Journal of International Money and Finance 29 (2010) 743–759

Contents lists available at ScienceDirect

Journal of International Money

and Finance
journal homepage: www.elsevier.com/locate/jimf

Dependence structure between the equity market and the

foreign exchange market–A copula approach
Cathy Ning*
Department of Economics, Ryerson University, 350 Victoria Street, Toronto, ON M5B 2K3, Canada

a b s t r a c t

JEL code: This paper investigates the dependence structure between the
C32 equity market and the foreign exchange market by using copulas.
C51 In particular, several copulas with different dependence structure
C52
are compared and used to directly model the underlying depen-
G15
dence structure. We find that there exists significant symmetric
F30
upper and lower tail dependence between the two financial
Keywords: markets, and the dependence remains significant but weaker after
Copulas the launch of the euro. Our findings have important implications
Tail dependence for both global investment risk management and international
Dependence structure
asset pricing by taking into account joint tail risk.
GARCH
Stock return
 2009 Elsevier Ltd. All rights reserved.
Foreign exchange rate return

1. Introduction

Studying the co-movements across financial markets is an important issue for risk management and
portfolio management. There is a great deal of research focusing on the co-movements of international
equity markets. Chakrabarti and Roll (2002) find that the correlations increased from the pre-crisis to
the crisis period in both Asian and European stock markets. They also find that the diversification
potential was bigger in Asia than in Europe in the pre-crisis period, but this was reversed during the
crisis. Other examples of research on the co-movements of equity markets can be found in Karolyi and
Stulz (1996), Longin and Solnik (2001), Forbes and Rigobon (2002), while the methodology used is
along the line of correlations and conditional correlations. Since the limitations of correlation-based
models as identified in Embrechts et al. (2002), research has started to use copulas to directly model

* Tel.: þ1 416 979 5000x6181; fax: þ1 416 598 5916.

E-mail address: qning@ryerson.ca

0261-5606/$ – see front matter  2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jimonfin.2009.12.002
744 C. Ning / Journal of International Money and Finance 29 (2010) 743–759

the dependence structure across financial markets. Works along this line include Mashal and Zeevi
(2002), Hu (2006) and Chollete et al. (2006), who report asymmetric extreme dependence between
equity returns, i.e., the stock markets crash together but do not boom together. While the above
literature focuses on the dependence structure and co-movements in equity markets via copulas,
Patton (2006a) also employs copulas to model the asymmetric exchange rate dependence and finds
that the mark and yen exchange rates are more correlated when they are depreciating against the US
dollar than when they are appreciating.
While there is extensive literature studying the co-movements between the international equity
markets and some literature on modeling the dependence structure between the exchange rates via
copulas, there is little literature on using copulas to study the co-movements across markets of
different asset types, such as the equity and foreign exchange markets. In this paper, we consider both
equities and foreign exchange rates in our study since the foreign exchange market is by volume one of
the largest financial markets and currency is an important asset in international financial portfolios. In
the literature, Giovannini and Jorion (1989) include exchange rates as assets in their portfolios. For
global investors who wish to diversify portfolios internationally, the co-movements and dependence
structure between equities and exchange rates would have important implications for their cross
market risk management. There has been extensive research (both theoretical and empirical) on the
relationship and co-movements between these two markets. Theoretical research includes the ‘‘flow-
oriented’’ models of exchange rate (see Dornbusch and Fischer (1980)) and the ‘‘stock oriented’’ models
of exchange rate (see Branson (1983) and Frankel (1983)). All these models argue that the stock market
impacts the exchange rate and vice versa. Empirical studies of the interaction or causality relationship
between the stock price and the exchange rate lead to mixed results (positive correlation, negative
correlation, existence or nonexistence of causality, causality one way or the other).
In this paper, we endeavor to investigate the dependence between the equity returns and the
exchange rate returns, by using a relatively new technique: copulas. The methodology we use in this
paper differs in a fundamental way from most of the methods used in the literature in analyzing
dependence between the financial markets, which is also sometimes called co-movement. We will use
dependence or co-movement interchangeably in this paper. The questions we intend to answer are:
what is the dependence structure between these two assets? Is there any extreme value dependence1?
Is the dependence symmetric or asymmetric? By answering these questions, we hope to better
understand the co-movements of stock-currency markets and the risks associated with the depen-
dence structure between markets.
A copula is a function that connects the marginal distributions to restore the joint distribution. The
advantage of using copulas in analyzing the co-movement concerned is multifold. First, copulas allow
us to separately model the marginal behavior and the dependence structure. This property gives us
more options in model specification and estimation. Second, the copula function can provide us not
only the degree of the dependence but also the structure of the dependence. It allows for the tail
dependence and asymmetric dependence. Linear correlation does not give the information about tail
dependence and the symmetry property of the dependence. Third, unlike correlation, copulas do not
require elliptically distributed random variables of the interest. As a result, they are especially useful
when modeling the dependence between asset returns (especially from high frequency data). Finally,
copulas are invariant to increasing and continuous transformations. This property is very useful as
transformation is commonly used in economics and finance. For example, the copula does not change
with returns or logarithm of returns. This is not true for the correlation, which is only invariant under
liner transformations.
To study the stock-foreign exchange dependence, we specify both the marginal models for the
returns and a joint model for the dependence. We employ the AR-t-GARCH models for the marginal

1
There has been growing interest in the research on extreme events and market co-movement, both theoretically and
empirically. In addition to Longin and Solnik (2001), Mashal and Zeevi (2002), Hartmann et al. (2003), Hu (2006), Chollete et al.
(2006), and Cappiello et al. (2006), DeVaries and Jansen (1991) examine the 1987 US stock market crash via the tail probability
of the US stocks. Dungey and Tambakis (2005) analyze international financial contagion during financial market extremes.
Hong et al. (2008) develop a theoretical model based on the communication process between advisors and investors to explain
dot-com bubble. Chollete (2008) proposes a theoretical model and find that extremes are endogenous.
 C. Ning / Journal of International Money and Finance 29 (2010) 743–759 745

distributions of each stock index return and exchange rate return series. For the joint model, we choose
three copulas with deferent dependence structure: the normal copula, which is symmetric but with
zero tail dependence; the student-t copula, which retains the correlation dependence and also has
symmetric nonzero tail dependence; and the SJC copula, which allows for asymmetric tail dependence
and nests symmetry as a special case. We use AIC, BIC, and the ‘hit’ test in Patton (2006a) to check for
the copula model specifications and goodness-of-fit tests.
The financial markets considered are the G5 countries (US, UK, Germany, Japan, France) which
include 5 stock markets and 4 exchange rates. We examine the dependence before and after the launch
of the euro. We find that there exists significant positive tail dependence between the stock market and
the foreign exchange market in each country for both the pre- and post-euro periods. Unlike the co-
movements across international stock markets, the tail dependence is symmetric between the stock
market and the foreign exchange market. This result is also different from tail dependence across
foreign exchange markets or across equity-bond markets. For instance, Hartmann et al. (2003) find
asymmetric extreme co-movements in currency markets in both industrial and emerging countries
using an extreme value theory approach. As well, Cappiello et al. (2006) report asymmetric extreme
dependence across equity-bond markets.
One of our findings is that the equity-currency tail dependence is decreased after the launch of the
euro, especially for the eurozone countries such as Germany. This may be explained by the fact that the
uncertainty in the foreign exchange market is reduced after multiple currencies were replaced by
a common currency, euro. The countries we examined are either in the eurozone (Germany and France)
or have strong trade relationships with the eurozone (the United Kingdom and Japan). Moreover, the
dependence of the German pair and the French pair is actually the dependence between the local stock
market and the euro after the adoption of the euro.
Our finding of significant tail dependence in the equity-currency pairs has important implications in
risk management and asset pricing. First, left tail dependence indicates the potential of a simultaneous
large loss in both the equity and foreign exchange markets. This joint downside risk has been well
documented in the literature of equity market. It is formally discussed and measured in Chollete et al.
(2006), where they find significant downside risk in G5 and Latin American stock markets. Equity-
foreign exchange rate joint downside risk is important to global investors when investing in foreign
stock markets. For example, an extreme market event involving a 20% loss in the London stock market
would, under equity-foreign exchange tail dependence, imply the potential of a large depreciation, for
instance 10%, in the exchange rate. Consequently, a US investor who invested in the London stock
market would lose 28% in US dollars. Thus the existence of lower tail dependence implies a much
higher downside risk in foreign stock market investment than the case of no tail dependence. This may
also partially explain ‘‘investor home bias puzzle’’.
Second, tail dependence allows investors to measure the probability of simultaneous extreme losses.
The likelihood of extreme joint losses suggests a higher than normal value-at-risk (VaR), which is an
important factor for risk management. Ignoring the tail dependence would under estimate the VaR.
Third, tail dependence is extremely important for safety-first agents investing globally. Susmel
(2001) discussed the safety-first criterion and tail dependence in the context of investment in
emerging stock markets. A safety-first investor minimizes the chances of a large loss, a loss that may
drive her or her firm out of business. Equity-foreign exchange tail dependence measures the likelihood
of large loss in foreign market investment. Thus a safety-first investor would especially care the size of
tail dependence.
Finally, this finding should also affect the pricing of assets. Discussed in Poon et al. (2004), tail
dependence, which gives the probability of joint occurrence of the most extreme values, is a true
measure for systematic risk in times of financial crisis. Global investors should be compensated for
exposure to such systematic joint risk in stock-foreign exchange markets during market extremes. In the
literature, this type of joint extreme risk has not been considered in the asset pricing model. We hope
that this work will also improve our understanding of risks associated with the extreme events and our
results will lead to the possible revision of the asset pricing models by picking up the tail dependence.
The remainder of the paper is structured as follows. Section 2 provides a brief review of copula
concepts. Section 3 specifies the models and the estimation. In Section 4, we describe the data and
discuss the results. Section 5 concludes.
746 C. Ning / Journal of International Money and Finance 29 (2010) 743–759

2. The copula concept and measures of dependence

A copula is a multivariate cumulative distribution function whose marginal distribution is uniform

on the interval [0,1]. The importance of the copula is that it can capture the dependence structure of
a multivariate distribution. This is justified by the fundamental fact known as Sklar’s (1959) theorem.
For the purpose of this paper and simplicity, we consider the bivariate case.
Sklar’s Theorem. Let FXY be a joint distribution function with margins FX and FY. Then there exists
a copula C such that for all x, y in R,

FXY ðx; yÞ ¼ CðFX ðxÞ; FY ðyÞÞ: (1)

If FX and FY are continuous, then C is unique; otherwise, C is uniquely determined on RanFX  RanFY.
Conversely, if C is a copula and FX and FY are the cumulative distribution functions (CDFs), then the
function FXY defined by Eq. (1) is a joint distribution function with margins FX and FY.
From Sklar’s theorem, we notice that a joint distribution FXY can be decomposed into its univariate
marginal distributions FX and FY, and a copula C, which captures the dependence structure between the
variables X and Y. As a result, copulas allow us to model the marginal distributions and the dependence
structure of a multivariate random variable separately.
In addition to linear correlations, there are various other measures of dependence, among which
Kendall’s s and Spearman’s r are two scale free measures of rank dependence and are commonly
studied with copula models.
A useful dependence measure defined by copulas is the tail dependence, which measures the
probability that both variables are in their lower or upper joint tails. Intuitively, upper (lower) tail
dependence refers to the relative amount of mass in the upper (lower) quantile of the distribution. An
important property of a copula is that it can capture the tail dependence. Furthermore, the tail
dependence between X and Y, as one of the copula properties, is invariant under strictly increasing
transformation of X and Y. The left (lower) and right (upper) tail dependence coefficients are defined as

Cðu; uÞ
li ¼ limu/0 Pr½FY ðyÞ  ujFX ðxÞ  u ¼ limu/0 ; (2)
u

1  2u þ Cðu; uÞ
lr ¼ limu/1 Pr½FY ðyÞ  ujFX ðxÞ  u ¼ limu/1 ; (3)
1u
where li and lr˛[0, 1]. If li or lr are positive, X and Y are said to be left (lower) or right (upper) tail
dependent. Different copulas usually represent different dependence structures with the association
parameters indicating the strength of the dependence. For example, Gaussian copula has zero tail
dependence while Clayton copula has left tail dependence and no right tail dependence. Further
examination of copulas and measures of dependence can be found in Joe (1997) and Nelson (1999).

3. Model specification and estimation

In order to study the dependence structure between the stock and the exchange rate return series,
we need to specify three models: the models for the marginal distribution of each stock and exchange
rate return series, and the model for the joint distribution of the two series by copula.

3.1. Marginal models

It is well documented in the literature that the daily asset returns show fat-tails and conditional
heteroscedasticity. As usual, the error variance is unknown and must be estimated from the data. The
generalized autoregressive conditional heteroscedasticity (GARCH) model is a common approach to
model time series with conditional heteroscedastic errors. Besides, Bollerslev (1987), among others,
has found that the student’s t distribution fits the univariate distribution of the daily exchange rate
 C. Ning / Journal of International Money and Finance 29 (2010) 743–759 747

returns quite well. Many asset returns also show autoregressive characteristic. As a result, AR(k)-t-
GARCH(p, q) model has been documented to be successful in capturing these stylized facts of asset
returns. This type of model and its variants have been used in Bollerslev (1987) and Patton (2006a). We
adopt this model for our return series. To verify the marginal distributions are indeed not normal, we
use the Jarque-Bera statistic normality tests for each return series. The test results indicate that the
normality is strongly rejected for all returns considered. The order of the autoregressive terms k is
determined by specifying the maximum being 10 and deleting the insignificant (with significant level
of 5%) autoregressive terms. Hence the marginal model can be specified as follows:
X
ri;t ¼ Intercept i þ ARi;k  ri;tk þ 3i;t0 (4)
k

X X
s2i;t ¼ Arch 0i þ GarchðpÞi s2tp þ ArchðqÞi 32i;tq (5)
p q

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
DoF
 3 wiidtDoF
s2i;t  ðDoF  2Þ i;t
where ri,t is the returns for the ith asset at time t.s2i;t is the variance of 3i,t and DoF is the degree of
freedom for the t distribution.
Since the joint copula model is a function of the marginal distributions, its correctness and
usefulness depends on the correct specification and validity of the marginal models. For this reason, we
perform several model misspecification tests, including Lagrange Multiplier (LM) tests and Kolmo-
gorov–Simirnov (K–S) tests employed in Patton (2006a), in order to confirm the empirical adequacy of
the marginal models. Details of the tests are described in Section 4.3.

3.2. Joint copula models

We consider several copula models with deferent tail dependence structure, namely, the Gaussian,
t, SJC, and symmetric-SJC copulas.
The bivariate Gaussian copula has the form as follows:
 
Cðu; y; rÞ ¼ Fr F1 ðuÞ; F1 ðyÞ; r ;

where the variables u and v are the CDFs of the standardized residuals from the marginal models, and
0  u, v  1. Fr is the bivariate normal distribution function with the correlation coefficient r, and F1
is the inverse of the univariate normal distribution function. Gaussian copula has zero tail dependence.
The t-copula is defined as
 
Cy;r ðu; yÞ ¼ ty;r ty1 ðuÞ; ty1 ðyÞ ;

where ty,r is the bivariate student-t distribution with degrees of freedom (DoF) y and the correlation
coefficient r. ty1 is the inverse of the univariate student-t distribution. The advantage of a t-copula is
that it retains the linear correlation measure r and in addition to that, it introduces a parameter, y, that
determines the extent of symmetric extreme dependence.
It is well documented that equity markets crash together but do not boom together, indicating
a positive lower and zero upper tail dependence. Since the literature has no empirical results to guide
us on the dependence structure between the stock market and the foreign exchange market, we
require the copula to be flexible in modeling the tail dependence in both directions, allowing the
asymmetric dependence with the symmetric dependence just a special case. Thus we also choose the
748 C. Ning / Journal of International Money and Finance 29 (2010) 743–759

Symmetrized Joe-Clayton (SJC) copula, which allows both upper and lower tail dependence and
symmetric dependence as a special case.
The SJC copula is a modification of the Joe-Clayton copula of Joe (1997). It is defined as
 
CSJC ðu; yjlr ; li Þ ¼ 0:5 CJC ðu; yjlr ; li Þ þ CJC ð1  u; 1  yjli ; lr Þ þ u þ y  1 (6)

where CJC ðu; yjlr ; li Þ is the Joe-Clayton copula defined as

h !
ig h ig g1 1=k
k k
CJC ðu; yjlr ; li Þ ¼ 1  1 1  ð1  uÞ þ 1  ð1  yÞ 1 (7)

where k ¼ 1/log2(2lr), g ¼ 1/log2(li), and li˛(0,1), lr˛(0,1).

By construction, the SJC copula is symmetric when li ¼ lr. We call this special case a symmetric-SJC
copula here after.
To compare the performance of copula models, we use the information criteria of AIC and BIC.
Moreover, we employ the ‘hit’ test described in the appendix of Patton (2006a) to test for goodness-
of-fit of the competing copula models. The idea of the ‘hit’ test is to compare the number of obser-
vations in each bin of a joint empirical histogram with what would be expected under the null
hypothesis that the true density is well-specified by the copula model. The advantage of this test is
that it can test the overall goodness-of-fit of the models in all regions as well as, in each individual
region of the distribution. In our case, we are particularly interested in the tail regions of the
distribution.

3.3. Estimation

There are usually two approaches to estimate a parametric copula model, namely one stage full
maximum likelihood (ML) and inference for the margins (IFM). The ML approach jointly estimates the
parameters in the marginal models and the parameters of the copula model simultaneously. The IFM
method proposed by Joe and Xu (1996) breaks the estimation into two steps: at the first step, we
estimate the parameters in the marginal distributions; at the second step, given the estimated
marginal parameters, we estimate the copula parameters. Joe (1997) proves that under regular
conditions, the IFM estimator is consistent and asymptotically normal. Next, we give a brief discussion
of the two estimation approaches.
Compared with the ML, the IFM method is less computationally intensive. Moreover, the large
number of parameters in the simultaneous ML estimation could make numerical maximization of the
likelihood function difficult. Since it is computationally easier to obtain the IFM estimator, it is naturally
worthwhile to compare the efficiency of the IFM estimator with the ML estimator. Joe and Xu (1996)
compared the efficiency of the IFM with the ML by simulation. They found that the ratio of the mean
square errors of the IFM estimator to the MLE is close to 1. Joe (1997) also points out that the IFM
method is highly efficient compared with the ML method. Theoretically, ML estimator should be the
most efficient, in that it attains the minimum asymptotic variance bound. However, for the finite
sample, Patton (2006b) found that the IFM was often more efficient than the ML, and in most cases not
less efficient. As a result, IFM is the main estimation method employed in estimating the copula
models.
Since our models involve a large number of parameters, we adopt the IFM method for our esti-
mation as well. We first estimate the marginal AR(k)-t-GARCH(p, q) models by maximum likelihood.
Then we estimate the copula parameters given the estimated parameters in the marginal models.2 To
judge how well our estimator performs in the finite sample, we also conduct a Monte Carlo simulation
study in Section 4.4.

2
The log likelihood function for the SJC copula is available from the author upon request.
 C. Ning / Journal of International Money and Finance 29 (2010) 743–759 749

Table 1
Summary statistics.

Variable Mean Std Dev Skewness Kurtosis

Pre-euro
r_uk 0.0372 0.8955 0.1913 5.6535
r_jp 0.0101 1.3841 0.4959 8.1078
r_gm 0.0342 1.0800 0.8751 14.8288
r_fr 0.0418 1.0561 0.2747 9.1200
r_us 0.0689 0.8116 0.5831 11.1050

r_pdus 0.0075 0.6496 0.2605 6.7721

r_jpus 0.0089 0.7677 1.1723 12.7899
r_gmus 0.0055 0.7013 0.0544 5.2502
r_frus 0.0047 0.6783 0.2261 6.9493

Post-euro
r_uk 0.0004 1.1240 0.2087 5.3564
r_jp 0.0074 1.3778 0.2354 4.5685
r_gm 0.0062 1.2187 0.2586 5.7519
r_fr 0.0095 1.2452 0.1993 5.3587
r_us 0.0003 1.1406 0.0031 5.4948

r_pdus 0.0036 0.5143 0.0322 3.6624

r_jpus 0.0022 0.6320 0.1124 5.3192
r_gmus 0.0077 0.6160 0.0572 3.7483
r_frus 0.0077 0.6159 0.0505 3.7569

r_uk, r_jp, r_gm, r_fr, r_us are returns of stock indices from United Kingdom, Japan, Germany, France and the US respectively.
r_pdus, r_jpus, r_gmus, r_frus are the returns for exchange rates of British pound, Japanese yen, German mark, French franc
against the US dollars respectively.

4. Data and results

4.1. Data

We use daily data from 1/1/1991 to 18/09/2008. The data are from the five largest developed coun-
tries: US, UK, Germany, France and Japan. We choose data from 1991 to 2008 since there have been more
frequent currency and stock market crises since nineties, for example, European Exchange Rate Mech-
anism (ERM) currency crisis of 1992–1993, Mexico economic crisis in 1994–1995, Asian Financial Crisis in
1997, Russian financial crisis in 1998, the dot-com bubble crash in 2000–2001, the 2001–2002 Argentine
economic crisis, and the global financial crisis in 2007–2008. The data end on 18/09/2008 as it is the last
date that the data for German mark and French franc3 exchange rates are available from DataStream.
The stock market index from each country should represent the stock market of that country. We
use the DataStream calculated stock market indices expressed in US dollars from each country. And
foreign exchange rates are expressed in US dollars per local currency. The codes for data from the
DataStream are listed in Appendix A. The returns are calculated as 100 times the logarithm differences
of the indices or the exchange rates between the day t and the day t  1.
We examine our data in two periods: a period before (pre-euro) and a period after (post-euro) the
adoption of the euro. Table 1 gives the summary statistics of the returns. The table shows that all
the means of the returns are small relative to their standard deviations, showing relatively high risks.
The skewness of returns is different from zero with most returns slightly skewing to the left. All returns
show excess Kurtosis ranging from 3.66 to 14.80. And the kurtosis is generally smaller in the post-euro
period, implying a thinner tail of returns. Both the skewness and the excess kurtosis indicate that the
return series are not normally distributed. Our Jarque-Bera test (not presented in the table to save
space) strongly rejects the normality of the returns.

3
The euro was introduced to world financial markets as an accounting currency on January 1, 1999. Since then, German mark
and French franc were replaced by the euro and their exchange rates to the euro were fixed. The exchange rate for the German
mark-USD and French franc-USD are backed from the exchange rate for the euro-USD after January 1 of 1999.
750 C. Ning / Journal of International Money and Finance 29 (2010) 743–759

In Table 2, we present the linear correlations, the Kendall’s tau, and Spearman’s rho rank correla-
tions between the stock and the exchange rate return pairs. We observe that the pair wise correlations
are all significantly positive, indicating that the increase (decrease) of the local stock market is asso-
ciated with the appreciation (depreciation) of the local currency. The Kendall’s Taus for our pairs of
interest are all significantly positive; showing the probability of concordance is significantly higher
than the probability of discordance. The Spearman’s Rhos for the pairs in each country are also
significantly positive, indicating strong rank correlations. The values of Taus and Rhos are consistent
with each other and the linear correlation. The Japanese pair has the strongest dependence, followed
by the German pair or the UK pair, and the weakest is in the French pair. Further, the correlations are
smaller in the post-euro than in the pre-euro period.
In order to see the dependence structure from the data, we calculate an empirical copula table (see
Knight et al. (2005)). To do this, we first rank the pair of return series in ascending order and then we
divide each series evenly into 10 bins. Bin 1 includes the observations with the lowest values and bin 10
includes observations with the highest values. We want to know how the values of one series are
associated with the values of the other series, especially whether lower returns in stock market are
associated with lower returns in foreign exchange market. Thus we count the numbers of observations
that are in cell (i, j). The dependence information we can obtain from the frequency table is that: if the
two series are perfectly positively correlated, we would see most observations lie on the diagonal; if
they are independent, then we would expect that the numbers in each cell are about the same; if the
series are perfectly negatively correlated, most observations should lie on the diagonal connecting the
upper-right corner and the lower-left corner; If there is positive lower tail dependence between the
two series, we would expect that more observations in cell (1,1). If positive upper tail dependence
exists, we would expect large number in cell (10, 10).
We present the dependence table in Table 3. For the pre-euro UK pair, cell(1,1) is 71, which means
out of 2007 observations, there are 71 occurrences when both British pound and UK stock returns lie in
their respective lowest 10th percentiles (1/10th quantile). Cell (10, 10) for pre-euro UK pair is 59,
indicating 59 occurrences when both series are in their highest 10th percentiles (9/10th quantile)
respectively. Numbers in other cells of the pre-euro UK pair are much smaller than those in these two
cells. This is the evidence of both upper and lower tail dependence. Comparing cell (1,1) and cell (10,10)
of all pairs, we note obvious evidence of both upper and lower tail dependence. And the dependence
seems symmetric. If we compare the joint frequencies for the same pair in the two periods, we find that
the joint extreme frequencies, relative to the total sample size, decrease for all pairs. This change seems
mostly serious for the German pair, from pre-euro of 66 to post-euro of 44 in cell (1,1), and from pre-
euro of 70 to post-euro of 49 in cell (10,10). Thus there is some evidence of reduced tail dependence
after the launch of the euro.
We then plot the dates of the top and the bottom 1/10th joint quantiles of returns in Fig. 1 in order to
see when most of the joint extremes occurred in our sample and whether they are related to extreme
events in financial markets. Clearly, for the European pairs, the dates of extremes are crowded during
91–93, 00–01, 08, which correspond to the periods for ERM currency crisis, dot-com bubble, and the

Table 2
Correlation measures.

Pairs Pearson correlation Kendall’s Tau Spearman’s Rho

Pre-euro
UK pair 0.3945 0.2245 0.3241
Japanese pair 0.4715 0.3172 0.4536
German pair 0.4157 0.2683 0.3856
French pair 0.2906 0.1641 0.2388

Post-euro
UK pair 0.2526 0.1764 0.2568
Japanese pair 0.3649 0.2522 0.3623
German pair 0.2123 0.1533 0.2223
French pair 0.2112 0.1500 0.2184
Euro pair 0.2362 0.1839 0.2667

This table gives different correlation measures for each stock-exchange rate return pair.
 C. Ning / Journal of International Money and Finance 29 (2010) 743–759 751

Table 3
Empirical copula for stock returns and exchange rate returns.

Pre-euro Post-euro
UK pair 71 23 17 19 15 16 9 13 9 8 56 37 24 21 25 14 21 27 16 9
40 22 29 23 18 21 13 15 15 5 49 33 31 21 25 15 18 19 22 17
21 23 27 26 26 22 19 16 14 7 31 33 26 28 22 23 24 18 26 19
6 22 26 31 20 22 18 21 16 18 28 32 32 28 20 35 21 24 14 16
16 28 16 19 29 20 12 26 20 15 20 19 31 35 31 41 20 25 18 10
13 17 21 15 25 20 27 23 21 19 13 24 24 32 29 28 38 19 20 24
12 16 19 19 23 25 34 19 17 16 13 24 21 21 26 30 35 28 25 27
6 18 17 17 18 23 22 26 33 21 9 16 22 27 25 24 28 36 30 33
9 15 17 21 17 16 21 28 24 33 15 12 21 21 15 22 26 24 43 41
6 17 12 10 10 16 25 14 32 59 16 20 18 16 22 19 19 30 36 54
German pair 66 26 21 15 17 16 12 11 7 9 48 36 28 23 21 22 15 20 17 20
47 34 24 20 19 18 12 9 9 9 43 28 23 25 32 21 21 17 24 16
20 30 35 25 21 23 16 11 9 11 33 31 29 28 23 25 18 22 22 19
11 19 27 31 21 21 26 17 22 5 27 35 33 33 22 30 20 13 17 20
12 16 20 27 29 19 23 18 22 15 20 24 25 31 32 39 24 22 16 17
14 20 19 20 21 30 17 26 17 17 16 28 28 27 29 13 36 33 21 20
13 20 22 12 16 17 26 27 30 17 15 19 13 28 25 30 40 35 21 24
5 9 17 22 21 23 32 19 29 24 14 20 25 22 28 25 25 32 31 28
8 11 11 18 21 23 16 32 37 24 15 13 24 16 24 20 26 30 45 37
4 16 5 10 15 11 20 31 19 70 19 16 22 17 14 26 25 26 36 49
French pair 59 27 19 16 7 15 18 12 13 14 46 31 34 22 22 24 16 15 21 19
31 41 17 16 19 11 19 17 15 15 48 29 22 25 26 17 24 22 22 15
21 20 24 20 30 22 19 17 15 13 32 34 28 32 27 28 16 18 17 18
16 23 20 17 26 26 23 18 19 12 24 32 25 26 32 28 20 25 20 18
14 17 25 17 24 20 21 25 22 16 22 21 28 29 24 29 29 22 26 20
19 13 19 28 21 26 22 19 20 14 22 28 21 32 30 21 29 33 21 14
10 14 23 24 21 22 21 20 25 20 14 18 20 32 25 26 32 31 28 24
12 22 20 25 19 19 25 22 17 20 10 22 26 18 25 33 33 30 25 28
8 12 21 21 21 19 18 27 27 27 11 18 27 20 19 29 28 34 27 37
10 12 13 16 13 21 14 24 28 50 21 17 19 14 20 16 23 20 43 57
Japanese pair 66 41 25 18 17 11 8 6 8 0 65 37 30 27 21 15 16 16 11 12
37 27 28 27 19 18 16 13 10 6 50 45 21 28 21 20 17 13 21 14
24 25 25 26 32 25 13 16 12 3 38 35 34 35 25 17 17 14 17 18
22 32 29 25 20 16 16 15 13 12 21 22 36 43 27 27 21 21 18 14
15 19 21 20 25 28 33 19 9 12 19 30 18 26 38 27 31 25 23 13
7 17 17 24 20 22 32 22 29 11 12 23 28 27 36 33 37 27 19 9
8 14 16 21 22 21 33 25 23 17 13 25 25 22 22 29 35 35 23 21
3 10 14 18 17 26 20 31 32 30 12 10 20 21 20 25 26 36 43 37
10 7 16 13 18 23 19 32 24 39 13 12 24 14 25 29 29 34 33 37
8 9 10 8 11 11 10 22 41 71 7 11 14 7 15 29 21 29 42 75

Total observations are 2501. In each table, the ranks for the exchange rate returns are on the horizontal axis in an ascending
order while the ranks for the stock returns are on the vertical axis in an ascending order.

ongoing global financial crisis. There are a few extremes around the 97 Asian crisis but not as often as
during the other three crises. For the Japanese pair, the joint extremes occurred most often around 97–
98 Asian financial crisis, 00–01 dot-com bubble, and then the 08 global financial crisis, some around
the ERM crisis but much less frequent than the European pairs. Thus the joint extreme returns are
coincidental with financial extreme events.

4.2. Results of the marginal models

We first estimate the marginal models: the AR(k)-t-GARCH(p, q) type models for each asset return
series. k is set to a maximum of 10, and the insignificant (with significant level of 5%) autoregressive
terms are deleted. We experiment on GARCH terms up to p ¼ 2 and q ¼ 2. The estimates of the marginal
models are presented in Table 4. We test the normality of the error term in the AR equation and the null
hypothesis of normality is strongly rejected for all series with the p-values of the Jarque-Bera statistic
being less than 0.0001 (not listed in table to save space).
752 C. Ning / Journal of International Money and Finance 29 (2010) 743–759

UK pair German pair

a b

J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J-
91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08

French pair Japanese pair

c d

J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J- J-
91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08

Fig. 1. The dates of joint lowest and highest 1/10th quantile of returns.

The marginal models for the pre-euro period are reported in Panel A of Table 4. There is evidence of
a long memory in most returns with significant AR(1)–AR(10) terms. For the stocks, GARCH(1,1) is
generally able to capture the conditional heteroscedasticity. For the exchange rates, higher GARCH
terms are required to better model the conditional heteroscedasticity. This is reflected in the signifi-
cance of the GARCH2 term for the British pound, the German mark and the French franc. The degrees of
freedom of the t distribution are all small, ranging from 3.9 to 7.8, indicating that the error terms are not
normal.
The marginal models for the post-euro period are presented in panel B of Table 4. First, the stock
returns generally show a longer memory with a significant AR(5) in common (except for the Japanese
stock returns) while the currencies generally have a shorter memory with a significant AR(1) term in
common. Second, GARCH(1,1) is able to capture the conditional heteroscedasticity for all post-euro
returns. Third, the degrees of freedom of t distribution are relatively larger than the pre-euro period,
ranging from 6.5 to 14.2, again showing the evidence of thinner tails in the post-euro period than in the
pre-euro period.

4.3. Goodness-of-fit tests for marginal models

As noted earlier, the joint copula model requires the correct specification of the marginal distri-
butions. If the marginal distributions are not correct, their probability transforms will not be i.i.d.
uniform (0, 1), and hence the copula model will be mis-specified. Patton (2006a) employs Lagrange
Multiplier (LM) tests to test for serial independence of the probability transforms, and then Kolmo-
gorov–Simirnov (K–S) tests to test if they are uniform (0, 1). We use these tests for goodness-of-fit test
of our marginal distributions. Specifically, the LM independence tests are applied to the first four
 C. Ning / Journal of International Money and Finance 29 (2010) 743–759 753

Table 4
Estimation of marginal models.

Int. AR1 AR5 AR6 AR10 Arch0 Arch1 Arch2 Garch1 Garch2 TDFI DoF
Panel A: Pre-euro
Stocks
US 0.0733 0.0507 0.0475 0.0049 0.0453 0.9481 0.1892 5.3
(0.0138) (0.0219) (0.021) (0.0021) (0.0089) (0.0101) (0.0217)
UK 0.0551 0.0478 0.0057 0.0315 0.9623 0.128 7.8
(0.0175) (0.0222) (0.0032) (0.0070) (0.0089) (0.0188)
German 0.0625 0.046 0.019 0.0656 0.9177 0.1525 6.6
(0.0189) (0.0228) (0.0064) (0.0119) (0.0141) (0.0151)
French 0.0519 0.0413 0.0638 0.8974 0.1332 7.5
(0.0204) (0.0142) (0.0134) (0.0223) (0.016)
Japanese 0.0191 0.0669 0.0298 0.0842 0.9037 0.1813 5.5
(0.0233) (0.0223) (0.0098) (0.0137) (0.0146) (0.0222)

FX
Pound 0.0074 0.0466 0.0467 0.0049 0.0621 0.9337 0.2548 3.9
(0.0109) (0.0212) (0.0188) (0.0020) (0.0133) (0.0126) (0.0286)
Mark 0.0036 0.0493 0.0061 0.0486 0.9417 0.2012 5.0
(0.0132) (0.02) (0.0029) (0.0106) (0.0126) (0.0266)
Franc 0.0053 0.0056 0.0383 0.0335 0.9202 0.2202 4.5
(0.0121) (0.0029) (0.0109) (0.010) (0.0164) (0.0245)
Yen 0.0222 0.0472 0.0445 0.0111 0.0424 0.9395 0.2478 4.0
(0.0131) (0.0203) (0.0198) (0.0041) (0.0104) (0.0143) (0.0226)

Int. AR1 AR3 AR5 AR6 AR8 Arch0 Arch1 Garch1 TDFI DoF
Panel B: Post-euro
Stocks
US 0.0427 0.0372 0.0369 0.0052 (0.0664) 0.9329 0.1175 10.3
(0.0157) (0.0219) (0.0201) (0.0024) (0.0098) (0.0094) (0.0169)
UK 0.0481 0.0434 0.0516 0.0438 0.0224 0.1030 0.8812 0.0728 8.5
(0.0157) (0.0208) (0.0210) (0.0204) (0.0062) (0.0134) (0.0151) (0.0182)
German 0.0629 0.0433 0.0400 0.0519 0.0217 0.0851 0.9018 0.0973 13.7
(0.0207) (0.0212) (0.0206) (0.0199) (0.0070) (0.0122) (0.0143) (0.0162)
French 0.0663 0.0464 0.0512 0.0213 0.0866 0.9006 0.0705 14.2
(0.0181) (0.0210) (0.0203) (0.0068) (0.0116) (0.0135) (0.0155)
Japanese 0.0250 0.0500 0.0663 0.9083 0.0935 10.7
(0.0247) (0.0166) (0.0121) (0.0171) (0.0167)

FX
Pound 0.0077 0.0042 0.0292 0.9554 0.0854 11.7
(0.0098) (0.0022) (0.0078) (0.0139) (0.0221)
Mark 0.0149 0.0511 0.0010 0.0263 0.9717 0.0786 12.7
(0.0106) (0.0211) (0.0007) (0.0057) (0.0060) (0.0226)
Franc 0.0149 0.0511 0.0010 0.0264 0.9715 0.0799 12.5
(0.0106) (0.0210) (0.0007) (0.0057) (0.0061) (0.0227)
Yen 0.0091 0.0558 0.0063 0.0444 0.9413 0.1541 6.5
(0.0106) (0.0200) (0.0026) (0.0092) (0.0126) (0.0197)

‘‘Int.’’ and ‘‘FX’’ refer to the intercept and foreign exchange rates, respectively. Numbers in parentheses are standard errors. TDFI
is the inverse of the degrees of freedom of T distributions. DoF is the degrees of freedom of T distributions.

moments of uex and ystock, defined as the probability transforms of the CDFs of the standardized
marginal distributions of the two returns. For this, we regress ðuex;t  uex Þk and ðystock;t  ystock Þk on 10
lags of both variables, for k ¼ 1,2,3,4. The test statistic (T  20)R2 from each regression follows an
asymptotic c220 distribution under the null hypothesis.
Table 5 presents the results for the LM and K–S tests. The p-values from the LM tests are from 5% to
90%, with most of them much higher than 5%, implying that we cannot reject the null hypothesis that
the marginal CDFs are serially independent. The p-values from the K–S tests are from 13% to 96%,
indicating the hypothesis that each of the probability transforms of the two marginal distributions is
uniform (0, 1) is not rejected. Thus our marginal models pass the LM and K–S tests at 5% level. These
results provide significant evidence that our marginal models are not mis-specified. Hence our copula
model can correctly capture the dependence structure of the two returns.
754 C. Ning / Journal of International Money and Finance 29 (2010) 743–759

4.4. Results of the joint copula models

Table 6 reports parameter estimates for the Gaussian, Student-t, SJC, and symmetric-SJC copulas. For
all pairs, the dependence parameters, i.e., the correlation coefficient r in both Gaussian and t-copulas,
the degree of freedom (DoF) v in the t-copula and tail dependence parameters li and lr in the SJC and
symmetric-SJC copulas are positive and strongly significant. r from the Gaussian or t-copulas is close to
the linear correlation coefficient from the data. This is not a surprise since both Gaussian and t-copulas
belong to elliptical copula family, and the coefficient r in these two copulas is just the usual linear
correlation coefficient given the elliptical margins (t distributions in our case). The DoFs of the t-copula
are from 6 to 13, indicating substantial extreme co-movements and tail dependence in each pair. The
significance of the DoFs in t-copula also implies that Gaussian copula which does not allow for tail
dependence, is not sufficient in modeling the dependence of the stock-currency pairs. The estimated
tail dependence coefficients li and lr indicate that the Japanese pair has the highest tail dependence for
both pre- and post-euro periods. The German pair ranks the second in terms of tail dependence, fol-
lowed by the UK pair and the French pair for the pre-euro period. For the post-euro period, the UK pair
is in the second place, followed by the tied German pair and French pair.
Since li(lr) measures the dependence between the stock returns and exchange rate returns when
both of them are in extremely small (large) values, the significance of li(lr) means that the stock market
crashes (booms) when the local currency depreciates (appreciates) heavily against US dollars and vice
versa. This finding is consistent with the basic international finance theory. When a country’s stock
market is booming, investors believe that it is a good place for investment; therefore they will purchase
that country’s currency to buy stocks there. Hence the demand of the currency increases, which leads
to the appreciation of the currency. This phenomenon happens in extreme cases as investors are more
sensitive to the extreme events.
Comparing the dependence before and after the launch of euro, we find that both the linear and tail
dependence in all pairs are smaller, especially for the German pair. We explain this by the alleviation of
exchange rate risk caused by the replacement of a common currency, euro, to 16 individual currencies.
A common currency reduces the uncertainty in the exchange rate market with many currencies.
Moreover, for the post-euro period, the dependence between the German stock and the German mark
is actually the dependence between the German stock and the euro. Since the euro is a currency for the
16 eurozone countries, its co-movement with one of the 16 stock markets, the German stock market,
would be weaker than German mark’s co-movement with the German stock market in the pre-euro

Table 5
Goodness-of-fit test for marginal distributions.

Copula Margins 1st moment LM test 2nd moment LM test 3rd moment LM test 4th moment LM test K–S test
Pre-euro
UK stock 0.20 0.24 0.25 0.28 0.96
British Pound 0.65 0.75 0.84 0.89 0.39
Japanese stock 0.34 0.25 0.17 0.12 0.93
Japanese yen 0.05 0.17 0.25 0.23 0.92
German stock 0.50 0.39 0.20 0.08 0.55
German Mark 0.05 0.10 0.12 0.08 0.83
French stock 0.62 0.82 0.89 0.90 0.40
French Franc 0.07 0.11 0.10 0.07 0.47

Post-euro
UK stock 0.28 0.09 0.19 0.06 0.23
British Pound 0.13 0.14 0.22 0.23 0.56
Japanese stock 0.09 0.20 0.12 0.08 0.92
Japanese yen 0.05 0.07 0.07 0.07 0.57
German stock 0.15 0.09 0.08 0.05 0.13
German Mark 0.20 0.12 0.08 0.08 0.17
French stock 0.77 0.43 0.10 0.09 0.19
French Franc 0.10 0.14 0.19 0.15 0.18

Numbers in the table are p-values of LM tests and K–S tests, respectively.
 C. Ning / Journal of International Money and Finance 29 (2010) 743–759 755

Table 6
Estimation of copula parameters and tail dependence.

Pairs Parameters Gaussian copula t-copula SJC copula Symmetric-SJC

Pre-euro period
UK r 0.3332 (0.0189) 0.3307 (0.0217)
n 6.0908 (1.0237)
ll 0.1616(0.0286) 0.1667(0.0168)
lr 0.1722(0.0295)
AIC 232.2784 L274.6233 270.3939 272.3436
BIC 226.6790 263.4245 259.1951 266.7442
German r 0.4058(0.0174) 0.4107 (0.0189)
n 7.0027(1.2358)
ll 0.1879(0.0304) 0.2259(0.0167)
lr 0.2648(0.0280)
AIC 356.0020 L396.8470 387.8573 387.2324
BIC 350.4026 L385.6482 376.6585 381.6330
French r 0.2470(0.0205) 0.2458 (0.0238)
n 6.0915 (1.0390)
ll 0.1287(0.0269) 0.1015(0.0162)
lr 0.0727(0.0275)
AIC 122.6694 L164.3350 153.9687 154.5041
BIC 117.0700 L153.1362 142.7699 148.9047
Japanese r 0.4703(0.0158) 0.4712 (0.0167)
n 13.2372(3.7919)
ll 0.2893(0.0279) 0.2662(0.0161)
lr 0.2449(0.0278)
AIC 497.1574 L507.6164 489.6678 490.7316
BIC 491.5580 L496.4176 478.4690 485.1322

Post-euro period
UK r 0.2873(0.0178) 0.2885 (0.0190)
n 10.4941(2.4935)
lL 0.0933(0.0266) 0.1187(0.0149)
lU 0.1434(0.0262)
AIC 211.3460 L229.1370 224.1213 224.8166
BIC 205.5215 217.4881 212.4724 L218.9921
German r 0.2519(0.0182) 0.2550 (0.0199)
n 9.2131 (2.0185)
ll 0.0833(0.0237) 0.0940(0.0142)
lr 0.1062(0.0259)
AIC 161.9898 L187.6447 179.6383 181.3259
BIC 156.1654 L175.9958 167.9895 L175.5015
French r 0.2604(0.0181) 0.2600 (0.0177)
n 10.9831(2.4913)
ll 0.0753(0.0231) 0.0960(0.0143)
lr 0.1199(0.0261)
AIC 173.6533 L191.8429 187.5119 188.3297
BIC 167.8289 180.1940 175.8630 L182.5053

Japanese r 0.3734(0.0162) 0.3782 (0.0179)

n 10.3372(2.4252)
ll 0.1619(0.0279) 0.1857(0.0152)
lr 0.2086(0.0262)
AIC 373.5385 L394.5947 374.7026 375.6061
BIC 367.7141 L382.9458 363.0537 369.7816

This table gives the estimates of the copula parameters, AIC and BIC in each copula. For the symmetric-SJC copula, only one tail
dependence parameter is reported since ll ¼ lr. Numbers in parentheses are standard errors.

period. This explanation also applies to the French pair. For the UK pair, though the United Kingdom is
not in the eurozone, many high-street banks have reported that 90% of their international trade is
conducted in euro. Thus the function of the British pound in the international trade is less important
after the adoption of euro. Therefore its co-movement with the UK stock market is weaker. Similar
756 C. Ning / Journal of International Money and Finance 29 (2010) 743–759

explanations apply to the Japanese pair. The EU (especially Germany and the United Kingdom) is the
third largest trade partner of Japan. Many trades between Japan and the EU have been conducted in
euro after the adoption of the euro.
Since the values of li and lr are very similar in the SJC copula, we next test whether the tail
dependence is symmetric. We use a likelihood ratio test to test for the null hypothesis: li ¼ lr. The test
result is presented in Table 7. All the p-values from the test are greater than 0.10. Therefore we cannot
reject the null hypothesis that the upper and lower tail dependence are the same. This means that the
dependence between the stock market and the foreign exchange market is the same at the time of
crashing and booming. This finding is different from the finding for the dependence structure between
stock markets documented in the literature: stock markets are more dependent at the time of crashing
than booming.
As we find that the dependence is symmetric, the t-copula and symmetric-SJC copula should be
more appropriate for the modeling. We provide more discussion of this in the next section.
To judge how well the parametric estimator performs in the finite samples, we employ a Monte
Carlo simulation study with a sample size of 250, 500, 1000, 2000, 3000 respectively, corresponding to
the data size of about 1, 2, 4, 8, and 12 years, and the number of replications of 1000. The true
parameters are the estimated values for the UK pair in the pre-euro period. Table 8 reports the result.
Overall, the bias and root mean square error (RMSE), which is a measure accounting for both bias and
variance, decrease with the increase of the sample size. For some cases when the bias is already
extremely small for sample size of 2000, the bias for sample size of 3000 is slightly larger than the
former. For the Gaussian, SJC, and symmetric-SJC copulas, the estimates for all sample sizes in the
simulations are acceptable. However, for the t-copula, a sample size of at least 1000 is required for an
acceptable estimation. 2000 observations are similar to the sample size in our empirical work: 2007 for
the pre-euro and 2051 for the post-euro period. The bias for the r, li and lr with sample size of 2000 are
very small ranging from 0 to 0.0015, while the bias for n is 0.212 given the true value of n being 6.09. The
RMSE for sample size of 2000 is also very small relative to their true values, and they are similar to
the asymptotic standard errors in our estimation. Thus our sample size is reasonably large for the
estimation.
Our main finding is the existence of extreme co-movements (tail dependence) between the stock
market and the exchange rate market. The extreme co-movements are symmetric, implying both
markets boom and crash together. The dependence decreased after the launch of the euro. This finding
improves our understanding of the market dependence. The significance of the tail dependence
implies that the stock market and exchange rate tend to experience concurrent extreme shocks. This
has important implications for risk management across these two markets during extreme events as
the significant tail dependence implies higher than normal joint risk, hence a higher VaR, a common
measure of risk in risk management. Furthermore, the finding is also important for international asset
pricing since the exposure to the joint tail risk should be compensated and thus should be included in
pricing international assets.

Table 7
Likelihood ratio test for symmetric tail dependence.

Pairs p-value of likelihood

ratio test
Pre-euro
UK pair 0.82
German pair 0.11
French pair 0.23
Japanese pair 0.33

Post-euro
UK pair 0.11
German pair 0.43
French pair 0.12
Japanese pair 0.14

This table gives p-values of likelihood ratio tests. A value of over 0.05 indicates that the null of symmetric tail dependence is not
rejected at a 5% level.
 C. Ning / Journal of International Money and Finance 29 (2010) 743–759 757

Table 8
Simulations.

Copulas T ¼ 250 T ¼ 500 Sample Size T ¼ 3000

T ¼ 1000 T ¼ 2000
Gaussian copula True value r ¼ 0.3332
_
Mean of r 0.3312 0.3344 0.3336 0.3331 0.3327
Standard error 0.0553 0.0374 0.0261 0.0201 0.0157
Bias 0.0020 0.0012 0.0004 0.0001 0.0005
RMSE 0.0553 0.0374 0.0261 0.0201 0.0157

t-copula True value r ¼ 0.3307

n ¼ 6.0908
_
Mean of r 0.3276 0.3322 0.3314 0.3310 0.3302
_
Mean of y 10.5285 7.0961 6.5957 6.3028 6.2007
Standard error 0.0639 0.0431 0.0303 0.0228 0.0178
16.5529 4.4159 2.0611 1.1532 0.9144
Bias 0.0031 0.0015 0.0007 0.0003 0.0005
4.4377 1.0053 0.5049 0.2120 0.1099
RMSE 0.0639 0.0431 0.0303 0.0228 0.0178
17.1374 4.5289 2.1220 1.1725 0.9209

SJC-copula True value ll ¼ 0.1616

_
lr ¼ 0.1722
ll
Mean of _ 0.1660 0.1644 0.1632 0.1631 0.1624
Mean of l r 0.1718 0.1712 0.1720 0.1722 0.1729
Standard error 0.0805 0.0562 0.0406 0.0285 0.0243
0.0781 0.0575 0.0392 0.0289 0.0231
Bias 0.0044 0.0028 0.0016 0.0015 0.0008
0.0004 0.0010 0.0002 0.0000 0.0007
RMSE 0.0806 0.0563 0.0406 0.0285 0.0243
0.0781 0.0575 0.0392 0.0289 0.0231

Sym-SJC-copula True value

_
of l ¼ 0.1667
Mean of l 0.1746 0.1701 0.1680 0.1665 0.1678
Standard error 0.0402 0.0307 0.0235 0.0165 0.0136
Bias 0.0079 0.0034 0.0013 0.0002 0.0011
RMSE 0.0410 0.0309 0.0236 0.0165 0.0136

This table evaluates the performance of parameter estimates using simulations of various sample sizes.

4.5. Goodness-of-fit tests for copula models

We use AIC, BIC, and the ‘hit’ test in Patton (2006a) to test the four copula models estimated. The AIC
and BIC is reported in Table 6 and the hit test is presented in Table 9.
The AIC and BIC from the t, SJC, and symmetric-SJC are very close, and are much smaller than those
from the Gaussian copula. Thus copulas with tail dependence fit the data better than the Gaussian
copula which does not allow for tail dependence. For the two nested copulas, the symmetric-SJC copula
performs slightly better than the SJC copula in almost all cases. Thus the copula with the symmetric tail
dependence has a better fit than the copula with asymmetric tail dependence. For the two copulas with
symmetric tail dependence, the t-copula performs better than the symmetric-SJC copula in terms of
AIC, but the result is mixed in terms of BIC. Thus the goodness-of-fit of these two copulas are not
distinguishable in our data. This is consistent with our finding of symmetric tail dependence in the
previous section.
In Table 9, we present the result of the joint ‘hit’ test that the models are well-specified in all regions
simultaneously and the individual hit test that the models are well-specified in the joint lower and
upper 10% regions. The test results on other individual regions are available on request. The t, SJC, and
symmetric-SJC copulas pass the joint ‘hit’ tests in all regions and the individual ‘hit’ test in each region
for all pairs at 5% level. The Gaussian copula is rejected by the ‘hit’ test in some regions for some pairs.
From the p-values of the ‘hit’ test, it seems that the symmetric-SJC copula performs better than the SJC
copula. Again the evidence for the symmetric-SJC and t-copulas is mixed. This indicates again that
758 C. Ning / Journal of International Money and Finance 29 (2010) 743–759

Table 9
Hit test.

Pairs Gaussian copula t-copula SJC copula Sym-SJC Gaussian copula t-copula SJC copula Sym- SJC
A. Hit test in all regions simultaneously.
Pre-euro Post-euro
UK Pair 0.0068 0.2627 0.1449 0.1465 0.024 0.4114 0.3533 0.4151
German Pair 0.0682 0.2010 0.1826 0.1875 0.4057 0.7728 0.6951 0.7156
French Pair 0.0543 0.8003 0.6302 0.6080 0.0739 0.1158 0.0919 0.1171
Japanese Pair 0.5887 0.7491 0.7871 0.8143 0.0674 0.3468 0.3468 0.3599

Pairs Gaussian copula t-copula SJC copula Sym-SJC Gaussian copula t-copula SJC copula Sym- SJC
B. Hit test in the joint lower and upper 10% regions.
Pre-euro Post-euro
UK pair Lower 0.0354 0.1015 0.1019 0.1045 0.3558 0.4340 0.3850 0.4230
Upper 0.3848 0.6258 0.6222 0.6254 0.5929 0.8675 0.7899 0.8892
German pair Lower 0.0971 0.2381 0.2036 0.2582 0.4190 0.7537 0.7167 0.7545
Upper 0.4538 0.9031 0.959 0.9447 0.4412 0.4722 0.4311 0.4714
French pair Lower 0.0175 0.2349 0.3302 0.1963 0.1095 0.1985 0.1749 0.2066
Upper 0.1115 0.408 0.2627 0.383 0.5639 0.5059 0.3485 0.4764
Japanese pair Lower 0.7789 0.8682 0.8001 0.8605 0.2953 0.4101 0.406 0.4052
Upper 0.3393 0.4968 0.5573 0.6053 0.2673 0.4315 0.4282 0.4484

Numbers in the table are p-values of ‘‘Hit’’ tests. A number over 0.05 indicates that the null is not rejected.

copulas with symmetric tail dependence have a better fit and confirms our finding of symmetric tail
dependence in the previous section.

5. Conclusion

In this paper, we examine the extreme co-movements between the stock and the exchange rate
markets by directly modeling their dependence structure via the use of the copulas. The symmetric tail
dependence is found to be significant in all the stock-currency return pairs analyzed in this study for
both pre- and post-euro periods. This finding is very important for global investors in their risk
management during extreme market events. The finding also implies that the Gaussian dependence
hypothesis that underlies most modern financial applications may be inadequate. Picking up the tail
dependence could lead to a more realistic assessment of the linkage between financial markets and
possibly more accurate risk management and pricing models.

Acknowledgements

I am grateful to Michael Melvin, an anonymous referee, and John Knight for their constructive
comments and suggestions. I thank Maurice Roche, Stephen Sapp, and Tony Wirjanto for their helpful
suggestions and comments. Thanks are also due to participants in the Canadian Economics Association
2006 annual meeting, Canadian Econometrics Study Group 2006 conference, and the Far Eastern
Econometric Society 2007 meeting for their comments. All the remaining errors are mine.

Appendix A. DataStream codes for data

Codes in DataStream for stock market indices and exchange rates are: TOTMKUS for the US, TOT-
MUK$ for UK, TOTMBD$ for Germany, TOTMFR$ for France, TOTMJP$ for Japan, BRITPUS for USD/UK
pound, WGMRKUS for USD/German mark, FRNFRUS for USD/French franc, JAPYNUS for USD/Japanese
yen.

References

Bollerslev, T., 1987. A conditional heteroskedastic time series model for speculative prices and rates of return. Review of
Economics and Statistics 69, 542–547.
Branson, W.H., 1983. Macroeconomic determinants of real exchange risk. In: Herring, R.J. (Ed.), Managing Foreign Exchange
Risk. Cambridge University Press, Cambridge, England.
 C. Ning / Journal of International Money and Finance 29 (2010) 743–759 759

Cappiello, L., Engle, R.F., Sheppard, K., 2006. Asymmetric dynamics in the correlations of global equity and bond returns. Journal
of Financial Econometrics 4, 537–572.
Chakrabarti, R., Roll, R., 2002. East Asia and Europe during the 1997 Asian collapse: a clinical study of a financial crisis. Journal of
Financial Markets 5, 1–30.
Chollete, L., De la Pena, V., Lu, C., 2006. Security Comovement: Alternative Measures, and Implications for Portfolio Diversifi-
cation. Unpublished Working paper; Columbia University and NHH.
Chollete, L., 2008. The Propagation of Financial Extremes. Unpublished Working Paper; NHH.
DeVries, C., Jansen, D., 1991. On the frequency of large stock returns: putting booms and busts into perspective. Review of
Economics and Statistics 73, 18–24.
Dornbusch, R., Fischer, S., 1980. Exchange rates and the current account. American Economic Review 70 (5), 960–971.
Dungey, M., Tambakis, D., 2005. Identifying International Financial Contagion: Progress and Challenges. Oxford Press, Oxford,
England.
Embrechts, P., Mcneil, A., Straumann, D., 2002. Correlation and dependence in risk management, properties and pitfalls. In:
Dempster, M.A.H. (Ed.), Risk Management: Value at Risk and Beyond. Cambridge Univ. Press, pp. 176–223.
Forbes, K., Rigobon, R., 2002. No contagion, only interdependence: measuring stock market co-movements. Journal of Finance
57 (5), 2223–2261.
Frankel, J.A., 1983. Monetary and portfolio-balance models of exchange rate determination. In: Bhandari, J., Putnam, S. (Eds.),
Economic Interdependence and Flexible Exchange Rates. MIT Press, Cambridge, MA.
Giovannini, A., Jorion, P., 1989. The time variation of risk and return in the foreign exchange and stock markets. Journal of
Finance 44 (2), 307–325.
Hartmann, P., Straetmans, S., De Vries, C., 2003. A global perspective on extreme currency linkages. In: Hunter, W.C., Kaufman, G.
G., Pomerleano, M. (Eds.), Asset Price Bubbles: Implications for Monetary, Regulatory and International Policies. MIT Press,
Cambridge, MA.
Hong, H., Scheinkman, J., Xiong, W., 2008. Advisors and asset prices: a model of the origins of bubbles. Journal of Financial
Economics 89, 268–287.
Hu, L., 2006. Dependence patterns across financial markets: a mixed copula approach. Applied Financial Economics 16,
717–729.
Joe, H., 1997. Multivariate Models and Dependence Concepts. In: Monographs on Statistics and Applied Probability, vol. 73.
Chapman and Hall, London.
Joe, H., Xu, J.J., 1996. The Estimation Method of Inference Functions for Margins for Multivariate Models. Technical Report No.
166. Department of Statistics, University of British Columbia.
Karolyi, G.A., Stulz, R.M., 1996. Why do markets move together? an investigation of U.S.-Japan stock return comovements.
Journal of Finance 51 (3), 951–986.
Knight, J., Lizieri, C., Satchell, S., 2005. Diversification when it hurts? The joint distribution of real estate and equity markets.
Journal of Property Research 22, 309–323.
Longin, F., Solnik, B., 2001. Extreme correlation of international equity markets. Journal of Finance 56, 649–676.
Mashal, R., Zeevi, A., 2002. Beyond Correlation: Extreme Co-movements Between Financial Assets. Working Paper. Columbia
Business School.
Nelson, R.B., 1999. An Introduction to Copulas. Springer, New York.
Patton, A.L., 2006a. Modelling asymmetric exchange rate dependence. International Economic Review 47 (2), 527–556.
Patton, A.L., 2006b. Estimation of multivariate models for time series of possibly different lengths. Journal of Applied Econo-
metrics 21 (2), 147–173.
Poon, S., Rockinger, M., Tawn, J., 2004. Extreme value dependence in financial markets: diagnostics, models, and financial
implications. Review of Financial Studies 17, 581–610.
Sklar, A., 1959. Fonctions de Riépartition á n Dimensions et Leurs Marges. Publications de l’Institut de Statistique de l’Université
de Paris 8, 229–231.
Susmel, R., 2001. Extreme observations and diversification in Latin American emerging equity markets. Journal of International
Money and Finance 20, 971–986.