THE VOLATILITY OF WORLD CRUDE OIL PRISES
Abstrak
Peran minyak dalam ekonomi adalah
sangat penting. Artikel ini mengukur ketidakpastian harga
minyak dunia berdasarkan pada
standar deviasi bersyarat. Artikel ini berfokus pada volatilitas
harga minyak mentah di pasar
Inggris, Texas, dan Dubai, selama periode Januari 1980 sampai Mei
2010. Hasil analisis menemukan bukti
bahwa dampak leverage yang asimetris tidak ditemukan.
Hasil analis juga menemukan bahwa
proses volatilitas return terhadap rata-ratanya hanya terjadi di
Dubai. Temuan ini memiliki beberapa
implikasi penting bagi Indonesia. Pemerintah dapat
menggunakan dinamika harga minyak di
pasar Dubai sebagai patokan untuk mengatur anggaran
negara dalam rangka mewujudkan
kesinambungan fiskal.
Keywords: Harga minyak,
volatilitas, asymmetric leverage, fiskal berkesinambungan
JEL classification numbers: C22, Q43
INTRODUCTION
Oil is arguably
the most influential physical
commodity in the
world and plays a prominent
role in an
economy. It is not a surprise,
therefore, that
the price of oil changes attracts
a considerable
degree of attention for
many decades.
Various attempts have been
undertaken to
explain the behaviour of the
oil price as well
as to assess the macroeconomic
consequences of
oil price shocks especially
since oil crisis
in 1970s.
The oil price
shocks was repeated in
early 2000s. The
wide price fluctuations in
2000s, when crude
oil price index has increased
272 percent
between January 2000
and March 2008,
and fluctuations by more
than USD 20 a
barrel in mid 2008 reinforce
the idea that oil
prices are volatile. The
lates one, in
early February 2011, the world
oil price touched
USD 100 per ba rrel.
The volatility of
oil prices has
prompted
governments, especially in developing
countries, to
intervene in the oil
market in various
ways. Most countries in
the world have
been conducting some policies
including
price-smoothing schemes for
end users, fuel
tax adjustments, price controls,
and subsidies for
lower income class,
or even
incentives for diversification away
from oil.
At the same
period, the world witnessed
the most marked
commodity price
boom of the past
century. The price of metals,
food grains, and
other commodities
rose sharply, and
over a sustained period.
Like commodity
booms in earlier decade,
this one was
associated with strong global
growth, but was
exceptional in its duration
and in the range
of commodities affected.
By mid 2008,
metals and minerals were
296 percent
higher and internationally
traded food
prices 138 percent higher —
mainly due to
higher grain prices.
The high food
commodity and oil
prices have
significant political impacts.
Haiti, for
example, faced serious internal
governmental
problems. China, Vietnam,
and India imposed
some protections to their
international
trades. Indonesia, among others,
desired to
develop the national food and
energy security
(Sugiyanto, 2008). Coupled
with the
financial crisis that erupted in September
2008 and the
subsequent global economic
downturn, some
developing countries
have suffered
dramatic increase in poverty
incidence (World
Bank, 2009).
The recent sharp
increase in oil
price raises the
question as to the nature i.e.
permanent or
temporary. Knowledge of oil
price fluctuation
under market-oriented energy
policy is very
important. The greater
oil price
volatility would increase a household’s
income risk and a
potential output
loss for
business. For the government, the
greater oil price
volatility would increase
subsidies. In
short, the oil price shocks
would deteriorate
the whole economy by
meant various
channels (Rodriguez and
Sanchez, 2005).
In the case of
Indonesia, oil price is
set by the
government. It is under government
subsidy since
1970s. Despite the fact
that Indonesia is
exporting oil, the country
also imports oil
from other countries. The
surplus of
importing value over the exporting
value makes
Indonesia a net oil importing
country. Despite
these facts, the repercussions
from price
increase in the world
market could not
be avoided from spillover
to the local
market.
Being a
government control item,
the event of oil
price surge has inflicted a
soaring fuel
subsidy bill to the government.
This situation
pressured the Indonesia’s
government to
review its policy on oil
prices and
finally decides implement oil
price increase in
the local market. The government’s
decision to
slowly liberalize the
local oil market
has triggered mixed responses
from the public,
particularly
households and
business units.
The main
objective of this paper is
to understand the
nature of dependence of
the conditional
variance on past volatility
in oil prices.
The conditional standard deviation
is interpreted as
a measure of uncertainty.
The rest of this
paper proceeds as
follows. The next
section describes the oil
prices behaviour.
This is followed by exploring
previous
empirical evidences. The
methodological
framework and the data are
delivered in the
proceeding section; the penultimate
section discusses
empirical results;
and the last
section concludes and
points to some
directions for future research.
Some policy
implications for Indonesia
are also drawn.
Oil Prices
Fluctuation
The world oil
price fluctuation has very
long history.
Crude oil prices behave much
as any other
commodity with wide price
swings in times of
shortage or oversupply.
The crude oil
price cycle may extend over
several years
responding to changes in demand
as well as OPEC
and non-OPEC
supply. Table 1
below presents some major
factors that have
influenced the world oil
markets and
therefore the oil price.
Let us start
explaining the oil price
dynamics within
1970s. In 1972, the price
of crude oil was
about USD 3.00 per barrel,
increased 50
percent compared to the beginning
decade. By the
end of 1974, the
price of oil had
quadrupled to over USD
12.00. After
embargo, the world crude oil
price was
relatively flat ranging from USD
12.21 per barrel
to USD 13.55 per barrel.
The Volatility of World … (Kuncoro) 3
Table 1: Some Major Influencing Factors on the World Oil Markets and Oil Price
No. Year Moment
1 1973-1974 · Oil embargo began
(October 19-20, 1973)
· Oil embargo ended (March 18, 1974)
2 1979-1982 · Iranian
revolution; Shah deposed
· OPEC raised prices 14.5% on April 1, 1979 and OPEC
raised prices 15%
· Iran took hostages; President Carter halted imports
from Iran
· Saudis raised marker crude price from 19 $/bbl to 26
$/bbl
· Kuwait, Iran, and Libya production cut drop OPEC oil
production to 27
million b/d
· Saudi Light raised to USD 28/bbl, Saudi Light raised
to USD 34/bbl
· First major fighting in Iran-Iraq War
3 1983-1986 · Libya initiated
discounts
· OPEC cut prices by USD 5/bbl and agreed to 17.5
million b/d output
· Norway, United Kingdom, and Nigeria cut prices
· OPEC accord cut Saudi Light price to USD 28/bbl
4 1990-1991 · Iraq invaded
Kuwait
· Operation Desert Storm began
· Persian Gulf war ended
5 1996-2001 · U.S. launched
cruise missile attacked into southern Iraq following an Iraqi
supported invasion of Kurdish safe
haven areas in northern Iraq.
· Prices rose as Iraq’s refusal to allow United Nations
weapons inspectors
into "sensitive" sites
raises tensions in the oil-rich Middle East.
· OPEC raised its production ceiling. This was the first
increase in 4 years.
· World oil supply increased by 2.25 million barrels per
day in 1997, the
largest annual increase since 1988.
· Oil prices continued to plummet as increased
production from Iraq coincides
with no growth in Asian oil demand
due to the Asian economic crisis
and increases in world oil inventories
following two unusually warm
winters.
· Oil prices tripled between January 1999 and September
2000 due to
strong world oil demand, OPEC oil
production cutbacks, and other factors,
including weather and low oil stock
levels.
· Oil prices fell due to weak world demand (largely as a
result of economic
recession in the United States) and
OPEC overproduction.
· Oil prices declined sharply following the September
11, 2001 terrorist
attacks on the United States,
largely on increased fears of a sharper
worldwide economic downturn (and
therefore sharply lower oil demand).
6 2002-2010 · Political
instability within various oil producing nations
· Rising costs of oil
· Speculator entered the oil market
· Global financial crisis
· European sovereign debt crisis
Source:
http://www.eia.doe.gov/emeu/cabs/chron.html, http://www.wtrg.com and Kuper
(2002)
4 ___________ ___________________________________________
In 1979 and 1980,
events in Iran
and Iraq led to
another round of crude oil
price increases.
The Iranian revolution resulted
in the loss of 2
to 2.5 million barrels
per day of oil
production between November
1978 and June
1979. The combination
of the Iranian
revolution and the Iraq-Iran
War caused crude
oil prices to more than
double increasing
from USD 14 in 1978 to
USD 35 per barrel
in 1981.
From 1982 to
1985, OPEC attempted
to set production
quotas low
enough to
stabilize prices. These attempts
met with repeated
failure as various members
of OPEC produced
beyond their quotas.
During most of
this period, Saudi Arabia
acted as the
swing producer cutting its
production in an
attempt to stem the free
fall in prices.
Crude oil prices plummeted
below USD 10 per
barrel by mid-1986 in
accordance with
world economic recession.
The price of
crude oil spiked in
1990 with the
lower production and uncertainty
associated with
the Iraqi invasion of
Kuwait and the
ensuing Gulf War. From
1990 to 1997
world oil consumption increased
6.2 million
barrels per day. Asian
consumption
accounted for all but 300,000
barrels per day
of that gain and contributed
to a price
recovery that extended into 1997.
Declining Russian
production contributed
to the price
recovery.
The price
increases came to a rapid
end in 1997 and
1998 when the impact of
the economic crisis
in Asia was either ignored
or severely
underestimated by
OPEC. In
December, 1997 OPEC increased
its quota by 2.5
million barrels per day (10
percent) to 27.5
MMBPD effective January
1, 1998. The
rapid growth in Asian economies
had come to a
halt. In 1998 Asian Pacific
oil consumption
declined for the first
time since 1982.
The combination of lower
consumption and
higher OPEC production
sent prices into
a downward spiral. In response,
OPEC cut quotas
by 1.25 million
b/d in April and
another 1.335 million in
July. Price
continued down through December
1998.
With minimal Y2K
problems and
growing US and
world economies the price
continued to rise
throughout 2000. Russian
production
increases dominated non-OPEC
production growth
from 2000 forward and
was responsible
for most of the non-OPEC
increase since
the turn of the century. In the
absence of the
September 11, 2001 terrorist
attack this would
have been sufficient to
moderate or even
reverse the trend. In the
wake of the
attack crude oil prices plummeted.
Spot prices for
the U.S. benchmark
West Texas
Intermediate were down 35
percent by the
middle of November. The
oil prices were
moving into the USD 25
range by March,
2002.
During much of
2004 and 2005 the
spare capacity to
produce oil was under a
million barrels
per day. A million barrels
per day is not
enough spare capacity to
cover an
interruption of supply from most
OPEC producers.
In a world that consumes
over 80 million
barrels per day of petroleum
products that
added a significant risk
premium to crude
oil price and was largely
responsible for
prices in excess of USD 40-
50 per barrel.
Throughout the
first half of 2008,
oil regularly
reached record high prices. On
February 29,
2008, oil prices peaked at
USD 103.05 per
barrel, and reached USD
110.20 on March
12, 2008, the sixth record
in seven trading
days. Prices on June 27,
2008, touched USD
141.71/barrel, for August
delivery in the
New York Mercantile
Exchange (after
the recent USD
140.56/barrel), In
January 2009, oil prices
rose temporarily
because of tensions in the
Gaza Strip. From
mid January to February
13, oil fell to
near USD 35 a barrel. As of
May 2010, crude
oil prices have started to
decline again due
to the 2010 European
sovereign debt crisis.
On May 17, 2010 the
price for a
barrel of crude oil fell below
USD 70 a barrel
to USD 69.41.
The Volatility of World … (Kuncoro) 5
While the
evolution of commodity
prices is
relatively stable, that of oil prices
is more volatile
(Reigner, 2007). Various
attempts to
explain the behaviour of the oil
price have been
undertaken in the past few
years. Three main
approaches can be identified
in this vast
literature: first, Hotelling’s
(1931) notion of
oil as exhaustible
resource; second
the ascertainment that the
global
macroeconomic situation is an important
factor, and,
thirdly, the notion that
additional
factors such as OPEC announcements
as well as
speculation affect
the price of oil.
Regarding the
first approach, Hotelling’s
(1931) seminal paper
proposes the
notion that oil
is exhaustible and that the
price of oil, in
optimum, grows at the rate
of interest.
Various extensions of this rule
have been
suggested and are still subject of
scientific
debates, see e.g. Sinn (2008). In
particular Krautkraemer
(1998), however,
provides evidence
of frequent failure of
empirically
testing Hotelling-type hypotheses.
Dvir and Rogoff
(2009) epitomize this
skepticism: they
apply the storage rather
than a Hotelling
resource extraction model
in order to model
oil price behaviour.
Papers such as
Slade (1982) and
Pindyck (1999)
deal with oil price behaviour
in the very long
run. These papers deal
with the question
as to whether the price of
oil follows a
deterministic trend. While
Slade (1982)
finds evidence of quadratic
trends in real
oil prices, Pindyck (1999) argues
that the oil
price fluctuates around a
long-run trend.
The trend itself is - due to
changes in
demand, extraction costs and
new site
discoveries – stochastically fluctuating
over time.
Livornis (2009) provides
an excellent
survey of this literature and
expresses a less
pessimistic view on the
significance of
the Hotelling rule.
In contrast to
this line of research,
Krichene (2002)
and Dees et al. (2007) argue
that the price of
oil is determined by
global economic
conditions and employ
demand and supply
frameworks in order to
explain the oil
price. Krichene (2002) uses
a structural
multiple equation model of the
global oil market
and focuses on the calculation
of demand and
supply elasticity.
Among the more
salient findings of this
paper is that
short-run demand and supply
of oil is very
price inelastic and that longrun
oil supply
elasticity significantly decreased
after the first
oil crisis 1973/74.
Dees et al.
(2008), in contrast, use a
country-by-
country approach and explicitly
incorporate
geological factors as well as
OPEC behaviour in
their oil supply function.
The model is
generally able to reproduce
responses of the
global oil market to
changes in OPEC
behaviour. The papers by
Kaufmann et al.
(2004) and Dees et al.
(2008) also focus
on the role of OPEC behaviour,
but do not
explicitly model oil
supply. Both
papers make use of an error
correction
approach and show that variables
such as OPEC capacity
utilization
and OPEC quotas
Granger cause real oil
prices but not
vice versa.
While these
results are more of very
general
character, Kaufman and Ullmann
(2009) show that
the 2008 oil price hike
can be explained
by a combination of fundamental
factors and
speculative behaviour,
and Miller and
Ratti (2009), finally, provide
evidence of the
existence of oil price
bubbles.
The unstable
world oil price pumps
dozen empirical
studies dealing with its
impacts on
economic activity in all aspects.
Sadorsky (1999),
among others, tested the
relationship
between oil price and stock
market. In
developing countries, Sari
(2006)
simultaneously examined the link of
oil price, stock
returns, interest rates, and
output in Turkey.
Gronwald et al. (2009)
analyzed the oil
price fluctuation in Kazakhstan
related to
economic growth.
Mohammad (2010)
observed the impact of
oil prices
volatility on export earning in
Pakistan. Aliyu
(2009) connected the oil
price to exchange
and inflation rates in Ni6
___________ ___________________________________________
geria. In general
the found a negative impact
generated from
oil price volatility.
Bacon and Kojima
(2008) investigated
the degree of oil
price volatility
Ghana, Chile,
India, Philippine, and Thailand
during July
1999-March 2007. They
observed some
adverse impacts on exchange
rates and fiscal
condition. Dealing
with world oil
price fluctuation, they point
out some policies
including the role of
hedging,
strategic stocks, price-smoothing
scheme, and
reducing the importance of oil
consumption to
achieve energy security.
In the case of
Indonesia, the world
crude oil price
is used as basic assumption
to set up budget
state in current year. Kuncoro
(2010) found that
the increase in oil
price marginally
induces fiscal stance for
about 0.02
percent. His study implied that
the primary
balance surplus is vulnerable to
maintain fiscal
sustainability. This finding
would suggest
that price smoothing based
on long-term
trends would have imposed a
considerable
fiscal drain.
To summarize, the
price of oil is affected
by numerous
factors and subject to a
considerable
degree of volatility. Hamilton
(2008) nicely
summarizes these findings:
“Changes in the
real price of oil have historically
tended to be
permanent, difficult
to predict, and
governed by very different
regimes at
different points in time”. Thus,
deriving future
predictions is a very difficult
task. In any
case, expecting the oil
price to begin a
stable increase in the near
future would definitely
be hazardous.
METHODS
The brief
literature review above suggests
the potential for
some interesting hypotheses
about potential
linkages among energy
commodities,
macroeconomic variables,
and more
importantly dependency across
energy markets.
The purpose of this section
is to develop an
analytical framework
within which
these can be clearly stated as
a set of formal
propositions. We focus on
the oil market.
From an
econometric point of view,
neglecting the
exact nature of the dependence
of the variance
of the error term conditional
on past
volatility will result in loss
of efficiency.
The ARCH models are developed
to model
time-varying conditional
variances (see
Bollerslev et al., 1994).
ARCH models
consist basically of two
equations, one
for the mean and one for the
conditional
variance. The mean equation
can be univariate
or may contain other
variables
(multivariate). GARCH model
addresses the
issues of heteroscedasticity
and volatility
clustering by specifying the
conditional
variance to be linearly dependent
on the past
behaviour of the squared
residuals and a
moving average of past
conditional
variance. Formally, the model
can be expressed
as follows:
yt
= bxt + et
(1)
The mean equation
may also include the
conditional
variance or the conditional
standard
deviation (ARCH-in-Mean models).
The specification
for the conditional
variance may
allow for asymmetric effects.
Here we start
with a symmetric univariate
specification.
In applications
using monthly data
the error
variance depends on past volatilities
going back a
number of periods. For
these
applications GARCH (Generalised
ARCH) models are
developed. The
GARCH model
depicts conditional variance
of a price series
to depend on a constant,
past news about
volatility and the
past forecast
variance. The GARCH(p,q)
model has p ARCH terms and q GARCH
terms (the values
of p and q are determined
by the Schwarz
Information Criterion):
_ _ − − 2
= + 2 + 2
t t p t q s w a e
b s (2)
It is commonly
assumed that the innovations
_t
are Gaussian. If this assumption
is violated the
usual standard errors are
not consistent
and the quasi-maximum likeThe
Volatility of World … (Kuncoro) 7
lihood
covariances and standard errors described
by Bollerslev and
Wooldridge
(1992) have to be
used.
The simplest
GARCH model is the
GARCH(1,1) model
that in many applications
provides a good
description of the
data. The error
variance depends on all past
volatilities with
geometrically declining
weights as long
as bt < 1. Well-defined
conditional
variances require that the parameters
w, _, b are non-negative. In many
applications the
estimates for _ + b in the
GARCH(1,1) model
are close to unity,
which means that
the model is not covariance
stationary. In
that case the model can be
used only to
describe short-term volatility.
It is notable
that in the symmetrical
model, the
conditional variance is a function
of the size and
not of the sign of
lagged residuals.
One way to allow for
asymmetries is the
Threshold GARCH
(TARCH) model:
t q t p
t t p
ln( ) [ / ]
ln( ) /
2
2
b s g e s
s w a e s
(4)
The coefficient g in the last term of equations
(3) and (4)
measures the leverage effects.
In theory there
may be many leverage
effects, Eviews
only allows for one. In this
model, good news
(_t < 0) and bad news (_t
> 0) have
different effects on the conditional
variance. Good
news has an impact of _,
while bad news
has an impact of (_ + g).
According to
Swaray (2002), the
strength of
ARCH-class models as compared
with time-series
models, lie in their
ability to allow
the conditional variance of
underlying
processes to vary over time. Also
the information
that is used in forming conditional
expectations is
similar to that used
to predict the
conditional mean (i.e. variables
observed in
previous periods). Hence,
the GARCH model
maintains the desirable
forecasting
properties of a traditional timeseries
but extends them
to the conditional
variance (Holt
& Aradhyula, 1990).
RESULTS
DISCUSSION
Data of world
crude oil prices are presented
by UK Brent
(light blend), WTI Midland
Texas, and Dubai
(medium) in USD per barrel
(fob). The sample
periods chosen for this
study extend from
January 1980 to the May
2010. The total
observation is 365 sample
points. The data
are provided by the International
Financial
Statistics (IFS) online service
(International
Monetary Funds, 2010).
The raw data are
then transformed into first
log-differenced
to obtain volatility measurement.
Figure 1 delivers
the crude oil
prices volatility
in three markets.
Table 2 presents
the elementary statistics
covering mean,
median, and extreme
values. The
average of first log-differenced
is close to each
other, around 2 percent for
the three
markets. However, the median values
are far enough
from the respective mean
especially in
Texas and Dubai. Similarly,
the absolute
(maximum and minimum) values
are not identical
to each other. Those
preliminary
indicate non normal distribution.
We will re-check
more convincingly later.
The Table also
delivers standard
deviation ranging
from 0.082 to 0.089. Statistically,
a set data is
said to be relatively
volatile if its
CV (ratio of standard deviation
to its mean) is
more than 50 percent.
Based on the
empirical rule, the crude oil
price in UK is
the most volatile indicated
by the highest
CV, followed by that in Dubai
and Texas
markets. This finding supports
to the
theoretical background in the
previous section
that the oil prices are not
stable.
8 ___________ ___________________________________________
The
log-differenced oil prices are
asymmetrically
distributed (bell-shaped)
indicated by the
high value of Jarque-Bera
tests. The null
hypotheses that the series
data is normally
distributed can be rejected
in 95 percent
confidence level. The lower
tail of the
distribution is thicker than the
upper tail
(indicated by the negative values
of skewness in
Texas and Dubai) and the
tails of the
distribution are thicker than the
normal (indicated
by the kurtosis coefficient
greater than the
thick tails can be
modelled by
assuming a “conditional”
normal
distribution for returns; where conditional
normality implies
that returns are
normally
distributed on each month, but
that the
parameters of the distribution
change from month
to month. Also, as evidenced
in Table 2, the
volatility (standard
deviation) of oil
price returns exhibits
“clustering” i.e.
bursts of high volatility
separated by
periods of relative tranquility.
The correllograms
of the logdifferenced
oil prices and of
the squared
log-differenced
oil prices for 12 lags suggests
strong dependence
in the mean of
variance. There
is only a few insignificant
in the longer
lags but substantial dependence
in the
volatility. This time-varying
nature of
variance is referred to in statistics
as
heteroscedasticty. The persistence of
volatility is an
indication of autocorrelation
in variances.
The Ljung-Box Q-statistic test can
be used to check
for autocorrelation in
variance. Under
the null hypothesis that a
time series is
not autocorrelated, Q(p) is
distributed c2(p), where p denotes the number
of
autocorrelations used to estimate the
statistic. For p = 12, the Q(p) statistic for
squared oil price
returns is 53.3, 58.7, and
94.8
respectively, which rejects the hypothesis
that variances of
monthly returns
are not
autocorrelated. They seem that the
price volatility
in the three oil markets is
persistent at
least in one year.
The price
volatility in the three oil
markets typically
is indifferent each other
presented by the
correlation matrices. The
correlation is
high even close to unity. The
highest oil price
volatility correlation is
more than 0.94
between Dubai and UK.
The oil price
volatility in Dubai market is
lowest correlated
with that in Texas (0.89)
compared to the
others. The long distance
between Dubai and
Texas might be the
source of
explanation.
-.4
-.3
-.2
-.1
.0
.1
.2
.3
.4
.5
1980 1985 1990
1995 2000 2005 2010
VUK
-.5
-.4
-.3
-.2
-.1
.0
.1
.2
.3
.4
1980 1985 1990
1995 2000 2005 2010
VTX
-.4
-.2
.0
.2
.4
.6
1980 1985 1990
1995 2000 2005 2010
VDB
Source: Data processed.
Figure 1: Crude Oil Prices Volatility
The Volatility of World … (Kuncoro) 9
Table 2: Descriptive Statistics
VUK VTX VDB
Mean 0.001772 0.001894 0.001936
Median 0.001748 0.000287 0.005566
Maximum 0.466400 0.391112 0.521421
Minimum - 0.313472 - 0.395148 -
0.335434
Std. Dev. 0.089002 0.081742 0.086547
Skewness 0.042534 - 0.393327 -
0.026617
Kurtosis 5.926260 6.838412 8.323111
Jarque-Bera 129.9819 232.8422
429.7982
Probability 0.000000 0.000000
0.000000
CV (%) 5022.69 4315.84 4470.40
Source: Data calculation.
Table 3: Causality Test
Null Hypothesis: Obs F-Statistic
Probability
VTX does not Granger Cause VUK 352 1.80577
0.04616
VUK does not Granger Cause VTX 1.66045 0.07431
VDB does not Granger Cause VUK 352 1.28950
0.22293
VUK does not Granger Cause VDB 0.70846 0.74325
VDB does not Granger Cause VTX 352 1.61119
0.08687
VTX does not Gra nger Cause VDB 1.78951 0.04874
Source: Data estimation.
Correlation does
not necessary present
causation. The
traditional Granger test
could be employed
to identify the direction
of causality. The
test is done for 12 lags as
suggested from
partial autocorrelation. Table
3 identifies how
great the oil price volatility
in one market
affects to the oil price
volatility in the
other markets. Regardless
to the
significance, Table 3 preliminary
suggest the
existence of oil price volatility
co-movement.
Does the high
volatility of the data
mean non stationary?
Table 4 shows the
results of
Augmented Dickey-Fuller (ADF)
and
Phillips-Perron (PP) unit root tests for
the underlying
data series in levels and first
differences.
According to Swaray (2002)
the
Phillips–Perron (PP) test can be more
appropriate in
this case because of the evidence
of
heteroscedasticity assumed in the
error process of
the price series examined.
We assume that
the level of the oil price is
not stationary.
Formal unit-root
tests (including a
constant, no
trend, and 12 lags) to log-oil
price data reject
the hypothesis of a unit
root at 5% (the
ADF test statistic equals -
1.7, 5% critical
value equals –2.8693). The
similar results
are obtained by implementing
PP unit root
tests. However, these tests
have only little
power if errors are not homogeneous
(Kim and Schmidt,
1993). Furthermore,
the power of unit
root tests depends
more on the span
of the data, which
in our case is
only 30 years, than on the
number of
observations (Perron and Shiller,
1985). Moreover,
the presence of structural
breaks reduces
the power of unit root tests
also (Perron,
1989). More details on unit
roots, structural
breaks, and trends can be
found in Stock
(1994).
The same method
imposed to the
log-differenced
oil price data gives the opposite
conclusion. The
ADF test statistic
equals from -13.1
to -14.6 and the PP test
statistic ranges
from -12.3 to 14.2 implying
the series data
have a unit roots. The occur10
___________ ___________________________________________
rence of unit
roots in the price series of
these commodities
gives a preliminary indication
of shocks having
permanent or
long lasting
effect, thus making it very difficult
for traditional
price stabilization policies
to survive.
Stationary is
required to perform
co-integration.
Co-integration is an important
concept to
analyze the data behaviour.
Using Johansen’s
maximum likelihood approach
(Johansen, 1988;
1991), we test the
bivariate among
the three oil price markets
volatility with 4
lags in all the cases. The
trace and
Max-Eigen value (_ max) statistics
for testing the
rank of co-integration
are shown in
Table 5.
The results of
both tests deny the
absence of
co-integrating relation oil prices
volatility
series. Furthermore, both tests
suggest the
presence of one co-integrating
equation at 5
percent level or better between
the non
stationary prices of crude oil
which means that
the linear combinations
of them are
stationary and, consequently,
prices tend to
move towards this equilibrium
relationship in
the long-run. This is
complement to the
result of correlation and
causality
analysis.
Furthermore, does
the stationary of
oil prices change
imply that it will return to
its mean value?
The following section presents
empirical results
for a monthly time
series data. The
results of GARCH estimation
model will
clearly answer this question.
The Schwarz
Information Criterion for
GARCH model
suggests that a = 1 and b =
1. The GARCH
model results are in Table 6.
The ARCH Lagrange
Multiplier test
indicates that
there is no autoregressive conditional
heteroscedasticity
up to order 12 in
the residuals. An
alternative test is the Ljung-
Box Q-statistic
of the standardized squared
residuals. At the
twentieth lag Q equals from
7.4 to 13.8,
indicating that the standardized
squared residuals
are serially uncorrelated.
From these tests,
we conclude that the
GARCH volatility
model is adequate.
Table 4: Unit Root Tests
ADF Test PP Test
Level t-stat 5% level t-stat 5% level
Log (OP UK) -1.676441 - 2.869285
1.459409 -2.869263
Log (OP TX) -1.766282 -2.869285
-1.450294 -2.869263
Log (OP DB) 1.771291 2.869285
1.272295 2.869263
First log-diff. t-stat 5% level
t-stat 5% level
VUK -14.64803 -2.869285 -14.19856
-2.869285
VTX -13.78728 2.869285 13.25763 2.869285
VDB -13.09016 2.869285 12.28922 2.869285
Source: Data estimation.
Table 5: Multiple Co-integration Tests
Hypothesized
Eigenvalue
Trace 5 Percent 1 Percent
No. of CE(s) Statistic Critical
Value Critical Value
None ** 0.315899 313.6462 29.68
35.65
At most 1 ** 0.246502 177.3519 15.41
20.04
At most 2 ** 0.190216 75.7447 3.76
6.65
Notes: (1) *(**) denotes rejection
of the hypothesis at the 5%(1%) level, (2) Trace test indicates 3
cointegrating equation(s) at both 5%
and 1% levels.
Source: Data estimation.
The Volatility of World … (Kuncoro) 11
Table 6: GARCH Model Estimates
VUK VTX VDB
Coeff. Z-stat Coeff. Z-stat Coeff. Z-stat
Constant - 0.002380 -0.67879
-0.003018 - 1.25735 0.005142 1.25949
w 0.000450 3.25528
0.000106 1.64696 0.003935 10.52062
a 0.348587 6.75545
0.351480 7.23237 0.501573 7.96304
b 0.647337 13.07202
0.703594 17.81267 -0.022678 - 0.39833
Diag. test Value Prob. Value Prob.
Value Prob.
a + b = 1 0.01415
0.9054 3.57608 0.0594 42.0631 0.0000
0.01415 0.9053 3.57608 0.0586
42.0631 0.0000
J-B test 20.70801 0.0000 15.70913
0.0000 132.11880 0.0000
ARCH
LM(12)
0.97428 0.47312 0.73163 0.72034
1.04797 0.40418
11.73505 0.46719 8.88609 0.71263
12.59086 0.39947
Q(12) 11.3580 0.4980 7.4229 0.8289
13.8170 0.3130
Source: Data estimation.
The Wald test for
(a + b = 1)
clearly indicates
that the volatility process
does not return
to its mean mainly in UK
and Texas. The F and c2 values are 0.01 for
UK and 0.06 for
Texas respectively. Those
are enough to
reject the null hypotheses
that (a + b = 1). For Dubai, the coefficient
b even is insignificant. The F and c2
values
are quite greater
to accept the null hypotheses.
This means that
the model can be used
only to describe
short-term volatility especially
in UK and Texas
in order to predict
in the near
future.
.00
.05
.10
.15
.20
.25
.30
1985 1990 1995
2000 2005 2010
Conditional
Standard Deviation
Source: Data processed
Figure 2a: Conditional Standard Deviation
of VUK
.00
.05
.10
.15
.20
.25
.30
1985 1990 1995
2000 2005 2010
Conditional
Standard Deviation
Source: Data processed.
Figure 2b: Conditional Standard Deviation
of VTX
.05
.10
.15
.20
.25
.30
.35
.40
1985 1990 1995
2000 2005 2010
Conditional
Standard Deviation
Source: Data processed.
Figure 2c: Conditional Standard Deviation
of VDB
12 ___________ ___________________________________________
Table 7: Asymmetric GARCH Model Estimates
VUK VTX VDB
TARCH Coeff. Z-stat Coeff. Z-stat Coeff. Z-stat
Constant -0.003480
- 0.93677 -0.004041 1.49348 0.004327 1.03028
w 0.000416 3.11633 9.46E-05 1.54344 0.003915 10.36967
a 0.281078 4.25853 0.282623 3.56680 0.399368 7.35491
b 0.667451 13.51501 0.717981 17.13441 -0.017900 0.31028
g 0.102887 0.85241 0.112852 1.06394 0.193605 1.13149
Test: g = 0 Value Prob. Value Prob. Value Prob.
F 0.726604 0.3946 1.131976 0.2881 1.280276 0.2586
c2
0.726604 0.3940 1.131976 0.2874 1.280276 0.2578
EGARCH Coeff. Z-stat Coeff. Z-stat Coeff. Z-stat
Constant 0.00190
- 0.5324 0.00315 - 1.1725 0.00535 - 1.5892
w -0.98086 - 4.4482 0.73476 - 3.8901 1.06589 - 7.7229
a 0.51025 6.7925 0.49732 6.1763 0.52547 9.1020
b 0.88365 25.1959 0.93344 32.7686 0.86583 42.7314
g -0.05553 - 0.9699 0.06908 - 1.3464 0.08179 - 1.5509
Test: g = 0 Value Prob. Value Prob. Value Prob.
F 0.940666 0.3328 1.812796 0.1790 2.405213 0.1218
c2
0.940666 0.3321 1.812796 0.1782 2.405213 0.1209
Source : Data estimation.
Volatility is
plotted in Figure 2 that
shows the
conditional standard deviation of
the GARCH (1,1)
model. Because the volatility
process does not
return to its mean
value, the
conditional standard deviation
graph contour in
UK and Texas rather fluctuates
without clear
basic pattern. On the
contrary, even
though also fluctuates, the
conditional
standard deviation graph contour
in UK quite
rather flats based on the
basic value a = 0.5015. Consequently, the
standard
deviation of oil price in Dubai is
relatively more
predictable than that in UK
and Texas.
As mentioned
earlier, in the symmetrical
model the
conditional variance is a
function of the
size and not of the sign of
lagged residuals.
TARCH and EGARCH
models take into
account the sign of lagged
residuals. The
results for the TARCH
(1,1,1) and
EGARCH (1,1) models are presented
in Table 9. In
general, the results of
TARCH and EGARCH
models statistically
have no different
from GARCH models as
presented in
Table 7.
The individual
tests using Z, F, and
c2
for g conclude that all of the leverage
effect terms is
not significantly positive
(even with a
one-sided of 5 percent level
test) so there
does not appear to be an
asymmetric
effect. In these models, good
news (_t
< 0) and bad news (_t
> 0) have no
different effects
on the conditional variance
*). The absence of leverage effect that
can normally be
found on financial markets
might be due to
that commodity markets
are more prone to
volatility when the price
goes up and when
the price goes down as
what can be
observed in the financial markets.
In term of
forecasting, the asymmetric
effects imply
that the prediction of
the oil price in
the near future is then relatively
easy without
considering bad news
and bad news. In
other words, the conditional
variance and
standard deviation are
controllable so
that the prediction value is
asymptotically
will be more accurate. Furthermore,
hedging cost
associated with the
change in oil
prices risk would be minimized.
Finally, the
optimal position for all
*) We do not report results the tests for the TARCH
and EGARCH models
completely since leverage
effects are not
significant. They can be available
on request to the
author.
The Volatility of World … (Kuncoro) 13
players in the
oil market would be achieved
in the frame of
market efficiency.
CONCLUSION
In this paper we
tried to understand the nature
of dependence of
the conditional variance
on past
volatility in oil prices. The
volatility is
measured by the first logdifferenced.
The measure of
uncertainty we
choose is the
within-month high-low range
of the
conditional standard deviations.
Time-varying
conditional variances are estimated
using univariate
(G)ARCH models.
GARCH models
depend on the frequency
of the data, so
we also examine
monthly time
series for the period January,
1980 to May, 2010
representing 365 observations.
We focus on
volatility of the world
crude oil prices
in UK, Texas, and Dubai
markets. We found
that the preferred model
is a symmetric
GARCH (1,1) model.
Asymmetric
leverage effects are not found
in the three
markets. In fact, the positive
shocks are more
dominant than the negative
shocks. However,
the volatility process
returns to its
mean only in Dubai.
Those findings
have some important
implications for
Indonesia. The main
policy
recommendation to emerge from this
paper is that any
effort invested in reducing
the oil
dependency of the Indonesian economy
is more than
justified. Moreover, it is
worth considering
a tightening of the stabilization
fund which would
lead to a less
fragile economic
development. Second, the
resurgence of
energy price crises should
redirect energy
security policy towards the
development and
adoption of energysaving
technology, such
as gas, coal, solar
panels, wind
turbines, hydropower, biomass,
and other
renewable energy.
Third, as a net
oil importer country,
Indonesia faces a
dilemma when the world
crude oil price
increases. In one hand, the
central
government revenue increases substantially
due to oil and
gas taxes. On the
other hand, the
central government has to
spend more
subsidies to avoid the increase
of domestic fuel
prices. In this case, the
government could
use the dynamics of oil
price in Dubai
market as a benchmark to
set up her state
budget in order to realize
fiscal
sustainability.
The volatility of
oil prices is interesting
to be explored
further. This study
used a univariate
GARCH model. More
advance research
could utilize the multivariate
GARCH to capture
volatility persistence
across markets.
It is also advisable to
use high
frequency data i.e. daily data in
the longer time
horizon to catch uncertainty
among oil,
commodity, and stock markets.
There is no doubt
that in the globalization
era, oil,
commodity, and stock markets are
increasingly
integrated.
REFERENCES
Aliyu, S.U.R.
(2009), “Impact of Oil Price Shock and Exchange Rate Volatility on Economic
Growth in
Nigeria: An Empirical Investigation,” Research Journal of
Internatıonal
Studıes, 11, 4-15.
Bacon, R. and M.
Kojima (2008), “Coping with Oil Price Volatility,” Energy Sector Management
Assistance
Program, Special Report 005/08, The World Bank, Washington
Bollerslev, T.
1986), “Generalized Autoregressive Conditional Heteroskedasticity,” Journal
of Econometrics, 31(3), 307–27.
Bollerslev, T.,
R.F. Engle, and D.B. Nelson (1994), “ARCH Models,” Handbook of
Econometrics, 4, 2959-3038.
14 ___________ ___________________________________________
Bollerslev, T.
and J.M. Wooldridge (1992), “Quasi-Maximum Likelihood Estimation and
Inference in
Dynamic Models with Time Varying Covariances,” Econometric
Reviews,
11(2), 143-72.
Dees, S., A.
Gasteuil, R.K. Kaufmann, and M. Mann (2008), “Assessing the Factors behind
Oil Price
Changes,“ European Central Bank Working Paper No. 855.
Dees, S., P.
Karadeloglou, R.K. Kaufmann, and M. Sanchez (2007), “Modelling the World
Oil Market: Assessment
of a Quarterly Econometric Model,” Energy Policy, 35,
178-91.
Dvir, E. and K.S.
Rogoff (2009), “Three Epochs of Oil,” NBER Working Paper No.
14927.
Engle, R.F. and
A.J. Patton, (2001), “What Good is a Volatility Model?” Quantitative Finance,
1, 237-245.
Gronwald, M., J.
Mayr, and S. Orazbayev, (2009), “Estimating the Effects of Oil Price
Shocks on the
Kazakh Economy,” IFO Working Paper No. 81, October.
Hamilton, J.D.
(2009), “Understanding Crude Oil Prices,” Energy Journal, 30, 179-206.
Holt, M. and S.V.
Aradhyula, (1990), "Price Risk in Supply Equations: An Application of
GARCH Time-Series
Models to the U.S. Broiler Market,” Southern Economic
Journal, 57, 230–242.
Hotelling, H.
(1931), “The Economics of Exhaustible Resources,” Journal of Political
Economy, 39, 137-75.
International
Monetary Fund, (2010), IFS Online Database, Available at
http://www.imfstatistics.org/imf/logon.aspx,
Accessed June, 26.
Johansen, S.
(1988), “Statistical Analysis of Cointegrating Vectors,” Journal of Economic
Dynamics and
Control, 12, 231–254.
Johansen, S.
(1991), “Estimation and Hypothesis Testing of Cointegrating Vectors in
Gaussian Vector
Autoregressive Models,” Econometrica, 59, 1551–1580.
Kaufmann, R.K.
and B. Ullman (2009), “Oil Prices, Speculation, and Fundamentals: Interpreting
Causal
Relationships among Spot and Future Prices,“ Energy Economics,
31, 550-558.
Kim, K. and P.
Schmidt, (1993), “Unit Root Tests with Conditional Heteroscedasticity,”
Journal of
Econometrics, 59(3), 287-300.
Krautkraemer,
J.A. (1998), “Nonrenewable Resource Scarcity,” Journal of
Economic Literature,
36, 2065-107.
Krichene, N.
(2002), “World Crude Oil and Natural Gas: A Demand and Supply Model,”
Energy Economics, 24, 557-76.
Kuncoro, H.
(2010), “State Budget Sustainability and Its Implication for Financial System
Stability,” Paper
Presented in the Discussion on Financial System Stability, Central
Bank of
Indonesia, November, 29.
Kuper, G.H.
(2002), "Measuring Oil Price Volatility,” CCSO Working Papers No.
2002/08,
Groningen University.
The Volatility of World … (Kuncoro) 15
Livornis, J.
(2009), “On the Empirical Significance of the Hotelling Rule,” Review of Environmental
Economics and
Policy, 3, 22-41.
Miller, J.I. and
R.A. Ratti (2009), “Crude Oil and Stock Markets: Stability, Instability, and
Bubbles,” Energy Economics, 31, 559-568.
Mohammad, S.D.
(2010), “The Impact of Oil Prices Volatility on Export Earning in Pakistan,”
European Journal
of Scientific Research, 41(4), 543-550.
Perron, P.
(1989), “The Great Crash, the Oil Price Shock and the Unit Root Hypothesis,”
Econometrica, 57, 1357-1361.
Perron, P. and
R.J. Shiller (1985), “Testing the Random Walk Hypothesis: Power versus
Frequency of
Observation,” Economics Letters, 18(3), 381-386.
Pindyck, R.S.
(1999), “The Long-run Evolution of Energy Prices,” Energy Journal, 20, 1-
27.
Reigner, E.
(2007), “Oil and Energy Price Volatility,” Energy Economics, 29, 405-27.
Rodriguez, R.J.
and S.M. Sanchez (2005), “Oil Price Shocks and Real GDP Growth: Empirical
Evidence for Some
OECD Countries,” Applied Economics, 37(2), 201-228.
Sadorsky, P.
(1999), “Oil Price Shocks and Stock Market Activity,” Energy Economics,
21(5), 449-469.
Sari, R. (2006),
“The Relationship between Stock Returns, Crude Oil Prices, Interest Rates,
and Output:
Evidence from a Developing Economy,” Empirical Economics
Letters,
5(4), 205-220.
Sinn, H.W.
(2008), “Public Policies against Global Warming: a Supply Side Approach,”
International Tax
and Public Finance, 15, 360-394.
Slade, M.E.
(1982), “Trends in Natural-Resource Commodity Prices: An Analysis of the
Time-Domain,” Journal of Environmental Economics and Management, 9, 122-37.
Stock, J.H.
(1994), “Unit Roots, Structural Breaks, and Trends,” Handbook of Econometrics,
4, 2739-841.
Sugiyanto, C.
(2008), “Rivalitas Pangan dan Energi,” Warta Ekonomi, February.
Swaray, R.
(2002), “Volatility of Primary Commodity Prices: Some Evidence from
Agricultural
Exports in
Sub-Saharan Africa,” Discussion Papers in Economics, University
of York.
World Bank (2009), Commodities at the
Crossroads..
sumber ; ROWLANDPASARIBU.WORDPRESS.COM
