A time-varying copula approach to oil andstock market dependence: the case oftransition economies

Abstract

We employ the time-varying copula approach to investigate the condi-tional dependence between the Brent crude oil price and stock markets inthe Central and Eastern European (CEE) transition economies. Our resultsshow evidence of a positive dependence between the oil and the stock mar-kets of the six CEE countries, which is indicative of a contagion betweenthose markets, regardless of the changes in the oil price or the CEE stockindex. Moreover, the dependence patterns in both the center and left tailsof the return distributions change over time, particularly during the heartof the financial crisis, and are best described by the the Survival Gumbelcopulas. The empirical evidence also suggests that the lower tail dependenceis much stronger than that of the upper tail, highlighting the importance ofcontagion during severe contractionary business cycles. Among the samplemarkets, Poland is shown to be particularly sensitive in this regard, whileHungary and Slovenia are the least sensitive.

JEL classification: C51, C58, F37, Q41, Q47.

Keywords: Copulas, oil prices, stock markets, transition economies.

1

1. Introduction

There is a strong presumption in the economic and financial literature

that oil prices impact stock markets negatively, mostly due to their con-

nection with inflation and precautionary oil demand (Sadorsky, 1999; Papa-

petrou, 2001; Ciner, 2001; Hooker, 2002; Barsky and Kilian, 2004). Studies

have been carried out on the oil price-stock market nexus for the major

industrial countries, Latin American countries, African countries, emerging

market economies and oil producing countries but the literature has yielded

mixed results. This literature shows no consensus that there is a defined rela-

tionship between oil prices and stock markets across countries. For example,

Apergis and Miller (2009) examine whether structural oil-market shocks have

an impact on stock prices in eight developed countries and conclude that de-

veloped stock markets do not react significantly to oil price changes. In

contrast, Park and Ratti (2008) find that oil price shocks have a statistically

significant impact on real stock returns in the U.S. and thirteen European

countries. They however find little evidence of asymmetric effects on real

stock returns for oil importing European countries of positive and negative

shocks of the oil price. Nandha and Faff (2008) provide evidence that the

rise in the price of oil has a negative impact on all industries but not the oil

and gas industries. Kilian and Park (2009) find that the response of stock

returns to oil prices may differ depending on the cause of the oil price shock.1

1Several other studies address the relationship between oil prices and stock returns atthe sector level (e.g., Sadorsky, 2001; Boyer and Filion, 2007; El-Sharif et al., 2005; Nandha

2

The literature, however, has not produced clear evidence on the oil-equity

relationship in Central and Eastern European (CEE) countries, which are dif-

ferent among themselves. The FTSE and MSCI groups recognize the Czech

Republic, Hungary and Poland as advanced emerging markets, while the

others are just emerging markets. The CEE economies are net oil importers,

vulnerable to oil supply disruptions, higher inflation, and some like Hungary

have excessive price regulations imposed by their governments (Mahanty et

al., 2010). They also differ in terms of size of GDP, GDP per capita, shares of

industry and agriculture in GDP, public debt, current account balance, and

compensation per employee, among others (Havlik, 2012). Therefore, under-

standing the oil-stock market relationship for these transitional economies

is important for investors and policy makers of those countries because of

the difference in relative contribution of oil to their energy consumption, oil

intensity of their economic sectors, and composition of economic structure.

The CEE countries have different relative energy dependence on oil, coal,

natural gas and nuclear energy.2 When it comes to oil consumption in these

and Faff, 2008; Nandha and Brooks, 2009; Arouri and Nguyen, 2010; Hammoudeh et al.,2010). Their results indicate that the reaction of sector returns to changes in oil pricesdiffers sensitively across sectors and that the presence of the oil assets in a portfolio ofsector stocks permits to improve the portfolio’s risk-return characteristics.

2The contributions of the primary sources of energy sources to total energy consump-tion differ from one CEE country to another during the period 2006-2010. In the CzechRepublic, coal dominates energy consumption followed by crude oil (Hugyecz, 2011). Nat-ural gas followed by crude oil dominates the consumption in Hungary. In Poland, coalstands out first and far, followed by a long distance by oil, somewhat similar to the CzechRepublic. In Romania, similar to Hungary, natural gas comes first by a big shot, followedby crude oil and then coal. Bulgaria is also similar to the Czech Republic because coaldominates energy consumption followed by oil and natural gas. This country relies on

3

countries as a whole, more than 60% of oil is used in the transportation sec-

tors, 10% in the industrial sector (steel, iron, aluminum etc.), and 16% in

non-energy use sectors (Hugyecz, 2011; IEA, 2010). This suggests that oil

affects most of the economic sectors of these countries, and that the behavior

of their economic players such as households and business accounts for most

of oil consumption. The oil-macroeconomy relationship underscores the im-

portance of the level of economic activity in determining oil consumption.

This also highlights the strong dependency of many companies on the quan-

tity and price of oil and the relationship between oil prices and stock prices

during bull and bear markets. The recent literature has shown that energy ef-

ficiency in the CEE economies has stalled since the 2007/2009 global financial

crisis (IEA, 2010). This study will focus on the oil-stock market relationship

for six major transition markets in the CEE region, namely Bulgaria, Czech

Republic, Hungary, Poland, Romania, and Slovenia.

Under normal economic conditions, changes in oil prices can affect the

CEE’s companies from both the supply and demand sides. From the supply

side, surges in oil prices increase the cost of production, thereby affect the

profitability of those companies. From the demand side, increases in the cost

of oil decrease consumers’discretionary expenditures which in turn reflect

negatively on the profitability of the companies. This applies to Romania

which produces some oil, as well as to Slovenia which basically produces no

nuclear energy for about 19% of its primary energy consumption.

4

oil. The fact that the CEE economies are net importers (see, Figure 2) implies

that they are more vulnerable to oil supply distribution than oil exporters,

which increases their sensitivity to oil price changes. Moreover, while some

of these economies depend on coal or natural gas as a second major energy

source to satisfy their industrial energy demand, they all depend to a varying

degree on imported oil as the major source of the transportation or surface

fuel which again highlights their oil susceptibility. As indicated earlier, more

than 60% of oil consumption in these CEE countries is used in the trans-

portation sector, while only 15% is used in the industrial sector. Having said

all that, the dependence structure between oil and CEE stock returns can

change under extreme conditions in the tail distributions because it can be

affected by other factors including strong herding, differential market power

and excessive price regulations, imposed by governments.

The objectives of this study are: i) to examine the time-varying depen-

dence between crude oil prices and stock returns in the six CEE transition

countries; ii) to discern the strength of relative oil-stock market interdepen-

dence during bullish and bearish market phases in these countries, which is

marked by the most recent financial crisis; and iii) to discuss the implications

of the empirical results on the future development of stock markets in these

six CEE countries, conditionally on the degree of their dependence on crude

oil as well as on oil price movements.

Our article contributes to the related literature in several important as-

pects. We first make use of a time-varying copula (TVC) approach to investi-

5

gate the dependence structure between oil and stock market returns through

time. Indeed, not allowing for time-varying parameters in the dependence

distribution generates a bias toward evidence of tail dependence. Similarly,

considering only tail dependence may falsely lead to evidence of asymmet-

ric relation between the returns. Empirically, return series are modeled by

GARCH-type processes with suitable marginal distributions, and appropriate

copula functions are then fitted to filtered return series in order to gauge their

dynamic interdependence. By doing so, it is possible to capture the potential

nonlinearities in the oil-stock market relationships as well as some well-known

empirical stylized facts of their return distributions such as volatility persis-

tence, fat tail behavior and asymmetric impacts of return innovations on

volatility (Sadorsky, 2006; Regnier, 2007; Arouri et al., 2011), while avoiding

the drawbacks of linear measures of interdependence such as Pearson corre-

lation (Jondeau and Rockinger, 2006). We are also able to examine both the

degree and nature of return dependence at extreme levels, i.e., the possibility

of joint extreme variations in the dynamics of oil and stock returns. Last but

not least, the use of a more recent dataset, spanning the period from Decem-

ber 1, 2005 to August 20, 2012, enables us to account for several episodes of

important fluctuations in oil and stock prices, especially over the 2007-2009

global financial crisis where extreme comovements are expected.

To the best of our knowledge, copulas have been previously used in Ger-

man and Khoroubi (2008), Zohrabyan (2008), and Nguyen and Bhatti (2012)

6

to examine the oil-stock market interactions.3 German and Khoroubi (2008)

analyze the diversification effect of including crude oil futures contracts into

a portfolio of stocks, while accounting for time-to maturity for the futures

contract. They use copula functions to examine the benefits of accounting

for the “maturity effect”on portfolio diversification and find that distance

maturity has a more pronounced negative correlation between the WTI fu-

tures prices and the S&P 500 index, regardless of changes in the oil price.

Zohrabyan (2008) employs copula functions in examining the dependence be-

tween oil prices and stock returns for three CEE countries (Poland, Czech

Republic, and Hungary), among other developed and developing countries.

His final conclusion is that copula results are sensitive to different oil price

scenarios and before and after the establishment of the euro zone. Nguyen

and Bhatti (2012) focus on the relationship between oil prices and stock mar-

kets in China and Vietnam which is an oil producer. The authors observe

the presence of left tail dependency between oil prices and the Vietnamese

stock market, but no such dependence for the Chinese market.

None of the above-mentioned studies uses a TVC approach and allows

for the choice of appropriate copulas among the most commonly-used copula

families. However, the study by Wen et al. (2012) is the most related to ours

in that it uses time-varying copulas to investigate whether contagion exists

3Copula models have also been recently used to examine the comovement of crude oilmarkets (Reboredo, 2011) as well as conditional dependence structure between oil pricesand exchange rates (Aloui et al., 2012; Wu et al., 2012).

7

between oil and stock markets (WTI crude oil price, the S&P’s 500 index,

and Chinese stock market indices), but their focus is more on the downside

dependence patterns. Their results show significantly increasing dependence

between the WTI oil and stock markets after the failure of Lehman Brothers,

supporting the presence of contagion between these markets, with weaker

contagion for China than the United States.4

Using daily data for the Brent crude oil index and stock market indices

in the six CEE transition economies (Bulgaria, Czech Republic, Hungary,

Poland, Romania, and Slovenia) in this study, we mainly find that oil and

these CEE stock markets exhibit time-varying interdependence in both the

center and lower tails of the return distributions, according to the three

families of copula models used. The dynamic dependence between the oil and

CEE markets is positive, whereby underscoring the importance of the global

business cycle, other factors that come into play under extreme conditions

such as strong herding and excessive price regulations or the global demand in

the Kilian and Park (2009) sense in moving these markets. The evidence also

suggests strong evidence of the lower tail dependence but the absence of the

upper tail dependence, highlighting the importance of contagion and possibly

herding during severe contractionary business cycles. This is underlined by

4In a related study, Weil (2011) applies copula and goodness-of-fit (gof) tests to esti-mate the VaR and expected shortfall (ES) for 12,000 bivariate portfolios of stocks, com-modities and FX futures. The analysis of three state-of-the-art approaches for testing acopula-model’s goodness-of-fit showed that none of the tests is able to identify the optimalparametric form unequivocally. See, for instance, Zhang and Guégan (2008), and Fombyet al. (2012) for more discussions regarding other applications of time-varying copulas.

8

the high time variations during the period 2009-2010 which includes the

heart of the recent global financial crisis. Poland is shown to be particularly

sensitive in this regard, while Hungary which has excessive government price

regulations and Slovenia are the least sensitive. Oil trading imbalances seem

to be the most important driver of the dependence patterns.

The remainder of the article is structured as follows. Section 2 briefly

reviews the related literature focusing on the linkages between oil and stock

markets around the world. Section 3 describes the TVC approach and the

estimation strategy. Section 4 presents the data used. Section 5 reports and

discusses empirical results. Section 6 provides some concluding remark.

2. Review of the literature

The introduction part has presented a brief review of the relationships

between oil prices and the stock markets of major industrial countries, Eu-

ropean countries, African countries and Latin American countries. This lit-

erature falls short of discussing these relationships for the CEE countries. In

this section, we rather provide a short survey of the most important studies

in the literature dealing with the relationships between oil and stock markets,

and with the dynamics of stock markets for CEE countries.

There are several influential studies that investigate the impact of oil

prices on financial variables. Chen et al. (1986) consider multivariate mod-

els of the determinants of returns by including a range of macroeconomic

variables and oil prices as explanatory variables for the USA. They find no

9

statistically significant relationship between oil price and stock returns. Con-

trary to Chen et al. (1986), Jones and Kaul (1996) examine the effect of real

oil prices on real returns for the United States, Canada, UK and Japan,

and document that in all four countries, the real oil price has a statistically

significant and negative effect on real returns. Subsequent studies including

Sadorsky (1999), Papapetrou (2001), and Ciner (2001), among others, pro-

vide evidence to support the Jones and Kaul (1996)’s initial findings. Some

recent studies find, however, convincing evidence of positive impact of crude

oil changes on stock market returns (e.g., Narayan and Narayan, 2010; Ono,

2011).

The relationship between oil prices and stock sectors has also been re-

cently examined by several studies which mostly apply the standard VAR/VEC

model. Sadorsky (2001) and Boyer and Filion (2007) show that oil price

increases positively affect stock returns of Canadian oil & gas companies.

El-Sharif et al. (2005) focus on the oil and gas sector returns in the United

Kingdom and reach similar findings. Their results also point to a weak link

between non-oil and gas sectors and oil price changes. Using data of thirty-

five global industries, Nandha and Faff (2008) provide evidence that the rise

in the price of oil has a negative impact on all industries but not oil and gas.

The results of Nandha and Brooks (2009) suggest that changes in oil prices

are an important determinant for stock returns of the transport sectors in

developed countries of their sample, but not in the Asian and Latin American

countries. At the firm level, Cong et al. (2008), and Narayan and Sharma

10

(2011) examine the relationship between oil prices and firm returns of dif-

ferent countries, and find strong dependence between these two variables.

Mohanty et al. (2010) focus on the link between oil prices and the stock

returns of oil and gas firms in the CEE countries. Their results indicate no

significant relation between oil prices and the stock returns over the period

1998-2010, but the oil price exposures of some oil and gas companies change

over time and across firms when subperiods are analyzed. The authors at-

tribute their findings to the systematic risk factors including both global risk

and local risks at the country, industry, and firm level.

As to studies exploring dynamics of the CEE stock markets, they mainly

focus on the impact of the accession of the CEE countries to the European

Union on their stock markets and on the relationships between CEE stock

markets and other markets in Europe, United States and emerging economies.

Most of these studies discuss the relations using standard linear frameworks

such as the VAR and VEC models. Few studies managed to use nonlin-

ear techniques. These linear models therefore do not permit to capture the

nonlinearity of the relationships, especially over certain periods of financial

stresses and crises like the 2007-2009 crisis. For instance, using symmetric

cointegration techniques, Jochum et al. (1999) analyze the behavior of the

Eastern European stock price indices, but with particular emphasis on the ef-

fects of the 1997/98 emerging market crisis on these indices. The authors find

less cointegration among these indices after the crisis than before, with the

Russian market having the dominant role during the crisis. Voronkova (2004)

11

explores the long-run cointegeration relationships between the emerging cen-

tral European stock markets and finds these relationships to be stronger

than was reported before when instability is taken into account. The author

also finds an equilibrium relationship with the developed markets, suggest-

ing that the central European markets have become more integrated with

the world markets. Using a smooth transition logistic trend model, Chelley-

Steeley (2005) investigates whether the stock markets in Hungary, Poland,

the Czech Republic and Russia have become less segmented and concludes

that they have a consistent increase in their comovements with some of the

other eastern European and developed markets. The author also finds that

Hungary is the country that has become the most integrated.

Johnson et al. (1994) examine the impact of increasing economic and

monetary integration within the European Community on the risk/return

characteristics of its equity markets. Scholtens (2000) explores the experi-

ences of the central European countries in setting up a financial system. The

author argues that these countries tend to catch up with the western Euro-

pean countries much faster in the case of their banking systems than in terms

of their stock markets. Differently, Gilmore et al. (2005) construct optimal

portfolios to investigate the diversification benefits for U.S. and German in-

vestors in three central European stock markets. Their results suggest that

diversification benefits are statistically significant for the U.S. investors, but

not for German investors. Middleton et al. (2008) examines the potential

benefits from diversifying into eight CEE stock markets and find evidence

12

of substantial benefits that accrue more from the geographical spread than

from the industrial mix of the equities included in the portfolio. Allen et

al. (2010) investigate the implications for European investors of investing

in twelve central and eastern European stock markets after the European

Union expansion and conduct a Markowitz effi cient frontier analysis of these

markets pre- and post-EU expansion. For an EU based investor, the findings

are not all good as revealed in the Markowitz analysis. Demian (2011) ana-

lyzes the long-run cointegeration relationship between the financial markets

of the Czech Republic, Estonia, Hungary, Poland, Romania and Slovakia,

and discerns the impact of their accession to the European Union on their

relationship with the markets of the developed countries. They conclude that

the EU accession has a minor impact on those relationships.

Syllignakis and Kouretas (2008) investigate whether the volatility of stock

returns of ten emerging markets of the new members of the European Union

has changed due to their accession to this bloc. The authors find that the

high volatility of stock returns of the new emerging stock markets of the new

EU members is associated largely with the 1997-1998 Asian and Russian fi-

nancial crises. They also find that there is a transition to the low volatility

regime as they approached the EU accession in 2004. Tudor (2008) finds

no evidence of significant changes in volatility both on Bucharest Stock Ex-

change and on Budapest Stock Exchange after the 2007 EU accession by

Romania and Bulgaria and the 2004 accession by ten new members. Harri-

son and Moore (2012) focus on forecasting stock market volatility in central

13

and eastern European countries. The authors test the predictive power of 12

volatility forecasting models and their results show that models that account

for asymmetric volatility consistently outperform all the other models.

3. Empirical model

We use a time-varying copula approach to examine the dependence struc-

ture between oil and the stock markets over time for the six CEE countries.

Following Grégoire et al. (2008), a rolling window procedure is adopted to

estimate the dependence parameter of copula models as well as the tail de-

pendence coeffi cients. To reduce the computational cost of this procedure,

we choose a window length of 250 days which corresponds to approximately

one trading year. Methodologically, we begin with modeling the margin of

the return series by fitting appropriate ARMA-GARCH specifications to the

data and extracting the standardized residuals. We then apply the empir-

ical cumulative distribution function (ECDF) to the standardized residuals

and estimate the selected copula models. This semiparametric approach is

repeated for each of the 250-day windows until the end of our estimation

period.

3.1. Models for marginal distributions

Tomodel the margin of return series, we combine an ARMA(m,n) process

with a standard GARCH(1, 1) model of Bollerslev (1986), which is probably

the most commonly used financial time series model that has inspired a

14

number of more sophisticated extensions. This combination allows not only

to succesfully characterize some important stylized facts of financial returns

such as volatility clustering and time-varying volatility, but also to obtain

approximate i.i.d (independent and identically distributed) residuals that

are suitable for further statistical analyses (Grégoire et al., 2008). Many em-

pirical studies have found that the GARCH(1, 1) model, albeit its simplicity,

is usually suffi cient to provide good estimates of the conditional volatility of

most macroeconomic and financial variables (see, e.g., Bollerslev et al., 1992).

Furthermore, Aloui et al. (2012) show that empirical results from copula

models are not sensitive to the choice of GARCH specifications when copula

parameters are estimated by the Canonical Maximum Likelihood (CML),

which is also used in this study.

Given a time series yt, an ARMA(m,n)−GARCH(1, 1) model can be

written as

yt = µ+m∑i=1

aiyt−i +n∑j=1

bjεt−j + εt,

εt = σtzt, (1)

σ2t = ω0 + αε2t−1 + βσ2t−1

where ω0 > 0, α ≥ 0, β ≥ 0, µ is a constant term of the conditional mean

equation; zt is a sequence of i.i.d. random variable with zero mean and unit

variance; and σ2t denotes the conditional variance of return series at time t,

which depends on both past return innovation (εt−1) and past conditional

15

variances (σ2t−1). We use the AIC and BIC criteria to determine the optimal

lag length for the conditional mean (ARMA) process and these criteria select

m = 1 and n = 0 for all the markets, except for Bulgaria where m = 2 is

chosen, and for Poland and the Brent oil market where m = 0 is chosen.5

3.2. Copula models for cross-market dependence

Introduced by Sklar (1959), copulas are a powerful tool for modeling a

large range of dependence structures. They have become increasingly popu-

lar in finance over the past ten years. A number of past studies have applied

copula-based models to measure dependence structure of financial data for

the purposes of better understanding derivatives pricing and portfolio man-

agement issues (e.g., Chan-Lau et al., 2004; Ning, 2010; Aloui et al., 2011;

Choe and Jang, 2011).

Formally, copulas are functions that link multivariate distributions to

their univariate marginal functions. They can be defined as follows

Definition 1. A d-dimensional copula is a multivariate distribution functionC with standard uniform marginal distributions.

Theorem 1. Sklar’s theoremLet X1, ..., Xd be random variables with marginal distribution F1, ..., Fd

and joint distribution H, then there exists a copula C: [0, 1]d → [0, 1] suchthat:

H(x1, ..., xd) = C(F1(x1), ..., Fd(xd)) (2)

Conversely if C is a copula and F1, ..., Fd are distribution functions, thenthe function H defined above is a joint distribution with margins F1, ..., Fd.

5Detailed results are not reported here to conserve space, but can be made availableunder request addressed to the corresponding author.

16

Copula functions provide an effi cient way to create distributions that

model correlated multivariate data. As far as the measure of interdepen-

dence is concerned, one can construct a multivariate joint distribution by

first specifying marginal univariate distributions, and then choosing a copula

to examine the correlation structure between the variables. Bivariate distrib-

utions as well as distributions in higher dimensions are possible. Copulas can

also be used to characterize the dependence in the tails of the distribution.

Two measures of tail dependence related to copulas are known as the upper

and the lower tail dependence coeffi cients. They are indeed very helpful for

measuring the tendency of markets to crash or boom together.

Let X and Y be random variables with marginal distribution functions

F and G. Then the coeffi cient of lower tail dependence λL is

λL = limt→0+

Pr[Y ≤ G−1(t)∣∣X ≤ F−1(t)

](3)

which quantifies the probability of observing a lower Y assuming that X is

lower itself. In the same way, the coeffi cient of upper tail dependence λU can

be defined as

λU = limt→1−

Pr[Y > G−1(t)∣∣X > F−1(t)

](4)

There is a symmetric tail dependence between two assets when the lower

tail dependence coeffi cient equals the upper one, otherwise the tail depen-

dence is asymmetric. The tail dependence coeffi cients provide a way for

ordering copulas. One would say that copula C1 is more concordant than

17

copula C2 if λU of C1 is greater than λU of C2.6

The copula models we consider fall into the three families of copulas:

elliptical (Gaussian and Student-t), Archimedean (Gumbel and Clayton) and

extreme value (Tawn) families. They are briefly presented below.

The Gaussian copula: The bivariate normal copula is defined by

C(u, v) = φθ(φ−1(u), φ−1(v))

=

φ−1(u)∫−∞

φ−1(v)∫−∞

1

2π√1− θ2

exp(−s2 − 2θst+ t2

2(1− θ2))dsdt

where φθ is the standard bivariate normal distribution with linear correla-

tion coeffi cient θ restricted to the interval (−1, 1), φ represents the univariate

standard normal distribution function.

The Student-t copula: The bivariate Student-t copula is defined by

C(u, v) =

t−1υ (u)∫−∞

t−1υ (v)∫−∞

1

2π√1− θ2

(1 +s2 − 2θst+ t2

υ(1− θ2))−

υ+22 dsdt

where t−1υ (u) denotes the inverse of the CDF of the standard univariate

Student-t distribution with υ degrees of freedom.

The Gumbel copula of Gumbel (1960) is an asymmetric copula with higher

6See Joe (1997) and Nelsen (1999) for detailed discussions of copulas functions andtheir properties.

18

probability concentrated in the right tail. It is given by

C(u, v) = exp{−[(− lnu)θ + (− ln v)θ]1/θ}

where the dependence parameter θ can take any value in (1,+∞).

The Tawn Copula or the mixed model of the Gumbel and independence

copula, is an extreme value copula expressed as

C(u, v) = uv exp{−θ lnu ln vln(uv)

}, 0 ≤ θ ≤ 1 (5)

where the dependence parameter θ can take any value in (−∞,+∞).

The Clayton copula which has been introduced by Clayton (1978) and is

expressed as

C(u, v) = (u−θ + v−θ − 1)−1/θ, , θ ∈ [−1,∞)\{0}

We also consider the Survival Gumbel and Survival Clayton copulas, which

can be viewed as a mirror image of the density of the Gumbel and Clayton

copulas.

All in all, our copula models allow to capture various dependence struc-

tures, ranging from independence to extreme dependence. While the Gaussian

copula is symmetric and has no tail dependence, the Student-t copula can

19

capture extreme dependence between variables. Unlike the elliptical copu-

las, the Archimedeans such as the Gumbel and Clayton copulas are used to

capture asymmetry between lower and upper tail dependences. The Clayton

copula exhibits greater dependence in the negative tail than in the positive

tail, whereas the Gumbel copula exhibits greater dependence in the upper

tail than in the lower tail. The Tawn copula or the asymmetric logistic model,

dating back to Tawn (1988), adds more flexibility to the Gumbel family of

copulas.

3.3. Choice of suitable copula models and estimation strategy

We estimate the parameters of the copula using a semi parametric two-

step estimation method, namley the Canonical Maximum Likelihood, or

CML (Cherubini et al., 2004). In the first step, we estimate the marginals

FX and GY non parametrically via their empirical cumulative distribution

functions (ECDF) FX and GY defined as

FX(x) =1

n

n∑i=1

1{Xi < x} and GY (y) =1

n

n∑i=1

1{Yi < y} (6)

In the implementation, FX and GY are rescaled by n/(n + 1) to ensure

that the first order condition of the log-likelihood function for the joint dis-

tribution is well defined for all finite number of observations, n. We then

transform the observations into uniform variates using the ECDF of each

marginal distribution and we estimate the unknown parameter θ of the cop-

20

ula as

θCML = argmaxθ

n∑i=1

ln c(FX(xi), FY (yi); θ) (7)

Under suitable regularity conditions, the CML estimator θCML is con-

sistent, asymptotically normal, and fully effi cient at independence. Further

details can be found in Genest et al. (1995).

In order to compare copula models, we use the goodness-of-fit (GOF) test

of Genest et al. (2009) which is based on a comparison of the distance be-

tween the estimated and the empirical copula. Specifically, the test statistics

considered use the Cramér—Von Mises distance as

Sn =

∫Cn(u)2dCn(u) (8)

Large values of the statistic Sn lead to the rejection of the null hypothesis

that the copula C belongs to a class C0. In practice, we require knowledge

about the limiting distribution of Sn which depends on the unknown para-

meter value θ. To find the p-values associated with the test statistics we

use a multiplier approach as described in Kojadinovic and Yan (2011). The

highest p-values thus indicate that the distance between the estimated and

empirical copulas is the smallest and that the copula in use provides best fit

to the data.

21

4. Data and stochastic properties

We use the daily closing price data for the Brent crude oil index and MSCI

(Morgan Stanley Capital International) stock market indices of the six CEE

transition economies (Bulgaria, Czech Republic, Hungary, Poland, Romania

and Slovenia) over the period from December 1, 2005 to August 20, 2012,

totalizing 1753 observations. The beginning of this sample period is dictated

by the availability of the data for Romania and our desire to have a balanced

data for all the countries to facilitate their comparison. This study period

is of particular interest for our investigation because the crude oil and CEE

stock markets may exhibit interdependence not only in the center but also in

the tails of the distributions, given the occurrence of the US Subprime crisis

in 2007, the resulting global financial crisis in 2008-2009 and the ensuing

euro-zone debt crisis. The choice of daily data is motivated by the fact

that extreme comovements between markets are more likely to occur at high

frequency levels. Data are expressed in US dollars to avoid the undesirable

impacts from exchange rate movements.7 The oil and stock market data are

obtained from the US Energy Information Administration (EIA) and MSCI

databases, respectively. For our empirical analysis, we consider log-returns

that are computed as rt = ln(Pt/Pt−1) with Pt being the index or the price at

time t. The time-variations of return series over the study period are plotted

7It is also more likely to have asymmetry and regime switching with weekly data thanwith daily data. Furthermore, it is well known that weekly results are sensitive to theselected day of the week and they provide less dynamics and correlations than daily data.

22

in Figure 1. It can be seen that daily returns are fairly stable during the

period preceding the start of the recent global financial crisis (i.e., December

2005 to the third quarter of 2008 which corresponds to the summer meltdown

of financial markets). All return series exhibit higher variability afterwards.

Table 1 presents the descriptive statistics and stochastic properties of our

return series. The average return is negative for all the series, except for

Czech Republic and Brent crude oil. Surprisingly, the unconditional volatil-

ity, measured by standard deviation, is relatively similar across oil and CEE

stock markets. Skweness coeffi cients are positive only for two cases (Hun-

gary and Brent crude oil). Excess kurtosis ranges from 3.487 to 17.571.

Normality of the unconditional return distributions is strongly rejected by

the Jarque-Bera test. These findings clearly show that the probability of

observing extremely negative and positive realizations for our return series is

higher than that of a normal distribution. The Ljung—Box statistics of order

12 suggest the existence of serial correlation for almost all series. Finally,

ARCH effects are found in all cases, thus supporting our decision to filter

daily returns with a GARCH-type model.

Table 2 displays petroleum production, consumption, imports, reserves,

and energy intensity for the six countries. Romania is the largest petro-

leum producer, while Slovenia is the smallest on the number of barrel basis.

Poland which has the largest population shows the greatest consumption,

while Slovenia which has a population of about two millions consumes the

least oil. Corresponding to their relative populations, Poland is the largest oil

23

MSCI Bulgaria

Q1 Q1 Q1 Q1 Q1 Q1 Q12006 2007 2008 2009 2010 2011 2012

­0.1

50.

05

MSCI Czech Republic

Q1 Q1 Q1 Q1 Q1 Q1 Q12006 2007 2008 2009 2010 2011 2012­0

.15

0.15

MSCI Hungary

Q1 Q1 Q1 Q1 Q1 Q1 Q12006 2007 2008 2009 2010 2011 2012

­0.2

00.

20

MSCI Poland

Q1 Q1 Q1 Q1 Q1 Q1 Q12006 2007 2008 2009 2010 2011 2012

­0.1

00.

10

MSCI Romania

Q1 Q1 Q1 Q1 Q1 Q1 Q12006 2007 2008 2009 2010 2011 2012

­0.3

00.

10

MSCI Slovenia

Q1 Q1 Q1 Q1 Q1 Q1 Q12006 2007 2008 2009 2010 2011 2012

­0.0

80.

08

Brent

Q1 Q1 Q1 Q1 Q1 Q1 Q12006 2007 2008 2009 2010 2011 2012

­0.1

50.

15

Figure 1: Daily returns on MSCI stock indices and Brent crude oil

24

Table 1: Descriptive statistics and stochastic properties of daily returns

Panel A: summary statisticsMin Mean ×103 Max Std Dev Skew. Ex. kurtosis

Bulgaria -0.183 -0.984 0.114 0.020 -1.430 12.319Czech Rep. -0.167 0.059 0.197 2.098e-02 -0.152 12.201Hungary -0.203 -0.264 0.203 0.028 0.026 5.997Poland -0.134 -0.070 0.142 0.024 -0.212 3.487Romania -0.316 -0.314 0.125 0.025 -1.351 17.571Slovenia -0.099 -0.221 0.095 0.015 -0.363 6.158Brent -0.168 0.438 0.181 0.022 0.016 6.462Panel B: statistical tests

Q(12) Q2(12) J-B ARCH(12)Bulgaria 113.144∗∗ 1368.996∗∗ 11606.120∗∗ 489.656∗∗

Czech 44.471∗∗ 1221.649∗∗ 10805.954∗∗ 394.377∗∗

Hungary 64.138∗∗ 911.930∗∗ 2607.873∗∗ 331.875∗∗

Poland 11.561 733.930∗∗ 893.799∗∗ 301.993∗∗

Romania 23.122∗ 36.417∗∗ 22933.168∗∗ 27.547∗∗

Slovenia 50.100∗∗ 1246.548∗∗ 2787.362∗∗ 456.704∗∗

Brent 12.793 309.890∗∗ 3027.677∗∗ 155.213∗∗

Notes: The table displays summary statistics for MSCI stock indices and crude oil returns.

The sample period is from December 1, 2005 to August 20, 2012. Q(12) and Q^2(12) are

the Ljunk-Box statistics for serial correlation in returns and squared returns for order 12.

JB is the empirical statistic of the Jarque-Bera test for normality. ARCH is the Lagrange

multiplier test for autoregressive conditional heteroskedasticity. * and ** indicate the

rejection of the null hypotheses of no autocorrelation, normality and homoscedasticity at

the 5% and 1% levels, respectively.

importer, while Slovenia imports the least oil. Romania possesses the largest

proven reserves, while Slovenia has basically no proven reserves. Bulgaria

has the highest oil consumption per one unit of GDP, while Hungary has the

lowest oil intensity. In terms of energy intensity, Bulgaria has the highest

intensity while Slovenia displays the lowest. Figure 2 shows the net crude

oil trading balances, as measured by the difference between total exports

and total imports, for the six countries over the period 2005-2010. It can be

25

seen that all the countries in our sample are net oil importers. Poland is the

largest oil importer, followed by far by Romania and the Czech Republic.

These evolving patterns reflect the consumption and production situations

for each country.

Table 2: Energy profile, population and GDP of the sample CEE countries

Bulgaria Czech Rep. Hungary Poland Romania SloveniaProduction 2.920 13.010 27.640 28.340 105.050 0.005Consumption 112.700 198.980 141.000 576.600 218.230 52.930Imports 109.780 185.970 113.360 548.260 113.180 52.925GDP 33.690 145.570 109.430 398.140 116.520 39.750Imports/GDP 3.259 1.278 1.036 1.377 0.971 1.331Consumption/GDP 3.345 1.367 1.288 1.448 1.873 1.332Reserves 0.015 0.015 0.027 0.096 0.600 0.000Energy Intensity 23973.90 11025.30 9440.40 10808.70 12910.40 8495.90Population 7.149 10.202 9.992 38.464 21.959 2.003

Notes: Oil production, oil consumption and oil imports are in thousand barrels per day

in 2011. Proven oil reserves are in billion barrels. Primary energy intensity is BTU per

year U.S. dollars for 2009. Rear GDP is in billions of 2005 U.S. dollars. Population is in

millions for year 2010. The oil and energy data are obtained from the EIA website of the

U.S. Department of Energy, while the real GDP data is in the 2005 base year dollars and

accessed from the US database.

5. Empirical results

5.1. Conditional dependence structure

The conditional dependence structure between the Brent oil price and

each of the six CEE stock markets is estimated through a three-step proce-

dure. We first filter the returns using appropriate ARMA(m,n)-GARCH(1, 1)

processes, defined previously (see, subsection 3.1).8 The objective is to get

8The estimation results of ARMA(m,n)-GARCH(1, 1) are not reported here to conservespace, but can be made available under request addressed to the corresponding author.

26

−400

−300

−200

−100

0

2005 2006 2007 2008 2009 2010Time

Net

Oil

Bal

ance

s

CEE countries

Bulgaria

Czeck Republic

Hungary

Poland

Romania

Slovenia

Figure 2: Net oil balances (oil exports minus oil imports) in thousand barrels per days

approximate i.i.d residuals, while controlling for the effects of conditional

heteroscedasticity. Second, we estimate the marginal distributions of the

filtered returns non-parametrically using their empirical cumulative distri-

bution functions. We finally make use of the CML method to determine

the unknown parameter θ of the seven copula models: Gaussian, Student-t,

Gumbel, Survival Gumbel, Tawn, Clayton and Survival Clayton copulas. As

suggested by Chen and Fan (2006), this semiparametric modelling approach

creates additional flexibility in that it offers the possibility to combine various

27

return-generating models with a rich variety of available copula families.

Table 3 reports the estimated values of the dependence parameters for

each pair of oil-stock markets with respect to copula models in use. As

expected, these dependence parameters are highly significant whatever the

copula function is, thus showing evidence of integration between the two

markets regardless of changes in those markets. This finding, which is entirely

explained by the oil dependence profiles in Figure 2, further suggests that the

price and index fluctuations are likely to comove over time. Moreover, the

dependence parameters are positive in all cases, which reveals that increases

in the price of oil coincide with an appreciation of stock prices in all selected

CEE countries. The fact that both the oil price and the stock market indices

tend to be cyclically and positively related to the global business may explain

this substantial comovement. Kilian and Park (2009) attribute the positive

correlation to positive shocks to the global demand for industrial commodities

that cause both higher real oil prices and higher stock prices.

Table 4 presents the Cramér-Von Mises statistics as well as the p-values

of the goodness-of-fit test proposed by Genest et al. (2009). Recall that for

this test, the null hypothesis stating that the estimated copula provides the

best fit to the data is rejected for the p-values that are less than the con-

ventional significance level. The results show that for all considered pairs,

the Survival Gumbel copula yields the smallest distance between the fitted

and empirical copula, and consequently the highest p-value for the conducted

goodness-of-fit test. Besides, the hypothesis of an appropriate fit of the other

28

Table3:Estimatesofcopuladependenceparameters

Gaussian

Student-t

GumbelS.Gumbel

Tawn

Clayton

S.Clayton

Bulgaria

0.228

(0.021)∗∗

0.234

(0.023)∗∗(v=17.120

(7.617)∗)

1.138

(0.018)∗∗

1.163

(0.019)∗∗

0.322

(0.039)∗∗

0.287

(0.032)∗∗

0.214

(0.031)∗∗

CzechRep.

0.366

(0.021)∗∗

0.365

(0.020)∗∗(v=74.537

(112.497))

1.254

(0.022)∗∗

1.277

(0.025)∗∗

0.521

(0.036)∗∗

0.475

(0.038)∗∗

0.388

(0.033)∗∗

Hungary

0.358

(0.020)∗∗

0.358

(0.021)∗∗(v=23.067

(14.731))

1.257

(0.021)∗∗

1.276

(0.024)∗∗

0.530

(0.036)∗∗

0.467

(0.037)∗∗

0.395

(0.032)∗∗

Poland

0.390

(0.020)∗∗

0.390

(0.020)∗∗(v=19.198

(10.271))

1.290

(0.022)∗∗

1.309

(0.025)∗∗

0.584

(0.035)∗∗

0.523

(0.040)∗∗

0.442

(0.032)∗∗

Romania

0.269

(0.021)∗∗

0.268

(0.023)∗∗(v=16.162

(7.063)∗)

1.163

(0.019)∗∗

1.196

(0.022)∗∗

0.364

(0.039)∗∗

0.358

(0.036)∗∗

0.247

(0.032)∗∗

Slovenia

0.239

(0.022)∗∗

0.240

(0.023)∗∗(v=22.015

(13.462))

1.147

(0.018)∗∗

1.167

(0.020)∗∗

0.342

(0.038)∗∗

0.298

(0.034)∗∗

0.223

(0.031)∗∗

Notes:ThetabledisplaystheestimatedcopuladependenceparametersfortheGaussian,Student-t,Gumbel,SurvivalGumbel,

Tawn,ClaytonandSurvivalClaytoncopulamodelsfortheCEEstockmarketsandBrentcrudeoil.Standarderrorsaregiven

inparentheses.*and**indicatethesignificanceofthecoefficientsatthe5%

and1%

levels,respectively.

29

Table4:Distancebetweenempiricalandestimatedcopulas

Gaussian

Student-tGumbelS.Gumbel

Tawn

Clayton

S.Clayton

Bulgaria

0.056

(0.000)∗∗

0.050

(0.000)∗∗

0.176

(0.000)∗∗

0.049

(0.007)∗∗

0.182

(0.000)∗∗

0.090

(0.000)∗∗

0.306

(0.000)∗∗

CzechRep.

0.036

(0.040)∗

0.035

(0.031)∗

0.151

(0.000)∗∗

0.030

(0.091)

0.164

(0.000)∗∗

0.113

(0.000)∗∗

0.359

(0.000)∗∗

Hungary

0.038

(0.046)∗

0.035

(0.032)∗

0.145

(0.000)∗∗

0.027

(0.138)

0.152

(0.000)∗∗

0.111

(0.000)∗∗

0.344

(0.000)∗∗

Poland

0.043

(0.016)∗

0.038

(0.025)∗

0.128

(0.000)∗∗

0.038

(0.028)∗

0.131

(0.000)∗∗

0.136

(0.000)∗∗

0.336

(0.000)∗∗

Romania

0.045

(0.008)∗∗

0.043

(0.007)∗∗

0.140

(0.000)∗∗

0.018

(0.512)

0.154

(0.000)∗∗

0.038

(0.054)

0.273

(0.000)∗∗

Slovenia

0.023

(0.255)

0.023

(0.271)

0.092

(0.000)∗∗

0.020

(0.384)

0.097

(0.000)∗∗

0.060

(0.003)∗∗

0.205

(0.000)∗∗

Notes:ThistabledisplaysthedistancebetweentheempiricalandtheestimatedcopulaaccordingtotheCramér-VonMises

statistic.Inbrackets,theresultsforthep-valuesarebasedonmultiplierapproach(KojadinovicandYan,2010).*and**

denotetherejectionofthecopulamodelatthe5%

and1%

levels,respectively.Boldfacenumbersindicatethelowestdistance

fortheconsideredcopulas.

30

copula models is rejected at the conventional significance level in almost all

cases. Overall, our findings suggest that the dependence between the return

series is positive and can be described as the best by the asymmetric Sur-

vival Gumbel copula, which implies that negative returns are more correlated

than positive returns. In other words, the correlation is stronger during regu-

lar contractionary business cycles than regular expansionary business cycles.

Since the oil-stock market pairs under consideration only exhibit extreme co-

movements in the left tails of the return distributions, the Survival Gumbel

copula is then the best candidate for capturing this kind of conditional de-

pendence structure. Intuitively, the relatively high dependence in the lower

tail may have to do with contagion, weak hedging, stronger herding and lack

of market power during economic stress.

Table 5: Tail dependence coeffi cients

λl λuBulgaria 0.185 0.000Czech Rep. 0.280 0.000Hungary 0.278 0.000Poland 0.302 0.000Romania 0.215 0.000Slovenia 0.189 0.000

Notes: This table presents the estimates of the lower and upper tail dependence parameters

obtained from the best fitting copula models for each MSCI-crude oil pairs.

To quantify the extreme dependence, we compute the tail dependence co-

effi cients implied by the estimated parameters of the Survival Gumbel copula

since it provides the best fit (Table 5). The copula chosen for our pairs of

31

crude oil and stock markets is asymmetric with positive lower tail depen-

dence (λl > 0) and zero upper tail dependence (λu = 0). This implies that

during severe contractionary business cycles or slumps, the contagion is the

strongest between the Brent oil price and the stock market indices. We note

that Poland shows the strongest positive dependence with crude oil, while

the lowest degree of extreme dependence is obtained for Bulgaria and Slove-

nia. This finding is effectively expected as Poland is the largest oil importer

among the sample countries, while Bulgaria and Slovenia are the smallest oil

importers (see, Table 2).

The zero dependence in the upper tail could be due to structural breaks or

regime shifts which change the relations between oil prices and the CEE stock

indices in the high volatility regime (Liu et al., 2012). Governments’excessive

price regulation as it is the case in Hungary (Mahanty et al., 2010) can also be

a reason for zero dependence during the boom conditions. Hedging against

higher oil prices can as well weaken the correlation between oil prices and the

stock indices. Moreover, the inability of companies to pass through higher

oil prices to the customers also diminishes the relationship between the CEE

stocks and oil prices.

5.2. Time-varying pattern of oil-stock market comovement

To examine the possible evolution of the dependence over time, a time-

varying copula approach was applied for the considered series. Following

Grégoire et al. (2008), we use a rolling window approach to estimate the de-

32

pendence parameter of the copula and the tail dependence coeffi cients. Due

to the computational cost of this procedure, we initially choose a window

length of 250 days (i.e., 250 return observations) which corresponds to ap-

proximately one trading year. Again, we focus on modeling the conditional

return distribution instead of unconditional returns. In a first step, we fit a

GARCH(1,1) model for each return series and extract the standardized resid-

uals. Then, we apply the empirical cumulative distribution function (ECDF)

for the standardized residuals and estimate the Gumbel copula dependence

parameters. This semiparametric approach is repeated for each new window

constructed from the remaining 1502 trading days between November 17,

2006 and August 20, 2012.

Figure 3 shows a plot of the estimated dependence parameters of the

Survival Gumbel copula for one-year rolling window periods for the six CEE

countries. One can observe that all estimated dependence parameters ex-

hibit a time-variation, taking on values between 1.004 and 1.743. Moreover,

the dependence between the return series seems to increase to a higher level

after the period of the financial crisis, in particular during 2009 and 2010,

which witnessed the deepest point of the global financial crisis. This find-

ing suggests that joint extreme losses tend to occur more frequently during

this kind of periods of time. Note that the period 2009-2010 in Europe is

marked by severe economic slowdowns and uncertainties due to the negative

shock transmission from the US subprime crisis and to the beginning of fears

about the public debt situation. Unsurprisingly, the cross-market comove-

33

Czech Republic

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

1.10

1.50

Hungary

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

1.1

1.5

Bulgaria

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

1.05

1.25

Poland

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

1.1

1.5

Romania

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

1.05

1.45

Slovenia

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

1.05

1.25

Figure 3: Time-varying dependence parameters of the Survival Gumbel copula for therelationship between crude oil and stock markets (1-year rolling window)

34

Czech Republic

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

0.15

0.35

Hungary

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

0.15

0.35

Bulgaria

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

0.05

0.25

Poland

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

0.15

0.35

Romania

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

0.05

0.25

Slovenia

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32007 2008 2009 2010 2011 2012

0.10

0.30

Figure 4: Time-varying lower tail dependence coeffi cient of the Survival Gumbel copulafor the relationship between crude oil and stock markets (1-year rolling window)

35

ment increased over that stressful period. It is now a common knowledge that

correlations increase during economic stress because of heightened herding

behavior and the elevated impacts of changes in aggregate macroeconomic

factors.

Figure 4 shows the evolution of the lower tail dependence coeffi cients of

the Survival Gumbel copula. As expected, the extreme dependence structure

between the variables is not constant over time. We find that there are

time periods when the lower tail dependence coeffi cients are approximately

zero, indicating that there is little or no relationship between returns in the

left tail (bearish markets) and other "stormy" time periods with a higher

probability of joint extreme losses. In all cases, the hypothesis of increasing

extreme tail dependence after the last financial crisis is verified. Moreover,

the extreme dependence between the considered returns seems to increase to

a high level in the second half of 2010 and the part of 2012 that is covered by

the data, the period of political instability in North Africa and the Middle

East, intensified debt crisis in the euro zone, and rocky economic recovery in

the United States. This finding confirms the fact that asset returns tend to

become more correlated during periods of turbulence or financial crisis.

5.3. Robustness check

5.3.1. Alternative rolling windows

A potential drawback of the rolling window approach is that empirical

results may be sensitive to the length of the rolling windows. While a longer

36

rolling window is likely to give more reliable estimation results, it may neglect

the potential of strong parameter instabilities which can only be captured by

shorter rolling window. For instance, several studies employing the rolling

regression method to examine the relationships between oil and stock market

returns show that the empirical results are reasonably robust to small changes

in the estimation window (Basher and Sadorsky, 2006; Hammoudeh and

Nandha, 2007).

We now turn to check the sensitivity of our results with respect to a sig-

nificant change in the size of the rolling window. Without a loss of generality,

we decide to choose two alternative rolling window lengths of 500 and 750

trading days which correspond approximately to 2-year and 3-year periods,

respectively. We then estimate the dependence parameters as well as the

lower tail coeffi cients of the Survival Gumbel copula for our oil-stock market

pairs, using these new rolling window sizes. The obtained results, displayed

in Figures 5-8, are not different from those of the 250-day rolling window.

They do indicate that all the dependence parameters estimated with the 500-

and 700-day rolling windows exhibit time-varying patterns, and experience

a clear tendency of increased comovement between the late 2008 and the

late 2010. Some high dependence levels are also observed during 2010 and

the part of 2012. In addition, the extreme dependence structure between oil

and CEE market returns is not constant over time and there is evidence to

suggest that the lower tail dependence has increased following the onset of

the global financial crisis.

37

Czech Republic

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

1.15

1.35

Hungary

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

1.15

1.35

Bulgaria

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

1.05

1.25

Poland

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

1.15

1.55

Romania

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

1.05

1.25

Slovenia

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

1.10

1.30

Figure 5: Time-varying dependence parameters of the Survival Gumbel copula for therelationship between crude oil and stock markets (2-year rolling window)

38

Czech Republic

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

0.15

0.35

Hungary

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

0.15

0.35

Bulgaria

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

0.10

0.30

Poland

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

0.15

0.35

Romania

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

0.05

0.25

Slovenia

Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q3 Q1 Q32008 2009 2010 2011 2012

0.10

0.26

Figure 6: Time-varying lower tail dependence coeffi cient of the Survival Gumbel copulafor the relationship between crude oil and stock markets (2-year rolling window)

39

Czech Republic

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

1.18

1.34

Hungary

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

1.20

1.40

Bulgaria

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

1.10

1.26

Poland

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

1.20

1.40

Romania

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

1.10

1.30

Slovenia

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

1.12

1.20

Figure 7: Time-varying dependence parameters of the Survival Gumbel copula for therelationship between crude oil and stock markets (3-year rolling window)

40

Czech Republic

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

0.20

0.28

0.36

Hungary

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

0.18

0.34

Bulgaria

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

0.12

0.28

Poland

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

0.20

0.36

Romania

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

0.12

0.28

Slovenia

Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q32008 2009 2010 2011 2012

0.14

0.22

Figure 8: Time-varying lower tail dependence coeffi cient of the Survival Gumbel copulafor the relationship between crude oil and stock markets (3-year rolling window)

41

5.3.2. Sensitivity of the results to weekly data

Some studies on the relationships between oil and stock markets, includ-

ing for example Arouri and Nguyen (2010), suggest that the weekly data may

better capture the interaction of oil and stock price changes than the daily

data because their use can significantly reduce the potential biases from the

bid-ask effect and the non-synchronous trading days, among others. On the

other hand, the monthly data are not appropriate for the study of oil-stock

market comovements as they may ignore the asymmetric responses of stock

returns to oil price shocks.

Accordingly, we repeat our estimation procedure using weekly data for

the same period.9 The obtained results reveal three main facts. First, the

dependence parameters estimated from the weekly data are positive for all

the pairs of oil and CEE stock markets (Table 6). This finding is entirely

in line with the conclusion we have made on the daily data, regardless of

copula models. Second, the results of the GOF tests, reported in Table 7,

typically suggest the Clayton copula as the best-fitted copula, while the Sur-

vival Gumbel copula is the best model when the daily data are used. Indeed,

the Clayton copula cannot be rejected for all the oil-stock market pairs and

it also provides the lowest distance between the empirical and estimated cop-

ulas in five out of the six cases (Bulgaria, Czech Republic, Hungary, Poland

9Weekly returns are also tested for stationarity using the ADF and PP tests and theobtained results indicate that they are all stationary and thus suitable for further statisticalanalysis.

42

Table6:Estimatesofcopuladependenceparameterswithweeklydata

Gaussian

Student-t

GumbelS.Gumbel

Tawn

Clayton

S.Clayton

Bulgaria

0.220

(0.054)∗∗

0.220

(0.051)∗∗(υ=100.006

(168.601))

1.095

(0.041)∗∗

1.158

(0.050)∗∗

0.200

(0.095)∗

0.331

(0.088)∗∗

0.119

(0.071)

CzechRep.

0.174

(0.050)∗∗

0.182

(0.055)∗∗(υ=21.981

(30.783))

1.074

(0.037)∗∗

1.149

(0.042)∗∗

0.209

(0.089)∗

0.300

(0.070)∗∗

0.070

(0.068)

Hungary

0.174

(0.053)∗∗

0.174

(0.053)∗∗(υ=76.002

(211.003))

1.078

(0.037)∗∗

1.131

(0.044)∗∗

0.198

(0.089)∗

0.268

(0.079)∗∗

0.086

(0.065)

Poland

0.148

(0.051)∗∗

0.151

(0.056)∗∗(υ=11.704

(10.667))

1.073

(0.035)∗∗

1.124

(0.042)∗∗

0.237

(0.085)∗∗

0.253

(0.077)∗∗

0.056

(0.059)

Romania

0.159

(0.047)∗∗

0.162

(0.056)∗∗(υ=10.270

(6.402))

1.089

(0.039)∗∗

1.113

(0.041)∗∗

0.198

(0.090)∗

0.216

(0.073)∗∗

0.140

(0.067)∗

Slovenia

0.162

(0.048)∗∗

0.164

(0.057)∗∗(υ=7.898

(4.361))

1.095

(0.038)∗∗

1.123

(0.041)∗∗

0.246

(0.089)∗

0.224

(0.074)∗∗

0.137

(0.066)∗

Notes:ThetabledisplaystheestimatedcopuladependenceparametersfortheGaussian,Student-t,Gumbel,SurvivalGumbel,

Tawn,ClaytonandSurvivalClaytoncopulamodelsfortheCEEstockmarketsandBrentcrudeoilusingweeklydata.Standard

errorsaregiveninparentheses.*and**indicatethesignificanceofthecoefficientsatthe5%

and1%

levels,respectively.

43

Table 7: Distance between empirical and estimated copulas (weekly data)

Gaussian Student-t Gumbel S.Gumbel Tawn Clayton S.ClaytonBulgaria 0.032

(0.068)0.032(0.066)

0.071(0.000)∗∗

0.025(0.241)

0.084(0.000)∗∗

0.018(0.533)

0.111(0.000)∗∗

Czech Rep. 0.053(0.002)∗∗

0.050(0.002)∗∗

0.100(0.000)∗∗

0.034(0.071)

0.094(0.000)∗∗

0.020(0.439)

0.151(0.000)∗∗

Hungary 0.029(0.123)

0.029(0.105)

0.057(0.002)∗∗

0.020(0.479)

0.057(0.000)∗∗

0.011(0.940)

0.092(0.000)∗∗

Poland 0.045(0.009)∗∗

0.042(0.011)∗

0.066(0.001)∗∗

0.031(0.096)

0.061(0.001)∗∗

0.019(0.493)

0.104(0.000)∗∗

Romania 0.036(0.042)∗

0.036(0.036)∗∗

0.051(0.007)∗∗

0.030(0.126)

0.056(0.002)∗∗

0.026(0.242)

0.068(0.001)∗∗

Slovenia 0.022(0.387)

0.019(0.455)

0.031(0.125)

0.018(0.541)

0.030(0.142)

0.022(0.376)

0.050(0.020)∗

Notes: This table displays the distance between the empirical and the estimated copula

according to the Cramér-Von Mises statistic. In brackets, the results for the p-values are

based on multiplier approach (Kojadinovic and Yan, 2010). * and ** denote the rejection

of the copula model at the 5% and 1% levels, respectively. Bold face numbers indicate the

lowest distance for the considered copulas.

and Romania). The Survival Gumbel copula only provides the best fit for

Slovenia, followed closely by the Student-t and the Clayton copulas. How-

ever, the findings for the daily and weekly data remain consistent since both

the Clayton and Survival Gumbel copulas exhibit greater dependence in the

lower tail than in the upper tail. Finally, the tail dependence coeffi cients for

the weekly data are lower than those for the daily data (Table 8). This is

effectively expected because extreme comovements are more likely to occur

at higher frequency data, and as a result, the daily return data have a more

fat-tailed distribution than the weekly data.

44

Table 8: Tail dependence coeffi cients (weekly data)

λL λUBulgaria 0.123 0.000Czech Rep. 0.099 0.000Hungary 0.075 0.000Poland 0.065 0.000Romania 0.041 0.000Slovenia 0.146 0.000

Notes: This table presents the estimates of the lower and upper tail dependence parameters

obtained from the best fitting copula models for pairs of crude oil and CEE stock market

returns using weekly data.

6. Conclusions

Previous literature on the relationships between oil and stock markets

around the world falls short of adequately discussing the case of CEE coun-

tries. Moreover, these relationships are frequently assumed to be constant

and linear over time. In this paper, we use the time-varying copula approach

to address this issue for six major transition economies in the CEE region:

Bulgaria, the Czech Republic, Hungary, Poland, Romania and Slovenia. This

method offers the possibility to examine the degree and nature of return de-

pendence at extreme levels through time, such as the last financial crisis. It

also allows to capture the potential nonlinearities in the oil-stock market re-

lationships as well as some well-known empirical stylized facts of their return

distributions.

The results show evidence of a positive dependence between the oil and

the stock markets of the six CEE countries. They thus suggest that oil

and the CEE stock market indices do not provide diversification benefits

45

during extreme financial conditions such as the recent global financial crisis

and they should not be combined in diversified portfolios to reduce systemic

risk. This implication applies more to Poland than the other CEE countries.

This is interesting given the fact that Poland has been less vulnerable to the

global crisis than other CEE countries. It also proposes that CEE countries

should reduce their dependence on oil imports and adopt more energy effi cient

technology, particularly during crises.

Another important implication is that there is a contagion between those

markets during severe financial stress, regardless of the changes in the Brent

oil price or the CEE stock index. This finding suggests that portfolio man-

agers who combine oil and CEE stocks should hedge their portfolios with

risk-reducing tools from other asset classes. The contagion also implies that

the CEE stock market authorities should have circuit breakers and safety

nets to be ready to use during crisis. Moreover, the results suggest that

contagion is related to herding across asset classes and changes in the global

business cycles and global aggregate demand. If the oil price moves ahead of

the equity indices, then this price has a predictive information content for the

CEE markets, which could help both investors and policymakers. Oil price

increases may also not be associated with inflation, or that inflation caused by

oil prices, if occurs, is not harmful to those countries. Finally, the goodness-

of-fit tests select the suvirval Gumble copula as the best specification. This

finding thus shows strong dependence in the lower tail, which suggests that

those CEE markets are highly vulnerable during severe global recessions, as

46

do the oil prices. This is underscored by the high time variations during the

period 2009-2010 which includes the heart of the crisis. Poland which is the

largest oil importer is shown to be particularly sensitive in this regard, while

Hungary which is strapped by excessive price regulations, and Slovenia are

the least sensitive.

Acknowledgement : we would like to thank the two anonymous reviewers

for their invaluable and helpful comments, and also Tengdong Liu for his

help with the data. All remaining errors are ours.

References

[1] Allen, D.E., Golab, A., Powell, R., 2010. Volatility and correlations for

stock markets in the emerging economies of Central and Eastern Europe:

implications for European Investors. School of Accounting, Finance and

Economics, Edith Cowan University.

[2] Aloui, R., Ben Aïssa, M.S., Nguyen, D.K., 2011. Global financial crisis,

extreme interdependences, and contagion effects: The role of economic

structure? Journal of Banking and Finance, 35, 130—141.

[3] Aloui, R., Ben Aïssa, M.S., Nguyen, D.K., 2012. Conditional Depen-

dence Structure between Oil Prices and Exchange Rates: A Copula-

GARCH Approach. Journal of International Money and Finance 32,

719—738.

47

[4] Apergis, N., Miller, S.M., 2009. Do structural oil-market shocks affect

stock prices? Energy Economics 31(4), 569-575.

[5] Arouri, M.E.H, Nguyen, D.K., 2010. Oil prices, stock markets and port-

folio investment: Evidence from sector analysis in Europe over the last

decade. Energy Policy 38, 4528—4539.

[6] Barsky, R.B., Kilian, L., 2004. Oil and the macroeconomy since the

1970s’. Journal of Economic Perspectives 18(4), 115—134.

[7] Basher, S.A., Sadorsky, P., 2006. Oil price risk and emerging stock mar-

kets. Global Finance Journal 17, 224—251.

[8] Bollerslev, T., 1986. Generalized autoregressive conditional het-

eroskedasticity. Journal of Econometrics 31, 307—327.

[9] Bollerslev, T., Chou, R.Y., Kroner, K.F., 1992. ARCH modeling in fi-

nance. Journal of Econometrics 52, 5—59.

[10] Boyer, M.M., Filion, D., 2007. Common and fundamental factors in

stock returns of Canadian oil and gas companies. Energy Economics

29(3), 428—453.

[11] Chan-Lau, J.A., Mathieson, D.J., Yao, J.Y., 2004. Extreme contagion

in equity markets. IMF Staff Papers 51, 386-408.

[12] Chelley-Steeley, P.L., 2005. Modeling equity market integration using

48

smooth transition analysis: a study of Eastern European stock markets.

Journal of International Money and Finance 24(5), 818-831.

[13] Chen, X., Fan, Y., 2006. Estimation of copula-based semiparametric

time series models. Journal of Econometrics 130, 307-335.

[14] Chen, N.F., Roll, R., Ross, S.A., 1986. Economic forces and the stock

market. Journal of Business 59, 383-403.

[15] Cherubini, U., Luciano, E., Vecchiato, W., (2004). Copula Methods in

Finance. John Wieley & Son Ltd.

[16] Choe, G.H., Jang, H.J., 2011. Effi cient algorithms for basket default

swap pricing with multivariate Archimedean copulas. Insurance: Math-

ematics and Economics 48, 205—213.

[17] Ciner, C., 2001. Energy shocks and financial markets: nonlinear linkages.

Studies in Nonlinear Dynamics and Economics 5, 203-12.

[18] Clayton, D.G., 1978. A model for association in bivariate life tables and

its application in epidemiological studies of familial tendency in chronic

disease incidence. Biometrika 65, 141-151.

[19] Cong, R.C., Wei, Y.M., Jiao, J.L., Fan, Y., 2008. Relationships between

oil price shocks and stock market: an empirical analysis from China.

Energy Policy 36(9), 3544-3553.

49

[20] Demian, C-V., 2011. Cointegration in Central and East European mar-

kets in light of EU accession. Journal of International Financial Markets,

Institutions and Money 21(1), 144-155.

[21] El-Sharif, I., Brown, D., Burton, B., Nixon, B., Russel, A., 2005. Evi-

dence of the nature and extent of the relationship between oil prices and

equity values in the UK. Energy Economics 27(6), 819-830.

[22] Fomby, T.B., Gunther, J.W., Hu, J., 2012. Return dependence and the

limits of product diversification in financial firms. Journal of Money,

Credit and Banking 44, 1151—1183.

[23] Genest, C., Ghoudi, K., Rivest, L.-P., 1995. A semiparametric estima-

tion procedure of dependence parameters in multivariate families of dis-

tributions. Biometrika 82, 543—552.

[24] Genest, C., Rémillard, B., Beaudoin, D., 2009. Goodness-of-fittests for

copulas: a review and a power study. Insurance: Mathematics and Eco-

nomics 44, 199—213.

[25] Genest, C., Segers, J., 2009. Rank-based inference for bivariate extreme-

value copulas. Annals of Statistics 37, 2990—3022.

[26] German, H., Kharoubi, C., 2008. WTI crude oil futures in portfolio

diversification: The time-to-maturity effect. Journal of Banking and Fi-

nance 32(12), 2553-2559.

50

[27] Gilmore, C.G., McManus, G.M., Tezel, A., 2005. Portfolio allocations

and the emerging equity markets of Central Europe. Journal of Multi-

national Financial Management 15(3), 287-300.

[28] Grégoire, V., Genest, C., Gendron, M., 2008. Using copulas to model

price dependence in energy markets. Energy Risk 5, 58—64.

[29] Gumbel, E.J., 1960. Bivariate exponential distributions. Journal of

American Statistical Association 55, 698-707.

[30] Hammoudeh, S., Nandha, M., 2007. Systematic risk, and oil price and

exchange rate sensitivities in Asia-Pacific stock markets. Research in

International Business and Finance 21, 326—341.

[31] Hammoudeh, S., Yuan, Y., Chiang, T., Nandha, M., 2010. Symmetric

and asymmetric US sector return volatilities in presence of oil, financial

and economic risks. Energy Policy 38(8), 3922-3932.

[32] Harrison, P., Moore, W., 2012. Forecasting stock market volatility in

central and eastern European countries. Journal of Forecasting 31(6),

490-503.

[33] Havlik, P., 2012. Economic transitions in Central and Eastern Europe:

any lessons for the Arab spring? HERMES-IR, Technical report 2012-06.

[34] Hooker, M.A., 1996. What happened to the oil price-macroeconomy

relationship? Journal of Monetary Economics 38, 195—213.

51

[35] Hugyecz, A., 2011. The impact of the crisis on energy deamnd and energy

intensity in Central and eastern European countries. Eastern Journal of

European Studies 2(2), 109-117.

[36] IEA, 2010. World energy outlook 2010. OCDE, Paris.

[37] Jochum, C., Kirchgässner, G., Platek, M., 1999. A long-run relation-

ship between Eastern European stock markets? Cointegration and the

1997/98 crisis in emerging markets. Review of World Economic 135(3),

454-479.

[38] Joe, H., 1997. Multivariate models and dependence concepts. Chapman

& Hall Ltd.

[39] Johnson, R., Lindvall, J., Soenen, S., 1994. EC economic and mone-

tary integration: implications for European equity investors. European

Management Journal 12(1), 94-101.

[40] Jondeau, E., Rockinger, M., 2006. The Copula-GARCH model of condi-

tional dependencies: An international stock market application. Journal

of International Money and Finance 25, 827—853.

[41] Jones, C.M., Kaul, G., 1996. Oil and the stock markets. Journal of

Finance 51(2), 463-491.

[42] Kilian, L., Park, C., 2009. The impact of oil price shocks on the U.S.

stock market. International Economic Review 50(4), 1267-1287.

52

[43] Kojadinovic, I., Yan, J., 2011. A goodness-of-fit test for multivariate

multiparameter copulas based on multiplier central limit theorems. Sta-

tistics and Computing 21 (1), 17—30.

[44] Liu, T., Hammoudeh, S., Balcilar, M., 2013. Interrelationships among

U.S. financial risks, economic activity and oil in a regime-changing en-

vironment. Drexel University, Philadelphia, PA.

[45] Middleton, C.A.J., Fifield, S.G.M., Power, D.M., 2008. An investigation

of the benefits of portfolio investment in Central and Eastern European

stock markets. Research in International Business and Finance 22, 162-

174.

[46] Mohanty, S., Nandha, M., Bota, G., 2010. Oil shocks and stock returns:

The case of the Central and Eastern European (CEE) oil and gas sectors.

Emerging Markets Review 11, 358-372.

[47] Nandha, M., Brooks, R., 2009. Oil prices and transport sector returns:

An international analysis. Review of Quantitative Finance and Account-

ing 33(4), 393-409.

[48] Nandha, M., Faff, R., 2008. Does oil move equity prices? A global view.

Energy Economics 30(3), 986-997.

[49] Narayan, P.K., Narayan, S., 2010. Modelling the impact of oil prices on

Vietnam’s stock prices. Applied Energy 87, 356-361.

53

[50] Narayan, P.K., Sharma, S.S., 2011. New evidence on oil price and firm

returns. Journal of Banking and Finance 35(12), 3253-3262.

[51] Nelsen, R. B., 1999. An introduction to copulas. Springer, NewYork.

[52] Nguyen, C.C, Bhatti, M.I., 2012. Copula model dependency between oil

price and stock markets: Evidence for China and Vietnam. Journal of

International Financial Markets, Institutions and Money 22(4), 758-773.

[53] Ning, C., 2010. Dependence structure between the equity market and the

foreign exchange market: a copula approach. Journal of International

Money and Finance 29, 743-759.

[54] Ono, S., 2011. Oil price shocks and stock markets in BRICs. European

Journal of Comparative Economics 8(1), 29-45.

[55] Papapetrou, E., 2001. Oil price shocks, stock market, economic activity

and employment in Greece. Energy Economics 23(5), 511-532.

[56] Park, J., Ratti, R.A., 2008. Oil price shocks and stock markets in the

U.S. and 13 European countries. Energy Economics 30, 2587-2608.

[57] Reboredo, J.C., 2011. How do crude oil prices co-move? A copula ap-

proach. Energy Economics 33, 948-955.

[58] Regnier, E., 2007. Oil and energy price volatility. Energy Economics 29,

405-427.

54

[59] Sadorsky, P., 1999. Oil price shocks and stock market activity. Energy

Economics 21(5), 449-469.

[60] Sadorsky, P., 2001. Risk factors in stock returns of Canadian oil and gas

companies. Energy Economics 23, 17—28.

[61] Sadorsky, P., 2006. Modeling and forecasting petroleum futures volatil-

ity. Energy Econ. 28, 467-488.

[62] Scholtens, B., 2000. Financial regulation and financial system architec-

ture in Central Europe. Journal of Banking and Finance 24(4), 525-553.

[63] Sklar, A., 1959. Fonctions de répartition à n dimensions et leurs marges.

Publications de l’Institut de Statistique de l’Université de Paris 8, 229-

231.

[64] Syllignakis, M.N., Kouretas, G.P., 2008. Switching volatility in emerging

stock markets: Evidence from the new EU member countries.Working

Paper, Athens University of Economics and Business.

[65] Tawn, J.A., 1988. Bivariate extreme value theory: models and estima-

tion. Biometrika 75(3), 397—415.

[66] Tudor, C., 2008. Comparative analysis of stock market behavior after

European accession in Romania and Hungary: Some Hypotheses Tests.

The Bucharest Academy of Economic Studies, Bucharest, Romania.

55

[67] Voronkova, S., 2004. Equity market integration in Central European

emerging markets: a cointegration analysis with shifting regimes. Inter-

national Review of Financial Analysis 13(5), 633-647.

[68] Wei, G.N.F., 2011 Are copula-GoF-tests of any practical use? Empirical

evidence for stocks, commodities and FX futures. Quarterly Review of

Economics and Finance 51(2), 173-188.

[69] Wen, X., Wei, Y., Huang, D., 2012. Measuring contagion between energy

market and stock market during financial crisis: A copula approach.

Energy Economics 34(5), 1435-1446.

[70] Wu, C-C., Chung, H., Chang, Y-H., 2012. The economic value of

co-movement between oil price and exchange rate using copula-based

GARCH models. Energy Economics 34, 270-282.

[71] Zhang, J., Guégan, D., 2008. Pricing bivariate option under GARCH

processes with time varying copula. Insurance: Mathematics and Eco-

nomics 42(3), 1095—1103.

[72] Zohrabyan, T., 2008. Essay on time series and causality in financial

markets. Unpublished Ph.D. dissertation, Texas A&M University.

56