Exchange Rates: A Vine Copula Based GARCH Method Relationship Between Oil, Stock Prices and

ECOFIN 595
No. of Pages 14, Model 1G
17 May 2016
North American Journal of Economics and Finance xxx (2016) xxxxxx
1
Contents lists available at ScienceDirect
North American Journal of

Economics and Finance
j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / e c o fi n
6
7
Relationship between oil, stock prices and

exchange rates: A vine copula based GARCH
method
Riadh Aloui a, Mohamed Safouane Ben Assa b,
3
4
9
10
11
a r t i c l e
1
2 3
3
14
15
16
17
18
19
20
21
22
LAREQUAD & ISGS, University of Sousse, Rue Abedelaziz El Bahi, B.P. 763, 4000 Sousse, Tunisia
LAREQUAD & FSEGT, University of Tunis El Manar, B.P. 248, El Manar II, 2092 Tunis, Tunisia
i n f o
Article history:
Available online xxxx
Keywords:
Vine copulas
Dependence measures
Crude oil price
Stock index
Exchange rate
a b s t r a c t
In this paper, we apply a vine copula approach to investigate the
dynamic relationship between energy, stock and currency markets.
Dependence modeling using vine copulas offers a greater flexibility
and permits the modeling of complex dependency patterns for
high-dimensional distributions. Using a sample of more than
10 years of daily return observations of the WTI crude oil, the
Dow Jones Industrial average stock index and the trade weighted
US dollar index returns, we find evidence of a significant and symmetric relationship between these variables. Considering different
sample periods show that the dynamic of the relationship between
returns is not constant over time. Our results indicate also that the
dependence structure is highly affected by the financial crisis and
Great Recession, over 20072009. Finally, there is evidence to suggest that the application of the vine copula model improves the
accuracy of VaR estimates, compared to traditional approaches.
2016 Published by Elsevier Inc.
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
1. Introduction
44
Crude oil is one of the most important commodities in the current global world. Over the past
decade, the greater instability in energy markets and the persistence of oil prices at higher levels
45
Corresponding author. Tel.: +216 58 450 000; fax: +216 71 872 277.
E-mail addresses: riadh.aloui@isg.rnu.tn (R. Aloui), safouane.benaissa@univ-amu.fr (M.S. Ben Assa).
http://dx.doi.org/10.1016/j.najef.2016.05.002
1062-9408/ 2016 Published by Elsevier Inc.
Please cite this article in press as: Aloui, R., & Ben Assa, M. S. Relationship between oil, stock prices and exchange
rates: A vine copula based GARCH method. North American Journal of Economics and Finance (2016), http://dx.doi.
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
2
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
R. Aloui, M.S. Ben Assa / North American Journal of Economics and Finance xxx (2016) xxxxxx
are largely responsible of the slowing world economic growth (Aydin & Mustafa, 2011; Sanchez,
2011). Through the increasing importance of oil price in the economic activity, the study of the relationship between energy, stock and currency markets becomes of greater importance for policy makers, economists and investors.
During the last financial crisis, oil prices experienced very large fluctuations as a clear structural
change around the second quarter of 2008 is apparent in Fig. 1. In fact, the spot price of crude oil
had a very sharp increase, rising from 20$ per barrel at the beginning of 2002 to 147$ per barrel in
July 2008, surpassing its 1980 record high in constant prices. Recent unrest in North Africa and the
Middle East and fears about the spread of political instability to other major oil producing countries
have contributed to higher oil prices and added more instability to energy markets.
Economic theory suggests that oil shocks have a significant effect on the stock market activity and
exchange rate movements. Huang, Masulis, and Stoll (1996) argue that the impact of crude oil movements on stock markets can be completely explained by their effect on current and future real cash
flows. Many recent papers found that an increase in oil prices implies a decrease in stock returns
(Chiou & Lee, 2009; Miller & Ratti, 2009; Nandha & Faff, 2008; Park & Ratti, 2008). By now, this idea
has become widely accepted in the literature and seems to be virtually axiomatic. More recent studies
such as Arouri and Nguyen (2010) and Fayyad and Daly (2011) demonstrate that the impact of oil on
stock markets is sensitively different across economic sectors (e.g., oil versus non-oil industries) and
across countries (e.g., net oil-exporting versus net oil-importing ones). According to Bjornland (2009)
and Jimenez-Rodriguez and Sanchez (2005), a positive association between oil price movements and
stock market returns is expected in the case of an oil exporting country, as the countrys income will
increase. It follows an increase in expenditures and investments which in its turn create more employment opportunities and the value of stocks will go up.
Studying the relationship between energy and currency markets has also received considerable
attention in the literature. The importance of oil prices as an explanatory variable of exchange rate
movements has been well documented in Krugman (1983), Golub (1983) and Rogoff (1991). In fact,
the influence of high oil prices on export competition and price level of a country will lead to frequent
and uncertain changes in the exchange rate. Moreover, oil prices are denominated in U.S dollar, and so
fluctuations in the exchange rate cause changes to the crude oil supply, demand and price. Using different datasets, the existing empirical studies have mainly found that the oil price increase is associated with a dollar appreciation (Aloui, Ben Assa, & Nguyen, 2013; Ding & Vo, 2012; Wu, Chung, &
Chang, 2012). By contrast, some other studies demonstrate a negative relationship between oil prices
and the U.S. dollar exchange rates (Narayan, Narayan, & Prasad, 2008; Zhang, Fan, Tsai, & Wei, 2008).
The inconsistency in empirical findings can be explained by the distinct features of the investigated
countries and the different extent of the used datasets. In this paper, our objective is to investigate
whether the relationship between oil, stock and exchange rate is positive, negative or unclear. To overcome the limitation of pair dependence analysis, which is evident in the related literature, we examine
the relationship between oil, stock and exchange rate in a multivariate framework. As pointed out by a
number of studies, it is important to understand the dependence between several variables interacting
simultaneously, not in isolation of one another. The omission of one important variable in the
extended system can be misleading because the channel through which the two other variables are
connected is omitted from the incomplete system.
As documented, for example, by Jondeau and Rockinger (2006), Junker, Szimayer, and Wagner
(2006) and McNeil, Frey, and Embrechts (2005), the widely used measure of dependence, known as
the Pearson correlation coefficient, may not appropriately describe the type of dependence between
returns and, consequently, could lead to underestimate the joint risk of extreme events. In order to
overcome this problem, the use of the copula methodology may be a very promising solution to characterize the multivariate distributions of asset returns. While there is a large literature exploring
dependence using bivariate copulas, the choice is much more restricted in the multivariate case.
The two most popular choices allowing multivariate dependence to be modeled with a nonrestricted correlation matrix are the normal and the Student-t copulas. However, these models are
restrictive in the tail and they do not allow asymmetric dependence. Recently, Bedford and Cooke
(2001) and Bedford and Cooke (2002) introduced vine or pair-copula construction of multivariate distribution. These models are flexible graphical models enabling the extensions to higher dimensions
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
3
20
40
60
80
100
120
140
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Fig. 1. Dynamics of daily prices of West Texas Intermediate (US Dollar/Barrel).
114
using a cascade of bivariate copula. The great advantage of vine-copula is that we can select bivariate
copulas from a wide range of existing copula families.
In this paper, we apply a vine copula approach to shed new light on the dynamic relationship
between crude oil price movements, U.S. market stock prices and U.S. dollar exchange rate. Moreover,
on the basis of this approach, we attempt to identify changes in the structure of crude oil prices and
what implications these change have on the dependence between the three markets. Using daily timeseries of crude-oil spot prices, Dow Jones Industrial Average (DJIA) stock index and nominal exchange
rate for the trade weighted U.S dollar index, we mainly find a significant and symmetric relationship
between these variables. However, this relationship is not constant across sample sub-periods. More
importantly, we find that changes in the structure of crude oil returns affect significantly the connection between the considered return series.
We structure the rest of the article as follows. Section 2 discusses the economic relationship
between oil, stock prices and exchange rates. Section 3 describes the empirical methodology and
the estimation strategy. Section 4 presents the used data and discusses our empirical results. Section 5
provides some concluding remarks.
115
2. Relationships between oil, stock prices and exchange rates
116
Theoretically, oil price movements affect stock returns in several ways. The value of a companys
stock at any point in time can be measured by making the sum of all expected future cash flows discounted back to the present using the discount rate (Huang et al., 1996), meaning that oil price shocks
can affect stock returns directly via the expected cash flows or indirectly by impacting the discount
rate. Since energy is an essential input to the production process, then higher oil prices lead to the
increase of production cost and reduce in the amount of the expected profits for non-oil related
companies. At the same time, expected cash flow will drop and companys stock prices will be
affected. As the number of the companies with falling stock prices go higher; stock index, reflecting
the performance of the whole stock market, will go down. On the other side, oil price increase is
expected to raise the overall trade deficit for oil importing countries. A growing trade deficit will generate expectations of future depreciation of the current exchange rate accompanied by higher inflation
100
101
102
103
104
105
106
107
108
109
110
111
112
113
117
118
119
120
121
122
123
124
125
126
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
4
165
rate. Consequently, the rise in the inflation rate may cause the discount rate to rise and stock prices to
fall.
The interaction of inflation and monetary policy exerts also a great influence on interest rates.
Assuming that other prices are sticky downwards, higher oil prices allow the rise on the domestic
price level which in its turn implies higher inflation and interest rates. In addition, hurdle rate which
means the required rate of return that an investment manager demands to undertake a particular project, will tend to rise leading to a low level of investments. Consequently, profitable projects will be
turned down and the stock price will be affected.
Several empirical works by Arouri, Lahiani, and Bellalah (2010), Basher and Sadorsky (2006), Boyer
and Filion (2007), Hammoudeh, Dibooglu, and Aleisa (2004), Huang et al. (1996), Jones and Kaul
(1996), Miller and Ratti (2009), Papapetrou (2001), Park and Ratti (2008) and Sadorsky (2001) have
shown that oil price shocks affect stock returns.
The relationship between oil prices and exchange rates has also received much attention in literature (see, among others, Amano & van Norden, 1998; Chen & Chen, 2007; Golub, 1983). A frequently
given explanation is based on the potential impact of oil shocks in driving term of trade movements,
which would therefore justify the effect on the exchange rate. Amano and van Norden (1998) consider
a small open economy with two-sectors for tradable (oil) and non-tradable goods (labour). The output
price of the tradable sector is set on the world markets, while the real exchange rate is determined by
the output price in the non-tradable sector. Consequently, an increase in oil prices leads to a decrease
in the labor price in order to improve the competitiveness of the tradable sector. Assuming that the
production in the non-tradable sector is more energy-intensive, the output price of this sector will
tend to rise causing the real exchange rate to appreciate. Another explanation for the link between
oil prices and exchange rates focuses on the balance of payments and the international portfolio
choices (Golub, 1983). In this approach, a surge in oil prices generates wealth transfers from oil
importing economies to oil exporting ones (like OPEC), leading to adjustments in exchange rates.
The final impact of oil shocks on exchange rate depends on the distribution of oil imports across oil
importing economies and on portfolio choices of both OPEC and oil-importing countries. If wealth
reallocation due to oil price increase is the outcome of an excess supply of dollars, then the dollar will
depreciate. Extending this approach, Krugman (1983) uses a dynamic partial equilibrium framework
to model how OPEC uses the accumulated wealth of their oil exports in dollars. Assuming that OPEC
will progressively spend its surpluses to import more goods from developing countries, the long-run
effect of an oil price hike on the dollar exchange rate will depend on the weight of oil in the U.S. total
imports compared to the U.S. weight in OPECs imports. On the short run, the effect depends on the U.
S. weight in the global oil imports compared to their weight in the dollar-denominated assets held by
OPEC.
The above discussion highlights the fact that there are strong theoretical arguments for why oil
price shocks should affect stock prices and exchange rates. Eventually, empirical analysis is needed
to obtain new insights into the relationship between these three variables. Also, It would be of further
interest to explore the changing dynamics of this relationship after the last financial crisis.
166
3. Empirical methodology
167
168
In this section, the vine copula construction method and some useful concepts that are necessary to
understand this approach are explained.
169
3.1. Pair-copula construction
170
Introduced first by Joe and Xu (1996) and extended by Bedford and Cooke (2001, 2002), vine
copulas are flexible graphical models enabling the extensions to higher dimensions using a cascade
of bivariate copulas or pairs-copulas. The modeling scheme is based on a decomposition of a
multivariate probability density into dd 1=2 bivariate copula densities, which may be chosen
independently from the others to allow for a wide range of dependence structure. Two main classes
of pair-copulas have been treated in literature, C- and D-vine copula models. Let us illustrate the
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
171
172
173
174
175
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
176
177
178
179
181
182
183
184
pair-copula construction for three dimensions. Consider three random variables X X 1 ; X 2 ; X 3 with
marginal distribution functions F 1 ; F 2 and F 3 and corresponding densities. One possible representation
of the joint density is
f x1 ; x2 ; x3 f 1 x1 f x2 jx1 f x3 jx1 ; x2
According to the Sklar theorem, we know that the joint density can be decomposed further into
univariate marginal densities and a copula density. It follows for the conditional density of x2 given
x1 that
185
187
188
f x2 jx1
f x1 ; x2 c1;2 F 1 x1 ; F 2 x2 f 1 x1 f 2 x2
c1;2 F 1 x1 ; F 2 x2 f 2 x2
f 1 x1
f 1 x1
For three random variables X 1 ; X 2 and X 3 , we have
189
f x3 jx1 ; x2
191
f x2 ; x3 jx1 c2;3j1 Fx2 jx1 ; Fx3 jx1 f x2 jx1 f x3 jx1
f x2 jx1
f x2 jx1
c2;3j1 Fx2 jx1 ; Fx3 jx1 c1;3 F 1 x1 ; F 3 x3 f 3 x3
193
Thus, the three-dimensional joint density (1) can be represented in terms of bivariate conditional
copulas and marginal densities
194
196
f x1 ; x2 ; x3 c2;3j1 Fx2 jx1 ; Fx3 jx1 c1;2 F 1 x1 ; F 2 x2 c1;3 F 1 x1 ; F 3 x3 f 1 x1 f 2 x2 f 3 x3
192
204
For high-dimensional distributions, there are a large number of possible pair-copula decompositions. Therefore, Bedford and Cooke (2001) introduced a graphical model called regular vine to help
organize them. In this paper, we concentrate on two special cases of regular vines; the C- and Dvine (Kurowicka & Cooke, 2004). In a canonical vine structure, each tree has a unique node that is connected to all other nodes of the tree. The intuition behind this is that one variable plays an essential
role in the dependency structure, thus all other variables are connected to it. On the other hand, Dvines are uniquely characterized through their first tree which has a path structure. Therefore the
order of variables in the first tree defines the complete D-vine tree sequence.
205
3.2. Sequential estimation method
206
Having decided the structure of R-vine to be used and the copula families for each pair and conditional pair of variables, the parameters are estimated sequentially starting from the first tree via maximum likelihood (see, e.g., Czado, Min, Baumann, & Dakovic, 2009). The log-likelihood function for the
C-vine copula is given by
197
198
199
200
201
202
203
207
208
209
210
lCV hCV ju
212
213
214
N X
d1 X
di
X
logci;ijj1:i1 F ij1:i1 ; F ijj1:i1 jhi;ijj1:i1
k1 i1 j1
where hCV denotes the parameter set of the C-vine copula, F jji1 :im : Fukj juk;i1 ;...; uk;im and the marginal
distributions are uniform. Similarly, the log-likelihood function for the D-vine copula is
215
lDV hDV ju
217
218
219
N X
d1 X
di
X
logcj;jijj1:ji1 F jjj1:ji1 ; F jijj1:ji1 jhj;jijj1:ji1
k1 i1 j1
The parameters of the C- and D-vine copulas are estimated using maximum likelihood estimation
method.
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
6
220
4. Data and results
221
4.1. Data and stochastic properties
222
In this section, we empirically investigate the relationship between oil, stock and currency markets
over the period from January 4, 2000 to May 31, 2013. Daily spot prices on West Texas Intermediate
(WTI) are used to represent the energy market, since this benchmark is closely related to other crude
oil markers such as those for Brent and Dubai. For stock markets, the Dow Jones Industrial Average
(DJIA) index was chosen to represent the US stock market as it is a broad-based stock index. Moreover,
exchange rate corresponds to the trade weighted US dollar (TWEXB) index, measuring the movement
of dollar against the currencies of a broad group of major U.S. trading partners. Higher values of the
TWEXB index indicate an appreciation of the US dollar.
For our analysis, we consider log-returns that are computed as rt lnP t =Pt1 from the original
price series. The time-paths of return series over the study period are plotted in Fig. 2. According to
the ADF and PP tests, the logarithmics series are all stationary at the 1% significance level. The descriptive statistics of the log-return data are presented in Table 1. We can see that the average log-return is
positive for the crude oil and stock index. Regarding the variance, the WTI crude oil has the highest
variability compared to the other two variables. Furthermore, skeweness values are negative for crude
oil and US dollar index and positive for DJIA index indicating that it is more likely to observe large negative returns on crude oil and currency markets. Return series are leptokurtically distributed in view of
significant excess kurtosis. These findings clearly show that return series depart from normality and
that the probability of extremely negative and positive realizations for our returns is thus higher than
that of a normal distribution. The departure from normality is confirmed by the JarqueBera test. The
LjungBox Q-statistics of order 12 show the existence of autocorrelation in all return series. Moroever,
the LjungBox statistics of order 12 applied to squared returns are highly significant. Finally, the
results of the Lagrange Multiplier test for conditional heteroscedasticity point to the presence of ARCH
effects in the return data, thus supporting our decision to use a GARCH model to filter the daily
returns.
We first filter the returns using the GJR-GARCH model proposed by Glosten, Jagannathan, and
Runkle (1993) which has several advantages over standard GARCH model.1 The aim is to obtain
approximately i.i.d series suitable for copula estimation, while controlling the effects of conditional
heteroskedasticity and asymmetry. The resulting filtered returns are then transformed into uniform variates by applying the probability integral transform to each marginal distribution. The scatterplot of the
copula data (uniform variates) and the corresponding contour plots with standard normal margins are
displayed in Fig. 3. The estimated Kendalls tau are equals to 0.083, 0.16 and 0.082 for the WTIDJIA, WTI-TWEXB and DJIA-TWEXB pairs respectively. It follows that the dependence is negative for
two pairs: WTI-TWEXB and DJIA-TWEXB and positive for only one pair, the WTI-DJIA.
Dimann, Brechmann, Czado, and Kurowicka (2013) suggest selecting the vine structure using
maximum spanning trees with absolute values of pairwise Kendalls taus as weights. Using this tree
selection algorithm suggests to choose the variable WTI as root in the C-vine. The node order of the
first tree is determined as the following: WTI, TWEXB and DJIA. In the next step, adequate paircopula families associated with the C-vine structure selected in the previous step have to be identified.
We select a copula family among the Gaussian, Student-t, Clayton, Gumbel, Frank, Joe, BB1, BB6, BB7,
BB8, Survival Clayton, Survival Gumbel, Survival Joe, Survival BB1, Survival BB6, Survival BB7 and Survival BB8 copula, which cover a wide range of dependence structures.2 For pairs with negative dependence the choice of a copula model is limited to the Gaussian, Student-t, Frank and rotated version of the
Clayton, Joe, BB1, BB6, BB7 and BB8 copulas. The selection of bivariate copula models is based on the AIC
and the BIC information criterions corrected for the numbers of parameters used in the models (Manner,
2007; Brechmann, 2010). Since the choice of copula models in the first tree have a greatest influence on
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
1
The optimal lag length for the conditional mean and variance processes of the GJR-GARCH model was determined with respect
to the AIC and BIC.
2
Following Joe (1997), the two-parameters copulas namely ClaytonGumbel, JoeGumbel, JoeClayton and JoeFrank are
simply referred as BB1, BB6, BB7 and BB8, respectively.
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
7
-0.15
0.05
WTI crude oil
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2008
2009
2010
2011
2012
2013
2009
2010
2011
2012
2013
-0.08
0.08
DJIA index
2000
2001
2002
2003
2004
2005
2006
2007
-0.020
0.015
USD trade-weighted index
2000
2001
2002
2003
2004
2005
2006
2007
2008
Fig. 2. Daily returns on crude oil, stock index and USD trade-weighted index.
Table 1
Descriptive statistics.
Panel A
WTI
DJIA
TWEXB
Panel B
WTI
DJIA
TWEXB
Min
Mean
Max
Std Dev
Skewness
Ex. kurtosis
0.171
0.082
0.023
3.770e04
9.367e05
3.397e05
0.164
0.105
0.017
0.025
0.012
0.003
0.267
0.006
0.034
4.697
7.422
4.089
Q12
Q2(12)
J-B
ARCH12
40.796
61.078
32.968
1051.918
3306.721
1316.654
3149.051
7766.256
2356.926
413.988
976.909
492.051
Notes: The table displays summary statistics for daily crude oil, DJIA stock index and TWEXB returns. The sample period is from
January 4, 2000 to May 31, 2013. Q(12) and Q^2(12) are the LjunkBox statistics for serial correlation in returns and squared
returns for order 12. JB is the empirical statistic of the JarqueBera test for normality. ARCH is the Lagrange multiplier test for
autoregressive conditional heteroskedasticity.
Indicates the rejection of the null hypotheses of no autocorrelation, normality and homoscedasticity at the 1% level of
significance.
267
268
269
270
271
272
273
the global fit of the R-vine model, we apply two goodness of fit tests based on scoring approach of Vuong
(1989) and Clarke (2007). Both tests are likelihood-ratio-based tests for model selection using the KullbackLeibler information criterion. The results of the scoring test, the AIC and the BIC suggest choosing
the Student-t copula for all the pairs of the first tree. Furthermore, we also look at the k-function introduced by Genest and Rivest (1993) to check the adequacy of the selected bivariate copula family. Comparing empirical to theoretical k-functions in Fig. 4, we can see that the Student-t copula fits the
empirical data of the two pairs WTI-TWEXB and WTI-DJIA remarkably well.
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
8
Fig. 3. Pairs plot of the copula data formed from the transformed standardized residual returns with scatter plots (top, right)
and the corresponding contour plots (bottom, left).
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
Having selected adequate copula families for all variable pairs, we estimate the corresponding copula parameters using the sequential method. A preliminary bivariate independence test based on Kendalls tau (Genest & Favre, 2007) is performed to identify possible independent conditional variable
pairs. If the p-value of the test is larger than 5% then the independence copula is chosen. Otherwise
the sequential estimation method is left unchanged. To improve the estimation results, the parameters obtained from the sequential method are used as starting values to determine the corresponding
MLE estimates. Results of the parameters estimation are displayed in Table 2. We can see that all estimated parameters of the conditional and unconditional copulas are significant at 1% significance level.
The strongest negative dependence is between WTI and TWEXB as shown by Kendalls tau value. The
tail dependence estimates show that, in the Student-t copula, the unconditional pair WTI-DJIA shows
strong dependence in the tails. The dependence in the tail is also symmetric for the three pairs. Similarly, we fitted a D-vine copula model to the return data and reported the results in Table 3. As noted
above, the Student-t copula provide the best fit for all conditional and unconditional pairs. The C- and
D-vine structures share one unconditional pair in common, WTI-TWEXB and produce identical Kendalls tau and tail dependence estimates for this pair. The dependence is also negative between TWEXB
and DJIA. For the tail dependence measures, the D-vine specification substantially shows stronger
lower and upper tail dependence in the conditional pair WTI-DJIATWEXB. Thus, we can conclude
that given the TWEXB as the condition reduce by more than half the lower and upper tail dependence
between WTI and DJIA, i.e., the information of currency market can help investors reduce significantly
the tail dependence between stock and oil markets.
In order to compare the two fitted vine-copula models, we calculate the loglikelihood, AIC, BIC and
p-values for Vuong (1989) test in Table 4. According to the loglikelihood, Akaike and Bayesian Information criteria, the C-vine copula model produces better fit, with little difference between the two
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
0.4
0.2
0.0
WTITWEXB
0.6
(v)
0.0
0.2
0.4
0.6
0.8
1.0
0.8
1.0
0.2
0.4
(v)
0.0
WTIDJIA
0.0
0.2
0.4
0.6
v
Fig. 4. The empirical lambda function (black line) and its confidence bands (dashed lines) corresponding to independence and
comonotonicity (k 0) are presented together with the fitted lambda functions for the Student-t copula (grey line) for the two
pair of the first tree.
Table 2
C-vine copula estimation results.
Copula
Parameters (SE)
hTWEXBWTI
Student-t
hDJIAWTI
Student-t
hDJIATWEXBjWTI
Student-t
0.242
0:017
0:127
(0.018)
0:114
(0.018)
Kendalls
m 12:702
(3.039)
m 7:978
(1.245)
m 14:964
(4.407)
Tail dependence
0.156
kU kL 0:033e02
0.081
kU kL 2:707e02
0.072
kU kL 0:038e02
Notes: The table summarizes the C-vine copula estimation results over the overall sample. The values in parenthesis represent
the standard error of the parameters.
Indicates significance at the 1% level.
Table 3
D-vine copula estimation results.
Copula
Parameters (SE)
hWTITWEXB
Student-t
hTWEXBDJIA
Student-t
hWTIDJIAjTWEXB
Student-t
0:242
(0.017)
0:135
(0.018)
0:093
(0.018)
Kendalls
m 12:702
(3.039)
m 9:539
(1.862)
m 9:991
(1.884)
Tail dependence
0.156
kU kL 0:033e02
0.086
kU kL 0:36e02
0.059
kU kL 1:166e02
Notes: The table summarizes the D-vine copula estimation results over the overall sample. The values in parenthesis represent
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
10
Table 4
Comparison of the C-vine and D-vine.
LogLik
AIC
BIC
Vuong test
C-vine
D-vine
185.68
359.363
322.583
0.427
184.42
356.839
320.059
Notes: The table reports the loglikelihood value, the AIC, the BIC and p-value of the Vuong test for
the C-vine and D-vine copula models.
300
specified vine structures. Under the null hypothesis that the C- and D-vine copula models are statistically equivalent, the Vuong test, failed in distinguishing between the two models. We conclude that
the both vine specifications are suitable for describing the multivariate dependence between returns
and can provide additional insights due to their specific structures.
301
4.2. Structural changes in crude oil
302
In this section, we will investigate whether structural changes have occurred in the dynamic
relationship between oil, stock and currency markets. We only deal with the C-vine copula which is
simpler and more comprehensive. Since crude oil is selected as root variable for all unconditional
pair-copulas in the C-vine, we attempt to identify changes in the structure of crude oil prices and
when these changes occur. Moreover, we try to identify what implications these change have on
the dependence between the three variables of interest.
To test whether crude oil data contain one or more structural break, we considered the Bai and
Perron (2003) test allowing us to test for an unknown number of breakpoints at unknown dates.
The results of the test indicate that there are five potential breaks at the following dates:
18/01/2002, 26/10/2004, 18/01/2007, 12/02/2009 and 08/04/2011.3 In fact, oil price spikes after major
world events such as the Afghanistan war in 2002, the Great Recession over20072009, and the European
Debt crisis in 2011. To examine the potential impact of oil shocks on market interdependencies, we
divide our study period into six sub-periods as follow: from 4 January 2000 to 17 January 2002, from
18 January 2002 to 25 October 2004, from 26 October 2004 to 17 January 2007, from 18 January 2007
to 11 February 2009, from 12 February to 7 April 2011 and from 8 April 2011 to 31 May 2013. The
C-vine copula model is then estimated separately for each sub-period under the assumption that WTI
is the root variable. Estimation results are reported in Table 5.
The reported results show that the tail dependence, the level and the structure of dependence are
changing between pre-crisis and post-crisis periods. In fact, it can be seen that all pairs are independents during the first sub-period. After the financial crisis of 2009, we notice that all conditional and
unconditional pairs became significantly dependents.
In particular, the dependence between WTI and TWEXB is significantly negative for all sub-periods
except the first one. The negative dependence between the two variables reaches its lower level, during the post-global financial crisis, over 20092011. The Rotated Gumbel copula (90) provides the
best description of the dependence structure over this sub-period. Recall that the rotation by 90
and 270 degrees allows for the modeling of negative dependence.
The dependence between WTI and DJIA became apparent after the financial crisis of 2009. Kendalls
tau values are positives and equal to 0.318 and 0.358. It turns out that an increase in the price of crude
oil is associated with an appreciation of the stock prices. We note that this pair show the highest
degree of tail dependence during bear and bull markets. This is not surprising as we expect that the
tail dependence increases during periods of extreme turbulence.
297
298
299
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
3
A structural break is considered significant if its F-statistic scaled by the number of varying regressors is higher than the Bai
Perron critical value. Note that a constant and a one-period lagged value of the dependent variable are used as explanatory
variables in the linear regression and all the test results can be made available under request addressed to the Corresponding
author.
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
11
Table 5
Sub-periods estimation results.
Copula
4 January 2000 to 17 January 2002
hTWEXBWTI
Indep.
hDJIAWTI
Indep.
Indep.
hDJIATWEXBjWTI
18 January 2002 to 25 October 2004
Rotated Gumbel 90
hTWEXBWTI
hDJIAWTI
hDJIATWEXBjWTI
Indep.
Survival Gumbel
26 October 2004 to 17 January 2007

hTWEXBWTI
Rotated Gumbel 270
hDJIAWTI
Rotated Gumbel 270
hDJIATWEXBjWTI
Indep.
18 January 2007 to 11 February 2009

hTWEXBWTI
Gaussian
hDJIAWTI
hDJIATWEXBjWTI
Indep.
Frank
12 February 2009 to 8 April 2011

hTWEXBWTI
Rotated Gumbel 90
hDJIAWTI
Student-t
hDJIATWEXBjWTI
Rotated BB7 90
8 April 2011 to 31 May 2013

hTWEXBWTI
Student-t
hDJIAWTI
BB1
hDJIATWEXBjWTI
Frank
Parameters (SE)
Kendalls
-1.078
(0.026)
1.066
(0.025)
-0.072
0.062
1.108
(0.032)
1.086
(0.031)
Tail dependence
kL kU 0
kL 0:084; kU 0
0.098
kL kU 0
0.079
kL kU 0
0.307
(0.035)
0.537
(0.266)
0.198
kL kU 0
0.059
kL kU 0
1.500
(0.050)
0.479
(0.034)
1.175
(0.056)
0.333
kL kU 0
0.334
(0.043)
0.532
(0.104)
1.334
(0.262)
6.057
(1.591)
0.276
(0.062)
7.535
(2.726)
1.231
(0.065)
0.318
kL kU 0:158
0.191
kL kU 0
0.217
kL kU 0:003
0.358
0.146
kL 0:347; kU 0:244
kL kU 0
Notes: The table summarizes the C-vine copula estimation results over the four subperiods. The values in parenthesis represent
336
The conditional pair DJIA-TWEXBWTI shows a relatively small degree of dependence and fluctuates within a range of 0.191 and 0.062. The tail dependence coefficients are equal to zero in most
cases. We conclude that the information of oil market can help investors reducing significantly the tail
dependence between stock and currency markets.
337
4.3. Value at risk
338
In this section, we illustrate the use of the C-vine copula model in quantifying the risk of an equally
weighted portfolio, composed of the WTI, TWEXB and DJIA returns. We indeed consider the Value-atRisk (VaR) as the portfolios market risk measure and estimate it using Monte Carlo simulations.
Value-at-Risk is one of the most popular measures for market risk assessment, defined by the maximum loss in a portfolios value with a given probability over a given time period.
333
334
335
339
340
341
342
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
12
Table 6
VaR backtesting results
Normal
Historical Simulation
C-vine copula
Gaussian copula
Student-t copula
Expected number
Exceedences
1a
Kupiec test
16.97
16.97
16.97
16.97
16.97
47
27
22
24
23
0.01
0.01
0.01
0.01
0.01
5.098e-09
0.041
0.163
0.106
0.106
Notes: The table reports the VaR backtesting results obtained from the C-vine copula, Gaussian copula, Student-t copula, the
historical simulations and the normal method. It also presents the p-values for the Kupiec (1995) test of unconditional coverage.
364
Methodologically, our procedure for computing the VaR requires the following steps. First, we estimate the whole model (GJR-GARH + C-vine copula) using a window of 1697 observations. Then, we
simulate 10.000 random trials of uniform variates from the C-vine copula and transform them into
standardized residuals by inverting the marginal CDF of each series. Finally, we reintroduce the autocorrelation and heteroskedasticity observed in the original return series, compute the value of the considered portfolio and estimate the VaR. This procedure is repeated until the last observation, and we
compare the estimated VaR with the actual next-day change in the portfolios value. The estimation
and simulation from the C-vine copula are repeated only once in every 50 observations owing to
the computational cost of this procedure. However, at each new observation, the VaR estimates are
modified because of changes in the GJR-GARCH parameters.
For comparison purposes, we also estimate the VaR using four other approaches: the multivariate
gaussian copula, the multivariate Student-t copula, the normal and the historical simulation methods.
For the normal and historical simulation methods, the model parameters are updated for every observation. The results for the backtesting are reported in Table 6. Note that a model is said to be best suited for calculating VaR is the one with the number of exceedances closest to the expected number of
exceedances. We can see that The C-vine copula provides better VaR forecasts at the 99% confidence
level followed by the multivariate Student-t and Gaussian copulas. We also use the Kupiec (1995) proportion of failures (POF) test to check the robustness of the VaR estimates. According to the backtest
results, all copula based models perform well for the considered portfolio. However, the accuracy of
the VaR estimates from the historical simulation and normal methods are rejected, since the pvalues are inferior to the 5% significance level. Overall, our results thus confirm the relevance of the
C-vine copula model.
365
5. Conclusion
366
In contrast to previous literature, this paper investigates the multivariate dependence between
crude oil, exchange rate and stock returns using a vine copula approach. We first employ a
GJR-GARCH model with skewed-t distribution to filter the return series and construct their marginal
distributions. The C- and D-vine copulas are then fitted to filtered returns and their suitability is
compared. After detecting structural change in crude oil returns, the C-vine copula model is estimated
separately for each sub-period to identify what implications these changes have on the dependence
between the three markets. We finally show the implications of the empirical findings for risk management issues related to an equally weighted oil, stock and exchange rate portfolio within a VaR
framework.
Our results show that both vine specifications are well suited for modeling the multivariate dependence between the return series over the entire sample period. It can also be concluded that an
increase of crude oil price is associated with a depreciation of exchange rate and an appreciation of
stock market prices. Furthermore, given the information of exchange rate can help investors and portfolio managers reducing significantly the extreme dependence between oil and stock returns. The subperiod estimation results show that the level and the structure of dependence are not constant over
time. More importantly, the multivariate dependence between the considered return series is highly
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
13
387
affected by the financial crisis and Great Recession, over 20072009. We finally find that the C-vine
copula model leads to more accurate VaR forecasts than the traditional VaR approaches.
Future extensions of this work could focus on a much more flexible analysis of the high dimensional dependence by allowing the vine copula parameters to be dynamic and switch between different regimes. Our methodology can also be used in the context of computing optimal portfolio
allocation weights and optimal hedging ratios.
388
Acknowledgement
389
390
We are grateful to two anonymous referees for their constructive comments and suggestions. As
usual, all remaining errors are ours.
391
References
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
Aloui, R., Ben Assa, M. S., & Nguyen, D. K. (2013). Conditional dependence structure between oil prices and exchange rates: A
copula-GARCH approach. Journal of International Money and Finance, 32(2), 719738.
Amano, R., & van Norden, S. (1998). Oil prices and the rise and fall of the US real exchange rate. Journal of International Money
and Finance, 17(2), 299316.
Arouri, M. E. H., Lahiani, A., & Bellalah, M. (2010). Oil price shocks and stock market returns in oil-exporting countries: The case
of GCC countries. International Journal of Economics and Finance, 2(5), 132139.
Arouri, M. E. H., & Nguyen, D. K. (2010). Oil prices, stock markets and portfolio investment: Evidence from sector analysis in
Europe over the last decade. Energy Policy, 38, 45284539.
Aydin, L., & Mustafa, A. (2011). Economic impact of oil price shocks on the Turkish economy in the coming decades: A dynamic
CGE analysis. Energy Policy, 39, 17221731.
Bai, J., & Perron, P. (2003). Computation and analysis of multiple structural change models. Journal of Applied Econometrics, 18,
122.
Basher, S. A., & Sadorsky, P. (2006). Oil price risk and emerging stock markets. Global Finance Journal, 17, 224251.
Bedford, T. J., & Cooke, R. M. (2001). Probability density decomposition for conditionally dependent random variables modeled
by vines. Annals of Mathematics and Artificial Intelligence, 32, 245268.
Bedford, T. J., & Cooke, R. M. (2002). Vines-a new graphical model for dependent random variables. Annals of Statistics, 30(4),
10311068.
Bjornland, C. H. (2009). Oil price shocks and stock market booms in an oil exporting country. Scottish Journal of Political Economy,
2(5), 232254.
Boyer, M. M., & Filion, D. (2007). Common and fundamental factors in stock returns of Canadian oil and gas companies. Energy
Economics, 29, 428453.
Brechmann, E. C. (2010). Truncated and simplified regular vines and their applications (Diploma Thesis). Technische Universitt
Mnchen.
Chen, S. S., & Chen, H. C. (2007). Oil prices and real exchange rates. Energy Economics, 29(3), 390404.
Chiou, J. S., & Lee, Y. H. (2009). Jump dynamics and volatility: Oil and the stock markets. Energy, 34(6), 788796.
Clarke, K. A. (2007). A simple distribution-free test for nonnested model selection. Political Analysis, 15, 347363.
Czado, C., Min, A., Baumann, T., & Dakovic, R. (2009). Pair-copula constructions for modeling exchange rate dependence.
Technical report. Technische Universitt Mnchen.
Ding, L., & Vo, M. (2012). Exchange rates and oil prices: A multivariate stochastic volatility analysis. The Quarterly Review of
Economics and Finance, 52, 1537.
Dimann, J., Brechmann, E. C., Czado, C., & Kurowicka, D. (2013). Selecting and estimating regular copulae and application to
financial returns. Computational Statistics and Data Analysis, 59(1), 5269.
Fayyad, A., & Daly, K. (2011). The impact of oil price shocks on stock market returns: Comparing GCC countries with the UK and
USA. Emerging Markets Review, 12, 6178.
Genest, C., & Favre, A. C. (2007). Everything you always wanted to know about copula modeling but were afraid to ask. Journal of
Hydrologic Engineering, 12(4), 347368.
Genest, C., & Rivest, L.-P. (1993). Statistical inference procedures for bivariate archimedean copulas. Journal of the American
Statistical Association, 88, 10341043.
Glosten, L. R., Jagannathan, R., & Runkle, D. E. (1993). On the relation between the expected value and the volatility of the
nominal excess return on stocks. Journal of Finance, 48, 17791801.
Golub, S. (1983). Oil prices and exchange rates. The Economic Journal, 93, 576593.
Hammoudeh, S., Dibooglu, S., & Aleisa, E. (2004). Relationships among US oil prices and oil industry equity indices. International
Review of Economics and Finance, 13, 427453.
Huang, R. D., Masulis, R. W., & Stoll, H. R. (1996). Energy shocks and financial markets. Journal of Future Markets, 16, 127.
Jimenez-Rodriguez, R., & Sanchez, M. (2005). Oil price shocks and real GDP growth: Empirical evidence for some OECD
countries. Applied Economics, 37(2), 201228.
Joe, H. (1997). Multivariate models and dependence concepts.London: Chapman & Hall/CRC.
Joe, H., & Xu, J. (1996). The estimation method of inference functions for margins for multivariate models. Technical Report 166.
Department of Statistics, University of British Columbia.
Jondeau, E., & Rockinger, M. (2006). The copula-GARCH model of conditional dependencies: An international stock market
application. Journal of International Money and Finance, 25(5), 827853.
Jones, C., & Kaul, G. (1996). Oil and the stock markets. Journal of Finance, 51, 463491.
382
383
384
385
386
org/10.1016/j.najef.2016.05.002
ECOFIN 595
17 May 2016
14
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
Junker, M., Szimayer, A., & Wagner, N. (2006). Nonlinear term structure dependence: Copula functions, empirics, and risk
implications. Journal of Banking and Finance, 30, 11711199.
Krugman, P. (1983). Oil shocks and exchange rate dynamics. Exchange rates and international macroeconomics. University of
Chicago Press.
Kupiec, P. (1995). Techniques for verifying the accuracy of risk measurement models. Journal of Derivatives, 2, 7384.
Kurowicka, D., & Cooke, R. M. (2004). Distribution-free continuous bayesian belief nets. In Fourth international conference on
mathematical methods in reliability methodology and practice. Santa Fe, New Mexico.
Manner, H. (2007). Estimation and model selection of copulas with an application to exchange rates. METEOR research
memorandum 07/056. Maastricht University.
McNeil, A., Frey, R., & Embrechts, P. (2005). Quantitative risk management: Concepts, techniques and tools. Princeton University
Press.
Miller, J. I., & Ratti, R. A. (2009). Crude oil and stock markets: Stability, instability, and bubbles. Energy Economics, 31(4),
559568.
Nandha, M., & Faff, R. (2008). Does oil move equity prices? A global view. Energy Economics, 30, 986997.
Narayan, P. K., Narayan, S., & Prasad, A. (2008). Understanding the oil price-exchange rate nexus for the Fiji islands. Energy
Economics, 30, 26862696.
Papapetrou, E. (2001). Oil price shocks, stock markets, economic activity and employment in Greece. Energy Economics, 23,
511532.
Park, J. W., & Ratti, R. A. (2008). Oil price shocks and stock markets in the U.S. and 13 European countries. Energy Economics, 30
(5), 25872608.
Rogoff, K. (1991). Oil, productivity, government spending and the real yendollar exchange rate. FRB of San Francisco, 1991,
Working Papers 91-06.
Sadorsky, P. (2001). Risk factors in stock returns of Canadian oil and gas companies. Energy Economics, 23, 1728.
Sanchez, M. V. (2011). Welfare effects of rising oil prices in oil-importing developing countries. Developing Economies, 49(3),
321346.
Vuong, Q. H. (1989). Ratio tests for model selection and non-nested hypotheses. Econometrica, 57(2), 307333.
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, 270282.
Zhang, Y. J., Fan, Y., Tsai, H. T., & Wei, Y. M. (2008). Spillover effect of US dollar exchange rate on oil prices. Journal of Policy
Modeling, 30, 973991.
474
org/10.1016/j.najef.2016.05.002

Exchange Rates: A Vine Copula Based GARCH Method Relationship Between Oil, Stock Prices and

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Exchange Rates: A Vine Copula Based GARCH Method Relationship Between Oil, Stock Prices and

Încărcat de

Drepturi de autor:

Formate disponibile

ECOFIN 595

No. of Pages 14, Model 1G

Contents lists available at ScienceDirect

North American Journal of

Relationship between oil, stock prices and

Riadh Aloui a, Mohamed Safouane Ben Assa b,

No. of Pages 14, Model 1G

No. of Pages 14, Model 1G

Fig. 1. Dynamics of daily prices of West Texas Intermediate (US Dollar/Barrel).

2. Relationships between oil, stock prices and exchange rates

No. of Pages 14, Model 1G

3.1. Pair-copula construction

No. of Pages 14, Model 1G

For three random variables X 1 ; X 2 and X 3 , we have

f x2 ; x3 jx1 c2;3j1 Fx2 jx1 ; Fx3 jx1 f x2 jx1 f x3 jx1

c2;3j1 Fx2 jx1 ; Fx3 jx1 c1;3 F 1 x1 ; F 3 x3 f 3 x3

f x1 ; x2 ; x3 c2;3j1 Fx2 jx1 ; Fx3 jx1 c1;2 F 1 x1 ; F 2 x2 c1;3 F 1 x1 ; F 3 x3 f 1 x1 f 2 x2 f 3 x3

3.2. Sequential estimation method

logci;ijj1:i1 F ij1:i1 ; F ijj1:i1 jhi;ijj1:i1

logcj;jijj1:ji1 F jjj1:ji1 ; F jijj1:ji1 jhj;jijj1:ji1

No. of Pages 14, Model 1G

4. Data and results

4.1. Data and stochastic properties

No. of Pages 14, Model 1G

WTI crude oil

USD trade-weighted index

No. of Pages 14, Model 1G

No. of Pages 14, Model 1G

Indicates significance at the 1% level.

Indicates significance at the 1% level.

No. of Pages 14, Model 1G

4.2. Structural changes in crude oil

No. of Pages 14, Model 1G

26 October 2004 to 17 January 2007

Rotated Gumbel 270

18 January 2007 to 11 February 2009

12 February 2009 to 8 April 2011

Rotated BB7 90

8 April 2011 to 31 May 2013

Indicates significance at the 1% level.

4.3. Value at risk

No. of Pages 14, Model 1G

No. of Pages 14, Model 1G

No. of Pages 14, Model 1G

S-ar putea să vă placă și