Escolar Documentos
Profissional Documentos
Cultura Documentos
Energy Economics
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e n e c o
Oil price volatility and stock price uctuations in an emerging market: Evidence from
South Korea
Rumi Masih a,1, Sanjay Peters b,2, Lurion De Mello c,
a
b
c
a r t i c l e
i n f o
Article history:
Received 10 April 2007
Received in revised form 26 March 2011
Accepted 27 March 2011
Available online 13 April 2011
JEL classication:
F31
C22
C52
Keywords:
Emerging markets
Real stock price
Oil price shocks
Industrial production
Generalized variance decompositions
Impulse response functions
a b s t r a c t
How important are oil price uctuations and oil price volatility on equity market performance? What are the
policy implications if volatility turns out to be signicant? We assess this issue in an economics/nance nexus
for Korea using a VEC model including interest rates, economic activity, real stock returns, real oil prices and
oil price volatility. Our main aim is to capture the effects of crude oil prices on the Korean economy thoroughly
covering the period of the Asian Financial Crisis of 1997, which heavily affected the country, and the oil price
hikes in the early 1990s after the Gulf War. South Korea was the country most hit by the nancial crisis
together with Indonesia and Thailand. Results indicate the dominance of oil price volatility on real stock
returns and emphasize how this has increased over time. Oil price volatility can have profound effect on the
time horizon of investment and rms need adjust their risk management procedures accordingly. This
increase in dependency has been found in other net oil importing emerging equity markets. We test the
relationship between oil price movements and economic activity by using modern time series techniques in a
cointegrating framework. We expand the standard error correction model by examining the dynamics of out
of sample causality through the generalized variance decomposition and impulse response function
techniques. The evidence from persistence proles also gives important guidelines based on how fast the
entire system adjusts back to equilibrium. In addition, we nd the cointegrating relationship to be stable and
nd that the linear error correction model to be more favorable than an asymmetric 2 period Markov
switching model.
2011 Elsevier B.V. All rights reserved.
1. Introduction
Since the oil price shocks of 197374 and 197980, dozens of
academics and practitioners have explored the relationships between
oil price shocks and the macroeconomic variables. The recessionary
impacts of these oil prices shocks were too close for possible causal
links to be ignored, and considerable attention has been devoted to
study the macroeconomics of these events. Policymakers have to take
serious account of the developments in the oil market, as a rise in the
world price of oil imposes macroeconomic costs in two ways. First, to
the extent that oil is both an important input to production and
consumer goods (i.e. petrol and heating oil), results in a reduction in
economic activity as energy becomes more expensive. Second, rising
oil prices contribute directly to the level of ination, particularly in
976
977
yt = yt zt
y1 ; y2 ; ; yT = y1 ; 1Ly2 ; ; 1LyT
z1 ; z2 ; ; zT = z1 ; 1Lz2 ; ; 1LzT
c
T
be set equal to 13.5. This test can attain a signicant gain in power
over the traditional unit root tests. The critical values for Elliott et al.
(1995) in Table 1 of Appendix A are estimated from Monte Carlo
simulations. For nite sample correlations, Cheung and Lai (1995)
provide approximate critical values. In the non-deterministic case, the
use of c = 7 is recommended where the test DF-GLS basically
involves the same procedure as computing the DF-GLS test, apart
from the exception that the locally detrended process series (yt ) is
replaced by the locally demeaned series (yt ) and zt = 1. The
asymptotic distribution of the DF-GLS test is the same as that of the
conventional DF test.
ADF tests indicate the presence of a unit root in each series since
for no series can the null of nonstationarity be rejected. To allow us to
measure how persistent the unit root in the process is, we also
calculate a condence interval (CI) due to Stock (1991), who suggests
that reporting CIs may provide useful information regarding sampling
uncertainty. The condence interval estimates tend to suggest that
the unit root is quite persistent with all lower bounds quite clearly
above 0.80 for both ADF () and ADF (). We also supplement these
results from Sims Bayesian unit root procedure which seem to be
suggestive of a unit root with high value of a. Furthermore, Geweke
and Porter Hudack (GPH) tests for fractional integration, also quite
978
Xt = + i Xt1 + Xtk + t
i=1
11
As a means of investigating the robustness of these results derived from
conducting tests for the total sample, we also undertake a sub-sample analysis of
these tests taking the October 1987 crash as the break point. In order to save space,
these results have not been reported for pre- and post-crash samples but available
upon request. Results, in general, indicate that the unit root approximation seems to
be quite robust to the October 1987 crash, since these sub-sample results do not
change our conclusion from conducting the tests over the full sample that these
variables are integrated of at most order one. Once again, it is important to warn
readers that such results are very much vulnerable to low power due to poor
performance in small samples.
12
Due to one of the biases of the JJ procedure being the sensitivity of cointegration
rank to the order of the lag length used in the VAR, We chose the lag subject to the
Akaikes FPE criterion. In addition, results of a unique cointegrating vector were
insensitive to slight modications to lag length. Furthermore, there has been much
recent work documenting the potential for severe small sample bias in Johansen tests
(see Cheung and Lai (1995)).The scaling-up factor on the asymptotic critical values
suggested by Cheung and Lai's study does not alter our conclusion of cointegration
rank. Furthermore, their study favors the trace test in that: it shows little bias in the
presence of either skewness or excess kurtosis, and is found to be more robust to both
skewness and kurtosis than the maximal eigenvalue test. (Cheung and Lai (1995,
p.324)). In the light of this statement, the trace statistic of 215.64, further conrms our
initial conclusion of r equal to at most 1.
979
Table 2
VEC model estimates using real oil volatility.
Equation
rt
rolt
ipt
rsrt
Equation
rt
rolvt
ipt
rsrt
1, t 1
0.019
(0.019)
0.899
(0.913)
0.296
(0.092)
0.084
(0.088)
1.096
(1.215)
-1.183
(1.208)
0.924
(1.514)
0.643
(1.472)
0.005
(0.015)
0.012
(0.011)
0.13
0.661
0.532
21.421
0.558
344.022
0.035
0.004
(0.001)
0.216
(0.068)
0.011
(0.007)
0.003
(0.007)
0.064
(0.090)
0.132
(0.090)
0.005
(0.113)
0.171
(0.109)
0.002
(0.001)
0.000
(0.000)
0.17
0.052
0.289
16.56
0.493
6336.890
10.381
0.000
(0.000)
0.041
(0.052)
0.001
(0.005)
0.003
(0.005)
0.141
(0.069)
0.000
(0.000)
0.280
(0.085)
0.208
(0.083)
0.000
(0.000)
0.000
(0.000)
0.15
0.042
0.168
14.666
5.134
32.301
30.191
1.019
(0.160)
47.990
(7.55)
0.579
(0.759)
0.440
(0.729)
15.303
(10.043)
14.406
(9.992)
17.004
(12.516)
4.130
(12.151)
0.116
(0.122)
0.098
(0.094)
0.43
6.760
3638.812
4.819
0.059
0.714
4.463
1, t 1
0.013
(0.012)
2.035
(1.983)
0.303
(0.094)
0.110
(0.092)
0.225
(3.195)
3.032
(3.342)
0.408
(1.600)
0.456
(1.526)
0.002
(0.014)
0.010
(0.011)
0.12
0.661
52.231
19.668
0.667
366.331
0.611
0.001
(0.000)
0.212
(0.054)
0.007
(0.003)
0.000
(0.003)
0.356
(0.087)
0.076
(0.091)
0.029
(0.043)
0.011
(0.041)
0.001
(0.000)
0.000
(0.000)
0.32
0.024
0.043
28.079
2.434
7507.398
1.592
0.001
(0.001)
0.239
(0.108)
0.001
(0.005)
0.000
(0.005)
0.503
(0.174)
0.129
(0.182)
0.307
(0.087)
0.233
(0.083)
0.001
(0.000)
0.001
(0.001)
0.21
0.041
0.148
21.212
4.851
32.053
25.134
0.626
(0.108)
99.474
(17.030)
0.388
(0.807)
0.910
(0.791)
65.539
(27.441)
10.258
(28.707)
19.242
(13.747)
5.077
(13.108)
0.055
(0.117)
0.085
(0.095)
0.39
6.834
3854.178
3.262
1.512
0.369
1.795
rt 1
rt 2
rolt 1
rolt 2
ipt 1
ipt 2
rsrt 1
rsrt 2
2
R
^
RSS
2SC[12]
2FF[1]
2NOR[2]
2HET[1]
rt 1
rt 2
rolvt 1
rolvt 2
lipt 1
lipt 2
rsrt 1
rsrt 2
2
R
^
RSS
2SC[12]
2FF[1]
2NOR[2]
2HET[1]
Notes: The underlying VAR model is of order 3 and contains unrestricted intercepts and
restricted trend coefcients. Lag order was selected by Schwarz Bayesian Criterion (SBC).
Standard errors are given in parenthesis. The diagnostics are chi-squared 2 (degrees of
freedom) statistics for serial correlation (SC), functional form misspecication (FF), nonnormal error terms (NOR) and heteroskedastic error variances (HET).
Notes: The underlying VAR model is of order 3 and contains unrestricted intercepts and
restricted trend coefcients. Lag order was selected by Schwarz Bayesian Criterion (SBC).
Standard errors are given in parenthesis. The diagnostics are chi-squared 2 (degrees of
freedom) statistics for serial correlation (SC), functional form misspecication (FF), nonnormal error terms (NOR) and heteroskedastic error variances (HET).
high oil prices leading to the Gulf War in 2000, South Korea did suffer
a decline in 1989 spurred by a sharp decrease in exports and foreign
orders which affected the industrial sector. Poor export performance
resulted from structural problems embedded in the nation's economy,
including an overly strong won, increased wages and high labour
costs, frequent strikes, and high interest rates. The high labour costs
saw industries investing heavily in automation and robotics technology which resulted in an increase in industrial products through the
1990s and in early 2000 period.
It is important to test the stability of the error correction mechanism when weakly exogenous factors are present in the model. In this
study crude oil prices, interest rates and industrial production are
found to be weakly exogenous and the stability of these in the error
correction model are tested using the CUSUM and CUSUMSQ tests
given in Figs. 1 and 2 of Appendix A. These tests are based on the null
hypothesis that the cointegrating vector is the same in every period;
the alternative is simply that it (or the disturbance variance) is not.
The test is quite general in that it does not require a prior specication
of when the structural change takes place. Since our study is not based
on a denitive piece of information, namely when the structural
change takes place it is preferred to the Chow test which only works
best when a denitive piece of information is at hand.
We test the stability of crude oil prices and interest rates after
estimating the VECM model of these variables as these were found to be
exogenous in our model and shocks originating from them could have
caused some instability in the system. Both the plots of the cumulative
sum of recursive residuals fall between the critical bounds and therefore
suggest that there is no cause for alarm in the structural stability of the
VECM. The CUSUM plot is around zero and more importantly lies
between the error bounds. The supporting CUSUMSQ which is deemed
more powerful than CUSUM also supports that crude oil prices and
interest rates did not cause instability in the cointegrating vector.
In addition to the above stability tests, we ran two bivariate cointegration tests between real stock returns and oil prices and real stock
returns and oil price volatility. Both models were found to be cointegrated and we tested for asymmetric effects in the error correction
mechanism by utilising a 2 regime Markov switching model. In
regime-switching models (like TAR, STAR and SETAR; see e.g. Franses
and van Dijk, 2000) the regimes and the switching mechanism are
explicitly dened through the threshold variable and the threshold
level. However, an upfront specication of this variable and level
is not a trivial task as the regime switches in our study are likely
to result from a combination of different fundamental drivers like
industrial production, interest rates real oil price, real oil volatility. On
the other hand, in Markov Regime Switching (MRS) models the
switching mechanism between the states is assumed to be governed
by an unobserved (latent) random variable. MRS models do not
require an upfront specication of the threshold variable and level
and, hence, are less prone to modelling risk. This gives them an
advantage in terms of parsimony.
A Markov-switching mean-adjusted auto regression can be written as:
Yt X St = A + A1 Yt1 X St + A2 Yt2 X St +
+ AN YtN X St + t ; t eN 0; 2
p11
p12
p21
p22
=
p11
1p11
1p22
p22
7
980
Table 3
Generalized variance decompositions.
Horizon
rt
Rol
ip
rsr
0.83
1.01
1.07
1.11
10.31
13.38
13.63
13.82
1.11
3.11
3.52
3.80
84.00
77.37
69.92
59.23
Table 4
Generalized variance decompositions.
Horizon
rt
rolv
ip
rsr
1.97
2.14
2.15
2.17
8.39
11.43
12.13
12.53
0.41
0.33
0.29
0.27
89.71
70.33
59.88
49.81
981
Acknowledgements
We would like to thank the two anonymous referees for their
patience, comments and suggestions that greatly improved the
paper. We are very grateful for Julian Inchauspe for research
assistance.
The views expressed in this paper are those of the authors and not
necessarily shared by JP Morgan.
13
982
Response of ROLV
0.9
Response of LIP
0.016
0.8
0.014
0.7
0.00
-0.2
0.00
-0.4
0.00
-0.6
0.012
0.6
0.00
0.01
0.008
0.4
-1.0
-0.01
0.3
0.006
0.2
0.004
-0.01
0.1
0.002
-0.01
-0.01
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
-0.8
-0.01
-1.2
-0.01
9 11 13 15 17 19 21 23 25 27 29 31
-1.4
-1.6
-1.8
1
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
9 11 13 15 17 19 21 23 25 27 29 31
Response of ROLV
Response of LIP
Response of RSR
0.03
0.00E+00
0.025
-2.00E-03
-0.2
-4.00E-03
-0.4
0.3
0.02
0.2
-0.6
-6.00E-03
0.015
-0.8
0.2
-8.00E-03
-1
0.01
0.1
-1.00E-02
0.1
0.005
-1.20E-02
0.0
-1.40E-02
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
-1.2
-1.4
-1.6
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Fig. 1. Generalized impulse response function. Notes: The horizontal axis refers to months after shock. The vertical axis refers to standard deviations. Charts provide generalized impulse response functions (GIRF) of all variables in our model
when interest rates (R) and oil volatility (ROLV) are shocked. Dashed lines represent single standard error bounds around the point estimates. We can compare the above to Fig. 3.
0.5
Response of RSR
0.0
0.00
0.018
Response of ROLV
Response of LIP
Response of RSR
0.6
0.05
0.00E+00
-0.02
0.4
0.04
-1.00E-03
-0.04
0.2
0.04
-2.00E-03
-0.06
-0.08
0.0
-3.00E-03
0.03
-0.2
-4.00E-03
0.02
-0.4
0.02
-0.6
0.01
-0.8
0.01
-1.0
-0.12
-5.00E-03
-0.14
-6.00E-03
-0.16
-0.18
-7.00E-03
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
0.00
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
-1.2
9 11 13 15 17 19 21 23 25 27 29 31
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Response of ROLV
0
-0.1
0.002
5
0.0015
-0.2
-0.01
-0.2
-0.3
Response of RSR
7
0.0025
-0.005
-0.1
-0.015
0.001
0.0005
-0.3
-0.0005
-0.4
-0.02
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
-0.025
-0.001
-0.4
-0.5
Response of LIP
0.003
-0.1
0.03
9 11 13 15 17 19 21 23 25 27 29 31
-0.0015
-0.002
-1
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Fig. 2. Generalized impulse response functions (Continued). Notes: The horizontal axis refers to months after shock. The vertical axis refers to standard deviations. Charts provide generalized impulse response functions (GIRF) of all variables
in our model when log of industrial production (LIP) and real stock returns (RSR) are shocked. Dashed lines represent single standard error bounds around the point estimates. We can compare the above to Fig. 4.
983
984
Response of ROL
Response of LIP
Response of RSR
0.0
0.035
0.00
0.03
0.00
0.025
0.00
0.02
0.00
0.015
0.00
-1.2
0.01
-0.01
-1.4
0.005
-0.01
-0.2
0.9
0.8
0.7
-0.4
-0.6
-0.8
0.6
0.4
0.3
-1.6
0.2
0.1
0
-0.01
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
-1.8
-2.0
9 11 13 15 17 19 21 23 25 27 29 31
9 11 13 15 17 19 21 23 25 27 29 31
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Response of ROL
0.4
0.08
0.3
0.07
Response of LIP
0.006
0.005
0.4
0.004
0.3
0.2
0.06
0.003
0.05
0.002
0.001
-0.2
-0.4
0.2
0.04
0.2
0.1
0.03
-0.001
0.02
-0.002
-0.6
-0.8
-0.003
0.1
0.01
0.0
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Response of RSR
0.6
9 11 13 15 17 19 21 23 25 27 29 31
-0.004
-1
-0.005
-1.2
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Fig. 3. Generalized impulse response functions (Continued). Notes: The horizontal axis refers to months after shock. The vertical axis refers to standard deviations. Charts provide generalized impulse response functions (GIRF) of all variables
in our model when interest rates (R) and oil prices (ROL) are shocked. Dashed lines represent single standard error bounds around the point estimates. We can compare the above to Fig. 1.
-1.0
0.5
-0.08
-0.1
-0.12
Response of LIP
0.05
0.008
0.04
0.007
0.04
0.006
0.03
0.005
0.03
0.004
0.02
0.003
0.02
0.002
0.01
0.001
0.01
0.6
0.4
0.2
0.0
-0.2
0.00
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Response of RSR
0.8
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
-0.4
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Response of LO
0
-0.1
6
5
4
0.001
-0.01
-0.2
0.0005
-0.3
-0.015
-0.3
-0.0005
-0.4
-0.02
-0.4
-0.5
0.0025
0.0015
-0.2
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
-0.025
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Response of RSR
7
0.002
-0.005
-0.1
Response of LIP
0.003
-0.001
-0.0015
-1
-0.002
-0.06
Response of LO
0.009
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
-2
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
Fig. 4. Generalized impulse response functions (Continued). Notes: The horizontal axis refers to months after shock. The vertical axis refers to standard deviations. Charts provide generalized impulse response functions (GIRF) of all variables
in our model when log of industrial production (LIP) and oil prices (LO) are shocked. Dashed lines represent single standard error bounds around the point estimates. We can compare the above to Fig. 2.
985
986