T h e J o u r n a l o f D e v e l o p i n g A r e a s
Volume 50 No. 3 Summer 2016
THE RELATIONSHIP BETWEEN OIL AND
AGRICULTURAL COMMODITY PRICES IN
SOUTH AFRICA: A QUANTILE CAUSALITY
APPROACH
Mehmet Balcilar
Eastern Mediterranean University, Turkey; University of Pretoria, South Africa
Shinhye Chang
Rangan Gupta
Vanessa Kasongo
Clement Kyei
University of Pretoria, South Africa
ABSTRACT
The increase in agricultural commodity prices in the recent past has renewed interest in ascertaining
the factors that drive agricultural commodity prices. Though a number of factors are possible, higher
oil prices are thought to be the major factor driving up agricultural commodity prices, especially as
the demand for biofuels production increases. However, empirical evidence of this relationship
remain ambiguous and largely depends on the method used. For this reason, there is a need to
examine the relationship in the context of different methodologies. Furthermore, information on how
South African commodity prices respond to world oil price shocks is less certain. A good
understanding of the factors that drive local commodity prices will assist in making sound
agricultural policies. In this paper, the Granger causality test is applied to the mean to investigate the
causality between oil prices and agricultural (soya beans, wheat, sunflower and corn) commodity
prices in South Africa. Daily data spanning from 19 April 2005 to 31 July 2014 is used for Brent
crude oil, corn, wheat, sunflower and soya beans prices. Agricultural commodity prices were
obtained from the Johannesburg Stock Exchange, and the series of Brent crude oil prices from the
U.S. Department of Energy. Results from the linear causality test indicate that oil prices do not
influence agricultural commodity prices. However, owing to structural breaks and nonlinear
dependence between the variables of study, these results are misleading. As an alternative, the
nonparametric test of Granger causality in quantiles, as proposed by Jeong, Härdle and Song (2012)
is used. Through this test, we not only look at causality beyond the mean estimates but also accounts
for the structural breaks and nonlinear dependence present in the data. Additionally, the method
becomes more instructive in the case where the distribution of variables has fat tails. The findings
show that the effect of changes in oil prices on agricultural commodity prices vary across the different
quantiles of the conditional distribution. The highest impact is not at the median, and the impact on
the tails is lower compared to the rest of the distribution. The analysis shows that the relationship
between oil prices and agricultural commodity prices depends on specific phases of the market, and
therefore contradicts the neutrality hypothesis that oil prices do not cause agricultural commodity
prices in South Africa. This implies that policies to stabilize domestic agricultural commodity prices
must consider developments in the world oil markets.
94
JEL Classifications: C32, Q02, Q43
Keywords: Granger causality, South Africa, Nonparametric kernel, Quantile causality, Commodity
prices
Corresponding Author’s Email Address: rangan.gupta@up.ac.za
INTRODUCTION
What are the forces driving the upward trend of agricultural commodity prices (Corn,
Wheat, Sunflower, and Soya beans) in recent years? The answer to this question is very
important in order to decide on appropriate policy options and to examine investment
opportunities. According to Abbott, Hurt and Tyner (2008), the main drivers of increasing
agricultural commodity prices are the result of compound interactions among
macroeconomic factors such as crude oil prices, exchange rate, growing demand for food
and slowing growth in agricultural productivity, as well as the policy choices made by
nations. Although these factors are mutually reinforcing, high oil prices are thought to be
the major factor driving up the agricultural commodity prices (FAO, 2008, Mitchell, 2008
and OECD, 2008, Zhang et al., 2010). This is due to the strong linkage between energy and
agricultural markets, especially as the demand for biofuels production increases. Ethanol
and biodiesel are substitutes for gasoline and diesel, thereby the recent surge in agricultural
commodity prices are attributed to increasing usage of crops in production of biofuels
(Nazlioglu & Soytas, 2010). It is therefore very important to put a figure on price variability
of agricultural products, as negative price shocks have an exacerbating impact on the
economic growth of developing economies (Dehn, 2000). Moreover, the process of
globalization has led economies around the world to be interconnected more than ever.
Hence, a shock related to a change in any specific economic factor such as oil in one
country gets carried over across the world instantly. This is more so the case when the
economies where the shock originates from are major role players in shaping world
economic activities. In other words, a specific country is not only likely to be affected by
shocks which generated domestically, but also by external shocks.
A large body of empirical studies (Chenery, 1975; Hanson, Robinson, & Schluter,
1993; Baffes, 2007; Kaltalioglu & Soytas, 2009; Nazlioglu, 2011) have tried to understand
the relationship between oil prices and agricultural commodity prices, but the results still
remain ambiguous. For instance, empirical studies like Reboredo (2012), and Nazlioglu
and Soytas (2010) found no evidence that oil prices lead agricultural commodity prices.
Others such as Chen, Kuo and Chen (2010) showed that a rise in oil price significantly
increases agricultural commodity prices. Some studies have gone as far as claiming that
“food prices mirror oil prices” (Dancy, 2012). These results, however, rely on the
methodology that was employed or the sampling period of the data. Also, the most popular
method used to investigate the energy-food nexus is based on conditional causality in the
mean, developed by Granger (1969). This method assumes a linear data generating process
for the variables and constant parameters over time. However, evidence in the energy
literature shows that results from linear and nonlinear causality methods are different
(Bekiros and Diks, 2008; Kim et al., 2010). For this reason, there is need to examine the
oil-food relationship in the context of different methods. Furthermore, information on how
South African commodity prices respond to world oil price shocks is less certain. A good
understanding of the factors that drive local commodity prices will assist in making sound
agricultural policies.
95
South Africa is a net importer of crude oil, which is an important input to various
sectors. Any fluctuation in world oil prices would therefore have important consequences
for domestic agricultural prices. Industry data shows that coke and refined petroleum
accounted for about 10% of intermediate input costs into agriculture, forestry and fishing
in 2013 (Quantec, 2014). According to the department of agriculture, forestry and fishing
(2014), South Africa is a net importer of wheat, sunflower and soya beans, but is self-
sufficient in maize production. This means that domestic prices for wheat, sunflower and
soya beans will be strongly impacted by dynamics on the international market. It is
important to note that the government does not intervene in the grains market, but only sets
the policy.
In this paper, we investigate the causal relationship between world oil prices and
agricultural commodity prices in South Africa using a nonparametric test of Granger
causality in quantiles. We start with the unit root tests, and then conduct the standard linear
Granger causality test. Also, we conduct the BDS independence test and check for the
presence of structural breaks. In the presence of nonlinear dependence, structural breaks
and regime shifts, the standard linear Granger causality test will provide unreliable and
biased results. Therefore, our decision to use nonlinear causality test is based on the
possibility of nonlinear data generating process for our variables of study and the possible
presence of structural breaks in the data. In addition, the nonparametric test of Granger
causality in quantiles is able to pick up causality in the tails of the conditional distribution.
This becomes more instructive in the case where the distribution of variables have fat tails.
We ask the question "Do oil prices lead agricultural commodity prices in various
conditional quantiles?”. In this regard, we provide a holistic insight on how agricultural
commodity prices in South Africa respond to oil prices. Furthermore, we help the process
of evidence-based policy making with respect to agricultural and energy policies. To the
best of our knowledge, this is the first study to analyse the energy-food relationship using
quantile causality with South African data.
The rest of the paper is organised as follows: Section 2 presents the methodology
employed in this study, Section 3 discusses the data and empirical results, and finally,
Section 4 concludes our study.
METHODOLOGY
Linear Granger Causality Test
According to Granger (1969), causality between two stationary series
t
x
and
t
y
can be
defined using the concept of predictability.
t
x
is said to "Granger" cause
t
y
if past
realizations of
t
x
improve the prediction of
t
y
compared to predictions using historical
values of
t
y
only.
Assuming that the stationary series
t
x
and
t
y
are of length
n
, a formal test for Granger
causality between
t
x
and
t
y
requires estimating a
p
-order linear vector autoregressive
model
)(pVAR
of the form:
96
t
t
pt
pt
pp
pp
t
t
t
t
x
y
x
y
y
x
2
1
22,21,
12,11,
1
1
22,121,1
12,111,1
2
1
=
(1)
where
),(=
21
ttt
represents a white noise process with zero mean and covariance
matrix
.
p
is the optimal lag order of the process selected using a sequential likelihood
ratio
)(LR
test.
1
and
2
are constants and
s
are parameters.
Non-parametric Granger Causality Test
Granger developed the primary method for deducing causality in financial applications
.This method considers two time series and determines whether one predicts, or causes, the
other. However, variables like financial returns tend to have fat tailed or nonelliptic
distributions and this may render results of any analysis using conditional means uncertain.
Moreover, causality relationships in the tails may be quite different from causality
relationships at the center of the distribution (see Lee and Yang (2007)).
Previous research has shown that the correlations across financial variables
depend on the market regime (Lin, Engle, and Ito, 1994; Ang and Bekaert, 2002; Longin
and Solnik, 2001; Ang and Chen, 2002). Extreme market conditions usually result in
stronger financial co-movement across financial variables, and in contagion and volatility
spillovers. Also, Granger causality in quantile is important for risk management and
portfolio diversification (Hong, Liu and Wang (2009)), as well as for the robustness
properties of conditional quintile.
In instances where the causality only exists in certain regions of the conditional
joint distribution of the variables, basing Granger causality tests on conditional means
alone might be misleading. However, extending the linear Granger causality test to linear
quintile regression could overcome this difficulty. Lee and Yang (2007) developed linear
Granger tests in quintile that detect the existing causality relationships in the tails of the
conditional distribution. However, the linear causality tests may still fail to detect non-
linear causality relationships. Although Financial and economic variables usually are linear
in the conditional mean, which is an overall summary of the conditional distribution, their
behaviour tends to be extremely nonlinear in the tails of the distribution. To overcome the
issues arising from the nonlinearity of the relationship between variables, several papers in
the literature, such as Nishiyama et al. (2011), have proposed nonparametric Granger
causality tests based on the kernel density estimation. Jeong, Härdle and Song (2012)
developed a nonparametric test of Granger causality in quantile based on the kernel density
method. This paper fills the existing gap in the literature both in terms of the causality in
the conditional and nonlinearity of the relationship. The authors defined the Granger
causality in quantile as follows:
1.
does not cause
in the -quantile with respect to
if
(2)
2.
is a prima facie cause
in the -quantile with respect to
if
97
(3)
Where
is the th conditional quantile of
given , which depends on t and
Let consider
,
,
,
and
and
are the conditional distribution function
given
and
, respectively.
The conditional distribution
is assumed to be absolutely continuous in
for almost all
. If we denote
and
,
we have,
w.p.1
Consequently, the hypothesis to be tested based on definitions (2) and (3) are
(4)
(5)
Building on Zheng (1998)’s work, Jeong, Härdle and Song (2012) reduce the problem of
testing quantile restriction by using as distance the measure
,where
is the regression error term and
is the marginal
density function of
.This allows for testing quantile restriction as specifically testing
a particular type of mean restriction. The regression error
arises from the fact that the
null hypothesis in (3) can only be true if only if
or
equivalently
, where is the indicator function. Jeong,
Härdle and Song (2012) specify the distance function as
(6)
Where and the equality holds if and only if the null hypothesis
in equation (4) is
true, while holds under the alternative
in equation (5). From the result in Fan and
Li (1999), a feasible test statistic based on the distance measure in equation (6) has the
leading term that follows a second order degenerate U-statistic. Jeong, Härdle and Song
(2012) show that under the-mixing process, the asymptotic distribution of the statistic is
asymptotically normal.
Additionally, Jeong, Härdle and Song (2012) showed that the feasible kernel-based test
statistic based on has the following form:
(7)
