Você está na página 1de 14

LITERATURE REVIEW

1. Introduction

The relationship between world crude oil and agricultural commodity prices are well-
studied in economic literature. The examples of the studies are such as Busse, Brummer and Ihle
(2011), Ciaian and Kancs (2011), Du, Yu and Hayes (2011), Nazlioglu (2011), Chen, Kuo and
Chen (2010). However, study on relating the financial variable towards the energy-agricultural
commodity relationship still very limited in the literatures. As review, there are very few studies
try to incorporate financial variables such as exchange rate and interest rate in their study such as
Dauvin (2014), Hamulczuk and Klimkowski (2012), Nazlioglu and Soytas (2012, 2011), Bhar
and Hammoudeh (2011), Akram (2009) and Harri, Nalley and Hudson (2009). In common,
agricultural commodities such as corn, soybean, wheat and rice get more attention from the
researchers in link with the world crude oil1. This is because most of the authors argued that
corn, soybean, wheat and rice are energy incentive products, and energy price closely impact
those commodities (Ji & Fan, 2012; Nazlioglu & Soytas, 2011; Nazlioglu, 2011; Chen, Kuo &
Chen, 2010).

Other than above studies, there are few authors used some extra variables to represent the
energy price and agricultural commodities rather than the common used variables. For instance,
Ji and Fan (2012) used as heating oil, gasoline and natural gas to represent the energy price.
They argued that gasoline and natural gas price will have effect on the commodity prices because
of the increasing usage of bioethanol and biodiesel in gasoline and diesel. In the other hand,
Busse, Brummer and Ihle (2011) used rapeseed oil and soybeans oil together with the crude oil
to represent the energy price. In similarity, the same soybean oil (ethanol) has been used by
Natanelov et al. (2011), Saghaian (2010) and Zhang et al. (2010) in their study2.

1
Other then the common used agricultural commodity prices, commodities such as cotton, orange sugar, cocoa,
coffee, lean hog (Ji & Fan, 2012; Natanelov et al., 2011) also been used in few studies. In addition, Ciaian and
Kancs (2011) used banana, sorghum and tea in the estimation.
2
In addition, Zhang et al. (2010) used gasoline price as well in their study to link with the commodities prices such
as soybean, corn, rice, sugar and wheat. Their results indicate that there no any direct long-run relationship between
the energy price and agricultural commodities and there is a short-run relationship between ethanol and commodity
price.
In term of financial variables, there is no any clear evidence in indicating which financial
variable have significant impact on the energy-agricultural commodities relationship. Most of the
authors used financial variables such as exchange rate (Dauvin, 2014; Hamulczuk &
Klimkowski, 2012; Nazlioglu & Soytas, 2012; and Nazlioglu & Soytas, 2011) and interest rate
(Bhar & Hammoudeh, 2011 and Akram, 2009) in the study3. Most of the authors argued that the
exchange rate has a significant influence in explaining the price of agricultural commodity
directly or indirectly. The rest of this review is divided into four sub-sections which are
theoretical/ modeling frameworks; empirical testing procedures; empirical evidences; and
concluding remarks.

2. Theoretical/ Modeling framework

This part will discuss about the background on the theory between agricultural
commodity and oil prices and financial variables. The use of a supply-demand framework as a
foundation in analyzing the influence of commodity price is quite common in the literatures (for
instance, Akram, 2009; Bhar & Hammoudeh, 2011; Chen, Kuo & Chen, 2010; Ciaian & Kancs,
2011; Dauvin, 2014). The framework implied based on the Law of Demand and Supply, when a
price of a good is getting higher, move along with the supply curve will increase the supply,
while a greater supply (from S0 to S1, Figure 1) will tend to lower the prices (from P0 to P1,
Figure 1). On the other hand, when a price of good is lower will bring a greater demand by
moves along the demand curve, and a greater demand (from D0 to D1, Figure 2) will tend to
increase prices from P0 to P1 as shown in Figure 24.

Based on the fundamental of demand and supply, an increase in energy price causes
agricultural commodity prices to increase due to increasing usage of crops in production of bio-
fuels especially ethanol and biodiesel which are substitutes for gasoline and diesel (Chen, Kuo &
Chen, 2010; Ciaian & Kancs, 2011). This statement has been strengthen by Hanson et al. (1993)
previously by added that an increase in oil prices can be followed by an increase in input costs

3
Akram (2009) used both exchange rate and interest rate in linked with the energy-agricultural commodity
relationship and identified that there is a negative relationship between interest rate and commodity prices and
interest rate may be a useful indicator of the movement in commodity prices.
4
Refer to Ciaian & Kancs (2011), for the more discussion on the supply and demand illustration in the context of
agricultural commodities and oil price relationship.
and causes agricultural prices to rise. In the other hand, Akram (2009) added that the price
impact of shifts in demand and supply of commodities may be particularly large if the demand or
supply of commodities is relatively price inelastic, which is generally believed to be the case for
many commodities and especially crude oil. In the recent study of Dauvin (2014), stated that oil
price is the good traded in the international market, therefore its price is determined by world
demand and supply and not the domestic. In contrast Bhar and Hammoudeh (2011) added that
the prices are responsive to their own supply and demand but it not able to explain the financial
variables in the model.

Source: Developed by the researcher.

However, the use of simple supply-demand framework cannot entirely account the
factors influence the commodity prices. Based on the Law of one price for tradable goods, a
decline in the value of the dollar to raise the purchasing power of commodity demand of foreign
consumers and other way round able to affect commodity prices (Ridler & Yandle, 1972; Akram,
2009; Browne & Cronin, 2010). Hence, Akram (2009) developed the model to show the
relationship between the value of the dollar and commodity prices in dollars can be shown as
follow:
(1)
where and represent the commodity price in dollars, nominal dollar exchange rate in
terms of units of a foreign currency and commodity price in units of a foreign currency
respectively.

According to Abbott et al. (2008), the link between oil prices to agricultural commodity
prices can be also explained through the exchange rates due to oil trade is conducted mainly in
US dollars and may have impact on local currencies of all influences the agricultural commodity
prices in turn of imports or exports and local prices of the commodities. In addition, Harri,
Nalley and Hudson (2009) concluded that there are two links from oil prices to agricultural
commodity prices: a direct link from oil prices to commodity prices and an indirect link through
exchange rates which can be shown in Figure 3. The authors argued that the exchange rate affect
the commodity price by value of dollar and the oil price affect the commodity production by
input prices (Harri, Nalley & Hudson, 2009).

Figure 3: Expected linkages between oil, exchanges rates, and commodity prices

Source: Adopted from Harri, Nalley and Hudson (2009)

Rather than supply-and-demand factors, financialization of commodity markets is also


the cause of influence commodity price. One may argue that increasing in speculative and
investor activity due to a low real interest rate will reduces the cost of storage, therefore, the
investors will include the real commodities as hedge against inflation to reduce the risk and to
diversify portfolios to make a profit. Frankel (1986), on the other hand, added that commodity
prices tend to overshoot in response to interest rates changes, as exchange rates do in
Dornbusch's (1976) model and the relationship between commodity prices and interest rates can
be presented as follow:

) (2)

where is the expected revaluation of a commodity over the period, measured by


the expected price increase from period t to t+1, given information available at time. On the
other hand, i and s(i) represent the nominal interest rate and storage costs of a given commodity
net of convenience yield respectively.

3. Empirical testing procedures

This part will discuss about the common used method in the estimation by the literatures.
Discussions of those methods able to give some idea in develop the methodology for this study
later on.

3.1 Unit Root Test

Testing for the stationarity (on non-stationarity) is important in time series analysis to
identify the data series are deterministic or stochastic trend5. Nazlioglu and Soytas (2012) added
that determining the order of integration of the variables is an important step in an empirical
analysis since the non-stationary variables results in spurious regressions. The basic idea of unit
root test is to find the order of integration in the variables. Broadly speaking, to check the
stationarity of the variables to make sure the mean and variance value are constant over period

5
If the trend in a time series is a deterministic function of time, such as time and time-squared then it is
deterministic trend. If the trend is unpredictable then it is stochastic trend (Gujarati & Porter, 2009, pg. 745)
(Gujarati & Porter, 2009, pg. 740)6. Since oil price, agricultural commodity prices and financial
variables are in the form of time series data, this test popular among the researchers under this
area of study. Based on the review, the common tests that used to check the stationarity test are
the Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) test which tests the null of unit
root. Studies which used these two tests are Mensi et al. (2013), Ji and Fan (2012), Ciaian and
Kancs (2011), Natanelov et al. (2011) and Chen, Kuo and Chen (2010).

On the other hand, few of the authors used Kwiatkowski–Phillips–Schmidt–Shin (KPSS)


(Kwiatkowski et al., 1992) tests KPSS in their study such as Hamulczuk and Klimkowski
(2012), Ciner (2011), Nazlioglu (2011) and Nazlioglu and Soytas (2011)7. The KPSS tests are
used to test a null hypothesis of stationarity, differ from other unit root test such as ADF and PP
(significant KPSS statistic means rejection of stationarity) (Nazlioglu & Soytas, 2011). In
addition, there are few other unit root test has been used in the studies such as DF-GLS (Zhang et
al., 2010; Nazlioglu & Soytas, 2011) which is not very popular. Gujarati & Porter (2009) argued
that there are so many unit root test and choosing one of the suitable test is based on the power
(probability of rejection) and size (level of significant) of the test.

3.2 Cointegration Test

Basically, if two variables have a long-term, or equilibrium, equal trend and relationship
between them then it indicates those variables are co-integrated (Gujarati & Porter, 2009, pg.
762; Zhang et al., 2010). Zhang et al. (2010) added that even though there may be short-run
shocks causing price series deviation, if they are cointegrated there is a long-run linear relation
which ties the prices together. There are number of methods to test the cointegration has been
proposed in the literatures.

6
According to Gujarati & Porter (2009), if the time series data are non-stationary then we just able to see the effect
for the selected time period under consideration and not able to generalize it to other time period.
7
Nazlioglu (2011) and Nazlioglu & Soytas (2011) used the KPSS test together with the ADF and PP test in their
study.
3.2.1 Johansen Cointegration Test

This test is the widely used test to identify the cointegration relationship. For variables to
be co-integrated, two situations must be fulfilled:

i. The series for the individual variables must have the same statistical properties.
ii. The variables must be integrated in the same order. If a series is stationary after the
first differentiation, it is said to be integrated to order I (1).

Based on the review, few authors have been used this method in their study such as Hamulczuk
and Klimkowski (2012), Ciaian and Kancs (2011) Ciner (2011), Natanelov et al. (2011),
Saghaian (2010), Zhang et al. (2010) and Harri, Nalley and Hudson (2009). Natanelov et al.
(2011) argued that cointegration method can be applied when the price series are non-stationary
in levels but stationary in first differences using Vector Autoregressive and Vector Error
Correction Models (VEC model). Saghaian (2010) added that VEC model provide better
adjustments for non-stationarity and long-run relationships among variables. In the same time,
Saghaian (2010) argued that it is complex to exemplify the qualitative relationships among the
variables and the expected sign of unknown parameters using VEC model. But, VEC model is
better performing for the energy price and agricultural commodity price relationship.

3.2.2 ARDL Bound Testing Approach

Based on the review, this approach is not very popular among the researcher in
identifying the energy-agricultural commodity prices relationship. This is may be because as
review, most of the financial, energy and commodity prices are integrated in first differences and
not in mixed order. Based on the review, only one study used this approach i.e. Chen, Kuo and
Chen (2010) even this method known as better than the alternative Johansen cointegration test
because of its simplicity and suitability for mixed results modeling (Dergiades & Tsoulfidis,
2008). Based on the author, the rationale behind the usage of this method is that the data allowed
determining the particular short-run dynamics and this method is one of the widely used methods
for estimating energy demand relationship in time series context. According to Sara, Ewing and
Soytas (2008), there are few advantages for ARDL model compare to traditional Johansen
cointegration test.
i. It can be applied regardless of whether the regressors are I(0) or I(1).
ii. Can avoid pre-test problems often associated with standard cointegration analysis
(unit root pretesting is not important).
iii. The ARDL model can be established either in a long-run or the error correction
model (ECM) version of the model constructed.

3.3 Causality Test

Even though regression analysis deals with the dependence of one variable on other
variables, it does not necessarily imply causation (Gujarati & Porter, 2009, pg. 652). The
causality test follows after the cointegration test; if two series are co-integrated, then a Granger
causality test must be applied to determine the direction of causality (Dhungel, 2008). This test
finds out the causality relationship between the variables and whether X causes Y to conclude
how much of the current Y can be explained by past values of (X). Based on the review, there
are few method have been used in the studies to identify the causality relationship.

3.3.1 Granger Causality Test

This test is a traditional causality test which is commonly used after the cointegration test.
Granger causality test assumes that the information relevant to the prediction of the respective
variables is contained solely the time series data on those variables (Gujarati & Porter, 2009, pg.
653; Hamulczuk & Klimkowski, 2012). A lot of authors such as Hamulczuk and Klimkowski
(2012), Ji and Fan (2012), Ciaian and Kancs (2011), Ciner (2011), Natanelov et al. (2011),
Saghaian (2010) and Zhang et al. (2010) used this method in their study. Saghaian (2010) argued
that if the variables are cointegrated, then the causality testing should be based on a VEC model
rather than on an unrestricted VAR model. His statement supported by Ciaian and Kancs
(2011). On the other hand, Zhang et al. (2010) added that, the short-run relationship among the
variables able to be detecting using the Granger Causality test. In contrast, Nazlioglu (2011)
argued that the Wald statistic of Granger Causality test may lead to nonstandard limiting
distributions depending upon the cointegration properties of the VAR system. The author added
that in this manner the alternative Toda and Yamamoto test is better perform.

3.3.2 Non-Causality Toda and Yamamoto (1995) Test

Another alternative test to test the causality is Toda and Yamamoto test. There are two
applied this test to test the causality i.e. Nazlioglu (2011) and Nazlioglu and Soytas (2011)8.
According to Nazlioglu and Soytas (2011), there are two advantages in this test compare to the
traditional Granger causality i.e. (1) the test do not necessitate a pre-test for cointegration or
estimation of a cointegration vector; (2) It can be applied to series with arbitrary integration
orders where we can conduct a modified Wald (MWALD) test on the first optimum lag length
parameters of a lag augmented system. However, even this test provides a powerful means for
long-run Granger causality tests; it does not tell us how series respond when there is a shock in
one of the variables in the system (Nazlioglu & Soytas, 2011).

4.0 Panel Data Estimation

Based on the review, time series data was popular in the analysis (refer to the review
table). But, there is few authors used panel estimation in their study as well with the argument
that it able to increase the statistical power of the empirical analysis by combining information
from both time and cross-section dimensions (Dauvin, 2014)9. In the same time, Nazlioglu and
Soytas (2012) stated that by using the panel estimation they are able to capture the dynamic link
between the series, which allowing for any feedback effects to show them.

8
In the same study by Nazlioglu (2011), the author included another causality test known as Diks–Panchenko
nonlinear granger causality test to test for the nonlinear causality relationship among the variables. The finding of
the Diks-Panchenko test indicates that the changes in the oil prices have predictive power in determining the future
dynamics of the agricultural commodity prices.
9
Panel data is the combination of time series and cross sectional data, in other word its combination of space as well
as the dimensions (Gujarati & Porter, 2009, pg. 591).
4.1 Panel Unit Root Test

Usage of panel estimation found to be very not often in energy-agricultural relationship.


Based on the review, two studies used this method i.e. Nazlioglu and Soytas (2012) and Dauvin
(2014). The author argued that panel unit root test able to increase the statistical power in the
estimation. This statement supported the earliest statement of Al-Iriani (2006) which stated that
the traditional augmented Dickey-Fuller test (ADF) of unit root test is characterized by having a
low power in rejecting the null hypothesis of the series especially for short-spanned data; panel-
based unit root tests have higher power. Nazlioglu and Soytas (2012) used LLC (developed by
Levin et al., 2002) and IPS (developed by Im et al., 2003) unit root test to find out the degree of
integration between the variables. The authors added that both LLC and IPS tests are based on
the ADF test principle. In the same time, Dauvin (2014) performed the second generation panel
unit root tests which take into account of the potential cross-sectional dependence between the
panel individuals10.

4.2 Panel Cointegration Test

Panel cointegration test is the second step of the panel test whereby it is used to
investigate the long-run relationship between the variables. Few studies used this test such as
Nazlioglu and Soytas (2012) and Dauvin (2014). Nazlioglu and Soytas (2012) used panel
cointegration test developed by Pedroni (1999), where Pedroni (1999) developed seven
cointegration statistics to test null of no cointegration i.e. panel v-statistic, panel ρ-statistic; panel
PP-statistic, Panel ADF-statistic, group rho-statistic, group PP-statistic, and group ADF-statistic.
The first four statistics are known as the panel cointegration statistics based on the within
approach. The last three statistics are group panel cointegration statistics and are based on the
between approach (Lee, 2005).

10
The author argued that, real effective exchange rates are interdependent by construction, and given that energy
prices are much likely to be strongly affected by cross-correlations which not taking into account of the cross
dependence results in tests that suffer from severe size distorision and restricted power (Dauvin, 2014).
4.3 Panel Granger Causality Test

After identifying the existence of long-run relationship between the variables, the next
step is to find the causality effect among the variables. This causality effect can be identified
using the Panel Granger causality test. This test is the continuation of the Panel cointegration
test. Therefore, studies which used the aforementioned test will used this test as well. Nazlioglu
and Soytas (2012) stated that inferences from a causality test based on a vector auto regression
(VAR) model in first differences will be misleading when the variables are cointegrated. The
authors added that in this manner the panel Granger causality performs better by using the
VECM.

5.0 Impulse Response Functions

Impulse Response Functions (IRF) is used to trace out the response of the dependent
variable in the VAR system to shock in the error terms for several periods in the future (Gujarati
& Porter, 2009, pg. 789). Few researchers have used this method in their analysis such as Ciaian
and Kancs (2011), Nazlioglu and Soytas (2011) and Akram (2009). Ciaian and Kancs (2011)
stated that IRF helps in calculate the long-run price transmission elasticity and the size of the
response between the variables. Nazlioglu and Soytas (2011) added that the advantage of the IRF
is the test able to show us how series respond when there is a shock in one of the variables in the
systems which is not able to be identify using the causality test. In addition, the authors used the
generalized IRF approach in their study11.

4. Empirical evidences

Based on the reviews, result of the previous studies not consistent over the period. The
results also differ based on the methodology and variables that the researchers used in the
study12. Accordingly, Hamulczuk and Klimkowski (2012) identified that, there is a strong

11
Nazlioglu & Soytas (2011) argued that generalized IRF approach is better than the traditional approach since the
results is not subject to the order of the variables. The result of the traditional approach will change if we change the
order of the variables.
12
Refer to the Table 1 for further information on the methodology and variables used in the study.
relationship between the exchange rate and wheat price. This finding is same with the initial
study of Harri, Nalley and Hudson (2009) which indicated that exchange rate, corn and oil price
are interrelated. In continues, the same finding has been identified by Nazlioglu and Soytas
(2012) by stated that weak dollar positively impact the agricultural commodity prices. In the
same time, Nazlioglu and Soytas (2011) did a research in Turkey and identify that there is a
direct or indirect affect of Turkish exchange rate towards the agricultural commodities through
the oil price. Those four studies clearly show that exchange rate play an important role in
influencing the price of agricultural commodity prices. However there are few studies tend to use
other financial variables in their study to find the effect of those variables toward the agricultural
commodity prices.

Dauvin (2014) added that term-of-trade is an important driver of the real exchange rate.
In contrast, Ji and Fan (2012) used US dollar index rather than exchange rate in their study. The
authors identified that the US dollar index become weakened since the crisis by using the daily
data from 2006 to 2010. On the other hand, Ciner (2011) used the consumer price index to
capture the inflation effect on the commodity prices and identify that there is a unidirectional
causality from the commodities prices towards the inflation using the Granger causality test13.
The authors added that, based on the objective of the study to identify the component that used in
stabilizing the price, commodities prices are playing an important role in determining the
inflation. In the same time, Bhar and Hammoudeh (2011) concluded that the directional
relationship between the commodities and macro-financial variables are mixed and it’s
depending on the regimes. Bhar and Hammoudeh (2011) used interest rate and foreign exchange
value represents the financial variables and crude oil, gold, silver and copper represent the
commodity prices. Previously, Akram (2009) also used interest rate in his study and identified
that commodity price raise when the interest rate fall and the real value of the dollar depreciate.

In term of energy-agricultural commodity relationship, Nazlioglu and Soytas (2012)


stated that there is a strong impact of oil prices on the agricultural commodity prices. This
statement supported the early study of Ciaian and Kancs (2011) where those authors indicate that

13
The same variable i.e. consumer price index has been used by Browne & Cronin in their study. In addition they
used M2 and GDP to link with the commodity prices and identified that there is a long run relationship between
money, commodity prices and consumer price index by using Johansen maximum likelihood approach.
their results confirm the theoretical hypothesis that energy prices do after the agricultural
commodities. The authors added that the oil price affect all the agricultural commodity prices,
does not taking into account whether its energy incentive agricultural commodity or not14. In
addition, Busse, Brummer and Ihle (2011) used rapeseed and soybeans linked with the energy
price and identified that, the correlation between the return of rapeseeds and crude oil increasing
by using the daily frequency data from 1999 to 2009. The same correlation result obtained by the
study of Saghaian (2010) by using the monthly data from 1996 to 2008.

On the other hand, Nazlioglu (2011) came out with the contradict statement said that oil
price and agricultural commodity prices do not influence each others in the linear causality
analysis. But, the author added that there is a nonlinear causality feedback between the oil and
agricultural commodity prices such as corn, soybeans and wheat. Based on the author, the
rationale behind using the nonlinearity test is because the oil and agricultural commodity prices
seem to be higher than their historical levels, which means that the agricultural commodities may
have new price regimes and the prices are characterized by nonlinear manners (Nazlioglu, 2011).
The same finding as been identified previously by Saghaian (2010) and Zhang et al. (2010) using
the Granger causality VECM framework. Zhang et al. (2010) added that there is no any direct
long run relationship between oil and agricultural commodity prices and very limited short run
relationship among the variables.

There are some exceptional studies which are not discussed in this section; those have
been tabulated in Table 1 accordingly.

5. Concluding remarks

In conclusion, many studies used the time series method to find out the relationship
between the oil, agricultural commodity prices and financial variables. Examples of unit root test
methods in common include the ADF and PP tests. Then, the cointegration tests are used,

14
Ciaian & Kancs (2011) used nine agricultural commodities in their study and indentify that all nine commodities
are cointegrated with the crude oil using Johansen cointegration test.
followed by the causality test. Most findings managed to reach at a common point where the
variables were mostly stationary of I(1). These variables were often co-integrated and long-run
relationships did exist among them. Nevertheless, there are some disputes in the causality test
results. Besides the time series analysis, there are few studies used Panel analysis as well such
as, Dauvin (2014) and Nazlioglu and Soytas (2012). There are few uncommon method used in
this kind of studies such as, GARCH (Mensi et al., 2013), bivariate stochastic volatility model
(Du, Yu & Hayes, 2011), and Diks-Panchenko nonlinear granger causality test (Nazlioglu,
2011).

Many researchers had attempted to study the subject matter using different methods.
Through observation, most of the study used high frequency data such as daily, weekly and
monthly data for the prices. In the same time, the common used variables for represent the
financial variables are exchange rate, interest rate and CPI, and common used agricultural prices
are such as corn, wheat and soybean. Based on the review, there still huge gap in the literature on
energy-agricultural prices-financial variable relationship need to be cover up.

Você também pode gostar