Paper VolatilityTransmission
Paper VolatilityTransmission
Paper VolatilityTransmission
=
+ u + =
t t t
t t t
D
Y Y
1
(1)
where
) , ( ' =
o
t
s
t t
r r Y with
s
t
r and
o
t
r being the returns on stock market and oil price
indices at time t respectively;
u is a (22) matrix of coefficients of the form
|
.
|
\
|
= u
2
1
0
0
;
) , ( ' =
o
t
s
t t
with
s
t
and
o
t
being the residual terms from the mean equa-
tions of stock market and oil returns respectively;
) , ( ' =
o
t
s
t t
refers to a (21) vector of independently and identically distri-
buted errors;
2
The optimal number of lags for the VAR system was chosen on the basis of commonly-used infor-
mation criteria.
9
and ) , (
o
t
s
t t
h h diag D = with
s
t
h and
o
t
h being the conditional variances of
s
t
r and
t
o
r respectively. Their time-series dynamics are modeled in Equations
(2) and (3) as:
2
1
2
2 1
2
2
2
1
2
1 1
2
1
2
) ( ) (
o
t s
o
t s
s
t s
s
t s s
s
t
h h C h
+ + + + = (2)
2
1
2
2 1
2
2
2
1
2
1 1
2
1
2
) ( ) (
s
t o
s
t o
o
t o
o
t o o
o
t
h h C h
+ + + + = (3)
As it can be seen, the conditional variance of the stock market (resp., oil mar-
ket) depends not only on its own pasts and innovations, but also on those of the oil
market (resp., stock market). This particular feature thus permits the direct transmis-
sion of volatility and shocks from one market to another. We can also express the
conditional covariance,
so
t
h , as follows:
o
t
s
t
so
t
h h h = (4)
where is the conditional constant correlation.
3
Taken together, the characteristics of the proposed model allow us to capture
both return and volatility spillover effects between oil and stock markets. To the ex-
tent that the normality condition is often rejected for the majority of macroeconomic
and financial series, we use the quasi-maximum likelihood (QML) method to esti-
mate the models parameters.
4
In Section 5, we will show how the estimation results
from our VAR(1)-GARCH(1,1) model can be used to compute the optimal weights
and hedge ratios of an oil-stock diversified portfolio.
3
It is worth emphasizing that the model VAR-GARCH with dynamic conditional correlations has not
been analysed theoretically yet (McAleer et al., 2009).
4
See Ling and McAleer (2003) for further details about the asymptotic properties of the VAR-
GARCH model and its estimation procedure.
10
4. Data and preliminary results
Our study examines the return and volatility linkages for the six countries members
of the GCC (Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, and UAE) over the period
from June 7, 2005 to February 21, 2010.
5
Stock market indices are obtained from
MSCI database, while world oil price data are extracted from the Energy Information
Administration (EIA). Note that the Brent spot prices are used to represent the inter-
national crude-oil market since they usually serve as reference prices for pricing
crude oil and many other derivatives products using oil as underlying asset. Unlike
the majority of previous studies which employ low frequency data (yearly, quarterly,
monthly, and weekly), we use daily data in order to adequately capture the rapidity
and intensity of the dynamic interactions between oil and stock prices in the GCC re-
gion. All price data are denominated in US dollars to avoid the impacts of exchange
rates and to ease the comparison across countries. Daily returns are calculated from
daily price data by taking the natural logarithm of the ratio of two successive prices.
The statistical and stochastic properties of the data are summarized in Table 1.
In Panel A, we test for the presence of unit roots in the levels and first differ-
ences of oil and stock market price indices. The results from the Augmented Dickey
Fuller (ADF) tests indicate that the price series are integrated of order one or I(1),
while the returns series are stationary. We report, in Panel B, basic statistics of log
returns series. Average daily returns on GCC stock market indices are all negative
over our sample period under the effects of the recent global financial crisis 2007-
2009, sparked by the US subprime crisis. Stock market in Bahrain realized the worst
performance (-0.121%), followed by those in the UAE and Saudi Arabia. Inversely,
oil market experienced a positive average return, which is not surprising in view of
5
The data used in the majority of previous studies predate the end of 2005, and as a result they missed
the spectacular variations that have occurred in the GCC and oil markets over the last three years.
11
the increasing trend in the price of oil over the last decade. Skewness is negative for
all stock markets, and positive for oil market. This means that extreme negative and
positive returns are likely to realize for stock and oil markets respectively. Kurtosis
coefficients are important in size and highly significant, indicating that outliers may
occur with a probability higher than that of a normal distribution. Accordingly, the
Jarque-Bera test statistics strongly reject the null hypothesis of normality for all se-
ries.
Table 1
Descriptive statistics
Panel A: Augmented Dickey-Fuller (ADF) stationarity tests
Levels First difference
Bahrain -2.536
**
(a) -32.621
***
(c)
Kuwait -1.703 (b) -32.891
***
(a)
Oman -0.323 (a) -31.399
***
(a)
Saudi Arabia -0.776 (a) -33.879
***
(a)
UAE -1.231 (a) -31.186
***
(a)
Qatar -1.434 (a) -32.689
***
(a)
Oil -1.916 (c) -35.222
***
(a)
Panel B: Basic statistics of return series
Bahrain Kuwait Oman
Saudi
Arabia
UAE Qatar Oil
Mean (%) -0.121 -0.002 -0.015 -0.043 -0.081 -0.022 0.037
Max. (%) 11.502 8.752 10.750 16.226 18.326 11.258 16.413
Min. (%) -17.983 -14.871 -17.644 -21.570 -15.497 -13.171 -12.826
Std. dev 1.588 1.783 1.634 2.244 2.207 2.008 2.768
Skewness -1.978 -1.249 -1.479 -1.508 -0.754 -0.563 0.104
Kurtosis 27.293 14.262 24.017 18.291 14.499 10.261 7.394
Jarque-Bera 30519
***
6704
***
22692
***
12238
***
6776
***
2707
***
975
***
ARCH(20) 64.2
***
40.1
***
76.7
***
39.6
***
35.7
**
38.4
***
59.1
***
Correlation
with Brent oil
0.061 0.018 0.026 0.024 0.10 0.373 -
Notes: ADF test is carried out using log prices and returns series. (a) designates a model with neither
constant nor deterministic trend; (b) a model with constant and without deterministic trend; and (c)
model with constant and deterministic trend.
*
,
**
and
***
denote rejection of the null hypothesis re-
spectively at the 10%, 5% and 1% levels, respectively.
As also shown in Panel B, all the return series were found to have a leptokurtic
behavior (i.e., their distributions have fatter tails than corresponding normal distribu-
tions). This suggests that each of the mean equations should be tested for the exis-
tence of conditional heteroscedasticity. By applying the Engle (1982)s test, we ob-
serve that the null hypothesis of ARCH effects cannot be rejected at conventional le-
12
vels in all cases, thus confirming that GARCH modeling is adequate for capturing
any persistence in the financial volatility of stock and oil markets we consider.
We also compute the unconditional correlations between GCC stock market re-
turns and oil price changes. The results reveal that cross-market correlations are not
high, but on average positive. These correlations range from 0.018 (Kuwait-Brent) to
0.373 (Qatar-Brent). This finding indicates that the presence of oil asset in a portfolio
may lead to greater diversification benefits.
5. Empirical results and portfolio implications
In what follows, we will first discuss our findings related to return linkages and vola-
tility transmission between oil and stock markets in the GCC countries within the
empirical framework of the VAR(1)-GARCH(1,1) model. Recall that the appropriate
lag length for the VAR system was determined by the usual information criteria (BIC
and SIC). We will then use the estimation results to compute the optimal weights of
an oil-stock portfolio as well as the optimal hedge ratios.
5.1 Return and volatility dependencies
Our VAR(1)-GARCH(1,1) estimation results are reported in Tables 2-7 for the six
oil-stock market pairs. Regarding the interdependence of returns in mean equations,
we find that lagged oil price returns significantly affect stock market returns in three
out of six cases: Bahrain, Oman, and Qatar. The effect of oil on stock markets is pos-
itive for Oman and Qatar, but negative for Bahrain. Inversely, oil returns are only
significantly related to stock market returns in Bahrain, but the link is negative. The
lagged value of stock market returns is found to be significant in two stock markets
(Bahrain and Qatar), indicating that their returns are predicted from past realizations,
and thus they are not informationally efficient according to the weak-form efficiency
13
(Fama, 1991). Taken together, our findings suggest some evidence of predictability
in the dynamic variations of stock market returns with respect to previous returns on
oil and stock market indices for three GCC countries, Bahrain, Oman and Qatar.
Table 2
Estimates of VAR(1)-GARCH(1,1) for Bahrain
Variables Bahrain Oil
Mean equation
C
Bahrain(1)
Oil(1)
0.7387
***
(0.1209)
0.1743
**
(0.0708)
-0.1914
***
(0.0432)
0.2989
**
(0.1212)
-0.2359
***
(0.0863)
0.0781
*
(0.0457)
Variance equation
C
2
1
) (
s
t
2
1
) (
o
t
s
t
h
1
o
t
h
1
-0.077 (9.7740)
-0.0025 (3.3908)
-0.0373 (0.2574)
-0.9411
***
(0.1179)
0.2772 (0.2907)
0.2979 (0.2326)
-0.0019 (5.0112)
0.1752
***
(0.0404)
0.2830 (0.3308)
0.9567
***
(0.0710)
Constant conditional correlation
Bahrain
Oil
1.0000
0.1918
***
(0.0557) 1.0000
Log likelihood
AIC
SIC
-5253.606
8.7697
8.8417
Notes: The optimal lag order for the VAR model is selected using the SIC information criterion. Stan-
dard errors are given in parenthesis.
*
,
**
and
***
denote rejection of the null hypothesis at the 10%, 5%
and 1% levels, respectively.
As for the estimates of ARCH and GARCH coefficients in the conditional va-
riance equations, they are significant at conventional levels in most cases. The sensi-
tivity to past own conditional volatility (GARCH-term) appears to be significant for
all countries, except for Qatar. This finding typically suggests that past values of the
conditional volatility in a particular stock market can be employed to forecast future
volatility. The results also show that the current conditional volatility of GCC stock
markets also depends on past shocks affecting return dynamics since ARCH-terms
are highly significant for all countries considered, with an exception for Bahrain.
14
Figure 1. Time-variations of conditional volatility for GCC stock markets
A closer inspection of the above coefficients reveals that in general conditional
volatility does not change very rapidly as the ARCH-terms measuring the impact of
past shocks on conditional volatility are relatively small in size. On the other hand,
the GARCH-terms, which capture the impact of past volatility on current volatility,
are substantially large, indicating gradual fluctuations over time. These properties
can be further apprehended through Figure 1 where we plot the time-variations of
conditional volatility for the six stock markets under investigation. It should be final-
ly noted that the similar conclusions hold for oil price volatility.
15
Table 3
Estimates of VAR-GARCH for Kuwait
Variables Kuwait Oil
Mean equation
C
Kuwait(1)
Oil(1)
0.0374 (0.0518)
0.0511 (0.0368)
-0.0096 (0.0257)
0.1390
**
(0.0549)
-0.0359 (0.0437)
-0.0322 (0.0252)
Variance equation
C
2
1
) (
s
t
2
1
) (
o
t
s
t
h
1
o
t
h
1
-0.0004 (26.2900)
-0.0881
***
(0.0304)
-0.0144 (0.1989)
0.9519
***
(0.0242)
0.2180
***
(0.0682)
0.1605 (0.1086)
-0.0925
*
(0.0557)
0.2033
***
(0.0235)
0.2576
*
(0.1502)
0.9538
***
(0.0259)
Constant conditional correlation
Kuwait
Oil
1.0000
0.0685
**
(0.0283) 1.0000
Log likelihood
AIC
SIC
-4904.386
8.1886
8.2606
Notes: The optimal lag order for the VAR model is selected using the SIC information criterion.
Standard errors are given in parenthesis.
*
,
**
and
***
denote rejection of the null hypothesis at the
10%, 5% and 1% levels, respectively.
Turning out to the volatility spillover effects between oil and stock markets in
the GCC countries, we first observe that there is no direct transmission of volatility
from oil market to Bahraini and Qatari stock markets (Tables 2 and 5). For the four
remaining countries (Tables 3-4 and 6-7), the cross-volatility coefficients (return in-
novation and volatility) are significant at conventional levels. More precisely, past oil
shocks have significant effects on stock market volatility in Saudi Arabia, while past
oil volatility strongly affects stock market volatility in Kuwait, Oman and the UAE.
However, the effects of past shocks and past volatility of oil returns on volatility of
stock markets in GCC countries should be moderated since their estimated coeffi-
cients are much smaller than those of past own shocks and volatilities. Oil price vola-
tility is affected, on the one hand, by past market shocks in Kuwait, and on the other
hand by past stock market volatility in Oman and the UAE. Therefore, direct volatili-
ty spillover is more widespread from oil to stock markets in the GCC countries than
16
the inverse case. Investors should then keep a close watch on what is going on in the
oil market in order to make accurate investment decisions.
Table 4
Estimates of VAR-GARCH for Oman
Variables Oman Oil
Mean equation
C
Oman(1)
Oil(1)
0.0021 (0.0809)
0.0700 (0.074)
0.0959
***
(0.0348)
0.0522 (0.0830)
0.0104 (0.0687)
0.0067 (0.0358)
Variance equation
C
2
1
) (
s
t
2
1
) (
o
t
s
t
h
1
o
t
h
1
0.2360 (0.1450)
-0.3730
***
(0.0635)
0.0022 (3.0922)
0.9052
***
(0.0334)
0.2458
***
(0.0784)
0.3226
***
(0.0893)
-0.0265 (0.8341)
0.2252
***
(0.0317)
0.2350
*
(0.1346)
0.9551
***
(0.0180)
Constant conditional correlation
Bahrain
Oil
1.0000
0.3163
***
(0.0425) 1.0000
Log likelihood
AIC
SIC
-4878.672
8.1458
8.2178
Notes: The optimal lag order for the VAR model is selected using the SIC information criterion.
Standard errors are given in parenthesis.
*
,
**
and
***
denote rejection of the null hypothesis at the
10%, 5% and 1% levels respectively.
Table 5
Estimates of VAR-GARCH for Qatar
Variables Qatar Oil
Mean equation
C
Qatar(1)
Oil(1)
-0.0529 (0.3190)
0.2127
**
(0.0913)
0.2006
*
(0.1141)
0.2257 (0.2425)
0.0930 (0.0789)
0.3115
***
(0.0865)
Variance equation
C
2
1
) (
s
t
2
1
) (
o
t
s
t
h
1
o
t
h
1
2.5835 (7.3489)
0.4966
***
(0.1585)
-0.0447 (1.3988)
0.4950 (0.4584)
0.0662 (360.250)
2.4520 (7.6177)
-0.1189 (0.2522)
0.0088 (6.2466)
-0.2344 (1.0028)
0.4814 (4.9435)
Constant conditional correlation
Qatar
Oil
1.0000
0.2804
***
(0.0871) 1.0000
Log likelihood
AIC
SIC
-865.8705
10.4481
10.7630
Notes: The optimal lag order for the VAR model is selected using the SIC information criterion.
Standard errors are given in parenthesis.
*
,
**
and
***
denote rejection of the null hypothesis at the
10%, 5% and 1% levels, respectively.
17
Table 6
Estimates of VAR-GARCH for Saudi
Variables Saudi Arabia Oil
Mean equation
C
Saudi(1)
Oil(1)
0.0506 (0.0511)
-0.0125 (0.0433)
0.0297 (0.0259)
0.1328
**
(0.0585)
-0.0180 (0.0310)
-0.0238 (0.0296)
Variance equation
C
2
1
) (
s
t
2
1
) (
o
t
s
t
h
1
o
t
h
1
0.1434
***
(0.0365)
0.2895
***
(0.0163)
-0.0894
***
(0.0219)
0.9580
***
(0.0046)
-0.0006 (4.6107)
-0.3103
***
(0.0685)
0.0262 (0.1145)
0.2909
***
(0.0213)
0.0004 (8.2154)
0.9485
***
(0.0085)
Constant conditional correlation
Saudi
Oil
1.0000
0.0628
**
(0.0288) 1.0000
Log likelihood
AIC
SIC
-5192.234
8.6676
8.6947
Notes: The optimal lag order for the VAR model is selected using the SIC information criterion.
Standard errors are given in parenthesis.
*
,
**
and
***
denote rejection of the null hypothesis at the
10%, 5% and 1% levels, respectively.
Table 7
Estimates of VAR-GARCH for UAE
Variables UAE Oil
Mean equation
C
UAE(1)
Oil(1)
-0.1629 (0.1040)
0.0651 (0.0459)
0.0323 (0.0404)
0.1113
*
(0.0661)
0.0501 (0.0375)
0.0127 (0.0283)
Variance equation
C
2
1
) (
s
t
2
1
) (
o
t
s
t
h
1
o
t
h
1
0.3593
***
(0.1123)
0.2057
***
(0.0226)
-0.0030 (2.5591)
0.9456
***
(0.0230)
0.2548
**
(0.0992)
0.2577
*
(0.1396)
0.0005 (10.8785)
0.2193
***
(0.0273)
0.2176
**
(0.1011)
0.9421
***
(0.0238)
Constant conditional correlation
UAE
Oil
1.0000
0.1155
***
(0.0370) 1.0000
Log likelihood
AIC
SIC
-5292.651
8.8346
8.9067
Notes: The optimal lag order for the VAR model is selected using the SIC information criterion.
Standard errors are given in parenthesis.
*
,
**
and
***
denote rejection of the null hypothesis at the
10%, 5% and 1% levels, respectively.
As expected, the estimates for constant conditional correlations between oil
and stock markets in the GCC countries are all positive. They are small in general,
18
suggesting the existence of potential gains from investing in both stock markets and
oil.
Summarizing all, the empirical VAR(1)-GARCH(1,1) model appears to satis-
factorily capture the return linkages and volatility transmission for the six pairs of
markets we consider. Stock market returns are significantly influenced by oil returns
in three out of six cases, and they only affect oil returns in Bahrain. The analysis of
volatility interdependence shows significant volatility spillovers from oil to stock
markets, while oil volatility is sensitive to stock market volatility in only two coun-
tries.
5.2 Portfolio management with oil-risk hedging strategies
Our previous findings suggest that potential gains from diversification are substantial
by investing in both oil and stock markets. Their return and volatility spillovers re-
quire however portfolio managers to quantify the optimal weights and hedging ratios
in order to deal adequately with the oil risk. To illustrate this purpose, we now con-
sider a portfolio composed of oil and stocks for which we attempt to minimize the
risk without reducing expected returns. According to Kroner and Ng (1998), the op-
timal weight of holdings of the two assets (oil and the stock market index) is given
by:
s
t
os
t
o
t
os
t
s
t
t os
h h h
h h
w
+
=
2
,
(5)
and,
>
s s
<
=
1 , 1
1 0 ,
0 , 0
,
, ,
,
,
t os
t os t os
t os
t os
w if
w if w
w if
w (6)
where
t os
w
,
refers to the weight of oil in a one-dollar tow-asset portfolio at time t;
and
os
t
h the conditional covariance between oil and stock market returns at time t.
19
Therefore, the optimal weight of the stock market index in the considered portfolio is
t os
w
,
1 .
Table 8
Weights and hedge ratios
Bahrain Kuwait Oman Saudi UAE Qatar
t os
w
,
0.6172 0.6534 0.7028 0.5778 0.4686 0.3997
t os,