Understanding the Crude Oil Price: How Important Is the
China Factor?
Xiaoyi Mu* and Haichun Ye**
This paper employs monthly data on China’s net oil import from January
1997 to June 2010 to assess the role of China’s net import in the evolution of the
crude oil price. Based on a vector autoregression (VAR) analysis, we find that
the growth of China’s net oil import has no significant impact on monthly oil
price changes and there is no Granger causality between the two variables. The
historical decomposition indicates that shocks to China’s oil demand have only
played a small role in the oil price run-up of 2002–2008. We also calculate the
price changes implied by China’s net oil import growth from a longer-term supply
and demand shift perspective.
doi: 10.5547/ISSN0195-6574-EJ-Vol32-No4-4
“Surging Chinese demand is underpinning the recent spike in
the price of oil, figures from the International Energy Agency
(IEA) show. This ‘China factor’ has more bearing on oil prices
than the ‘risk factor’ coming from global tensions, some experts
say”
—CNN (2004)
“The price of crude oil could soar to $200 a barrel in as little
as six months, as supply continues to struggle to meet demand
. . . Soaring global demand for oil is being led by China’s
The Energy Journal, Vol. 32, No. 4. Copyright 䉷2011 by the IAEE. All rights reserved.
*
**
Corresponding author. Center for Energy, Petroleum and Mineral Law and Policy, University
of Dundee, Carnegie Building, Dundee, DD1 4HN, UK. E-mail: [email protected].
School of Economics, Shanghai University of Finance and Economics, 777 Guoding Road,
Shanghai 200433, China. E-mail: [email protected].
We thank Kevin Forbs, Ian Lange, Zhen Zhu, two anonymous referees, the journal editor James Smith
and other seminar participants at University of Stirling and University of Dundee for helpful comments, Audrey McLaughlin for helping us proofread the paper. Xiaoyi acknowledges the travel grant
from the Carnegie Trust. All errors remain our own.
doi: 10.5547/ISSN0195-6574-EJ-Vol32-No4-4
69
70 / The Energy Journal
continuing economic boom and, to a lesser extent, by India’s
rapid economic expansion.”
—BBC (2008)
1. INTRODUCTION
It is often asserted that the rising oil demand from China is one of the
main reasons for the increase in oil prices over the period of 2002–2008. Indeed
since China became a net importer in the world oil market in 1993, China’s oil
consumption has risen quickly. Because the domestic oil production in China has
remained largely flat, the increase in consumption is mainly satisfied by increases
in import. Figure 1 displays China’s net oil import which includes the net import
in both crude oil and refined petroleum products from January 1997 to June 2010.
During this period, China’s net import has increased by almost five times with
an annual average growth rate of 15.75 percent. It is probably this rapid growth
in China’s oil import that has attracted much attention from the media. Also shown
in Figure 1 is the front month futures price of West Texas Intermediate (WTI)
which is deflated using the US consumer price index(CPI) and expressed in January 2009 levels. While both series appear to have a common upward trend, how
much China’s import has contributed to the world oil price remains an open
question.
A systematic examination of the relationship between China’s import
and oil prices in the world market can help us disentangle various factors behind
oil price changes. The surge in crude oil prices from 2002 to mid-2008 has spurred
a new wave of heated debate over the causes and consequences of the oil price
shocks. Some of the significant contributions in the academic literature include
Kilian (2009), Smith (2009) and Hamilton (2009a, 2009b). Kilian (2009) distinguishes oil price shocks between oil supply shocks, aggregate demand shocks,
and precautionary demand shocks that are specific to the oil industry and argues
that the recent oil price run-up until mid-2008 is primarily driven by booming
aggregate demand. He finds that the demand-driven shocks have very different
effects on the real price of oil and tend to impact the real economic activity
differently from supply-driven oil price shocks. Hamilton (2009a) reviews several
strands of theories about oil prices including the cash-and-carry model, the futures
market theory and Hotelling’s scarcity rent theory and relates them to statistical
evidence. He concludes that the scarcity rent may have started to become an
important factor in the price of crude oil owing to the strong demand growth
from China, the Middle East and other emerging economies. Hamilton (2009b)
analyzed the causes and consequences of the oil price shock of 2007–2008 and
argues that it was caused primarily by a combination of strong demand growth
and stagnating production. Smith (2009) analyzes the global demand shift, nonOPEC and OPEC supply shifts relative to 1973–1975 levels and concludes that
a substantial part of the oil price rise since 2004 can be explained by a combination
of unexpected demand growth from China and other developing nations and a
Understanding the Crude Oil Price: The China Factor / 71
Figure 1: China’s Net Oil Import and the Real Price of Oil:
1997M1–2010M6
WTI
5.00
CNIMP
100
4.00
80
3.00
60
2.00
40
Jan-2010
Jan-2009
Jan-2008
Jan-2007
Jan-2006
Jan-2005
Jan-2004
Jan-2003
Jan-2002
Jan-2001
0.00
Jan-2000
0
Jan-1999
1.00
Jan-1998
20
Jan-1997
WTI Price (US$/B)
120
CNIMP (MMB/D)
6.00
140
Notes: WTI is the front-month futures price of West Texas Intermediate. The real oil price is obtained
by deflating WTI by the U.S. consumer price index (January 2009⳱100). CNIMP denotes China’s
net import of both crude oil and refined products. The solid line denotes the real price of oil. The
dotted line denotes China’s net oil import.
negative shift in oil supply due to higher factor costs. While both Hamilton
(2009a, 2009b) and Smith (2009) argue that the demand growth from China has
been an important factor, neither attempted to assess the relative importance of
the “China factor” in oil prices.
This paper employs monthly time series data on China’s net oil import
and the international benchmark crude oil prices over the period of January 1997
to June 2010 to assess the role of China’s demand growth in the world oil price
run-up. We focus on China’s net import for two reasons. First, data on net import
is readily available. Chinese Customs typically releases the data on monthly import and export of crude oil and refined petroleum products within two weeks
after the end of each month. As there is no official statistics on oil inventory
changes at monthly or weekly levels in China, the import and export statistics
has become almost the most important single barometer for industry analysts and
traders to gauge China’s oil demand. Second, since China is a net importer in the
world oil market during this period, changes in Chinese net import effectively
represent demand changes in the world oil market.
In the first part of our empirical analysis, we estimate a vector autoregression (VAR) model and perform impulse response analysis, forecast error variance decomposition and historical decomposition to investigate the interaction
between China’s oil demand and the real price of oil. In addition, we also conduct
72 / The Energy Journal
out-of-sample Granger causality tests to examine the causal relationship between
China’s oil demand and the real oil price. In general, our results suggest that the
growth rate of China’s net oil import has only a small impact on the real oil price
and that there is no Granger causality between the two variables. The second part
of our analysis, from a longer-term demand and supply shift perspective, answers
the question how much price change is required in order to increase the crude oil
supply to meet China’s growing demand based on plausible estimates of price
elasticity of crude oil supply. The result indicates, on average, the growth in
China’s net oil import has contributed to about 11–23 percent of the price increase
between 2002 and mid-2010 depending on assumed supply elasticities. Notably,
both the historical decomposition from the VAR analysis and the longer-term
demand and supply shift analysis suggest that the share of the real oil price change
attributable to China’s demand growth is lower in the price spike of 2008 than
the average estimates for the full sample period.
The rest of the paper is organized as follows. Section 2 describes our
data and reviews the empirical methodology used in this study. Section 3 reports
the empirical results from our VAR analyses. In Section 4 we conduct the longerterm demand and supply shift analysis. Section 5 offers concluding remarks.
2. DATA AND EMPIRICAL METHODOLOGY
2.1 Data
The nominal oil price data is the monthly averages of the daily settlement
price of WTI front-month futures and obtained from the Energy Information
Administration (hereafter, EIA) of the US Department of Energy.1 We deflate the
nominal oil price by the US consumer price index (CPI) and express it at the
January 2009 level. The unit root tests in Table 1 indicate that the logged real oil
price is stationary when a deterministic trend is included. We thus remove the
linear trend from the series and use the detrended logged real oil price (DTLRWTI)
in our empirical model.
We obtain China’s net import of crude oil and refined petroleum products
(liquid products only, measured in million barrels per day) from the General
Administration of Customs of China. The data spans the period January 1997–
June 2010. As shown in Table 1, while the augmented Dickey-Fuller (ADF) and
the Elliott-Rothenberg-Stock DF-GLS (DF-GLS) tests find some weak evidence
of the series being trend-stationary at the 10% level, the Ng-Perron and the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) tests show that the series is a unit root
process rather than trend-stationary. In this study we use the year-over-year
1. The futures price data used in this study is highly correlated with the U.S. refiners’ acquisition
cost for imported crude oil used by Kilian (2009). The sample correlation coefficient is 0.998. Using
Bai and Perron’s (1998) methodology for multiple structural changes, we find no evidence of structural
breaks in the logged real oil price during the sample period.
Understanding the Crude Oil Price: The China Factor / 73
Table 1: Unit Root Tests
Variable
Real oil price
China’s net oil import
World oil production
Level without trend
Level with trend
First difference
Level without trend
Level with trend
First difference
Seasonal difference
Level without trend
Level with trend
First difference
ADF
DF-GLS
NP
KPSS
–1.717
–3.433**
–6.586***
–0.726
–3.372*
–6.908***
–5.771***
–0.985
–2.037
–11.010***
–1.082
–3.101**
–5.319***
1.556
–2.927*
–0.323
–1.874*
0.441
–2.046
–1.998**
–1.621
–25.737***
–28.128***
1.144
–4.267
–0.750
–7.503*
–0.445
–7.117
11.341**
1.258***
0.092
0.040
1.390***
0.303***
0.163
0.134
1.319***
0.169**
0.044
Notes: The sample period is Jan 1998–Jun 2010. All variables are in natural logarithms. ADF, DFGLS, NP and KPSS refer to the augmented Dickey-Fuller test statistic, the Elliott-Rothenberg-Stock
DF-GLS test statistic, the Ng-Perron test statistic and the Kwiatkowski-Phillips-Schmidt-Shin test
statistic, respectively. The null hypothesis in the ADF, DF-GLS and NP tests is that the series has a
unit root while the null hypothesis in the KPSS test is that the series is stationary. Lag lengths in the
ADF, DF-GLS and NP tests are selected by the Akaike information criterion (AIC). The superscripts,
***, **, and *, denote the rejection of the null at the significance levels of 1%, 5% and 10%,
respectively.
The seasonal difference of China’s net oil import is the log difference of China’s net oil import between
month t and month t-12.
The time trend variable in logged real oil price is statistically significant at the 1% level in the ADF
test.
growth rate of China’s net oil import (CNIMPG, referred to as seasonal difference
in Table 1), which is defined as the log difference of China’s net oil import
between month t and month t-12.2 As such, we lose the first 12 observations and
the sample period used in our main empirical analysis runs from January 1998 to
June 2010.
To control the possibility that changes in crude oil supply drive the
relationship between China’ net oil import and the oil price, we include the percentage change in world crude oil production (WDPROG) in our empirical analysis. The world crude oil production data (measured in thousands barrels per day)
is also available from the EIA. As evident in Table 1, both the world oil production
growth and China’s net oil import growth are covariance stationary.
2.2 Empirical Model
To analyze the dynamic relationship between China’s net oil import
growth and the real price of oil, we estimate a three-variable vector autoregression
(VAR) model over the entire sample period as follows:
2. In addition, we also applied Johansen’s cointegration test on the logged oil price and the logged
China’s net oil import while allowing for a linear deterministic trend, and found no evidence that
these two series were cointegrated. Detailed results from Johansen’s cointegration test are not reported
but available upon request.
74 / The Energy Journal
3
BYt⳱A0Ⳮ
兺A Y
i
i⳱1
Ⳮet
(1)
t–i
where Yt⳱(WDPROGt ,CNIMPGt ,DTLRWTIt)⬘ and et is a vector of structural
innovations. Based on the Akaike Information Criterion (AIC), we use three lags
in the VAR model.
Pre-multiplied equation (1) by B – 1 and let ut denote the reduced-form
VAR residuals such that ut⳱ B – 1et. The structural innovations et can be recovered
from the reduced form residuals ut by imposing restrictions on B – 1. Following
Killian’s (2009) identification strategy, we apply the Cholesky decomposition to
the reduced-form residuals with the variables ordering WDPROGt ,CNIMPGt ,
DTLRWTIt . That is,
冤 冥冤
冥冤
uWDPROG
u11 0
0
t
ut⳱ utCNIMPG ⳱ u21 u22 0
uDTLRWTI
u31 u32 u33
t
etoil supply shock
etChina’s oil demand shock
etother oil demand shock
冥
(2)
In this model, first, we assume that the world oil production growth contemporaneously responds to only its own shocks (hereafter referred to as “oil supply
shocks”).3 Due to the long-lead time and capital intensive nature of petroleum
production projects, the price elasticity of crude oil supply in the short-term is
extremely low. Thus, it is reasonable to assume that crude oil production does
not respond to innovations in demand and prices within the same month. Second,
we assume that China’s net oil import growth is contemporaneously affected by
only oil supply shocks and shocks to China’s oil demand (referred to as “China’s
oil demand shock” hereafter), but not shocks to international crude oil prices.4
Last, we refer innovations to the real oil price that cannot be explained by either
oil supply shock or China’s oil demand shocks as other demand shocks, which
potentially represent all other countries’ oil demand shocks and the “precautionary
demand” shocks referred to by Kilian (2009). We assume that the real oil price
responds contemporaneously to all three types of shocks including oil supply
shocks, China’s demand shocks and also other demand shocks.
3. An example of oil supply shocks could be a disruption of oil production in one of the oil
producing countries such as Nigeria.
4. This restriction is plausible for two reasons. First, over the study period, the price of crude oil
sold domestically in China is indexed to international benchmark oil prices with one month’s lag; the
international oil price is unlikely to have an immediate impact on China’s import. Second, it takes
several weeks to ship oil from Middle East and western Africa, where China imports most of the
crude oil, to China. Even if the international oil price has an immediate impact on China’s import
decision, the cargo is unlikely to arrive at a Chinese port and to be accounted in the customs’ statistics
within a month.
Understanding the Crude Oil Price: The China Factor / 75
Based on these identified shocks, we then employ impulse responses
analysis, forecast error variance decomposition, and also historical decomposition
to investigate how the real price of oil is impacted by each of these shocks.5
2.3 Out-of-sample Granger Causality Tests
To further understand the role of China’s oil demand in the evolution of
the real oil price, we also conduct out-of-sample tests to check whether China’s
net oil import growth Granger causes changes in the real oil price. While the insample Granger causality tests have been widely employed in previous studies
on causality, the value of in-sample evidence of Granger causality may not be
very reliable in the sense that it could simply be an artifact of the specification
searches used in obtaining empirical models. In contrast, as pointed out by Ashley
et al. (1980), an out-of sample comparison of forecasting performance can yield
the maximum amount of information that is relevant to the hypothesis of Granger
causation and thus is more in the spirit of the definition of Granger causality.
Thus, we employ out-of-sample Granger causality tests here to study the causal
linkage between China’s net oil import growth and the real oil price. 6
The out-of-sample tests for Granger causality from China’ net oil import
growth to the detrended log real price of oil are implemented in two steps. In the
first step, we estimate both the restricted and unrestricted models for the real oil
price. Specifically, the unrestricted model for the real oil price is simply the last
equation in the VAR(3) model (Eq. (1)) that has the detrended log of the real oil
price as the dependent variable, while the restricted model for the real oil price
is the unrestricted model without the lagged values of China’s net oil import
growth variable. In the second step, formal statistical tests are employed to examine whether the out-of-sample mean squared forecast errors (MSFE) from the
unrestricted model are smaller than those obtained from the restricted one. If the
unrestricted model for the real oil price improves forecast accuracy over the restricted model by yielding significantly smaller MSFE, China’s net oil import
growth is said to have predictive power for the real oil price, and thus is considered to be evidence for China’s net oil import growth Granger causing movements
in the real oil price. Granger causality from the real oil price to China’s net oil
import growth is tested similarly.
In this study we consider five out-of-sample tests recently developed in
the literature of forecast evaluation: the Granger-Newbold (1976) test, the Die-
5. We believe our indentifying restrictions in equation (2) are reasonable. Nevertheless, our results
from the impulse response function and variance decomposition are robust to different orderings of
the variables.
6. We also conducted the in-sample F test for Granger causality between China’s net oil import
growth and real oil price based on the baseline VAR model and found no evidence for causation
between these two variables.
76 / The Energy Journal
bold-Mariano (1995) test, Clark-West (2006, 2007) test, McCracken’s (2007)
MSE-F test, and also the Clark-McCracken’s (2001) ENC-NEW test.7
2.4 The Bootstrap Method
Given that we have a relatively small sample, the statistical inference
based on asymptotic distributions may be problematic. We use the bootstrapping
method (with 5000 replications) to obtain the confidence intervals for impulse
responses and forecast error variance decompositions, and the p–values for rejecting the null hypothesis of equal MSFE for the restricted and unrestricted
models.8
Let ût, t⳱1, . . ,T, denote the OLS residuals from the VAR(3) model (Eq.
(1)). We first compute the centered residuals as ût – ū , where ū⳱T – 1 ût, and
obtain bootstrap residuals u*1 , . . , u*T by randomly drawing with replacement from
the centered residuals. From these bootstrap residuals, we then construct artificial
time series of world oil production growth (WDPROG), China’s net oil import
growth (CNIMPG) and the detrended log of the real oil price (DTLRWTI) using
the VAR(3) as the bootstrap data generating process (DGP).9
For each of these 5000 artificial datasets, we then construct the 95 (68)
percent confidence intervals for the impulse responses and forecast error variance
decomposition using the 2.5th (16th) and the 97.5th (84th) percentiles of their
empirical distributions as lower and upper bounds, respectively. The p-value for
each of the out-of-sample Granger causality test statistics is calculated as the
proportion of the generated test statistic values exceeding the test statistic value
obtained using the actual sample data.
兺
3. EMPIRICAL RESULTS FROM VAR
3.1 Results from Innovation Accounting
Figure 2 graphs the point estimates of impulse responses of the real oil
price to one-standard-deviation structural shocks along with their bootstrapped
95 percent and 68 percent confidence intervals. Given a positive one-standard
deviation structural shock to China’s oil demand (i.e. raising China’s net oil import), the real oil price first rises for about seven months and then declines gradually. Six months after shock, the real oil price rises by about 1.5 percent. Based
on the bootstrapped 95 percent confidence intervals, however, the positive impact
of China’s demand shock on the real price of oil is not statistically significant at
7. See Ashley and Ye (2010) for detailed review of these five out-of-sample Granger causality
tests.
8. We also applied the recursive-design wild bootstrap proposed by Goncalves and Killian (2004)
and obtained very similar results.
9. For simplicity, we fix the values of initial observations at their actual sample values.
Understanding the Crude Oil Price: The China Factor / 77
Figure 2: Impulse Responses of the Real Oil Price: Baseline VAR(3) Model
Panel A. Responses of the real oil price to one-standard deviation oil supply
shock
Panel B. Responses of the real oil price to one-standard deviation China’s oil
demand shock
Panel C. Response of the real oil price to one-standard deviation other demand
shock
Notes: The horizontal axis indicates the time horizon in terms of months after shocks. The vertical
axis shows the changes in the logarithm of the real oil price. The solid line denotes the point estimates
of impulse responses. The dashed lines and the dotted lines denote the bootstrapped 95 percent and
68 percent confidence intervals for the impulse responses based on 5000 replications, respectively.
78 / The Energy Journal
Table 2: Forecast Error Variance Decomposition for the Real Oil Price:
Baseline VAR(3) Model
Forecasting
Horizon (Months)
3
6
9
12
15
Oil supply shock
China’s demand shock
Other demand shocks
0.227
[0.084, 8.508]
0.188
[0.103, 9.559]
0.174
[0.116, 10.079]
0.183
[0.124, 10.143]
0.198
[0.127, 10.154]
0.537
[0.156, 10.287]
1.400
[0.206, 17.785]
2.230
[0.237, 22.800]
2.677
[0.251, 24.873]
2.837
[0.258, 25.545]
99.236
[86.255, 99.430]
98.412
[79.162, 99.310]
97.596
[74.125, 99.192]
97.137
[72.481, 99.163]
96.966
[72.017, 99.143]
Notes: The percentage share of total forecast error variance of the real oil price (in logarithm, detrended) attributed to each one of the three structural shocks at horizon h is obtained from the estimated
VAR(3) model that include world oil production, China’s net oil import growth and the logged real
oil price (detrended). Their 95% confidence intervals are constructed using the bootstrap method with
5000 replications and reported in brackets.
the five percent level.10 With respect to positive oil supply shocks (i.e. raising
world oil production), the real oil price drops slightly for the first seven months,
with a maximum fall of 0.75 percent two months after the shock. The 95 percent
confidence intervals for the responses of the real oil price to oil supply shocks,
yet again, suggest that the negative effect of oil supply shocks on the real oil
price is statistically insignificant. The last panel of Figure 2 shows the response
of oil price to other oil demand shocks. With a positive one-standard deviation
shock, the real oil price rises on impact for roughly eight months. The real price
of oil is expected to increase by 10 percent or so within the first three months
and then the positive impact starts to diminish slowly. As indicated by the 95
percent confidence intervals, the positive impact of other oil demand shocks on
the real oil price is statistically significant at the five percent level.
In Table 2 we report the percentage contributions of the three identified
shocks to the forecast error variance of the real oil price at various horizons.
China’s oil demand shock turns out to have limited explanatory power for the
movements in the real oil price. The proportion of the real oil price variation
accounted for by China’s oil demand shocks at the one-year horizon is 2.68
percent, with the 95 percent confidence interval extending from 0.25 percent to
24.87 percent. Over time the proportion of forecast error variance of the real oil
price due to China’s oil demand shock rises slightly but remains less than three
10. Based on the 68 percent confidence intervals, we notice that the impulse responses of real oil
price to China’s demand shock are statistically significant, at the 32 percent level, between the 9th
month and 11th month after the shock.
Understanding the Crude Oil Price: The China Factor / 79
Figure 3: Historical Decomposition of the Real Oil Price: Baseline VAR(3)
Model
Notes: The horizontal axis indicates the time period. The vertical axis indicates the historical contribution of each of the three shocks to the logged real oil price. The dotted line indicates the cumulative
effect of oil supply shocks on the real oil price. The solid line indicates the cumulative effect of
China’s oil demand shocks on the real oil price. The dashed line indicates the cumulative effect of
other demand shocks on the real oil price.
percent at any time horizons. As compared to China’s oil demand shock, oil
supply shock explains even less amount of forecast error variance of the real oil
price. The proportion of forecast error variance contributed by oil supply shock
is around 0.2 percent at any horizons. The majority of the variation in the real oil
price, unsurprisingly, is induced by other oil demand shock. It accounts for about
99 percent of the variation in the real oil price at the three-month horizon, with
the 95 percent confidence interval extending from 86.26 percent to 99.43 percent.
As time passes, its explanatory power for the movements in the real oil price
decreases somewhat yet still remains at the level of around 97 percent.
To gain further insight into the effects of the three identified shocks on
the behaviour of the real oil price over time, we plot in Figure 3 the historical
contributions the three shocks have made to fluctuations in the real price of oil.
Among the three shocks, oil supply shock has made the smallest contribution to
the fluctuations in the real oil price, with a size of less than three percent change
in the real oil price. Nonetheless, the result is consistent with Kilian’s (2009)
finding. As compared to the oil supply shock, China’s oil demand shock has had
a slightly larger effect on the evolution of the real oil price. After the 1997 Asian
80 / The Energy Journal
financial crisis (i.e. 1998–1999), China’s oil demand shock induced a fall in the
real oil price of roughly 10 percent. The largest positive effect of China’s oil
demand shock on the real oil price is observed in the middle of year 2000. From
November 1999 to July 2000, China’s oil demand shock contributed to an increase
of over 10 percent in the real oil price. Between the year 2002 and 2005, the real
oil price experienced small increases, not more than five percent, due to China’s
demand shock. When the real oil price spiked from mid-2007 to mid-2008,
China’s oil demand shock actually lowered the real oil price from its trend by
five percent. From late 2008 to 2009, China’s oil demand has quickly recovered,
perhaps as a result of its stimulus packages, and helped pull up the real oil price
by three to four percent. Consistent with our impulse responses and forecast error
variance decomposition analysis, the biggest contribution to the evolution of the
real oil price is from other oil demand shocks. During the period Jan 2007–mid
2008, other oil demand shocks caused the real oil price to rise dramatically by
about 65 percent.
3.2 Results from Out-of-sample Granger Causality Tests
To investigate the causal relationship between China’s net oil import
growth and the real price of oil, we perform a variety of out-of-sample Granger
causality tests based on the VAR(3) model (Eq. (1)). The first three observations
in the year of 1998 are reserved for creating lagged variables. The 81 sample
observations from April 1998 to December 2004 are used as the in-sample period
for model estimation, and the remaining 66 observations over the period from
January 2005 to June 2010 are reserved as the out-of-sample period for forecast
accuracy evaluation. In particular, we use recursive one-step-ahead forecast errors
in the forecast accuracy evaluation.
Table 3 reports the sample test statistics along with their p-values from
the out-of-sample tests of Granger causality between China’s net oil import
growth and the real oil price. The left column presents the testing results for the
null hypothesis of no Granger causality from China’s net oil import growth to the
real oil price, and the right column shows the results for the null hypothesis of
no Granger causality from the real oil price to China’s net oil import growth. The
reported p-values are for rejecting the null hypothesis and are obtained using the
bootstrapped sampling distributions of the listed test statistics. A p-value less than
five (ten) percent means that the null hypothesis of no Granger causality can be
rejected and thus there is evidence in favour of Granger causality running from
one variable to the other at the significance level of five (ten) percent.
With regard to the null hypothesis of nonexistence of Granger causality
from China’s net oil import growth to the real oil price, none of the out-of-sample
test statistics are significant at the ten percent level, indicating that China’s net
oil import growth does not have a significant amount of predictive power for the
movement of the real oil price. Thus, there is no out-of-sample evidence for
China’s net oil import growth Granger causing changes in the real oil price. The
Understanding the Crude Oil Price: The China Factor / 81
Table 3: Out-of-sample Granger-causality Tests: Baseline VAR(3) Model
Granger-Newbold (GN) test
Diebold-Mariano (DM) test
Clark-West (CW) test
McCracken (MSE-F) test
Clark-McCracken (ENC-NEW) test
H0: No Granger causality
from CNIMPG to
DTLRWTI
H0: No Granger causality
from DTLRWTI to
CNIMPG
–0.7614
(0.7113)
–0.6087
(0.6395)
–0.2465
(0.7904)
–1.3327
(0.5827)
–0.2435
(0.7766)
–2.0230
(0.9874)
–1.3988
(0.9464)
0.7790
(0.6995)
–16.8070
(0.9998)
4.1626
(0.3369)
Notes: Here IMPG denotes the growth rate of China’s net oil import, and DTLRWTI denotes the
deviation of the logged real oil price from its linear trend. The in-sample period is 1998M4–2004M12
with a total of 81 observations. The out-of-sample period is 2005M1–2010M6 with a total of 66
observations. Sample statistics are reported and their bootstrapped p-values are reported in parentheses.
results are fairly similar with respect to Granger causality running from the real
oil price to China’s net oil import growth. Again, none of the test statistics are
statistically significant at the ten percent level, meaning that there is no Granger
causality from the real oil price to China’s net oil import growth, either.
3.3 Robustness Checks
A potential concern over the above analysis is that we didn’t consider
the impact of demand growth from the rest of the world. In this subsection we
evaluate the robustness of our findings regarding the relationship between China’s
oil demand and the real oil price using alternative model specifications. One
robustness check is to use China’s share of oil consumption in the world to replace
China’s net oil import growth. The monthly world consumption data is also available from the EIA. Since there are no official statistics on oil inventory changes
at monthly frequencies in China, we follow practices by many industry analysts
and compute China’s oil consumption share of the world on the basis of China’s
apparent consumption which is the sum of domestic production and net import.11
11. For example, Platts releases its monthly calculation of China’s apparent demand between the
18th and 26th of every month via press release and via its website. In a news report on August 25,
2010, Bloomberg reports “China’s apparent crude demand growth may slow ‘noticeably’ in the third
quarter”. We are aware that Table 3a of the EIA short-term energy outlook contains data on China’s
oil consumption at monthly frequencies. However, a closer look reveals that until 2003 the data appear
to be derived from quarterly statistics as the monthly numbers within a quarter are all identical before
2004.
82 / The Energy Journal
Figure 4: Impulse Responses of the Real Oil Price: Using China’s Oil
Consumption Share
Panel A. Responses of the real oil price to one-standard deviation oil supply
shock
Panel B. Responses of the real oil price to one-standard deviation China’s oil
demand shock
Panel C. Response of the real oil price to one-standard deviation other oil
demand shock
Notes: The horizontal axis indicates the time horizon in terms of months after shocks. The vertical
axis shows the changes in the logarithm of the real oil price. The solid line denotes the point estimates
of impulse responses. The dashed lines and the dotted lines denote the bootstrapped 95 percent and
68 percent confidence intervals for the impulse responses based on 5000 replications, respectively.
Understanding the Crude Oil Price: The China Factor / 83
Figure 5: Impulse Responses of the Real Oil Price: Including ROW
Consumption Growth
Notes: The horizontal axis indicates the time horizon in terms of months after shocks. The vertical
axis shows the changes in the logarithm of the real oil price. The solid line denotes the point estimates
of impulse responses. The dashed lines and the dotted lines denote the bootstrapped 95 percent and
68 percent confidence intervals for the impulse responses based on 5000 replications, respectively.
We then estimate a VAR(2) model over the period March 1997-June 2010 using
the growth rate of world oil production (WDPROG), the detrended China’s oil
consumption share of the world (DTCNSHARE) and also the detrended log real
oil price (DTLRWTI).12 Another robustness check is to see whether our main
results are sensitive to the inclusion of oil consumption growth rate from the rest
of the world. The monthly data of oil consumption from the rest of the world
(referred to as “ROW” for short) is also drawn from the EIA and its growth rate
is added to our baseline VAR model. For each of these two VAR models, we then
conduct innovation accounting and out-of-sample Granger causality tests. In gen-
12. The monthly series of China’s oil consumption share of the world starts from January 1997.
Unit root tests finds that China’s oil consumption share of the world is trend stationary. We thus use
its detrended component in our VAR analysis. On the basis of AIC, two lags are included in the VAR
model.
84 / The Energy Journal
Figure 6: Historical Decomposition of the Real Oil Price: Alternative
Specifications
Panel A. Using China’s oil consumption share: 1997M3–2010M6
Panel B. Including other countries’ oil consumption growth: 1998M1–2010M6
Notes: The horizontal axis indicates the time period. The vertical axis indicates the historical contribution of each of the identified shocks to the logged real oil price. The dotted line indicates the
cumulative effect of the oil supply shock on the real oil price. The solid line indicates the cumulative
effect of China’s oil demand shock on the real oil price. The bubbled line indicates the cumulative
effect of the ROW oil demand shock on the real oil price. The dashed line indicates the cumulative
effect of other demand shock on the real oil price.
Understanding the Crude Oil Price: The China Factor / 85
Table 4: Forecast error variance decomposition for the real oil price:
alternative specifications
Panel A. Using China’s oil consumption share
Forecasting Horizon
(Months)
3
6
9
12
15
Oil supply shock
China’s demand
shock
Other demand shocks
0.057
[0.046, 6.359]
0.058
[0.045, 6.924]
0.059
[0.044, 7.033]
0.060
[0.043, 7.067]
0.060
[0.043, 7.068]
0.482
[0.047, 8.423]
0.709
[0.062, 13.444]
0.825
[0.069, 15.943]
0.881
[0.069, 16.940]
0.906
[0.069, 17.439]
99.461
[89.015, 99.671]
99.233
[84.574, 99.626]
99.115
[82.083, 99.622]
99.060
[81.264, 99.611]
99.034
[81.008, 99.611]
Panel B. Including other countries’ oil consumption growth
Forecasting
Horizon (Months)
Oil supply
shock
ROW demand
shock
China’s
demand shock
Other demand
shocks
3
0.289
[0.089, 8.639]
0.219
[0.127, 9.751]
0.288
[0.166, 10.333]
0.470
[0.198, 10.583]
0.641
[0.213, 10.662]
13.935
[2.871, 31.480]
18.335
[3.081, 42.572]
21.312
[3.275, 48.554]
22.910
[3.538, 50.624]
23.486
[3.608, 50.980]
0.473
[0.166, 9.814]
1.142
[0.214, 16.389]
1.923
[0.269, 21.061]
2.414
[0.294, 22.249]
2.617
[0.309, 22.482]
85.303
[63.793, 93.650]
80.303
[51.174, 91.948]
76.477
[43.643, 90.907]
74.206
[40.382, 90.390]
73.256
[39.004, 90.236]
6
9
12
15
Notes: Bootstrapped 95% confidence intervals are obtained based on 5000 replications and reported
in brackets.
eral, our robustness checks yield similar results to those from the baseline model.
That is, China’s oil demand shock has statistically insignificant impact on the real
oil price, and there is no Granger causality at either direction between the two
variables.
Figures 4 and 5 exhibit the impulse responses of the real oil price to
various shocks obtained from the VAR(2) model that uses China’s oil consumption share of the world and the VAR(3) model that includes ROW oil consumption
growth, respectively. In both cases the real oil price responds very little to China’s
oil demand shock. We also graph the historical decomposition of fluctuations in
the real oil price from the two VAR models in Panels A and B of Figure 6,
respectively. Again, we observe that the contribution of China’s oil demand shock
remain very small in both cases.
In Table 4 we report forecast error variance decomposition of the real
oil price due to each of the identified shocks based on the two VAR models. While
86 / The Energy Journal
Table 5 Out-of-sample Granger-causality Tests
Panel A. Using China’s oil consumption share
Granger-Newbold (GN) test
Diebold-Mariano (DM) test
Clark-West (CW) test
McCracken (MSE-F) test
Clark-McCracken (ENC-NEW) test
H0: No Granger causality
from DTCNSHARE to
DTLRWTI
H0: No Granger causality
from DTLRWTI to
DTCNSHARE
–1.2019
(0.7103)
–1.2845
(0.7339)
–0.8809
(0.7502)
–1.5476
(0.6339)
–0.5500
(0.7265)
–0.3153
(0.7844)
–0.2581
(0.7616)
1.0003
(0.6693)
–1.3750
(0.8040)
2.6506
(0.5123)
Panel B. Including ROW oil consumption growth
Granger-Newbold (GN) test
Diebold-Mariano (DM) test
Clark-West (CW) test
McCracken (MSE-F) test
Clark-McCracken (ENC-NEW) test
H0: No Granger causality
from CNIMPG to
DTLRWTI
H0: No Granger causality
from DTLRWTI to
CNIMPG
0.2598
(0.2721)
–0.2538
(0.4553)
0.1100
(0.6433)
–0.4807
(0.4235)
0.1003
(0.6571)
–0.7015
(0.7956)
–0.7134
(0.7934)
1.2680
(0.4935)
–6.9843
(0.9718)
6.4296
(0.1460)
Notes: DTCNSHARE denotes the detrended China’s oil consumption share of the world, and
DTLRWTI denotes the deviation of the logged real oil price from its linear trend. The in-sample period
is 1997M3–2004M12 with a total of 94 observations when China’s oil consumption share is used,
and 1998M4–2004M12 with a total of 81 observations when the ROW oil consumption growth is
included. The out-of-sample period is 2005M1–2010M6 with a total of 66 observations. Sample
statistics are reported and their bootstrapped p-values are reported in parentheses.
the explanatory power of China’s oil demand shocks for the variation in the real
oil price obtained from the VAR(3) model that includes ROW oil consumption
growth is very similar to that from the baseline VAR model, the proportion of
variation in the real oil price due to China’s oil demand shock becomes much
smaller in the VAR(2) model that uses China’s oil consumption share of the world,
accounting for less than one percent of the forecast error variance of the real oil
price at any horizons.
Table 5 reports the out-of-sample test results on Granger causality between China’s oil consumption share of the world and the real oil price from the
two VAR models. Again, there is no statistically significant evidence for a Granger
Understanding the Crude Oil Price: The China Factor / 87
causal relationship between the two variables as none of the test statistics have
p-values lower than ten percent.
4. ANALYSIS OF LONGER-TERM DEMAND AND SUPPLY SHIFT
In the above analysis, we investigate the interaction between the growth
rate of China’s net oil import and the deviation of the logged real oil price from
its linear trend. To the extent that China’s import has contributed to the trend of
oil prices, our statistical results may understate the impact of China’s oil import.
After all, China’s import growth accounts for nearly 30 percent of the increase
in world oil consumption between 2002 and 2008. To further evaluate the longerterm impact of China’s import growth on oil price changes, we calculate the
percentage changes in oil prices that are needed to bring supply up to meet China’s
growing demand based on some plausible estimates of supply elasticity.
By definition of elasticity, for a positive demand shock to the global oil
DQw/Qw
market (DQw), to restore equilibrium the price must rise by
where gs
gs – gd
and gd denote the elasticity of supply and demand, respectively, and Qw is the
equilibrium quantity demanded in the world. If China’s share in DQw is s, then
DQw/Qw
the price change attributable to China is s •
. Since the observed change
gs – gd
in China’s import should have incorporated the movement along the demand
curve, the change in oil price attributable to China’s import growth is
DCNIMP/Qw 13
. Given reasonable estimates of supply and demand elasticities,
gs
we can quantify the price changes implied by China’s net import growth.
Estimates of long-run supply elasticity typically range from 0.10 to 0.35.
For example, using an error correction model Krichene (2002) finds that the longrun supply elasticity of crude oil is 0.10 during 1973–1999. In an analysis on the
impact of China’s growing demand on US petroleum markets, the US Congressional Budget Office (hereafter, US CBO, 2006) adopts a five-year supply elasticity of 0.2. Smith (2009) uses 0.3 in his analysis of the long-run demand and
supply shifts in the oil market since 1970s. In our calculation, we assume the
five-year supply elasticity ranges between 0.2 and 0.1.14 In panel (a) of Figure 7,
we show the range of price changes (in percentage) implied by the five-year
changes in China’s net import from 2002 to June 2010. The lower and upper
bounds correspond to supply elasticities of 0.2 and 0.1 respectively. To eliminate
the influence of seasonal variations, both DCNIMP and Qw are calculated on a
12-month moving average basis. For example, the value of DCNIMP for June
13. In the appendix, we give a detailed graphic exposition. Also see Smith (2009).
14. We interpret five-year as an intermediate period as it takes about three to five years to develop
(e.g. drilling development wells and building production facilities) an oil field that has already been
discovered.
88 / The Energy Journal
Figure 7: Percentage Changes in Real Oil Prices
a. Price changes implied by China’s net oil import growth
Note: The lower bound indicates the price change implied by China’s net oil import growth over a
moving five-year period when the supply elasticity of crude oil is equal to 0.2 and the upper bound
depicts the implied oil price changes when oil supply elasticity is equal to 0.1.
b. Historical changes of real oil prices
Notes: This vertical line depicts the percentage change in the moving average of the real oil price for
12 months ending at each point relative to the moving average of the 12 months ending at the same
time five years ago.
Understanding the Crude Oil Price: The China Factor / 89
2003 is the difference between China’s net import averaged over the 12 months
ending June 2003 and that averaged over the 12 months ending June 1998. Similarly, Qw is the 12-month average of the world consumption at the beginning of
the five-year period.15 For example, in calculating DCNIMP/Qw for the month of
June 2003 the Qw is the average world consumption for the 12 months ending
June 1998. For comparison purposes, in panel (b) of Figure 7, we also plot the
historical five-year changes in prices (DP/P) where both DP and P are defined in
a similar way to DCNIMP and Qw.
Two features are worth commenting. First, for most of the time during
this period, the price change implied by growth in China’s net import is between
10 and 25 percent depending on the assumed supply elasticity. The mean implied
five-year price change due to China’s import growth is 11 percent for a supply
elasticity of 0.2 and 22 percent for a supply elasticity of 0.1. In comparison, the
average historical five-year price change during the same period is 96 percent. In
other words, approximately 11–23 percent of the historical price changes after
2002 are attributable to the growth in China’s net import under reasonable estimates of crude supply elasticity.
Second, although there is good correlation between the price changes
implied by China’s net oil import and the historical changes in real oil prices,
there are important differences.16 For example, when the real oil price spiked in
2008 the 12-month moving average of the real oil price in September 2008 was
202 percent higher than in September 2003. In contrast, the price increase implied
by China’s net import growth ranges between 12 and 24 percent over this period.
Therefore, about 6–12 percent of the price spike in mid-2008 can be attributed
to the growth in China’s net import. The result is consistent with our findings
from the VAR analysis that China’s oil demand shocks actually lowered the oil
price from its trend in the price spike of 2008.
This result, while suggesting the “China factor” indeed plays an important role in the crude oil price run-up after 2002, indicates that there are other
important factors responsible for the dramatic changes in crude oil price. Of
particular note, is that the world crude oil production remained largely flat between mid-2005 and early-2008 despite a more than 100% increase in the real
oil price. As argued by Hamilton (2009a) and Smith (2009), the failure for crude
production to respond to oil price increases appears to have less to do with oil
depletion than with restrained investment in some OPEC countries.
5. CONCLUSION
It is often asserted that China’s growing demand for oil is one of the
major reasons for the rapid rise in crude oil prices in the past decade. In this
15. If Qw is calculated in an arc elasticity fashion, that is, the average world consumption of the
beginning and the ending five-year period, the percentage in quantity DCNIMP/Qw would be smaller.
16. The correlation coefficient between the two series is 0.7.
90 / The Energy Journal
paper, we make use of monthly data on China’s net oil import from January 1997
to June 2010 to assess the relative importance of the “China factor” to the evolution of the real oil price during this period. In the first part of our analysis, we
examine the interaction between the growth rate of China’s net oil import and the
deviation of the real oil price from its linear trend under a VAR framework. We
find that the response of the real oil price to China’s oil demand shocks is small
and statistically insignificant and that only a small fraction of the forecast error
variance of the real oil price is attributable to China’s oil demand shocks. Furthermore, the historical decomposition indicates that the largest positive effect of
China’s oil demand shock on the deviation of the real oil price from its trend
occurred in 2000. Between 2002 and 2005, no more than five percent of the price
increase in the real oil price was induced by China’s demand shocks. When the
oil price spiked in 2008, China’s demand shocks actually lowered the oil price
from its linear trend. In addition, our out-of-sample tests find no evidence for
Granger causality between China’s oil demand and the real oil price.
The second part of our analysis calculates the price changes implied by
increases in China’s net oil import from a longer-term supply and demand shift
perspective. Under plausible assumptions of long-term price elasticity of crude
oil supply, approximately 11–23 percent of the historical price changes between
2002 and mid-2010 are attributable to the growth in China’s net oil import. Consistent with the result from the historical decomposition of the VAR analysis, the
contribution of the “China factor” to the real oil price is even smaller in the price
spike of 2008.
Our analysis casts doubt on the popular view that the demand growth
from China is the predominant reason for the dramatic oil price increase between
2002 and 2008. Notwithstanding, if China’s demand growth continues its trend,
it could play a bigger role in the future especially when it is combined with rigid
crude oil supply.
REFERENCES
Ashley, Richard, and Haichun Ye (2010). “On the Granger causality between median inflation and
price dispersion.” Working paper, Department of Economics at Virginia Tech University.
Bai, J. and P. Perron (1998). “Testing for and Estimation of Multiple Structural Changes”, Econometrica 66: 47–79.
BBC (2008). “Oil Price May Hit $200 a Barrel”, May 7, 2008, http://news.bbc.co.uk/1/hi/business/
7387203.stm (accessed Sep 10, 2010)
Clark, Todd, and Michael McCracken (2001). “Test of Equal Forecast Accuracy and Encompassing
for Nested Models.” Journal of Econometrics 105(1): 85–110.
Clark, Todd, and Kenneth West (2006). “Using Out-of-Sample Mean Squared Prediction Errors to
Test the Martingale Difference Hypothesis.” Journal of Econometrics 135(1–2): 155–186.
Clark, Todd, and Kenneth West (2007). “Approximately Normal Tests for Equal Predictive Accuracy
in Nested Models.” Journal of Econometrics 138(1): 291–311.
CNN (2004). “China factor driving oil prices”, May 24, 2004, http://edition.cnn.com/2004/BUSINESS/05/24/china.oil.demand/index.html (accessed Sep 15, 2010)
Gonclaves, Silvia, and Lutz Killian (2004). “Bootstrapping Autoregressions with Conditional Heteroskedasticity of Unknown Form.” Journal of Econometrics 123(1): 89–120.
Understanding the Crude Oil Price: The China Factor / 91
Granger, Clive, and Paul Newbold (1976). “Forecasting Transformed Series.” Journal of the Royal
Statistical Society Series B 38(2): 189–203.
Hamilton, James (2009a). “Understanding Crude Oil Prices.” The Energy Journal 30(2): 179–206.
Hamilton, James (2009b) “Causes and Consequences of the Oil Shock of 2007–08” Brookings Papers
on Economic Activity, Spring 2009, 215–261
Killian, Lutz (2009). “Not All Oil Price Shocks Are Alike: Disentangling Demand and Supply Shocks
in the Crude Oil Market” American Economic Review 99(3): 1053–1069.
Krichene, Noureddine (2002) “World Crude Oil and Natural Gas: A Demand and Supply Model.”
Energy Economics, 24(6): 557–576.
McCracken, Michael W. (2007). “Asymptotics for Out of Sample Tests of Granger Causality.” Journal
of Econometrics 140(2): 719–752.
Smith, James (2009). “World Oil: Market or Mayhem?” Journal of Economic Perspectives, 23(3):
145–164.
US Congressional Budget Office (2006). “China’s Growing Demand for Oil and Its Impact of U.S.
Petroleum Markets”. http://www.cbo.gov/ftpdocs/71xx/doc7128/04–07-ChinaOil.pdf (accessed
Sep 13, 2010)
APPENDIX: RESULT OF A DEMAND SHIFT
S
P1
C
P0
B
A
D1
D0
Q0
Q1
Q1’
This graph shows the resulting market equilibrium following a demand shift from
D0 to D1. The movement from A to B reflects this demand shift, although B is
not observable. From B to C represents the movement along the demand curve
and from A to C represents the movement along the supply curve. For an observed
equilibrium C, the price change from P0 to P1 can be calculated from the quantity
change from Q0 to Q1 and estimated supply elasticity.
92 / The Energy Journal