Beruflich Dokumente
Kultur Dokumente
Quarterly
Energy Economics Review
Volume 4, Number 12, Spring 2007
CONTENTS
Dynamics of Petroleum Markets in OECD Countries In a 2
Monthly VAR Model
Mehdi Asali
Allocation of CO2 emissions in petroleum refineries to 71
petroleum joint products: a case study
Alireza Tehrani Nejad M. - Valérie Saint-Antonin
Abstracts
Age Estimation of Car Fleet and its Impact on Fuel 101
Consumption in Iran: Efficiency vis-à-vis Renovation
Mohammad Mazraati
Impacts of Oil Price Shocks on Economic Variables in a 103
VAR Model
A. Sarzaeem
Recent Crude Oil Market Developments: Making 104
Structural Models
M. Zamani
A Survey of New Structure of LNG and Natural Gas 106
Industry in the World
E. Mansour Kiaee
Dynamics of Petroleum
Markets in OECD Countries
In a Monthly VAR Model
Mehdi Asali1
Abstract
This paper contains results of the first part of a study in which a Vector
Auto-Regression (VAR)/Vector Error Correction (VEC) model is developed
and estimated to investigate dynamics of petroleum markets in OECD. Time
series of the model comprises monthly data for the variables: demand for oil
in OECD, WTI in real term as a benchmark oil price, industrial production in
OECD as a proxy for income and commercial stocks of crude oil and oil
products in OECD for the time period of January 1995 to March 2007. The
detailed results of this empirical research are presented in different sections
of the paper, nevertheless, the general result that emerges from this study
could be summarized as follows: (i) There is convincing evidence of the
series being non-stationary and integrated of order one I(1) with clear signs
of co-integration relations between the series. (ii) The VAR system of the
empirical study appears stable and restore its dynamics as usual following a
shock to the rate of changes of different variables of the model taking
between 5 to 8 periods (months in our case). (iii) It appears that the null
hypothesis of ‘the expected mean of the series being insignificant from
zero’, cannot be rejected in 5% level for each of four time-series indicating
lack of statistical proof for presence of the deterministic trend in time-series
of concern, (iv) Normality tests and histograms of the series reveals that
while distribution of the samples in level differ from theoretical normal
distribution however this is not the case for the differenced time series and
for the residuals, on the other hand autocorrelation functions of the series are
consistent with unit root process .(v) We find the lag length of 2 as being
optimal for the estimated VAR model. (vi), Significant impact of changes in
the commercial crude and products’ inventory level on oil price and on
demand for oil is highlighted in our empirical study and in different
formulations of the VAR model indicating importance of changes in stocks’
level on oil market dynamics. (vii) Income elasticity of demand for oil
appear to be prominent and statistically significant in most estimated models
of the VAR system, while price elasticity of demand for oil is found to be
negligible and insignificant in the short-run.
1. Introduction
Structural changes in petroleum markets and its pricing systems in the
last decades and increasing exposure of the financial markets to
international oil trade have contributed to growing complexity of oil
markets dynamics and volatility of its prices. These complexities are
due to, and in turn intensify, uncertainties surrounding the oil markets
and most often than not undermine robust and accurate prediction of
the oil market developments. Growing demand for energy and oil
particularly in major developing countries with considerable pace,
increasing number of oil producing and exporting nations,
technological changes in` upstream and downstream activities,
integration of oil in the global commodity markets, movements of
open interest in global financial markets on oil trade, impacts of
geopolitical and environmental concerns on petroleum sector and
climate changes are amongst numerous factors effecting oil markets
and its prices. In fact, oil prices have been the most volatile price of all
in commodity markets in recent years.
Because of its significant share in energy mix, which is crucial
for well functioning of every modern society, its important role in the
world economy’s development, and also its exhaustibility, petroleum
markets and its dynamics have always been watched carefully not
only by major consumers and producers but also by economic agents
involved with commodity and financial markets and indeed by the
governments of the producing and consuming nations alike.
Understandably studies related to different aspects of economics of
petroleum have been expanding fast in different directions,
particularly after the oil shocks of 1970s and early 1980s. Turbulences
in oil markets of the late 1990s when oil price reduced drastically due
to the excess supply amid a recession in South East Asian countries,
and continuous upward trend in oil prices on back of persistent tight
markets dominated by growing demand for energy and oil particularly
in major developing countries in last few years, have added yet more
dimensions to dynamics of petroleum sector; further complicating its
analysis.
Although this study is confined to the oil markets in OECD
countries, nevertheless, given this region’s share in the global demand
for oil, investigating developments of petroleum markets in OECD
could contribute to our comprehension of the dynamics of global
petroleum markets. This is a fact that any significant changes in
demand for oil in OECD have always had far-reaching effects on
petroleum markets. Although OECD’s share of global demand for oil
has been continuously declining, due to a fast growing demand for
energy and oil in developing nations; particularly in China, India and
the Middle East, however, in 2006 OECD still was consuming about
60% of global oil supply and according to IEA (2007) OECD will
continue to dominate the oil markets in the foreseeable future with a
55% share of global demand in 2011. The same applies for the crude
and oil products stocks particularly when it comes to commercial
crude and product inventories.
Our study examines relationship between major variables of oil
market in OECD countries for the time period of 1995 to 2007
employing a monthly Vector Auto-regression (VAR)/Vector Error
Correction (VEC) model. The estimated model then is used to forecast
short term demand for oil in OECD. Typically, these relationships are
explored using simple correlation and deterministic trends which in
most cases could not yield consistent and robust results. Advantage of
the VAR modeling approach for our purpose in this study over
possible alternative econometric techniques will become clearer as we
proceed further. However, we have briefly pointed out here to the
shortcoming of the traditional econometric models in oil market
analysis.
Fig.Fig
1. 1.
OilOil Consumption in
Consumption in OECD
OECD and
andthe World
the (000
World b/d)b/d)
(000
90,000
80,000
70,000
60,000
50,000
40,000
30,000
20,000
10,000
0
1970 1975 1980 1985 1990 1995 2000 2005
products) on dynamics of oil markets, i.e. on oil price and through this
variable on demand for oil, thus in this study, our specification of the
four variable VAR model takes in account the effects of oil stock
changes on oil market developments. This is also important to note
that WTI in our VAR is in real term. In other words nominal oil price
is deflated by OECD consumer price index. One can find many
examples of VAR models of oil market treating oil price in real terms
(see for example R. Jimenez-Rodrigues and Marcelo Sanches, 2004).
In fact in this version of the VAR model we are mainly interested in
oil demand in OECD, which is suppose to be a function of real oil
price, rather than impacts of nominal oil price shocks on inflation and
economic activities in his region. In a next attempt we will extend the
VAR model to consider nominal oil prices and other variables as well.
This would be interested to compare the results of this version of the
VAR model with one where impacts of oil price on inflation and
economic growth are examined. The VAR model of four variables in
our study in its most general structure could be introduced as follows:
k k k k
Dt = α1 + ∑ β1i Dt −i + ∑ β 2iYt −i + ∑ β 3i Pt −i + ∑ β 4i St −i +ε 1t
i =1 i =1 i =1 i =1
k k k k
(2) Yt = α 2 + ∑ β 5i Dt −i + ∑ β 6iYt −i + ∑ β 7 i Pt −i + ∑ β 8i St −i +ε 2t
i =1 i =1 i =1 i =1
k k k k
Pt = α 3 + ∑ β 9i Dt −i + ∑ β10iYt −i + ∑ β11i Pt −i + ∑ β12i St −i +ε 3t
i =1 i =1 i =1 i =1
k k k k
St = α 4 + ∑ β13i Dt −i + ∑ β14iYt −i + ∑ β15i Pt −i + ∑ β16i St −i + ε 4i
i =1 i =1 i =1 i =1
Where, as defined before, Dt, Yt, Pt and St stands for Demand for
oil in OECD, an index of monthly industrial production in OECD,
WTI benchmark oil prices and oil stock (crude and product) in OECD
countries at time t and respectively. Coefficients of the model “ β i ”
and intercepts α i are to be estimated.
N 2 ( K − 3) 2
(3) J −B= [S + ]
6 4
Where S is the skewness, and K is the kurtosis (see: Eviews, 6,
2007). Under the null hypothesis of a normal distribution, the Jarque-
Bera statistic is distributed as χ 2 with 2 degree of freedom. The
reported Probability is the probability that a Jarque-Bera statistic
exceeds (in absolute value) the observed value under the null
hypothesis- a small probability value leads to the rejection of the null
hypothesis of normal distribution. So for example if in a test of
normality J-B statistic’s probability was reported to be about 0.04 we
could say the hypothesis of normal distribution is rejected in 5% but
not at the 1% significant level. In estimated VAR model, the J-B test
statistic is a multivariate extension of the test, which compare the third
and fourth moments of the residuals to those from normal distribution.
Usually Doornik and Hansen (1994) approach to normality test is
adopted.
From table (1) we can see that, probability of the normality of
the distributions of variables in first difference and also the VAR
residual is higher than that of the variables in level. This is also
evident from visual inspection of the histograms of the model’s
variables in level and first difference.
Generally speaking we cannot reject normality of these variables
in difference and residuals of the VAR model except for variable
LTSOECD- stock of crude and products in OECD. The test provides
normality of distribution of this variable in level form. Since the
normality test, along with AR and ARCH tests is usually considered a
test of mis-specification, so having found evidence of the residuals of
the VAR model being derived from normal distribution is encouraging
on the basis that this is an indication of proper specification of the
model.
While it is difficult to discern if a time-series is nonstationary by
visual examination of level data, however, it is always a good idea to
start with a visual inspection of the data and their time series
.08 .3
.2
.04
.1
.00
.0
-.04
-.1
-.08 -.2
-.12 -.3
1996 1998 2000 2002 2004 2006 1996 1998 2000 2002 2004 2006
a. Difference of (Log of) Monthly Demand for Oil OECD b. Difference of (Log of) Real WTI
.015 .04
.010 .02
.005
.00
.000
-.02
-.005
-.04
-.010
-.015 -.06
1996 1998 2000 2002 2004 2006 1996 1998 2000 2002 2004 2006
c. Difference of (Log of) Industrial Production Index OECD d. Difference of (Log of) Total Commercial Stock OECD
residual but to its size relative to most of the other residuals in the
regression. Some of most outstanding outliers for WTI series appear
to be around year 2000, 2001 and 2005. The Katrina hurricane in July
2005 in Golf of Mexico and the South East of the USA appears to be
the reason behind price violation of mid 2005 reflected in the time
series residuals.
Fig.Fig.4
3. WTI
WTIPrices:
Prices: Firstand 12-Month
First and 12-Month Difference
Difference in Logarithms
in Logarithms
-1
Prices Changes (in Logarithms)
-2
-3
1996 1998 2000 2002 2004 2006
-1
-2
-3
1996 1998 2000 2002 2004 2006
.008
.04
.004
.00 .000
-.004
-.04
-.008
-.08 -.012
1996 1998 2000 2002 2004 2006 1996 1998 2000 2002 2004 2006
.04
.1
.02
.0 .00
-.02
-.1
-.04
-.2 -.06
1996 1998 2000 2002 2004 2006 1996 1998 2000 2002 2004 2006
10
12
8
Density
Density
8 6
4
4
2
0 0
10.60 10.65 10.70 10.75 10.80 10.85 10.90 4.3 4.4 4.5 4.6 4.7
LRWTI LTSOECD
2.0 16
1.6
12
1.2
Density
Density
8
0.8
4
0.4
0.0 0
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 14.64 14.68 14.72 14.76 14.80 14.84 14.88
LIPIOECD
1.0
LDOECD
A u to c o rre la tio n
0.8
.8
0.6
A u to co rre latio n
.6
0.4
.4
0.2
.2
2 4 6 8 10 12 14 16 18 20 22 24
.0 Actual Theoretical
2 4 6 8 10 12 14 16 18 20 22 24
Actual Theoretical
LRWTI LTSOECD
1.0 1.0
0.8
A u to c o r re la tio n
A u to c o rre la tio n
0.5
0.6
0.4
0.0
0.2
0.0
-0.5
2 4 6 8 10 12 14 16 18 20 22 24
2 4 6 8 10 12 14 16 18 20 22 24
Actual Theoretical
Actual Theoretical
100
16
80
12
Density
Density
60
8
40
4
20
0 0
-.12 -.08 -.04 .00 .04 .08 .12 -.02 -.01 .00 .01 .02
DLRWTI DLTSOECD
7 40
6
30
5
4
Density
Density
20
3
2
10
1
0 0
-.3 -.2 -.1 .0 .1 .2 .3 -.08 -.04 .00 .04 .08
DLDOECD
DLIPIOECD
.6
.4
.4
Autocorrelation
.3
Autocorrelation
.2 .2
.1
.0
.0
-.2
-.1
-.4 -.2
2 4 6 8 10 12 14 16 18 20 22 24 2 4 6 8 10 12 14 16 18 20 22 24
DLRWTI
DLTSOECD
.2
.4
.1
A u to c o rre la tio n
A u to c o rre la tio n
.2
.0
.0
-.1
-.2 -.2
-.3 -.4
2 4 6 8 10 12 14 16 18 20 22 24 2 4 6 8 10 12 14 16 18 20 22 24
It appears that all four time series of the VAR model are
integrated of order one or in other word first differences of these
process are stationary which will be investigated along with other
statistical properties of the model in the next sections. The probability
distribution and autocorrelation functions (ACF) for the level and first
differences of the series are presented in figures (5) and (6).
Histograms summarize the frequency of occurrence that a variable
falls within a certain range of values. As such, histograms provide
insights into the mean, standard error and probability distribution of
given variable. Histograms of the demand for oil (D=LDOECD) and
Industrial production Index (Y=LIPIOECD) in OECD, and histograms
of (P=WTI) bench mark oil prices and total commercial stock
(S=LTSOECD) are presented in the top panel of Figure (5) and Figure
(6) respectively, with a plot of nominal distribution superimposed over
the histogram and a smooth line generated from the histogram.
It is apparent that none of the time series in level distributed
normally, hence the variables are skewed differently relative to the
normal distribution. The ACFs through 24 lags of the variables are
presented in the lower panel of the figures. The ACFs are highly
significant. The wavy cyclical correlogram with a seasonal frequency
for demand for oil and also commercial stocks suggest a seasonal
pattern for these variables. These graphical and statistical finding are
consistent with unit root processes.
Panels of Figure (6) depict histogram and ACFs for the first
differences of all the variables of the model. In each case, the
differencing transformation appears to have resolved the non-
stationarities in the four series for the most parts. In all cases, the
probability distributions of the differenced series appear to be
approximately normally distributed. Further, the auto- correlations are
also much smaller, and statistically insignificant at most lags. These
results lend support to the inference that the time series of our
VAR/VEC model are unit roots in level and integrated processes of
order one, I(1).
Therefore the problems associated with non-stationary data will
be a key consideration in the model specification and estimation that
follows.
Fig. 7. The Empirical and Normal Density of the Four VAR Residuals
Fig. 14 The Empirical and Normal Density of the Four VAR
Residuals
RESIDUALS LIPIOECD
RESIDUALS LDOECD
25 120
100
20
80
15
Density
Density
60
10
40
5
20
0 0
-.10 -.05 .00 .05 .10 -.02 -.01 .00 .01 .02
8 40
6 30
Density
Density
4 20
2 10
0 0
-.3 -.2 -.1 .0 .1 .2 .3 -.08 -.04 .00 .04 .08
error for one observation is large this dose not mean that the next error
term also will be large. Indeed, the fact that an term is positive should
have no implications for whether the next term is positive or negative.
Fig. 8. Autocorrelogram of the Four VAR Residuals
Fig. 15 Autocorrelogram of the Four VAR Residuals
.8 .8
.4 .4
.0 .0
-.4 -.4
-.8 -.8
2 4 6 8 10 12 14 2 4 6 8 10 12 14
.8 .8
.4 .4
.0 .0
-.4 -.4
-.8 -.8
2 4 6 8 10 12 14 2 4 6 8 10 12 14
( 6) X t = ΦDt + Π 1 X t −1 + Π 2 X t −2 + ε t
t = 1,...., T , and ..ε t ~ N p (0, Ω)
where Dt contains constant values, are reported below. An
inspection of the estimated coefficients reveals more significant
coefficients at lag 1 than lag 2. Also we realize that most of the
coefficients with large t-ratios are on the diagonal, implying a highly
autoregressive character of the variables. Also the significance of the
crude and product stocks in oil demand and oil price equations is
worth noting. These results support our earlier remark on the
importance of the crude and product stocks for oil price and oil
demand.
..............................The.Estimated.Model. X t = ΦD + Π 1 X t −1 + Π 2 X t − 2 + ε t
________________________________________________________________________
Dt 2.12 0.60 .. − 0.03......0.50 ......0.82 Dt −1
∗ ∗
P 6.8
+ − 0.25......1.05 . − 1.93 ....1.4 Pt −1 +
∗ ∗
(7 ) t =
S t 2.67 ∗ − 0.13∗.....0.01......0.83∗.. − 0.39 S t −1
Yt − 0.24 0.001.. − 0.02 ... − 0.002....0.94 Yt −1
∗ ∗
1.0 0.00066
0.109......1.0
Ω= ,..σˆ ε = 0.00502
0.051......0.26.......1.0 0.00022
0.76........0.086.......0.096......1.0 0.00002
10.80
10.76
.03 10.72
.02
10.68
.01
.00
-.01
-.02
-.03
95 96 97 98 99 00 01 02 03 04 05 06
-0.8
-1.2
.3 -1.6
.2 -2.0
.1 -2.4
.0
-.1
-.2
95 96 97 98 99 00 01 02 03 04 05 06
Fig.Fig.
9.c.10.c
Commerclal Stocks of Crude and Products. Fitted
Commercial Stocks of Crude and Products,
values and Residuals
Fitted values and Residuals
14.84
14.80
14.76
.06
14.72
.04
.02 14.68
.00
-.02
-.04
-.06
95 96 97 98 99 00 01 02 03 04 05 06
4.7
4.6
.015 4.5
.010
4.4
.005
.000
-.005
-.010
-.015
95 96 97 98 99 00 01 02 03 04 05 06
.02 .02
.01 .01
.00 .00
-.01 -.01
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of LP to LS Response of LP to LD
.100 .100
.075 .075
.050 .050
.025 .025
.000 .000
-.025 -.025
-.050 -.050
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of LS to LD Response of LP to LY
.100
.02
.075
.01
.050
.025
.00
.000
-.01
-.025
-.050
-.02
1 2 3 4 5 6 7 8 9 10
1 2 3 4 5 6 7 8 9 10
(c) if any of the eigenvalues are outsise the unit circle, then {Xt }is
explosive.
For the multivariate (four variable) demand for oil VAR(2) model,
we will have 4×2=8 roots, the model of which are reported directly
from the computer output in table (2) below (rounded to point three
digits):
1.0
0.5
0.0
-0.5
-1.0
-1.5
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
In this part of the paper we will estimate the model after having
examined the hypothesis of the variables of the system being I(1).
Given the time series of the model are I(1) we first will discuss its
short-run features by transferring data to a set of stationary time-
series. In the next part of the paper we will embark on a VEC
modeling of the time-series examining the co-integration relations and
the long-run characteristics of the system. On the basis of the
preceding discussion our analysis would proceed with a formal test of
the integration of the VAR model time-series. Two forms of the
Vol. 4, No. 12 / Spring 2007 / 31
Quarterly Energy Economics Review
the column LogL increase with p. The Hannan-Quinn (HQ) and the
Schwartz (SC) criterions select the order 1 while the Akaike (AIC)
and the Final Prediction Error (FPE) select the order 2. However, the
log-likelihood ratio statistics rejects order 1, but do not reject a VAR
of order 2.
The log-likelihood ratio statistics (LR column on table 5) are
computed for testing the hypothesis that the order of the VAR is (p)
against the alternative that it is P (the Maximum expected order), for
p=1,2… ,P-1. To test this hypothesis one should construct the relevant
log-likelihood statistics for these tests by using the maximized values
of the log-likelihood function given in the relevant column of the
result table. For example, to test the hypothesis that the order of the
VAR model is 2 against 3 the relevant log-likelihood ration statistics
is given by:
(15) LR (2 : 3) = 2( LogL3 − LogL2 )
Where LogLp =p=1,2,..,p refers to the maximized value of the
log-likelihood function for the VAR(p) model. Under the null
hypothesis, LR (2:3) is distributed asymptotically as a chi-squared
variate with m 2 (3 − 2) = m 2 degrees of freedom, where m is the
dimension of coefficient vector of the variables in an standard VAR(p)
model.
Table 5. Testsoflaglengthor Model Reduction for VAR
Lag LogL ..............LR ..............FPE ..............AIC ................SC ....................HQ
____________________________________________________________________________
0 736.1135 ....NA ............. 2.48e - 10 -10.76638 -10.68071 -10.73156
1 1450.608 1376.453 8.58e - 15 -21.03836 - 20.61003 * - 20.86430 *
2 1466.798 30.23674 8.56e - 15 * - 21.04115 * -20.27015 -20.72784
3 1478.728 21.57805 9.10e - 15 -20.98129 -19.86762 -20.52872
4 1495.090 28.63356 9.08e - 15 -20.98661 -19.53028 -20.39480
5 1513.493 31.12377 8.80e - 15 -21.02196 -19.22297 -20.29089
6 1526.956 21.97606 9.20e - 15 -20.98465 -18.84299 -20.11433
7 1546.722 31.10212 * 8.79e - 15 -21.04003 -18.55570 -20.03046
8 1556.867 15.36716 9.71e - 15 -20.95393 -18.12694 -19.80511
_____________________________________________________________________________
* indicates lag order selected by the criterion
LR : sequential modified LR test statistic (each test at 5% level)
FPE : Final prediction error
AIC : Akaike information criterion
SC : Schwarz information criterion
HQ : Hannan - Quinn information criterion
(16) ( )
AIC = log DetΣˆ + 2 × c × T −1
( )
BSC = log DetΣˆ + 2 × c × log(T )× T −1
( )
HQ = log DetΣˆ + 2 × c × log(log(T )) × T −1
The first terms on the right-hand side of the equations are the
log determinants of the estimated residual covariance matrix. The log
determinant of the estimated residual covariance matrix will decline as
the number of regressors, just as in a single equation ordinary least
squares regression. It is similar to the residual sum of squares or
estimated variance. The second term on the right-hand side acts as a
penalty for including additional regressors (c) which scaled by the
inverse of the number of observations (T). The lag length chosen is
the model with minimum value for the Statistics. The three tests do
not always agree on the same number of lags. In practice, the use of
the SBC is likely to result in selecting a lower order VAR model than
when using AIC. Notice that it is quite usual for the SBC to select a
lower order VAR as compared with the AIC. Also as was discussed in
our case and according to the statistics given on table (5) order 1 is
rejected by the log-likelihood ratio. In the light of these reasons we
choose the VAR (2) model.
(19) t = t −1
∆S t 0.001 − 0.17 ∗......0.03... − 0.14.... − 041. ∆S t −1
∗
∆Yt 0.002 − 0.01........0.02 ∗ − 0.01.... − 0.02. ∆Yt −1
− 0.14..... − 0.02.... − 0.03.......0.66 ∆Dt − 2 ε 1t
− 0.23..... − 0.06.......0.38.... − 0.24
∆Pt − 2 + ε 2 t
0.04........ − 0.00.......0.32 ∗.. − 0.11 ∆S t − 2 ε 3t
0.00......... − 0.00......0.03.... − 0.18 ∆Yt − 2 ε 41
∗
∗
.Significant.at.5%.level
1 0.00073
0.06.......1.0 0.00515
Ω = . ................σˆ ε =
0.72.......0.10......1.0 0.00022
0.021.....0.10......0.11.....1. 0.00002
.04 .1
.00 .0
-.04 -.1
-.08 -.2
1996 1998 2000 2002 2004 2006 1996 1998 2000 2002 2004 2006
.010
.02
.005
.00
.000
-.02
-.005
-.04
-.010
-.06 -.015
1996 1998 2000 2002 2004 2006 1996 1998 2000 2002 2004 2006
Density
Density
60
model:
20
40
10
20
0 0
-.08 -.04 .00 .04 .08 -.02 -.01 .00 .01 .02
∑φ 14 (i )
( 22) ρ (i ) = i =0
n
∑φ
i =0
44 (i )
changes of oil prices on demand for oil but the magnitudes of impacts
are different as a shock to the rate of changes of demand for oil has
greater effect on the rate of oil price changes vice versa. Also we
notice that almost all the shocks are faded in 5 to 7 time periods. So,
although a shock to the rate of changes of stocks (crude and products)
on oil price is almost twice as its impact on demand for oil (in
absolute terms), however, both impulse impacts fade away in
approximately the same time period.
Fig. 13. Dynamics Fig.
of the Short-run
Dynamics of the Model
Short-run(39) Response
Model (39) to Cholesky
Response
Oneto Cholesky One S.D. Innovations
S.D. Innovations ± 2 S.E. ± 2 S.E.
Response of D(LD) to D(LP) Response of D(LD) to D(LY)
.04 .04
.03 .03
.02 .02
.01 .01
.00 .00
-.01 -.01
-.02 -.02
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
.08 .08
.04 .04
.00 .00
-.04 -.04
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
.010
.004
.005
.002 .000
-.005
.000
-.010
-.002 -.015
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
( 6) X t = ΦDt + Π1 X t −1 + Π 2 X t − 2 + ε t
t = 1,...., T , and ..ε t ~ N p (0, Ω)
1.0 0.00066
0.05......1.0
Ω= ,..σˆ ε = 0.00502
0.75......0.10.......1.0 0.00022
0 .03 ........ 0 .13...... 0 .087...... 1 .0 0.00002
.02 .02
.01 .01
.00 .00
-.01 -.01
-.02 -.02
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of D(LP) to D(LD) Response of D(LP) to D(LS)
.08 .08
.06 .06
.04 .04
.02 .02
.00 .00
-.02 -.02
-.04 -.04
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
.010 .06
.005 .04
.000 .02
-.005 .00
-.010 -.02
-.015 -.04
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
∆ P 6. 8
+ − 0.25 .......0.05 .... − 1.94 .....1.42 ∆Pt −1 +
∗
(28) t =
∆S t 2.67 ∗
− 0.14 ...0.013 .... − 0.17 .... − 0.39 ∆S t −1
∗
∆Yt − 0.24 ...0.001 ...0.02 ... − 0.002 ....... 0.06 ∆Yt −1
∗
1.0 0.00066
0.05 ......1.0
Ω= ,..σˆ ε = 0.00502
0.75 ......0.10....... 1.0 0.00022
0 . 03 ........ 0 . 13 ...... 0 . 087 ...... 1 . 0 0.00002
Statistics .. for .individual .. var iables .......D.......... ....... P.......... .....S .......... ...Y
R 2 .......... .......... .......... .......... .......... .......0.36.......... ..0.15.......... 0.10....... 0.10
Adj − R 2 .......... .......... .......... .......... ....... 0.32.......... ..0.10.......... 0.05 ........0.05
F − statisic .......... .......... .......... .......... ....9.26.......... .2.9.......... ..1.9.......... .1.97
Common .sample .statistics ,. Akaik .Informatio n.Criterion ( AIC ), and .Schwartz .Criterion ( SC )
ˆ = −32 .86,.. AIC = −21, ,.SC = −20
Log ( L max) = 1534 .9,..... log Ω
__________ __________ __________ __________ __________ __________ __________ __
.02 .02
.01 .01
.00 .00
-.01 -.01
-.02 -.02
-.03 -.03
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
.08
.08
.04
.04
.00
.00
-.04
-.04 1 2 3 4 5 6 7 8 9 10
1 2 3 4 5 6 7 8 9 10
.010
.08
.005
.000 .04
-.005
.00
-.010
-.015 -.04
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
supporting the hypothesis that like many economic series, demand for
oil, oil prices, industrial production and stock of crude and products
are non-stationary time series. This can have important implication
with respect to estimation, forecasting and policy analysis as not
taking into account possible non-stationarity in the data could lead to
misleading results and failing to catch important properties of the data.
While this does not mean that VAR-VEC modeling approach are
always superior to the other modeling techniques, nevertheless, VAR-
VEC analysis can provide a useful benchmark for empirical
investigation of oil markets dynamics.
References
Atkins, J. F. (2005) “Final Report on VAR Estimation of Oil
Consumption, Price and GDP Relationship” University of Calgary
and PMA Department of OPEC Secretariat
Barrel, B. and O. Pomerantz (2004) “Oil Price and theWorld
Economy” NIESR, London
Barsky, R. and L. Kilian (2004) “Oil and the Macroeconomy Since the
1970s” Centre for Economic Policy Research, London
Bernanke, B. S., M. Gertler and M. Watson (1997) „Systematic
Monetary Policy and the Effect of Oil Price Shocks“ Brooking
Papers on Economic Activity.USA
Boswijk, H. P. (1995) “Identifying of Cointegrated Systems” Working
Papers, Tinbergen Institute and Department of Actual Science and
Econometrics, University of Amsterdam, the Netherlands
Box, G. E. P., G. M. Jenkins and G. C. Reinsel (1994) “Time Series
Analysis” Prentice Hall, New Jersey
Brown, S. P. A. and M. K. Yuecel (2001) “Energy Prices and
Aggregate Economics Activity” Federal Reserve Bank of Dallas.
Brown, S. P. A., M. K. Yuecel and J. Thompson (2003) “Business
Cycle: The Role of Energy Prices” Federal Reserve Bank of Dallas.
Cziraky, D. (2001) “Cointegration Analysis of the Monthly Time-
Series Relationship Between Retail Sales and Average Wages in
Appendix
1. Stationarity
From the preceding discussion it becomes clear that a time series
analysis of oil markets would require, from the outset, verification of
the classical econometric theory assumption of stationarity of the data
generating processes representing the market of concern. These
assumptions intend that means and variances of the process are
constant over time. However, graphs of monthly data of the petroleum
markets variables for the time period under study (1995-2007) some
of which are depicted in different panels of Fig. 2 and the historical
record of economic forecasting of these variables would suggest
otherwise.
Petroleum markets evolve, grow and change over time in both
real and nominal terms, some times dramatically- and economic
forecasts of oil prices, its supply and demand are often badly wrong,
whereas forecasting failure of this magnitude should occur relatively
infrequently in a stationary process (see Hendry, and Juselius, 1999).
A significant deterioration in forecast performance relative to the
anticipated outcome confirms that petroleum markets’ data are not
stationary because even poor models of stationary data would forecast
on average as accurately as they fitted, yet that manifestly does not
occur empirically.
Figure 1 shows the four time series of the petroleum market in
OECD that we are trying to establish the relationship between them.
The first panel reports the monthly time series of demand for oil
Vol. 4, No. 12 / Spring 2007 / 61
Quarterly Energy Economics Review
4.68
10.85 LDOECD
4.64 LIPIOECD
10.80 4.60
4.56
10.75 4.52
4.48
10.70
4.44
10.65 4.40
1996 1998 2000 2002 2004 2006 1996 1998 2000 2002 2004 2006
-0.4 14.85
LTSOECD
-0.8
14.80
LRWTI
-1.2
14.75
-1.6
14.70
-2.0
-2.4 14.65
1996 1998 2000 2002 2004 2006 1996 1998 2000 2002 2004 2006
WTI in Real Term Total Commercial Stock of Crude and Products: OECD
2. Addressing non-stationarity
As was mentioned non-stationarity is observed in economic time
series very frequently so much that it seems natural feature of
economic life. Legislative changes are one obvious source of non-
stationarity, often inducing structural breaks in time series, but it is far
from the only one. Economic growth, perhaps resulting from
technological progress, ensures secular trends in many time series.
Such trends need to be incorporated into statistical analysis. In time
series analysis of economic systems such as petroleum markets focus
usually are on a type of stochastic non-stationarity induced by
persistent accumulation of past effects, called unit-root process. Such
process can be interpreted as allowing a different ‘trend’ at every
point in time, so are said to have stochastic trends.
6. Model selection
The most difficult aspect of working with time series data is model
selection. The crucial point to understand is setting up models in terms
of components which have a direct interpretation. This enables the
researchers to formulate, at the outset, a model which is capable of
reflecting the salient characteristics of the data. Once the model has
been estimated, its suitability can be assessed not only by carrying out
diagnostic tests to see if the residuals have desirable properties but
also by checking whether the estimated components are consistent
with any prior knowledge which might be available. Thus for
example, if a cyclical component is used to model the oil trade cycle,
knowledge of the petroleum economics history of the period should
enable one to judge whether the estimated parameters are reasonable.
This is in the same spirit as assessing the plausibility of a regression
model by reference to the sign and magnitude coefficients.
Classical time series analysis is based on the theory of stationary
stochastic process, and this is starting point for conventional statistical
time series model building. It can be shown that most stationary
process can be approximated by a model from the class of
autoregressive moving average (ARMA) models. However, a much
wider class of models, capable of exhibiting non-stationary behavior,
can be obtained by assuming that a series can be represented by an
ARMA process after differencing. This is known as the autoregressive
integrated moving average (ARIMA) class of models and a model
selection strategy for such models was developed by Box and Jenkins
(1976).
The following criteria for a good model have been proposed in
the econometrics literature. (Hendry and Richard(1983), Mizon and
Richard(1986), Ericson and Hendry (1985) Mizon(1984) and
Haevey(1981a). They apply to both structural time series and pure
time series modeling.
(a) Parsimony A parsimonious model is one which contains a
relatively small number of parameters and other things being
equal, a simpler model is to be preferred to a complicated one.
a leading indicators.
A multivariate time series model offers the possibility of
allowing for interactions among the variables when forecasts are
made. As with univariate time series modeling, the effectiveness of a
multivariate model depends on the extent to which it remains stable
over time.
Having reviewed the main features of the VAR modeling
approach in context of our empirical study of the oil market dynamics
in OECD we now turn to specify the VAR model outline of which
was discussed above briefly.
Abstract
Oil refining is a joint production system because the production of one
oil product makes technically inevitable the production of other oil
products. Due to the complex nature of the process involved and the
vast number of joint product outputs that are strongly correlated, it is
very difficult to establish any meaningful CO2 emissions allocation
between oil products. Nevertheless, the allocation of petroleum
refinery energy use and the resultant CO2 emissions among different
oil products is necessary in “well to wheel” analysis in order to
evaluate the environmental impacts of individual transportation fuels.
In practice, allocation methods used so far for the petroleum-
based fuel are traditionally based on two fundamental approaches:
physical measures (mass, volume, energy contents or other relevant
parameters) or market value of individual oil products from a given
refinery. These methods are open to discussion on two points. Fist, an
a priori assumption about the allocation procedure (i.e., mass, volume,
energy contents, market value, etc.) is in some ways completely
1. alireza.tehraninejad@gmail.com
2. valerie.saint-antoni@ifp.fr
Vol. 4, No. 12 / Spring 2007 / 71
Quarterly Energy Economics Review
1. Introduction
Life Cycle Assessment (LCA) is one of the engineering methods that
has increasingly gained attention and is regarded today as an
important tool for environmental policy and strategic decision making.
1. Throughout this paper, by marginal (average) CO2 content of a given oil product we mean
the additional CO2 emissions associated with the marginal (average) production of that oil
product within the refinery.
1. The cetane number measures the speed at which diesel burns in an engine when subjected
to high temperature and pressure (Favennec, 1998).
2. The degree API is an arbitrary scale for the measurement of the density of crude oil. The
relationship between relative (with respect to water) density and degree API is given by the
formula: °AP I = 141.5/d −131.5 (Favennec, 1998).
optimal combination of inputs are such that all final product demands
are satisfied without any excess of production. This assumption is
justified within a static LP model in which no inventory or exportation
variable is defined. Moreover, we denote Β the (k × k) basic matrix
and β the set of basic index. Sections 3 and 4 describe how the optimal
LP solution could be used in order to yield non arbitrary prospective
and retrospective WTT information.
b
(2) 0
xB = B −1
0
f −1
where xB represents the vector of basic variables and B
corresponds to the inverse of the basis (a matrix) which is pre-
multiplied by the right-hand-side (RHS) vector. At the optimum, the
emission variable ε is always positive and can be expressed as
follows,
b
T −1 0
(3) ε = eε B
0
f
where eε is the ε th unit vector ( eεt = 0 for t ≠ ε and eεε = 1 ) and
eεT B −1 corresponds to the row of B −1 associated with the basic CO2
variable ε. Simplifying (3), we get:
(4) ε= ∑ α i bi + ∑ γ j f j ,
i∈M j∈ S
1. The exergy content of a system indicates its distance from the thermodynamic equilibrium.
The higher the exergy content, the farther from the thermodynamic equilibrium (definition
from, http://www.holon.se/folke).
be 3.5 times more than the total unallocated CO2 emissions due to the
total process-related emissions, i.e., ∑ j γ i f j .
Achieving retrospective LP-based coefficients requires the
reassignment of the total process-related emissions over the oil
products. We achieve this objective within a two stage procedure from
the optimal emission function in relation (4) (For a complete
discussion and a numerical example, see Tehrani Nejad M., 2007a). In
this regard, care should be taken not to fall in the realm of any
arbitrary or ad hoc measures.
where, by construction, ∑ i ∈M
aik + lk = 1. To separate the part of
each output in γj, we introduce the following (k−1)×(k−1) allocating
matrix Ωi and Ω l :
Ω i = diag ( aik , i ∈ M , k ∈ β , k ≠ ε )
(9)
Ω l = diag (lk , k ∈ β , , k ≠ ε ),
Where the coefficients aik are the average productivity of crude oils
associated with the I th row of B. Similarly, the coefficients lk
correspond to the loss coefficients associated with the material balance
row for losses in B.
Using definition (9), relation (8) can be rewritten as
(10) γ j = EBT ( ∑ Ωi + Ω l )ς j
i∈M
Where construction, ∑ i ∈M
Ωi + Ω l = diag (1, 1,..., 1)
Rearranging (10), we sp separate the part of oil product outputs
from process loss in γj.
part of oil products
part of loss
64748 6 474 8
(11)
γ j = ∑ EBT Ωiς j + EBT Ω lς j .
i∈M
Relation (11) relates each γj to the refinery's outputs (including the
process losses) through a realistic technical relationship which
emerges from the equilibrium behavior of the firm.
Now, using the decomposition relation (11) and doing some
algebraic manipulations the optimal emission function (5) can be
expressed as
αi θ 4
6447 448 647 8
(12) ε =∑ (α i + ∑ E B δ ij )bi + (∑ E B δ lj )l.
T T
(14) l = ∑ ϑi bi + ∑ υi f j ,
i=M j∈S
that relates the total process loss of the refinery to the oil
products b and the limited (fixed) unit processes x through the
marginal allocation coefficients, ϑi (i ∈ M ) and υi ( j ∈ S ) .
Relating the refinery loss to solely oil products needs
reallocating the loss contribution of scarce unit process over oil
products. By analogy to γ j , the marginal coefficient υ j can be
formulated by differentiating the loss constraint in (6) with respect to
x j as follows
dx dl
(15)
∑ lk k −
dx dx
= 0,
k∈β
k ≠l
j
{ j
=υ j
where the vector lk corresponds to the loss row in B, from which the (-
dxk
1) coefficient associated with l is omitted. The vector ( )
dx j
corresponds to the column of B −1 associated with the slack variable of
the jth scarce unit process, from which υ j is extracted. For notational
dxk
convenience, we set ( ) = ς~j .
dx j
By using the same procedure as the first stage, relation (15) can
be rewritten as
~
(16) υ j ≈ lBT ( ∑ Ωi )ς~j
i∈M
~
where the (k − 1) × (k − 1) allocating matrix Ωi is defined as
(17) ~
Ωi = diag (aik , i ∈ M , k ∈ β , k ≠ l),
ϑ
6447i 448
With ∑i∈M (ϑi + ∑ lBT δ ijn )bi .
j∈S
An exact reallocation scheme for υi ( j ∈ S ) requires normalizing
the input-output coefficients associated with each column of the
Vol. 4, No. 12 / Spring 2007 / 85
Quarterly Energy Economics Review
~
υ j = lBT ( ∑ Ωin )ς~j
i∈I
1. These are considered to be typical of the quality of crudes currently available in European
refineries (Saint-Antonin, 1998).
insurance and commission rate. Here, we used the average CIF prices
for year 2005 (Table 2).
The first standard process is the topping unit, in which crude oils
are distilled at atmospheric pressure and separated into various
fractions according to their boiling points. Light fractions (with a
boiling point lower than 180°C) are used to make light petroleum gas
(i.e., propane and butane), naphtha and gasoline whilst middle
fractions (distilled between 180 and 360°C) contribute to the
production of jet fuel and diesel oil. Straight-run cuts can not be used
directly to blend pools since their characteristics are too far from end-
product specifications.
Various process units operate in order to improve these
intermediate products. In particular, acatalytic reforming unit
upgrades the heavy naphtha cut into an aromatic gasoline component
with significantly higher octane numbers. This unit provides, as a
precious by-product, the hydrogen required for the hydro-treating and
desulfurization units. Moreover, two different isomerization units are
also modeled to convert n-paraffins into isoparaffins1 of substantially
1. Severity is a way of defining the operating conditions, which can be more or less severe, of
a process unit.
diesel in China are respectively 800 ppm and 2000 ppm, or 16 and 40 times higher
than European specifications. For our base case study (year 2005), we considered
the European specifications of the year 2005 for automotive fuels (Table 3).
1. In order to guarantee a 10 ppm sulfur level at the pump, generally refineries aim at a level
of 6 or 7 ppm to take into account the pipeline contamination issues.
1. Here, an optimal departure means to preserve the cost efficient equilibrium of the refinery.
the final simplex tableau, that is the optimal vector ϑi. . Table 7
summarizes these information in order to perform the first phase
calculations. Commercially available software for large scale models
include some especial commands to extract these coefficients. The
mathematical software we use is LAMPS (Linear And Mathematical
Programming System) and the available command is
TRANSFORMCOLUMNS (Advanced Mathematical Software Ltd.,
1991).
The quasi average CO2 contents α i (i ∈ M ) are calculated
according to relation (12) in Table 6. As it was expected, the relative
error term of the first step, є = θ l / ε , for the three simulations were
respectively 2%, 3% and 5%1 . We consider that these first-stage CO2
allocations are good estimations of the final average CO2 content
α iA (i ∈ M ) and the procedure can be successfully stop here.
Table 6. Evolution of the average CO2 contents
Oil products Base case Scenario 2008 Scenario 2010
Gasoline 0.302 -1.189 -0.931
1. The complete computations are available upon request from the first author.
Vol. 4, No. 12 / Spring 2007 / 95
Quarterly Energy Economics Review
DI = topping unit, RF = Reforming unit, IS = isomerization unit of type A IR = isomerization unit of type
B, PE = FCC feed Pretreatment, BT prod. = Bitumen production HA = revamp of the desulfurisation unit,
HO imp. = Heating oil importation JF = Jet fuel importation, HA + HX = Total capacity of
desulfurisation units.
7. Conclusion
In this paper we distinguished between prospective (marginal) and
retrospective (accounting) WTT analysis. We argued that prospective
analysis should be considered when the objective is to explore the
environmental effects associated with the marginal production of a
given automotive fuel. On the other hand, retrospective analysis is of
interest when the main objective is to evaluate the average
environmental impacts of a given automotive fuel in transportation
studies. It was also explained that, an exact prospective/retrospective
study for the production of automotive fuels requires to assess the
marginal/average contribution of gasoline and diesel to the total CO2
emissions generated within the refinery. Oil refining is one of the most
complex joint production system, and traditional WTT methods fail to
account for the complete interaction and substitution effects among
the process units.
In order to compute the marginal/average contribution of
automotive fuels at the gate of refineries, a practical method based on
linear programming was developed. We illustrated that the
marginal/average LP-based emission coefficients which emerge from
the optimal solution, as opposed to the ones computed by traditional
methods, include all consequences of the desired change on the
operation of the refinery as well as compositional changes of the oil
products. In other words, these emission coefficients embody the
physical and process relationships in the refinery system and provide a
more realistic estimates of the environ-mental impacts of automotive
fuels.
Using the LP model developed in Sections 3 and 4, we
estimated the marginal/average CO2 contribution of the petroleum
products for a typical European refinery. Then, three simulations for
years 2005, 2008 and 2010 were performed to evaluate the impact of
the sulfur reduction policy on the CO2 content of automotive fuels at
the gate of the refinery. Based on the obtained numerical results, the
following core conclusions can be highlighted. Due to the transport
and fiscal policies in most of the European countries, the demand for
automotive diesel, at the expense of gasoline, has been drastically
Vol. 4, No. 12 / Spring 2007 / 97
Quarterly Energy Economics Review
References
Advanced Mathematical Software Ltd., 1991. Linear And
Mathematical Programming Sys-tem (LAMPS), User Guide,
Version 1.66.
Azapagic, A., Clift, R., 1994. Allocation of Environmental Burdens by
Whole-System modeling – The Use of Linear Programming. In:
Allocation in LCA, Huppes and Schneider (Eds.), SETAC,
Brussels, 54-60.
Azapagic, A., Clift, R., 1995. Life Cycle Assessment and Linear
Programming – Environ-mental Optimization of Product System.
Computers Chemistry Engineering 19 (Suppl.), S229-S234.
Azapagic, A., Clift, R., 1998. Linear Programming as a tool in Life
Cycle Assessment. International Journal of LCA 6 (3), 305-316.
Azapagic, A., Clift, R., 1999a. Allocation of Environmental Burdens
Mohammad Mazraati1
Abstract
Volume of gasoline consumed in Iran is determined by such different
factors as gasoline price as well as efficiency, age, and number of cars
in use among many other structural and cultural variables. This paper
considers the average age and efficiency of cars and gasoline demand
models as a function of the age, efficiency, price and other
explanatory variables to prove that although renovation of the fleet
could have positive impact on fuel saving, it is yet to be the most
effective approach. It also shows that mandatory efficiency standards
for car manufactures and imported cars like the CAFÉ standards in
US, could lead to improved efficiency of car fleet, where lower cost
within a shorter period of time are incurred. The short term efficiency
elasticity of gasoline demand is -3.5 which proves the above
mentioned hypothesis. One percent increase in the efficiency of the
fleet would lead to about 3.5 percent decrease in gasoline demand in
the short run denoting a considerable fuel saving. The short term price
A. Sarzaeem 1
Abstract
In recent decades, volatility of oil prices has led to such consequences
as macroeconomic turbulences. Most of the researches look at the
matter from oil importer's point of view while oil producers are
usually neglected. This paper explains the relationship between oil
price shocks and economic growth and inflation by the econometric
methods like VAR model and OLS regression. Based on quarterly
data, an unrestricted autoregressive model is estimated in order to
gauge short run effects of oil shocks on different variables such as
exchange rate, money, government expenditure, inflation and GDP.
Long run effects are also measured by the aid of a co-integration
autoregressive model. Impulse-response technique is used to estimate
reaction of aforementioned variables to different shocks. At the end,
the paper proposes the economic policies based on statistical results.
Mehrzad Zamani1
Abstract
The recent oil market conditions, have confronted analysts with
serious challenges in their efforts to analyze oil price trends and the
reasons behind them. They have, so far, addressed the issue on the
basis of structural changes, periodical behavior and speculation. The
roots of those challenges, in their view, are embedded in the
influential development of fundamental elements on the price of oil.
OPEC's oil production in the 1990s, is a case in point, which
materialized the organization's price objectives very effectively. The
organization, however, has lost its price controlling mechanisms and
capabilities in recent years. The relationship of crude oil prices and
commercial stockpilings has also reversed. This study reviews
changes in the fundamental factors on the basis of econometric
models. According to results, OPEC's excess or surplus production
capacity in recent years (post 2003) has affected the relationship of oil
stockpilings and prices. OPEC's surplus production capacity had no
impact on oil prices prior to 2003, when it began to demonstrate its
influence. Thus, OPEC which set its members' production quotas in
1. Expert, Modeling and Long term energy studies Group, IIES, E-mail: m-zamani@iies.net
Abstract
Hegelian process of change dictates that when entity (thesis) is
transformed into its opposite (antithesis), the combination would be
resolved in a higher form (Synthesis). It seems that global liquefied
natural gas industry is undergoing such a process. The process of
restructuring traditional monopoly in LNG industry materialized by
market liberalization and privatization requires a new form of
structure that benefits from competitive markets. This paper discusses
the probable impacts of liberalization on the future developments of
LNG industry. The paper also examines the probability that whether
Hegel's invisible hand would be able to transform the traditional
monopoly in LNG industry into the restructured one involving viable
competitive markets.