −0.030 −0.259 −0.030 −0.178 −0.314 −0.033 −0.155

255
Std. deviation 0.150 0.041 0.010 0.006 0.007 0.015 0.004 0.009 0.003 0.014 0.004 0.006 0.010 0.003 0.011
Skewness 1.078 0.558 −0.165 −0.232 −0.116 −0.327 −0.291 −0.211 0.128 −1.768 0.166 −3.689 −6.036 −0.353 −0.771
Kurtosis 62.631 9.149 15.786 7.693 14.037 70.970 11.906 89.554 20.810 175.078 11.387 136.114 168.115 17.322 18.031
Jarque-Bera 821,131 9008 37,729 5129 28,105 1,065,579 18,369 1,727,782 73,167 6,831,895 16,248 4,099,047 6,321,144 47,422 52,655

p-Value (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Note: “Jarque-Bera” in the first column corresponds to the Jarque-Bera test statistics with p-values in parentheses.
Research in International Business and Finance 49 (2019) 251–268
 M.G. Chadwick

Table 2b
Summary statistics of daily log returns of the MSCI indices.
MSCI-BRL MSCI-CLP MSCI-COP MSCI-IDR MSCI-INR MSCI-KRW MSCI-PEN MSCI-RUB MSCI-SGD MSCI-THB MSCI-TRL MSCI-TWD MSCI-ZAR

Mean 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.000 0.000
Median 0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.000 0.000 0.000 0.000 0.000
Maximum 0.226 0.221 0.165 0.168 0.164 0.117 0.145 0.242 0.110 0.214 0.178 0.092 0.068
Minimum −0.183 −0.116 −0.130 −0.191 −0.155 −0.142 −0.165 −0.310 −0.098 −0.181 −0.197 −0.103 −0.122

256
Std. deviation 0.024 0.013 0.016 0.019 0.016 0.019 0.018 0.030 0.013 0.018 0.024 0.015 0.013
Skewness −0.010 0.358 −0.085 −0.144 −0.278 0.009 −0.189 −0.413 0.017 0.659 0.046 −0.037 −0.336
Kurtosis 10.993 21.661 11.718 12.641 11.098 8.607 10.470 14.716 9.741 15.450 9.191 6.300 7.763
Jarque-Bera 14,736 80,428 17,533 21,454 15,193 7252 12,903 31,816 10,479 36,150 8842 2513 5336

p-Value (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Note: “Jarque-Bera” in the first column corresponds to the Jarque-Bera test statistics with p-values in parentheses.
Research in International Business and Finance 49 (2019) 251–268
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

Table 4
Results of time-varying normal and Rotated-Gumbel Copula (dependence between FED-DFF and XRs).
Normal Copula Rotated-Gumbel Copula

ωN βN αN LL ωRG βRG αRG LL

XR-BRL −0.024 −0.100 0.851 −4.061 0.934 −1.249 0.803 −0.942

(0.000) (0.001) (0.007) (0.009) (0.008) (0.004)
XR-CLP −0.006 −0.084 0.601 −1.425 2.211 −2.406 0.595 −0.613
(0.000) (0.001) (0.012) (0.000) (0.000) (0.000)
XR-COP −0.047 −0.191 −1.990 −4.439 −1.043 1.371 −0.927 −3.363
(0.000) (0.001) (0.000) (0.000) (0.000) (0.000)
XR-IDR −0.042 −0.039 −0.848 −0.744 −0.045 0.632 −1.425 −18.523
(0.001) (0.001) (0.044) (0.001) (0.001) (0.002)
XR-INR −0.016 −0.017 1.253 −1.387 0.963 −0.765 −0.529 −0.503
(0.000) (0.000) (0.011) (0.005) (0.004) (0.004)
XR-KRW −0.071 −0.259 −1.212 −6.951 −0.620 1.029 −1.107 −6.779
(0.000) (0.001) (0.005) (0.000) (0.000) (0.000)
XR-PEN 0.002 −0.016 1.617 −0.755 0.015 0.425 −1.115 −4.394
(0.000) (0.000) (0.013) (0.011) (0.010) (0.003)
XR-RUB 0.003 −0.033 −0.140 −0.155 1.903 −1.608 −0.735 −5.265
(0.000) (0.001) (0.018) (0.000) (0.001) (0.002)
XR-SGD −0.012 −0.018 0.385 −0.231 −1.160 1.417 −0.749 −1.189
(0.000) (0.001) (0.032) (0.012) (0.010) (0.004)
XR-THB −0.086 −0.162 −0.646 −5.580 −0.903 1.208 −0.898 −1.983
(0.000) (0.001) (0.008) (0.012) (0.010) (0.004)
XR-TRL −0.067 −0.039 0.000 −2.945 −0.009 0.013 −0.008 0.038
(0.013) (0.001) (0.013) (0.043) (0.044) (0.018)
XR-TWD −0.001 −0.082 −0.282 −0.695 −0.846 1.202 −1.017 −1.515
(0.000) (0.001) (0.021) (0.009) (0.008) (0.004)
XR-ZAR −0.102 0.058 −1.225 −3.252 −0.131 0.158 −0.078 0.134
(0.001) (0.001) (0.014) (0.065) (0.069) (0.014)

Note: Standard errors are in parentheses. Bold cells are statistically significant at the 5% level. The notation here is as presented in Section 3 and
Appendix A. LL is the log likelihood.

dependence is extremely important from risk management, policy making and broad economic perspectives. Embrechts et al. (2002),
who introduced copulas into risk management, show that the standard Pearson correlation can be dangerously inaccurate as a risk
measure.

4. Results and discussion

I estimate both the static and time-varying versions of the bivariate copula models for the variables reported in Table 1. Results of
the estimations are reported in Tables 4–7, and I select the best copula model based on the maximum log likelihood.9
Table 4 reports the estimates for the Fed effective fund rate and exchange rate return pairs. The evolution of the dependence
between pairs is graphed in Appendix B. The results indicate that there is a significant dependence between US monetary policy and
the exchange rate returns of the emerging market countries that I study.10 Table 4 shows that, for eight countries, the time-varying
normal copula performs better than the asymmetric Rotated-Gumbel copula. For Indonesia, Peru, Russia, Singapore and Taiwan, I
observe asymmetric tail dependence with lower tail dependence but no upper tail dependence.11 Lower tail dependence for these
countries means extreme monetary policy easing (decreasing Fed funds rate), accompanied by local currency depreciations in these
emerging markets.
Table 6 reports the estimates for US monetary policy uncertainty proxied by the MOVE index and exchange rate return pairs. The
evolution of the dependence between pairs is graphed in Appendix D. The results indicate that there is a significant dependence
between US monetary policy uncertainty proxied by the MOVE index and the exchange rate returns of the emerging market countries
that I study. Interestingly, the Rotated-Gumbel copula is a better fit for nearly all pairs in this category. In addition, the findings
suggest that decreasing US monetary policy uncertainty is associated with currency depreciations in most of the emerging countries
studied. I believe that this result is very reasonable, as more capital will flow back into the US as bond volatilities decrease.
The graphs in Appendices B and D show clear periods of change in the dependency between exchange rates and US monetary
policy. Surprisingly, for nearly all countries, this relationship weakened after the 2008 global financial crisis. Conversely, the

9
Since my estimation sample is very long, the information criteria AIC and BIC choose the same models as the log likelihood, so I do not present
the information criteria results here.
10
I compare the static copula AIC and BIC with the time-varying copulas’ AIC and BIC. For every pair, the time-varying copulas perform better, so
I choose not to report the results of the static copulas here.
11
I do not report the SJC copula results and log likelihoods as there was no case in which the SJC copula performs better than the other two time-
varying copulas.

257
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

Table 5
Results of time-varying normal and Rotated-Gumbel Copula (dependence between FED-DFF and MSCIs).
Normal Copula Rotated-Gumbel Copula

ωN βN αN LL ωRG βRG αRG LL

MSCI-BRL −0.010 −0.021 1.776 −6.091 0.425 −0.865 1.037 −4.968

(0.000) (0.000) (0.002) (0.003) (0.002) (0.003)
MSCI-CLP −0.033 −0.095 0.725 −4.573 −1.095 1.389 −0.810 −1.819
(0.000) (0.001) (0.009) (0.009) (0.008) (0.003)
MSCI-COP −0.003 −0.277 −1.091 −6.824 0.396 −0.204 −0.675 −3.132
(0.001) (0.001) (0.011) (0.006) (0.006) (0.004)
MSCI-IDR −0.009 −0.021 1.715 −4.018 −0.176 0.709 −1.369 −10.960
(0.000) (0.000) (0.003) (0.002) (0.002) (0.002)
MSCI-INR −0.064 −0.005 0.000 −2.807 0.083 −0.633 1.269 −15.114
(0.001) (0.001) (0.013) (0.001) (0.001) (0.002)
MSCI-KRW −0.056 −0.059 −1.852 −1.246 −0.100 −0.561 1.560 −34.714
(0.000) (0.000) (0.004) (0.001) (0.000) (0.002)
MSCI-PEN −0.018 −0.024 1.424 −3.071 2.468 −2.381 −0.262 −0.230
(0.000) (0.000) (0.011) (0.007) (0.007) (0.004)
MSCI-RUB −0.023 −0.185 −2.007 −5.273 0.093 0.574 −1.685 −23.272
(0.000) (0.001) (0.000) (0.001) (0.001) (0.002)
MSCI-SGD −0.081 −0.036 −0.624 −2.636 −0.096 −0.552 1.487 −28.760
(0.001) (0.001) (0.033) (0.000) (0.000) (0.001)
MSCI-THB −0.120 −0.138 −0.170 −10.011 0.006 0.611 −1.511 −22.255
(0.001) (0.001) (0.012) (0.001) (0.001) (0.002)
MSCI-TRL −0.067 −0.043 −1.432 −1.128 0.009 −0.619 1.424 −27.923
(0.001) (0.001) (0.010) (0.001) (0.001) (0.002)
MSCI-TWD −0.001 −0.012 1.868 −1.786 −0.124 −0.539 1.547 −16.613
(0.000) (0.000) (0.001) (0.000) (0.000) (0.000)
MSCI-ZAR −0.034 −0.146 −1.848 −2.250 0.160 −0.723 1.482 −4.276
(0.001) (0.001) (0.002) (0.000) (0.000) (0.000)

Note: Standard errors are in parentheses. Bold cells are statistically significant at the 5% level. The notation here is as presented in Section 3 and
Appendix A. LL is the log likelihood.

Table 6
Results of time-varying normal and Rotated-Gumbel Copula (dependence between MOVE and XRs).
Normal Copula Rotated-Gumbel Copula

ωN βN αN LL ωRG βRG αRG LL

XR-BRL −0.002 0.013 1.979 −44.432 0.852 −0.853 0.001 0.063

(0.003) (0.002) (0.025) (0.005) (0.004) (0.004)
XR-CLP −0.039 0.028 1.681 −51.727 0.831 −0.831 0.000 0.320
(0.001) (0.000) (0.006) (0.004) (0.004) (0.002)
XR-COP −0.086 0.112 1.134 −55.206 0.867 −0.867 −0.001 0.199
(0.001) (0.001) (0.012) (0.005) (0.004) (0.003)
XR-IDR −0.253 −0.012 −0.658 −24.586 0.868 −0.868 0.001 0.007
(0.009) (0.001) (0.091) (0.005) (0.005) (0.004)
XR-INR 0.000 0.007 1.993 −16.095 0.831 −0.830 −0.003 0.074
(0.000) (0.000) (0.000) (0.006) (0.005) (0.004)
XR-KRW −0.249 0.176 −1.374 −20.709 0.841 −0.842 0.002 0.011
(0.001) (0.001) (0.008) (0.006) (0.006) (0.011)
XR-PEN −0.002 0.007 1.979 −25.636 0.827 −0.826 −0.004 0.132
(0.000) (0.000) (0.000) (0.005) (0.005) (0.004)
XR-RUB −0.003 0.025 1.921 −25.614 2.234 −2.399 0.394 −0.012
(0.000) (0.000) (0.001) (0.011) (0.010) (0.002)
XR-SGD −0.175 0.378 −1.262 −20.669 0.318 −0.554 0.533 −0.764
(0.001) (0.001) (0.006) (0.006) (0.005) (0.004)
XR-THB −0.004 0.021 1.899 −14.147 0.849 −0.848 −0.002 0.132
(0.000) (0.000) (0.001) (0.007) (0.006) (0.007)
XR-TRL 0.000 0.011 2.001 −46.331 0.844 −0.844 0.000 0.039
(0.000) (0.000) (0.000) (0.005) (0.005) (0.003)
XR-TWD −0.009 0.044 1.744 −13.760 0.796 −0.795 −0.002 0.032
(0.000) (0.000) (0.003) (0.009) (0.006) (0.008)
XR-ZAR −0.023 0.024 1.802 −50.241 0.835 −0.836 0.002 0.125
(0.000) (0.000) (0.003) (0.004) (0.005) (0.007)

Note: Standard errors are in parentheses. Bold cells are statistically significant at the 5% level. The notation here is as presented in Section 3 and
Appendix A. LL is the log likelihood.

258
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

Table 7
Results of time-varying Normal and Rotated-Gumbel Copula (dependence between MOVE and MSCIs).
MSCI-BRL −0.128 0.046 1.164 −77.533 0.827 −0.828 0.002 0.470

(0.002) (0.001) (0.012) (0.004) (0.004) (0.003)

MSCI-CLP −0.295 0.184 −0.380 −64.485 0.844 −0.845 0.000 0.379
(0.001) (0.001) (0.012) (0.004) (0.004) (0.002)
MSCI-COP −0.419 0.247 −2.023 −44.625 0.857 −0.831 −0.067 −0.182
(0.001) (0.001) (0.000) (0.004) (0.005) (0.007)
MSCI-IDR −0.016 0.007 1.734 −10.301 0.762 −0.812 0.141 −0.071
(0.000) (0.000) (0.006) (0.006) (0.006) (0.010)
MSCI-INR −0.011 0.015 1.807 −11.885 0.815 −0.813 −0.005 0.179
(0.000) (0.000) (0.002) (0.005) (0.005) (0.005)
MSCI-KRW −0.015 0.026 1.740 −16.999 0.828 −0.891 0.179 0.288
(0.000) (0.000) (0.002) (0.006) (0.005) (0.006)
MSCI-PEN −0.234 0.134 −0.015 −51.884 0.843 −0.842 −0.002 0.316
(0.003) (0.001) (0.021) (0.004) (0.004) (0.003)
MSCI-RUB −0.291 0.168 −0.385 −57.953 0.864 −0.864 0.000 0.103
(0.003) (0.001) (0.021) (0.005) (0.004) (0.004)
MSCI-SGD −0.007 0.012 1.912 −22.567 0.818 −0.819 0.002 −0.012
(0.000) (0.000) (0.001) (0.005) (0.005) (0.004)
MSCI-THB −0.018 0.011 1.742 −14.689 0.800 −0.800 0.000 0.002
(0.000) (0.000) (0.005) (0.006) (0.005) (0.005)
MSCI-TRL −0.193 0.177 −0.622 −23.013 0.834 −0.833 −0.004 −0.019
(0.001) (0.001) (0.013) (0.005) (0.004) (0.004)
MSCI-TWD −0.200 0.125 −1.391 −12.031 0.740 −0.654 −0.268 −0.294
(0.001) (0.001) (0.008) (0.005) (0.004) (0.003)
MSCI-ZAR −0.401 0.251 −1.401 −58.439 0.860 −0.859 −0.001 0.269
(0.001) (0.001) (0.006) (0.005) (0.004) (0.003)

Note: Standard errors are in parentheses. Bold cells are statistically significant at the 5% level. The notation here is as presented in Section 3 and
Appendix A. LL is the log likelihood.

dependence slightly strengthened towards the end of the sample period. This result is stronger for the co-movements between US
monetary policy uncertainty and exchange rates of the sampled emerging market countries, except for Indonesia. This weakening
dependence is especially obvious for Turkey. This leads me to conclude that most of these emerging market countries, which ex-
perienced different local currency shocks in the 1990s and early 2000s, eventually learned to deal with the shocks and spillovers
caused by advanced economies. Thus, after the 2008 global financial crisis, nearly all countries managed their exchange rate markets
more professionally.
Regarding the dependence between US monetary policy and emerging countries exchange rate markets, my findings are very
similar to those of Özmen and Yılmaz (2017), who report that the relationship between the exchange rate and interest rate differ-
entials vanishes during 2009–2013.12
I next examine the relationship between US monetary policy and monetary policy uncertainty, and the stock market return pairs
of the emerging market countries. Tables 5 and 7 show significant dependence between the stock markets of the countries in question
and US monetary policy. However, this dependence is always symmetrical according to my estimated copula results. Indeed, I obtain
very bad log likelihood results for the Rotated-Gumbel and SJC copulas. The evolution of each pair's dependence is graphed in
Appendices C and E. Appendix C shows that, except for Russia, dependence between emerging country stock markets and US
monetary policy has decreased significantly, especially after the global financial crisis.
Most studies conclude that quantitative easing enabled the US dollar to become the funding currency in large-scale carry trade
activity, with emerging markets as the target currencies. 13 Large capital outflows can disrupt financial markets so that real economic
activity may ultimately deteriorate in emerging markets. However, as observed from the evolution of the dependence between both
US monetary policy and monetary policy uncertainty, the dependence of emerging market stock markets on US monetary policy
weakened after the sub-prime crisis. This have been due to capital controls applied by emerging markets to eliminate the harmful
effects of capital inflows on their countries’ economies.
The weakening dependency between emerging countries’ financial markets and US monetary policy in the aftermath of the global
financial crisis, demands further explanation. After the global financial crisis, as interest rates hit the zero lower bound in advanced
economies, interest rate differentials, though still large, became unimportant. Another way to consider the decreasing dependency
concerns the nature of US monetary policy, which shifted within the same period. Most advanced economies started to implement
quantitative easing policies, which created a tremendous amount of liquidity that spread across the globe, enabling developing
economies to also benefit greatly from this flow of funds. This increased liquidity reversed the dependency in question, which peaked

12
See Özmen and Yılmaz (2017) for more details.
13
Ahmed and Zlate (2014) and Aizenman et al. (2016) claim that quantitative easing and unprecedentedly low US interest rates led to large short-
term capital inflows to a number of emerging markets, which in turn led several of them to impose capital controls, such as Brazil, Indonesia and
South Korea.

259
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

during the global financial crisis, which had caused a liquidity shortage around 2008.
After the global financial crisis, changes to the implementation of US monetary policy created different dependence structures for
each emerging country studied here because it was not only limited to the advanced countries’ quantitative easing policies. Rather, it
may also reflect the outcomes of unconventional monetary policy actions in emerging markets, especially Turkey and Brazil. Several
emerging markets reformed their monetary policy, using new tools to maintain financial stability. They also developed a new space
for macroprudential policies. This may also explain the disappearance of the dependence between the exchange rate and monetary
policy in the US.

5. Concluding remarks

Many emerging countries have been affected by the US Federal Reserve's tapering policy, initiated towards the end of 2012. This
policy caused substantial drops in stock market indices and large exchange rate depreciations through expectations of decreasing
capital inflows and carry-trade activity to emerging markets. Movements within the financial markets of these emerging countries
accordingly responded to the tapering. This response suggests that, in an era of financial globalization, the financial markets of
emerging economies cannot remain insulated from Fed policy stances.
In this study, I analyze the evolution of co-movements between US monetary policy and the financial markets of two groups of
emerging market countries, sometimes known as “Troubled” or “Fragile”. Various studies have already investigated the effect of US
monetary policy on the financial markets of emerging economies, including co-movements between US tapering policies and
emerging markets. However, these studies mostly focus on the post-subprime crisis era. While it is true that the subprime crisis had a
global spillover effect, emerging market countries have long been affected by US interest rate differentials through capital flows and
carry-trade. I therefore conduct my analysis over a longer period of more than a decade to observe how the dependency of each
vulnerable emerging market country to US monetary policy evolves over time.
My results reveal increasing dependency throughout the end of 1990s, especially for far-eastern countries. I also observe Russia's
crisis at the beginning of 2000 and other localized stressful periods for specific groups of emerging markets. Using the persistence
parameter obtained from the time-varying copulas of Patton (2006), I can clearly determine heterogeneity among different emerging
market countries, and differences between exchange rates and stock markets. One surprising result was that many of these emerging
market countries have become much less dependent on US monetary policy or monetary policy uncertainty than during the pre-
subprime crisis period. The factors accounting for the dependency and the persistence of this dependency deserve further exploration,
although I can conclude that the outlook is not all doom and gloom for emerging countries like the “Fragile Five” and “Troubled Ten”.
This research is valuable due to its considerations of different levels of dependence of fragile and/or vulnerable countries on US
monetary policy. I believe that this dependency influences the diversification policies of investors worldwide, thereby ultimately
altering risk premiums in fragile emerging market countries. In addition, if dependency increases then it imposes severe systemic
costs, so the global economy may require a coordinating agency to improve resource allocation. Such policy considerations have not
been considered by previous empirical research on heavy tails in international markets, which provided a further motivation for my
study. What is arguably more important economically is that there are aggregate consequences for elevated asset price and un-
certainty dependence during periods of heavy spillovers. I therefore conclude that it is essential to provide information about the
dependence structure across emerging countries over time.

Appendix A. Copula functions

Normal Copula: The normal copula can be defined as follows:

Φ−1 (u1) Φ−1 (u2) 1 −(r 2 − 2ρrs + s 2) ⎫
N (u1, u2 |ρ) = ∫−∞ ∫−∞ exp ⎧ drds, ρ ∈ (−1, 1)
2π (1 − ρ2 ⎨
⎩ 2(1 − ρ2 ) ⎬
⎭
where the dependence parameter, ρ, is the linear correlation coefficient. Its dynamic equation may be written as:
10
⎛ 1 ⎞
ρt = Λ ωN + βN ρt − 1 + αN ·
⎜ 10
∑ Φ−1 (u1, t − j)·Φ−1 (u2, t − j)⎟
⎝ j=1 ⎠
The normal copula is symmetrical with no tail dependence, i.e. λL = λU = 0. Kendall's tau (τ) can be computed from the correlation
coefficient as τ = (2/π) arcsinρ. Λ(·) is a logistic transformation to keep the parameters within their intervals.
Rotated-Gumbel Copula (Survival Gumbel copula): this copula has the following form:
RG (u1, u2 |θ) = u1 + u2 − 1 + G (1 − u1, 1 − u2 |θ),
where G corresponds to the Gumbel copula. The dependence parameter θ has a dynamic equation that can be written as:
10
⎛ 1 ⎞
θt = Λ ωRG + βRG θt − 1 + αRG·
⎜ 10
∑ |u1,t−j − u2,t−j |⎟
⎝ j=1 ⎠
The Rotated-Gumbel copula has only lower tail dependence, given by λL = 2 −21/θ, while Kendall's τ can be computed as
τ = 1 − θ−1

260
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

Symmetrized Joe-Clayton Copula: this copula, defined by Patton (2006), takes the following form:
1
SJC (u1, u2 |λU , λL) = ·(JC (u1, u2 |λU , λL) + JC (1 − u1, 1 − u2 |λU , λL) + u1 + u2 − 1)
2
where JC is the Joe-Clayton copula, also called the BB7 copula of Joe (1997), which can be presented as:
JC (u1, u2 |λU , λL) = 1 − (1 − {[1 − (1 − u1)κ ]−γ + [1 − (1 − u2)κ ]−γ − 1}−1/ γ )−1/ κ
with κ = 1/log2(2 − λU), γ =−1/log2(λL) and λU, λL ∈ (0, 1)

Appendix B. Dependence of emerging market XRs on FED-DFF

261
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

262
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

Appendix C. Dependence of emerging market MSCIs on FED-DFF

263
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

264
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

Appendix D. Dependence of emerging market XRs on MOVE

265
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

266
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

Appendix E. Dependence of emerging market MSCIs on MOVE

267
M.G. Chadwick Research in International Business and Finance 49 (2019) 251–268

References

Ahmed, S., Zlate, A., 2014. Capital flows to emerging market economies: a brave new world? J. Int. Money Financ. 48 (Part B), 221–248.
Aizenman, J., Binici, M., Hutchisonc, M.M., 2016. The transmission of federal reserve tapering news to emerging financial markets. Int. J. Central Bank. 12 (2), 317–356.
Embrechts, P., McNeil, A., Straumann, D., 2002. Correlation and dependence in risk management: properties and pitfalls. Risk Manage.: Value Risk Beyond 1, 176–223.
Fama, E.F., 1965. The behavior of stock-market prices. J. Bus. 38 (1), 34–105.
Joe, H., 1997. Multivariate Models and Multivariate Dependence Concepts. CRC Press.
Kryzanowski, L., Zhang, J., Zhong, R., 2017. Cross financial market correlations and quantitative easing. Financ. Res. Lett. 20, 13–21.
Mandelbrot, B., 1963. The variation of certain speculative prices. J. Bus. 36 (4), 394–419.
Nelsen, R.B., 2007. An Introduction to Copulas. Springer Science & Business Media.
Özmen, M.U., Yılmaz, E., 2017. Co-movement of exchange rates with interest rate differential, risk premium and fed policy in “fragile economies”. Emerg. Markets Rev. 33,
173–188.
Patton, A.J., 2006. Modelling asymmetric exchange rate dependence. Int. Econ. Rev. 47 (2), 527–556.
Sancetta, A., Satchell, S., 2001. Bernstein approximations to the copula function and portfolio optimization, Technical report. Faculty of Economics, University of Cambridge.
Sancetta, A., Satchell, S., 2004. The bernstein copula and its applications to modeling and approximations of multivariate distributions. Economet. Theor. 20 (3), 535–562.
Sklar, M., 1959. Fonctions de répartition à n dimensions et leurs marges. Université Paris, pp. 8.

268