SPILLOVER EFFECT BETWEEN FOOD AND AGRICULTURAL

COMMODITIES: A CASE STUDY OF PAKISTAN

Andleeb Ismail
Pakistan Institute of Development Economics, Islamabad
andleebismail_14@pide.edu.pk
andleeb779@yahoo.com

Munazza Jabeen
International Islamic University, Islamabad
Munazza.jabeen17@gmail.com

Abstract:

This study focuses on analyzing the return and volatility spillovers among the major food
and agricultural commodity prices by using the GARCH models. It explores the conditional
volatility dynamics of the food and agricultural commodity prices as well as volatility
transmission between them. It also examines whether the volatility in price series of one
commodity affects the price volatility of a substitute commodity i.e. rice and wheat and beef
and poultry using monthly data from April 1983 till April 2013. Findings of this study shows
that prices series are characterized with high volatility. Further price volatility of wheat
significantly affect the volatility of rice price series. Whereas, price of poultry do not affect the
mean prices of beef and vice versa. Additionally, volatility of poultry market is not transmitted
to beef market and vice versa.

JEL: Q22, Q110, C220

Key words: Food and agricultural commodities, Return and volatility spillovers, GARCH
models

1. Introduction

The prices of major food and agricultural commodities have risen dramatically from the
end of 2006 to middle of 2008 internationally and domestically. They have reached their
highest levels in nearly thirty years. In the second half of 2008, the prices upswing decelerated
and prices fell down rapidly during the financial crises and at the wake of economic recession.
These prices of food and agricultural commodities are characterized by price volatility with
booms and slumps and present serious challenges to market participants such as producers,
consumers and investors. Moreover the macroeconomic effects of large food and agricultural
price swings have been broad and far-reaching, including their effect on balance of payments,
imports and exports, government budget, inflation, and poverty (Roache, 2010).
A number of studies economists such as Abbot et al. (2008), Mitchell (2008), Cooke and
Robles (2009) and Gilbert and Morgan (2010a) have discussed the factors behind the price
fluctuations in food and agricultural commodities. These involved changes in supply and
demand factors. On the demand side, the fast economic growth in Asian economies and
particularly in China is emphasized. On the supply side, the underinvestment in agriculture as
well as low commodity inventory levels of recent years are mentioned as contributory factors.
In addition, a new factor has emerged in the form of a change in the use of food crops with the
increasing production of bio fuels. Other macroeconomic and financial factors are considered

1
Spillover Effect between Food and Agricultural Commodities…

to influence agricultural commodity price volatility include: changes in oil prices, changes in
the world money supply, changes in the value of the dollar since many agricultural commodity
prices that are traded are denominated in terms of the US dollar. Other factors which are also
quoted include climate change, trade policies, the feedback between price expectation and
market responses and speculation in futures and options trading in food commodity markets
(Mitchell, 2008; Cooke & Robles, 2009; Gilbert & Morgan, 2010a). Moreover, some studies
(Gilbert, 2006; Balcombe, 2009; Sumner, 2009; Gilbert & Morgan, 2010a; Huchet Bourdon,
2011) have shown that the prices of food and agricultural commodities were low in the 1960s,
higher in the 1970s and again lower in the 1980s and the 1990s but remained above the level
of the sixties. These studies have also found a persistent volatility in agricultural price series.
The report of Food and Agricultural Organization of the United Nations and the Organization
for Economic Co-operation and Development (2001) have also indicated that volatility in
agricultural prices has changed over the period of 2001-2010.
A number of studies have examined dynamic changes in prices of food and agricultural
commodities (Malliaris & Urrutia, 1996; Chatrath et al., 2002; Dahl & Iglesias, 2009). These
studies have shown the long run co-movements between prices of agricultural commodities
which are by the fact that agricultural commodities have some essential properties in common
such as geographical areas, seasonality and climatic situations as well as worldwide demand.
Accordingly, one agricultural commodity can be substituted or complementary. Many other
studies have discussed the volatility spillover effect between crude oil and food & agricultural
markets (Babula & Somwaru, 1992; Baffes, 2007; Ghaith & Awad, 2011; Elmarzougui &
Larue, 2013). Apergis and Rezitis (2003) and Khiyavi et al. (2012) have examined the
spillover effect amongst the volatilities of selected agricultural output prices, input prices,
producer prices and retail prices. Most of existing studies mainly focus on the volatility
spillover between energy and agricultural commodities markets or between output and input
prices of agricultural commodities. However, there is lack of studies about spillover effects
across prices of agricultural commodities.
This study focuses on analyzing the return and volatility spillovers among the major food
and agricultural commodity prices by using the GARCH models. It explores the conditional
volatility dynamics of the food and agricultural commodity prices as well as volatility
transmission between them. It also examines whether the volatility in price series of one
commodity affects the price volatility of a substitute commodity?
This paper is organized as follows: section 2 presents the literature review, section 3
introduces the model specifications and methodology, section 4 describes the data and
empirical results, section 5 provides conclusion.

2. Literature Review

There are many studies that have explored spillover effects in financial markets, focusing
on exchange rates, interest rates and bonds and in energy markets, mainly among crude oil and
gasoline (Skintzi & Refenes, 2006; Aloui, 2007; Arouri, Jouini & Nguyen, 2011 and Bubak,
Kocenda & Zikes, 2011). Many other studies have discussed the volatility spillover effect
between crude oil and agricultural markets and have concluded positive spillover between
them (Babula & Somwaru, 1992; Baffes, 2007; Ghaith & Awad, 2011; Elmarzougui & Larue,
2013).
However, only a few studies involve the analysis of price and/or volatility transmissions
(also called spillovers) among agricultural markets. Buguk, Hudson and Hanson (2003) have
investigated price volatility spillovers in US catfish markets. They have estimated univariate
exponential generalized autoregressive conditional heteroskedasticity (GARCH) models to
test price volatility spillovers in the catfish supply chain. They have found significant
unidirectional spillover from corn, soybean and menhaden prices to catfish feed, farm, and
 A. Ismail

wholesale catfish prices. Apergis and Rezitis (2003) have investigated volatility spillover
effects in Greek agricultural markets using a multivariate GARCH model. They have found
significant positive volatility spillover effects among agricultural input and retail food prices
and agricultural output prices. Rezitis (2003) have considered the prices of beef, lamb, pork
and poultry, to check the cross price spillover effect for substitute goods. Using GARCH model
his study has found significant spillover effect between selected four commodities.
Kim and Doucouliagos (2008) have utilized realized volatility and correlation estimates
for corn, soybean and wheat futures prices by employing realized volatility and co-variation
methods to estimate a vector autoregression (VAR) model. They have evaluated volatility
spillover effects through generalized impulse responses. The three estimated volatilities were
closely related over time based on the existence of volatility spillover effects from one
commodity to the others.
Zhang et al. (2010) have investigated the causality of fuel prices on agricultural commodity
prices. They estimated a VEC model with impulse response functions and error variance
decomposition analyses utilizing ethanol, gasoline, oil, corn and soybean prices. No long-run
relation existed among fuel (ethanol, oil and gasoline) prices and agricultural commodity (corn
and soybean) prices. In addition, although short run relations between fuel and agricultural
commodity prices were present they were not persistent
Wu, Guan, Myers (2010) have proposed a trivariate volatility spillover model to compare
three model specifications with different assumptions on the spillover effects from crude oil
futures price to corn cash and futures prices. They have specified three different models: a
constant spillover model (containing constant spillover parameters), an event spillover model
(including differing spillover parameters before and after the introduction of the Energy Policy
Act of 2005) and a substitution spillover model (containing time-varying spillover parameters
allowed to vary with the ratio of fuel ethanol consumption to gasoline consumption). The
trivariate model was estimated using T-GARCH (threshold) and BEKK-GARCH models to
account for asymmetric volatility effects and utilized error correction models as a proxy of the
mean equations for these GARCH processes. Volatility spillovers from crude oil prices to corn
cash and futures prices were detected, in the case of the constant spillover model and an
increase in the intensity of spillover effects since Energy Policy Act of 2005 in the case of the
event spillover model.
Saghaian (2010) has analyzed causal relationships across five US price series: corn,
soybeans, wheat, ethanol and crude oil and have obtained mixed results. That is, the VEC
model indicated that there were no causal links between energy and agricultural markets.
However, results of Granger causality tests indicated crude oil prices Granger cause corn,
soybeans, and wheat prices.
Serra, Zilberman and Gil (2011) have examined price and volatility spillover effect in the
Brazilian ethanol industry by using an error correction model and a multivariate GARCH
process in a single step and concluded ethanol price levels and volatility were positively related
to crude oil and sugar prices, in both the short and long run.
Du, Yu and Hayes (2011) have used a Bayesian analysis to investigate volatility spillover
effects from crude oil to agricultural commodity markets (i.e., corn and wheat). Two types of
models were estimated, a univariate stochastic volatility model with Merton jump and bivariate
stochastic volatility models. They confirmed the existence of volatility transmissions and
concluded that factors such as scalping, speculation, and petroleum inventories help to explain
crude oil price volatility.
Alom, Ward and Hu (2011) have used VAR and TGARCH model to investigate the mean
and volatility spillover effect between world oil prices and prices of food. They have concluded
a significant relationship between oil prices and food prices, further they found that linkage in
recent years have become stronger. Spillover effect was also observed between importer and
exporter countries.

3
Spillover Effect between Food and Agricultural Commodities…

Khiyavi et al. (2012) have used GARCH model to investigate the volatility spillover for
retail prices, producer price and input prices of poultry market. Output price of poultry industry
was observed to be more volatile than the input and retail price volatility. Additionally,
spillover effect from volatile agricultural input price to output prices of poultry and retail food
price spillover to agricultural output prices was observed.

3. Model Specification and Methodology

3.1. GARCH Models

Extending the framework of Engle (1982), Bollerslev (1986) generalized the ARCH (p)
model to GARCH (p, q) in which he added the p lags of past conditional variance into the
equation. The GARCH Model considers conditional variance depends not only on the past
values of squared error term but also on the past values of its conditional variance. Empirical
findings suggest that GARCH model is more parsimonious than ARCH model (Hamao et al.,
1990; Dehn, 2000; Agnolucci, 2009). An ARMA (p, q) – GARCH (p, q) is specified as

𝑟𝑡 = 𝑐 + ∑𝑙𝑝=1 𝛿𝑝 𝑟𝑡−𝑝 + ∑𝑚
𝑞=1 𝜑𝑞 𝜀𝑡−𝑞 + 𝜀𝑡 (1)

𝑝 2 𝑞 2
𝜎𝑡2 = 𝜔0 + ∑𝑖=1 𝛼𝑖 𝜀𝑡−𝑖 + ∑𝑗=1 𝛽𝑗 𝜎𝑡−𝑗 (2)

Where i=1,2…..p, j= 1, 2. . . q, 𝜔0 > 0, 𝛼𝑖 ≥ 0 and 𝛽𝑗 ≥ 0 ensure strict positivity of the

𝑝 𝑞
conditional variance. If ∑𝑖=1 α𝑖 + ∑𝑗=1 β𝑗 is lesser than 1 or closer to unity then volatility will
𝑝 𝑞
persist over time and if ∑𝑖=1 α𝑖 + ∑𝑗=1 β𝑗 is equal to 1, the volatility will persist forever.

The simple GARCH (1, 1) model is given as

𝜎 2 𝑡 = 𝜔0 + 𝛼1 𝜀𝑡−1
2 2
+ 𝛽1 𝜎𝑡−1 (3)

Engle, Lilien and Robins (1987) have extended the GARCH model to GARCH-M model
which allows the conditional mean to be a function of conditional variance so that the
conditional volatility can generate a risk premium which is part of the expected returns. The
ARMA (p, q) - GARCH-M (1, 1) model is specified as follows:

r𝑡 = 𝑐 + ∑𝑙𝑝=1 𝛿𝑝 r𝑡−𝑝 + ∑𝑚 2
𝑞=1 𝜑𝑞 ε𝑡−𝑞 + 𝛿σ 𝑡 + ε𝑡 (4)
𝜎 2 𝑡 = 𝜔0 + 𝛼1 𝜀𝑡−1
2 2
+ 𝛽1 𝜎𝑡−1 (5)

The coefficients δ in mean equation (5) measures the risk premium describing the nature
of relationship between returns and volatility. If δ is greater than zero and statistically
significant, it indicates that the return is positively related to its volatility. It implies a rise in
mean return is caused by an increase in conditional variance as a proxy of increased risk. If δ
is less than zero and statistically significant, it indicates that the return is negatively related to
its volatility. It implies a fall in mean return is caused by an increase in conditional variance
as a proxy of increased risk and δ=0 indicates no relation between return and risk.

3.2. EGARCH Model

The Exponential GARCH model was introduced by Nelson (1991) that incorporates
skewness or asymmetric effects. EGARCH model overcomes two major drawbacks of
 A. Ismail

symmetric GARCH model. It is specified to capture the leverage effect and relaxes the non-
negativity constraint. An EGARCH (p, q) model is expressed as follows:

2 𝑝 𝜀𝑡−𝑖 |𝜀𝑡−𝑖 | |𝜀𝑡−𝑖 | 2

𝑙𝑛𝜎𝜋𝑡 = 𝜔0 + ∑𝑖=1 𝛼𝑖 𝜃1 + 𝜃2 { − 𝐸( )} + 𝛽𝑗 𝑙𝑛𝜎𝑡−𝑗 ) (6)
√𝜎𝑡2 −𝑖 √𝜎𝑡2 −𝑖
√√𝜎𝑡2
−𝑖
[ ]
The EGARCH model indicates that the conditional variance is an exponential function,
therefore even if the parameters are negative, σ2t will be positive. There is thus no need to
impose non-negativity constraints on the model parameters. In the model, εt-i > 0 is the sign of
good news, and εt-i< 0 i.e. negative lagged error is the sign of bad news. The 𝜃1 reflects the
sign effect and 𝜃2 reflects the magnitude effect. If the asymmetry effect is present, then 𝜃1 <
0, while there is no asymmetry effect, if 𝜃1 = 0. When 𝜃1 < 0, positive shock increases volatility
less than negative shock. When 𝜃1 > 0, negative shock increases volatility less than positive
shock.

3.3. GJR-GARCH Model

The GJR- GARCH model was presented by Glosten, Jaganathan and Runkle in 1993. It
provides a dummy variable It which is an indicator for sign of error terms. Where I t (dummy
variable) = 1 if γi < 0, and 0 if γi > 0 .The GJR-GARCH (1, 1) model is specified as

𝑝 2 𝑞 𝑓 2
𝜎 2 𝑡 = 𝜔0 + ∑𝑖=1 𝛼𝑖 𝜀𝑡−𝑖 + ∑𝑗=1 𝛽𝑗 𝜎𝑡−𝑗 + ∑𝑘=1 𝛾𝑘 𝜀𝑡−𝑘 𝐼𝑡−𝑘 (7)

Where It (dummy variable) = 1 if γk < 0, and 0 if γk > 0. In the model, positive lagged error
εt-i > 0 is a sign of god news, and negative lagged error εt-i < 0 is the sign of bad news. If γk >
0, bad news increases volatility and indicates a leverage effect. The impact of shock is
symmetric if γk = 0, i.e. past bad news (negative shocks) impacts similarly on current volatility
as good news (positive shocks).

Following the methodology of Hamao et al. (1990) and Buguk et al. (2003) GARCH
models is applied with best fitted specifications. GARCH models is applied on monthly return
series. Residuals and residual squares of the each food and agricultural series are saved for
the modelling volatility spillover effect for substitute goods. Maximum likelihood estimation
method is used. ARMA (p, q) specifications are chosen keeping in view the ACF and PACF
with minimized information criterion.

3.4. Spillover Effect using GARCH Modelling

Price spillover occurs when a change in the prices of commodity at a specific location
increases supply and changes the price of that commodity in other locations through trade. It
may also significantly affect the price of a related commodity in the same location. This is
particularly relevant for products with low demand elasticity (Bantilan & Davis, 1991).

Spillover effect using GARCH model for substitute goods are considered for wheat and
rice, and for beef and poultry.

Mean and Variance equation for GARCH (1, 1) model for the volatility spillover effect of
rice on wheat is given as

5
Spillover Effect between Food and Agricultural Commodities…

r𝑤ℎ𝑒𝑎𝑡,𝑡 = 𝑐 + ∑𝑙𝑝=1 𝛿𝑝 r𝑤ℎ𝑒𝑎𝑡,𝑡−𝑝 + ∑𝑚

𝑞=1 𝛾𝑞 ε𝑤ℎ𝑒𝑎𝑡,𝑡−𝑞 + 𝜓1 𝑢𝑅𝐼𝐶𝐸,𝑡 + ε𝑡 (8)
2
𝜎𝑤ℎ𝑒𝑎𝑡,𝑡 = 𝜔0 + 𝛼1 ε2𝑤ℎ𝑒𝑎𝑡,𝑡−1 + 𝛽1 𝜎 2𝑤ℎ𝑒𝑎𝑡,𝑡−1 + 𝜓1 𝑢𝑅𝐼𝐶𝐸,𝑡
2
(9)

Mean and Variance equation for the spillover effect of residual of wheat on return series
of rice is given as

r𝑟𝑖𝑐𝑒,𝑡 = 𝑐 + ∑𝑙𝑝=1 𝛿𝑝 r𝑟𝑖𝑐𝑒,𝑡−𝑝 + ∑𝑚

𝑞=1 𝛾𝑞 ε𝑟𝑖𝑐𝑒,𝑡−𝑞 + 𝜓1 𝑢𝑤ℎ𝑒𝑎𝑡,𝑡 + ε𝑡 (10)
2
𝜎𝑟𝑖𝑐𝑒,𝑡 = 𝜔0 + 𝛼1 ε2𝑟𝑖𝑐𝑒,𝑡−1 + 𝛽1 𝜎 2𝑟𝑖𝑐𝑒,𝑡−1 + 𝜓1 𝑢𝑤ℎ𝑒𝑎𝑡,𝑡
2
(11)
Where, r𝑤ℎ𝑒𝑎𝑡 is the return series of wheat and r𝑟𝑖𝑐𝑒 is the return series of rice in time‘t’.
Residual of rice and residual of wheat are 𝑢𝑅𝐼𝐶𝐸 and 𝑢𝑤ℎ𝑒𝑎𝑡 respectively. Residual square of
2 2
rice and wheat are shown as 𝑢𝑅𝐼𝐶𝐸 and 𝑢𝑤ℎ𝑒𝑎𝑡 , 𝜎 2 𝑤ℎ𝑒𝑎𝑡 and 𝜎 2 𝑟𝑖𝑐𝑒 is the variance of the wheat
and rice series in time‘t’.

To measure the volatility spillover effect of beef on poultry mean and variance equation
for the GARCH (1, 1) model is given as

r𝑝𝑜𝑢𝑙𝑡𝑟𝑦,𝑡 = 𝑐 + ∑𝑙𝑝=1 𝛿𝑝 r𝑝𝑜𝑢𝑙𝑡𝑟𝑦,𝑡−𝑝 + ∑𝑚𝑞=1 𝛾𝑞 ε𝑝𝑜𝑢𝑙𝑡𝑟𝑦,𝑡−𝑞 + 𝜓1 𝑢𝑏𝑒𝑒𝑓,𝑡 + ε𝑡 (12)

2 2 2 2
𝜎𝑝𝑜𝑢𝑙𝑡𝑟𝑦,𝑡 = 𝜔0 + 𝛼1 ε𝑝𝑜𝑢𝑙𝑡𝑟𝑦,𝑡−1 + 𝛽1 𝜎 𝑝𝑜𝑢𝑙𝑡𝑟𝑦,𝑡−1 + 𝜓1 𝑢𝑏𝑒𝑒𝑓,𝑡 (13)

Mean equation and Variance equation for the spillover effect of poultry on beef

r𝑏𝑒𝑒𝑓,𝑡 = 𝑐 + ∑𝑙𝑝=1 𝛿𝑝 r𝑏𝑒𝑒𝑓,𝑡−𝑝 + ∑𝑚 𝑞=1 𝛾𝑞 ε𝑏𝑒𝑒𝑓,𝑡−𝑞 + 𝜓1 𝑢𝑝𝑜𝑢𝑙𝑡𝑟𝑦,𝑡 + ε𝑡 (14)

2 2 2 2
𝜎𝑏𝑒𝑒𝑓,𝑡 = 𝜔0 + 𝛼1 ε𝑏𝑒𝑒𝑓,𝑡−1 + 𝛽1 𝜎 𝑏𝑒𝑒𝑓,𝑡−1 + 𝜓1 𝑢𝑝𝑜𝑢𝑙𝑡𝑟𝑦,𝑡 (15)
Where, r𝑝𝑜𝑢𝑙𝑡𝑟𝑦 is the return series of poultry and r𝑏𝑒𝑒𝑓 is the return series of beef in time
period‘t’. Residual of poultry and beef are represented by 𝑢𝑝𝑜𝑢𝑙𝑡𝑟𝑦 and 𝑢𝑏𝑒𝑒𝑓 respectively.
2 2
Residual square of poultry and beef are shown as 𝑢𝑝𝑜𝑢𝑙𝑡𝑟𝑦 and 𝑢𝑏𝑒𝑒𝑓 respectively and the
2 2
variance of poultry is shown by 𝜎 𝑝𝑜𝑢𝑙𝑡𝑟𝑦 and 𝜎 𝑏𝑒𝑒𝑓 is the variance of the beef series for the
time‘t’.
This study used ARMA (p, q) model for the identification of the appropriate lag length
for conditional mean and variance specification. Additionally, ACF and PACF of return series
was also considered in the ARMA model. Maximum log likelihood estimation with standard
error based on a least square process is employed. Gaussian and student “t” distribution is
used to get the best fitted model. For the estimation of parameters BFGS-BOUNDS is applied.
This method is efficient for unconstrained optimization as followed by Broyden (1970).
BFGS converges for an optimum quadratic Taylor expansion. After finding appropriate lag
length of the return series residuals and residual squares of all the series are saved. Secondly,
the LM - ARCH test is applied to analyze the presence of ARCH effect in the residuals of
each series. Set of diagnostic tests are applied to check the best fitted lag length for the
GARCH models, based on the information criterion (Akaike criteria, Shibata criteria,
Schwartz Bayesian criteria and Hannan-Quinn criteria), that is to be minimized. The best
fitted ARMA model is the one for which GARCH model is converging towards the normal
distribution. Q-statistics and Q2-statistics of Ljung-Box-Pierce test is applied to check the
correlation and the volatility clustering. Following Tse (2002) to observe conditional
heteroskedasticity Residual Based Diagnostic (RBD) test is also applied. Conditional mean
and conditional variance is also saved for the volatility modelling of the food and agricultural
series under observation.
 A. Ismail

4. Data and Empirical Results

4.1 Data Description and Sources

In this study monthly data of selected food and agricultural commodities from Pakistan is
used. The period chosen is from April 1983 till April 2013 for rice, wheat, beef, and poultry.
The data is taken from IMF, World Bank, IFS and State Bank of Pakistan. Monthly data on
all the commodities are represented in US dollar.

Table 1. List of Variables

Variables Symbols Sample Data frequency Data sources Units

Period
Rice R 1983-2013 April 83 - April 13 World bank US Dollars per
Metric Ton
Wheat W 1983-2013 April 83 - April 13 World Bank US Dollars per
Metric Ton
Beef BEEF 1983-2013 April 83 - April 13 World bank US cents per Pound
Poultry PLTY 1983-2013 April 83 - April 13 IMF US cents per Pound
Note: Table 1 shows list of all the selected variables, their unit and sources.

For the analysis of the behavior of price series, returns are calculated which will be used to
measure volatility. Monthly returns are calculated by taking first difference of logged monthly
prices for each commodity:
𝑟𝑖𝑡 = ln( 𝓅𝑖,𝑡 /𝓅𝑖,𝑡−1 )
Where, 𝓅 is the price of commodity, where i is the commodity in time t.

4.2. Graphical Analysis

Figures show plots of monthly price series of beef, poultry, rice and wheat. Plots in figure
1 show the original monthly price series, which reveals general upward trend for the selected
period.

7
Spillover Effect between Food and Agricultural Commodities…

Figure 1. Monthly Price Series of Food/Agricultural Commodities

The plots of monthly return series are shown in figure 2. These clearly show the time
varying volatility and do not show any fix pattern. All the return series show high fluctuation
and return back to its mean slowly. It is clearly shown that the variance of return series of
prices is not fixed over time. The volatility clustering is exhibited in return series showing
periods of high and low volatilities, indicating ARCH effect.

Figure 2: Return Series of Food/Agricultural Commodities: The Returns shows no

definite pattern, and are reverting towards mean. These graphs shows that the return series are
characterized with volatility clustering.
 A. Ismail

In figure 3 squared returns indicates variation in volatility. Brief periods of high volatility
are more visible from the squared returns taken as measure of volatility. High order serial
correlation is observed through the graphical representation of squared returns. It discloses that
volatility is affected more by the periods of exciting returns.

Figure 3. Squared Return Series of Food/Agricultural Commodities

The graph of the density function for return series shows the histogram and distribution of
the data. Figure 4 shows the histogram of price return series of food and agricultural
commodities. It shows that the mean and median of monthly price returns are not significantly
different from zero and indicates a slightly increasing trend over time. It is observed that the
density function of the selected series shows stylized facts of non-normal distribution, high
peak and fat tails. Fat tail in a return series shows high probability of extreme events than the
Gaussian distribution.

9
Spillover Effect between Food and Agricultural Commodities…

Figure 4. Histogram of Price Return Series: Density plot shows that the distribution of the return
series is not normal and exhibiting the features of skewness, and leptokurtosis. Normal distribution of series is
shown by dotted line and solid lines shows the non-normal distribution, which is more peaked than the normal
density.

4.3. Descriptive Statistics of Food and Agricultural Price Return Series

Summary statistics of price return series (r𝑖𝑡 ) is shown in Table 2. The purpose is to
investigate the nature of the series, and it partly indicates which model is best for the analysis.
Volatility is measured by the standard deviation, all the series show a high value compared
with their mean value, and it suggests the series are highly volatile.
All the series are characterized with excess kurtosis and skewness, which is the feature of
high frequency data. A series is said to be normally distributed if the value of its skewness is
close to zero. Rice, wheat, beef and poultry are positively skewed1 are significant at 1% (see.
Table 2). Non normality of data distribution is clearly indicated for all the series. Secondly,
the value of excess kurtosis2 in Table 2 helps to analyze the peak of the return price series
distribution. The series of rice and beef have indicates high kurtosis. Higher value of kurtosis
indications that the data distribution of these series is leptokurtic 3.It also indicates that there is
a high probability of having extreme values in the distribution. The high value of the excess
kurtosis indicates that the data shows a heavier tail than the normal distribution.

1 𝐸[(𝓎−𝓊)3 ]
𝑆𝑘𝑒𝑤𝑛𝑒𝑠𝑠 = . When the value of skewness is positive and greater than zero, it is
𝜎3
positively skewed with a tail longer at right end.
2 𝐸[(𝓎−𝓊)3 ]
𝐸𝑥𝑐𝑒𝑠𝑠 𝐾𝑢𝑟𝑡𝑜𝑠𝑖𝑠 = − 3 𝑜𝑟 𝐹𝑖𝑠ℎ𝑒𝑟 𝐾𝑢𝑟𝑡𝑜𝑠𝑖𝑠
𝜎3
𝑤ℎ𝑒𝑟𝑒, 𝑦 𝑖𝑛𝑑𝑖𝑐𝑎𝑡𝑒𝑠 𝑡ℎ𝑒 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛 𝑤𝑖𝑡ℎ 𝑚𝑒𝑎𝑛 𝓊 𝑎𝑛𝑑 𝜎 𝑠ℎ𝑜𝑤𝑠 𝑠𝑡𝑎𝑛𝑑𝑎𝑟 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛
3
Leptokurtic is sharper than normal distribution with a heaver tail.
 A. Ismail

Table 2.Descriptive statistics for returns (Rt) of series

Standard Jurque-
Minimum Maximum Mean deviation Skewness Kurtosis Bera
Rice -0.2813 0.0019 0.0019 0.0624 1.2022 8.6085 1198.3000
Wheat -0.2193 0.0017 0.0017 0.0587 0.4165 2.4198 98.2390
Beef -0.1797 0.0014 0.0014 0.0371 0.0311 3.2890 162.3200
Poultry -0.0571 0.0032 0.0032 0.0223 0.6637 2.1792 97.6610
critical value 0.01 0.05 0.10
Jurque-Bera 9.21 5.99 4.61

Whereas, wheat and poultry shows a value less than 3 for kurtosis. A value lower than 3
shows that there is less probability of extreme values and these series are platykurtically
distributed4. It implies that the distributions of return series are not normal.
In Jarque Bera test (JB) the null hypothesis under consideration is that the series are
normally distributed. The null hypothesis of normality of JB statistics are rejected for all return
series as the calculated value in Table 2 is greater than the critical value at 1%, 5%, and 10%
level of significance suggesting that all the distributions of a return series are non-normal.
Overall descriptive statistics show that distributions of price return series are skewed,
leptokurtic and platykurtic. It concludes that the price return series of Pakistani market show
non normal distribution which is the main characteristic of the data set of most of the emerging
markets (Choudry, 1996).

4.4. Unit Root Test

Several tests are used in existing literature to test the stationarity of the return series;
Augmented Dicky Fuller test (1979), Phillips Perron test (1988), Schmidt and Phillips test in
1989 and Kwiatkowski Phillips Schmidt Shin test (1992). The unit root tests existing in the
literature use the null hypothesis (H0) that series is non stationary and alternate hypothesis (HA)
that it is stationary. Whereas, KPSS test conditions that the null hypothesis is that the series is
stationarity. Therefore, this research focuses on the stationarity of the return series by using
KPSS test at level and at first difference with constant. The results of unit root tests are shown
in Table 3.
All series are non-stationary at the level as the calculated values are greater than the
critical values at 1%, 5%, and 10% level of significance. At first differenced return series of
rice, wheat, beef and poultry are stationary at 1% level of significance.

4
Platykurtic distribution is flatter than normal distribution.

11
Spillover Effect between Food and Agricultural Commodities…

Table 3. Unit Root Test: Kwiatkowski–Phillips–Schmidt–Shin (KPSS)

Log Level Log First Difference
with with
Variables with constant constant and with constant constant and
trend trend
Rice 7.2194 2.0313 0.0930 0.0357
Wheat 7.9321 1.8281 0.1078 0.0283
Beef 5.7678 3.0097 0.2345 0.0378
Poultry 16.6182 1.1796 0.0370 0.0339
critical value 1% 5% 10%
with constant 0.739 0.463 0.347
with constant and trend 0.216 0.146 0.119
Note: All series are non-stationary at level and are stationary at first difference at 1% critical
value

4.5. Testing for ARCH Effect

After the visual analysis it is clear that the data is characterized with ARCH effect. Some
diagnostic tests are applied for the statistical evidences. The Ljung box Q-statistics are applied
to check the serial correlation in the return series (Ljung and Box, 1978). The null hypothesis
of no serial correlation was used in the test and tested residuals for lag 5, 10, 20 and 50 are
reported in table 4. Results significantly reject the null hypothesis, indicating that serial
correlation is present in residual series. This suggest AR process in mean equation.
To check for ARCH effect, Ljung Box’s Q2 statistics is also applied to the square residual.
2
Q statistics are tested up to lag 50 with the null hypothesis of no serial correlation. Results in
table 4 shows strong evidence of serial correlation on squared residuals for all the food and
agricultural commodities; hence data are characterized by volatility clustering.
Lagrange Multiplier ARCH (LM-ARCH) test is applied to the square residual of a return
series of rice, wheat, beef and poultry. The null hypothesis of no ARCH effect was tested and
the results are shown in Table 4. Concluding that square residual of the series has ARCH effect.
Strong indication for the rejection of null hypothesis is indicated.

4.6. Estimated GARCH Models

Due to the presence of ARCH effect and non-normal distribution of the return series, there
is a need of GARCH-type model for volatility. From the evidences found in literature it is
evident that family of GARCH models is appropriate to adjust the ARCH effect.
Information criteria of Akaike, Schwarz, Shibata and Hannan Quinn are used to select the
appropriate p and q for the better fit ARMA model. Primarily several models are estimated
with different p and q values, and the order for which information criterion is minimum is
selected (Bozdogan, 2000).The necessary condition of convergence of data and maximum
likelihood function value is considered.
The plots of the ACF and PACF are used to observe patterns in the return and squared
return series that helped in the selection of p and q lags for conditional mean and conditional
variance equations (see. Figure 5 & 6). The values of ACF and PACF that lie outside the
confidence interval will identify order for ARMA models. Additionally, it shows that the
return series are characterized with short memory. ACF and PACF of the squared return series
exhibit that the autocorrelation do not persist for a long term and die out fast.
 A. Ismail

Table 4.ARCH Effect Diagnostic

ARCH
TEST Q-STATISTICS Q-STATISTICS
ARCH (1-2) ARCH (1-5) ARCH (1-10) On raw data On squared data
Variables F(2,355) F(5,349) F(10,339) Q(5) Q(10) Q(20) Q(50) Q(5) Q(10) Q(20) Q(50)
35.4820 15.1630 7.5504 50.0978 65.8433 86.8228 144.2720 86.2966 88.8900 90.6369 100.7910
Rice
(0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0001)**
10.0560 4.1426 2.4165 26.5722 39.4901 49.3374 84.9637 25.4564 33.1676 52.4578 108.3830
Wheat
(0.0001)** (0.0011)** (0.0087)** (0.0001)** (0.0001)** (0.0002)** (0.0014)** (0.0001)** (0.0002)** (0.0001)** (0.0001)**
14.3070 7.1982 4.0405 24.5758 40.7535 64.9500 95.7265 34.0669 37.1655 40.2718 65.9114
Beef
(0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0001)** (0.0000)** (0.0000)** (0.0046)** (0.0651)**
Poultry 62.0850 24.7490 11.9960 108.3210 157.0560 259.0090 454.0520 100.2850 109.3790 111.7900 200.2760
(0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)** (0.0000)**

Note: p-values are shown in brackets, * shows the 5% level of significance and ** shows significance at 1%. Q-Statistics is the Ljung-Box statistics based on
standardized residual and square of standardized residual up to lag 50 with H0: no serial correlation. LM-ARCH (n) Lagrange multiplier test for ARCH effect up
to order n, its H0: series is not subject to ARCH effect.

13
Spillover Effect between Food and Agricultural Commodities…

Figure 5 Autocorrelation Functions (ACF) and Partial Autocorrelation Functions (PACF) of Monthly Return
Series

Figure 6. Autocorrelation Functions (ACF) and Partial Autocorrelation Functions (PACF) of Square of
Monthly Return Series
 A. Ismail

In order to find the best fitted model for each return series of food and agricultural commodity, initially we
estimated the n-GARCH, t-GARCH, EGARCH and GJR-GARCH models.
Final model and specifications for each commodity is selected after applying the diagnostic tests of residuals
(normality test, information criterion, log likelihood value, Q- statistics, Q2 -statistics, ARCH test, Residual Based
Diagnostic). Results of final GARCH (1, 1) models that best fitted the individual series are presented in Table 5 and
diagnostic test for each model is exhibited in Table 6. Results5 of the final model are discussed in the next section.
Maximum likelihood estimation method is used to compute the covariance matrix. ARMA models of maximum
(3, 3) order are estimated for food and agricultural commodities. Considering the ACF, PACF and minimum
information criterion best fitted specification for tea chosen is ARMA (2, 3) t-GARCH (1, 1). Parameter for the mean
and variance equations are statistically significant for the selected model.
Specification chosen for the return series of rice after satisfying minimum information criterion is ARMA (2, 1),
t-GARCH (1, 1). Parameters of mean and variance equation show statistical significance. Coefficients of
unconditional variance is 3.080723 approximately the same as sample variance. Sum of α 1, β1 equal to 0.94722,
indicating positive constrain and stationarity. High volatility persistence is shown by the high value of α 1, β1. Return
series of rice is positively skewed with a small value of excess kurtosis. The high value of JB test is the indication of
the non-normal

Table 5: GARCH (1, 1) Best Fitted Model for Individual Series

Rice Wheat Beef Poultry
Model ARMA (2, 1) ARMA (1, 0) ARMA (2, 1) ARMA (1, 2)
t-GARCH t-GARCH t-GARCH t-GARCH
Specifications
(1, 1) (1, 1) (1, 1) (1, 1)
Mean Equation
C 0.0010 -0.0006 0.0009 0.0042
(0.6524) (0.8528) (0.6466) (0.6466)
AR(1) 0.7034 0.2454 0.2102 0.3971
(0.0007) ** (0.0000) ** (0.0003) ** (0.0091) **
AR(2) -0.2803 -0.0722
(0.0000) ** (0.1988)
MA(1) -0.3932 0.2362
(0.0683) (0.1169)
MA(2) 0.1236
(0.2045)
Variance
Equation
C 3.0807 2.0485 1.8495 0.0000
(0.0454) * (0.1912) (0.1176) (1.0000)
α(1) 0.2616 0.1143 0.1413 0.0697
(0.0060) ** (0.0380) * (0.0492) * (0.0064) **
β(1) 0.6856 0.8382 0.7258 0.9270
(0.0000) ** (0.0000) ** (0.0000) ** (0.0000) **
Student (DF) 4.0170 4.5461 5.5598 9.6869
(0.0001) ** (0.0011) ** (0.0005) ** (0.0637)
α(1)+β(1)|
0.9472 0.9524 0.8671 0.9967
α(1)+β(1)+γ(1)[1]
Log Likelihood 581.4410 551.8820 710.7930 969.1440
Akaike Criteria -3.1858 -3.0327 -3.9100 -5.3397
Schwarz Criteria -3.0994 -2.9679 -3.8344 -5.2533
Shibata Criteria -3.1867 -3.0332 -3.9107 -5.3406
Hannan-Quinn -3.1514 -3.0069 -3.8799 -5.3054
Note: p-values are shown in brackets, * shows the 5% level of significance
and ** shows significance at 1%.

5
N-GARCH and t-GARCH

15
Spillover Effect between Food and Agricultural Commodities…

Table 6: Diagnostic Test for GARCH (1, 1) Best Fitted Model for Individual Series
Rice Wheat Beef Poultry
Skewness 0.6728 0.5453 -0.3827 -0.2412
p-value (0.0000) ** (0.0000) ** (0.0029) ** (0.0607) *
Normality Test Excess 2.6685 1.6670 3.4803 0.8413
Kurtosis (0.0000) ** (0.0000) ** (0.0000) ** (0.0010) **
133.9700 59.5220 190.4700 14.1070
Jarque-Bera
(0.0000) ** (0.0000) ** (0.0000) ** (0.0009) **
4.4758 0.8377 4.8625 6.2789
Q(5)
(0.1067) (0.9333) (0.1821) (0.0433)*
0.6728 0.5453 -0.3827 -0.2412
Q(10)
Q- Statistics on (0.0000) ** (0.0000) ** (0.0029) ** (0.0607)
On raw data 2.6685 1.6670 3.4803 0.8413
Q(20)
(0.0000) ** (0.0000) ** (0.0000) ** (0.0010) **
133.9700 59.5220 190.4700 14.1070
Q(50)
(0.0000) ** (0.0000) ** (0.0000) ** (0.0009) **
4.4758 0.8377 4.8625 6.2789
Q(5)
(0.1067) (0.9333) (0.1821) (0.0433)*
6.4029 9.2229 3.6111 9.8962
Q(10)
Q- Statistics on (0.6022) (0.3238) (0.8904) (0.2724)
On squared data 17.0717 20.4041 6.8837 13.9000
Q(20)
(0.5182) (0.3105) (0.9911) (0.7356)
41.5005 59.9610 26.4571 48.7972
Q(50)
(0.7347) (0.1153) (0.9951) (0.4408)
0.6835 0.6196 0.4627 2.1238
ARCH(1-2)
(0.5055) (0.5387) (0.6300) (0.1211)
0.5947 1.2302 0.4138 1.3827
ARCH Test ARCH(1-5)
(0.7041) (0.2944) (0.8391) (0.2301)
0.6439 0.8186 0.3382 0.5057
ARCH(1-10)
(0.7759) (0.6109) (0.9702) (0.8858)
-0.0601 -0.3412 2.4091 50.0934
RBD( 2)
(1.0000) (1.0000) (0.2998) (0.0000) **
1.6597 6.8904 4.1237 -22.8016
RBD Test RBD( 5)
(0.8939) (0.2289) (0.5318) (1.0000)
8.7646 9.0309 5.9376 4.0252
RBD( 10)
(0.5546) (0.5292) (0.8205) (0.9462)
Note: * shows the 5% level of significance and ** shows significance at 1%. Q-Statistics is the Ljung-Box statistics based on standardized residual
and square of standardized residual up to lag 50 with H0: no serial correlation. LM-ARCH (n) Lagrange multiplier test for ARCH effect up to order
n, its H0: series is not subject to ARCH effect. JB (Jarque Bera) test H0: series is normal. RBD is the Residual based diagnostics.

data distribution. No serial correlation appears till the level of 20 lags. Q2 –statistics and ARCH test shows no volatility
clustering and ARCH effect. The p-value of the coefficients of RBD test signpost no heteroscedasticity up to lag 10.
For wheat coefficients for the parameters of mean and variance equation of wheat series is significant, estimated
through maximum likelihood are shown in table 5 and table 6. ARMA (1, 0), t-GARCH (1, 1) is the best fitted model
for wheat return series that satisfied the minimum information criterion (AIC, SIC, SBC, HQIC). Sum of α, β equals
to 0.95244 which is the indication of stationarity (α1 + β1 < 1) and as it is greater than zero it shows positive constrain.
Moreover, the high value of α1, β1 suggest the existence of high volatility. The variance of the return series is almost
same to the unconditional variance i.e. 2.048525. GARCH (1, 1) modelling of wheat series shows positively skewed
 A. Ismail

data with platykurtic distribution (which is the indication of less extreme value). High JB value in wheat series is the
indication of the distribution is non normal. The test of residual and square residuals shows high significance for not
rejecting the null hypothesis of no serial correlation and no volatility clustering. ARCH test reports no ARCH effect
in the series till lag 10. The significance of the coefficients of RBD test suggests no heteroscedasticity.
Specifications for beef which show significance are ARMA (2, 1), t-GARCH (1, 1). Parameters of unconditional
mean and the variance equation show the significant estimates (see. Table 5 & 6). The positivity constraint (α1 + β1
≥ 0) for the GARCH (1, 1) model is detected since α1 + β1 = 0.86712 which is greater than zero. The sum of α1 and β1
is less than 1 for the satisfaction of stationarity constraint. The sum of α 1 + β1 indicates that shocks to volatility are
high. Beef series is negatively skewed and its distribution is leptokurtic (showing a high probability of extreme values).
Non normality of distribution is concluded from the high JB value. A result for GARCH (1, 1) model shows high
significance for no serial correlation, no volatility clustering, no ARCH effect and no heteroscedasticity.
An appropriate specification for the return series of poultry is ARMA (1, 2), t-GARCH (1, 1) chosen after fulfilling
the minimum information criterion. Parameters for unconditional variance and mean equation are significant. The
positivity constraint (α1 + β1 ≥ 0) for the GARCH (1, 1) model is detected as α1 + β1 = 0.99673. The totality of α1, β1
is less than 1 which satisfies the stationarity restriction (α1 + β1 <1). The sum of α1 + β1 is close to 1 that indicating
high shocks to volatility. Return series of poultry is negatively skewed; playtokurticly distributed which shows non
normal distribution. Poultry series shows some signs of correlation and volatility clustering at lag 5. LM-ARCH test,
do not reject the null hypothesis of no ARCH effect. No heteroscedasticity was indicated from the results in Table (5
& 6).

4.7. Return and Volatility Spillover Effect of Related Goods GARCH (1, 1) Models

This study examines the return and volatility spillover effect for substitute goods. After choosing the appropriate
GARCH model, residual and residual squares of the series are saved. Residual of the substitute commodity is added
in the mean equation and Residual Square is added in the variance equation to check its impact on the return series.
Again, GARCH (1, 1), EGARCH (1, 1) and GJR-GARCH (1, 1) models are estimated, considering the results of
skewness, kurtosis, JB, minimum information criterion and appropriate model is selected. Best fitted model is the one
with maximum likelihood, no correlation, no volatility clustering, no ARCH effect and no heteroscadisity. The main
objective is to see if the volatility is transmitted from one commodity to another substitute commodity.
There are mixed empirical evidences about the transmission variation in price of one commodity across its close
substitute, like wheat and rice. Evidences by McGuirk and Mundlak (1991) observed positive price transmission
between wheat and rice, while Swamy and Binswanger (1983) and Cornelisse, Kuijpers and Salam (1987) concluded
negative price transmission between wheat and rice. The wheat and rice are harvested in different seasons, so they do
not impact on the production level of each other. Whereas, the current prices do depend on the commodity last year
production, stock, other seasonal and input factors.
Table 7 shows the results for return and volatility spillover effect models. For spillover effect of residual of rice
series on the return series of wheat, the best fitted specification is ARMA (1, 0), t-GARCH (1, 1) for which parameters
of mean and variance equation are significant. The results shows that the coefficient for residual and residual square
of rice are insignificant, indicating that the volatility of rice do not affect the volatility of wheat series. High value of
β shows that wheat series is less reactive to market changes. Sum of α, β is less than one and therefore it satisfies the
stationarity constrain but the high value also represent the high volatility due to shock. Spillover model of rice on
wheat is positively skewed and platykurtically distributed. Jarque Bera is too high indicating that the distribution is
not normal. Q-statistics and Q2–statistics provide evidence of no serial correlation and no volatility clustering. LM-
ARCH test was applied, which shows no ARCH effect and RBS indicates no heterosadesity in the model.
Next, we estimate the spillover effect of wheat on rice. The best fitted specification chosen is ARMA (2, 3), n-
GARCH (1, 1) with highly significant mean and variance parameters. Parameter for residual of wheat is insignificant.
Results in Table 7 shows that the volatility of wheat significantly affect the volatility of rice series, but the positivity
of the coefficient of wheat shows that these commodities are not substitutes. Sum of α1, β1 equals 0.97595 which is
less than one, so the model is stationary but the high value also indicate high volatility due to shock. High value of JB
test shows non normal distribution. Results are evident of no serial correlation, no volatility clustering and no ARCH
effect. Additionally, RBS for residual shows no heteroscedasticity (see. Table 8).

17
Spillover Effect between Food and Agricultural Commodities…

Table 7: Return and Volatility Spillover Effect

rice on wheat wheat on rice poultry on beef beef on poultry
Model ARMA (1, 0) ARMA (2, 3) ARMA (2, 0) ARMA (0, 3)
Specifications t-GARCH (1, 1) n-GARCH (1, 1) t-GARCH (1, 1) t-GARCH (1, 1)
Mean Equation
C -0.0004 0.0019 0.0008 0.0043
(0.9052) (0.5419) (0.6581) (0.0021) **
Residual 0.0432 0.0084 -0.0850 -0.0028
coefficient (0.3686) (0.8356) (0.2388) (0.8446)
0.2368 1.0000 0.2107
AR(1)
(0.0000) ** (0.0000) ** (0.0003) **
-0.9377 -0.0759
AR(2)
(0.0000) ** (0.1778)
-0.7038 0.6237
MA(1)
(0.0000) ** (0.0000) **
0.5706 0.3907
MA(2)
(0.0000) ** (0.0000) **
0.3377 0.1775
MA(3)
(0.0000) ** (0.0016) **
Variance
Equation
1.8217 0.8090 2.1382 0.0000
C
(0.2073) (0.2139) (0.0967) (1.0000)
Residual sq -0.0001 0.0247 -0.0399 -0.0004
coefficient (0.9925) (0.0161) * (0.4851) (0.6394)
0.1031 0.2337 0.1453 0.0611
α(1)
(0.0477) * (0.0000) ** (0.0512) * (0.0057) **
0.8577 0.7423 0.7105 0.9375
β(1)
(0.0000) ** (0.0000) ** (0.0000) ** (0.0000) **
4.2807 5.7529 9.7874
Student (DF)
(0.0009) ** (0.0010) ** (0.0730)
α(1)+β(1) 0.9608 0.9760 0.8559 0.9986
Log Likelihood 552.3130 571.1450 711.4170 970.7090
Akaike Criteria -3.0240 -3.1119 -3.9023 -5.3373
Schwarz Criteria -2.9376 -2.9932 -3.8052 -5.2293
Shibata Criteria -3.0249 -3.1137 -3.9035 -5.3388
Hannan-Quinn -2.9896 -3.0647 -3.8637 -5.2943

Note: * shows the 5% level of significance and ** shows significance at 1%.

 A. Ismail

Table 8: Diagnostic Test for Return and Volatility Spillover Effect

rice on wheat on poultry on beef on
wheat rice beef poultry
0.5350 0.5821 -0.3360 -0.1989
Skewness
(0.0000) ** (1.8969) (0.0090) ** (0.1220)
Excess 1.6778 0.0000 3.0359 0.8107
Normality Test
Kurtosis (0.0000) ** (0.0000) ** (0.0000) ** (0.0016) **
59.4000 74.3030 145.0200 12.2310
Jarque-Bera
(0.0000) ** (0.0000) ** (0.0000) ** (0.0022) **
1.1223 0.0000 4.3171
Q(5)
(0.8907) (0.2011) (0.1155)
8.3299 9.5245 12.5272 21.5201
Q(10)
Q- Statistics on (0.5012) (0.0899) (0.1291) (0.0030)**
On raw data 17.8639 23.6640 28.2595 62.6229
Q(20)
(0.5315) (0.0710) (0.0582) * (0.0000)**
40.9793 58.1556 68.0036 170.5780
Q(50)
(0.7854) (0.0902) (0.0302)* (0.0000)**
6.5071 2.5310 2.0767 9.4738
Q(5)
(0.0893) (0.4697) (0.5566) (0.0236)*
9.2441 8.0353 3.6659 10.6797
Q(10)
Q- Statistics on (0.3221) (0.4300) (0.8859) (0.2205)
On squared data 19.7553 16.9218 7.3679 14.6304
Q(20)
(0.3467) (0.5285) (0.9866) (0.6872)
58.4606 54.0353 28.5279 46.9340
Q(50)
(0.1433) (0.2548) (0.9885) (0.5165)
0.7747 0.5215 0.4169 2.4322
ARCH(1-2)
(0.4616) (0.5941) (0.6594) (0.0893)
1.2652 0.5240 0.4107 1.6122
ARCH Test ARCH(1-5)
(0.2785) (0.7582) (0.8413) (0.1561)
0.8364 0.8155 0.3431 0.5576
ARCH(1-10)
(0.5938) (0.6139) (0.9686) (0.8480)
-0.1242 1.1410 2.3574 31.4493
RBD( 2)
(1.0000) (0.5652) (0.3076) (0.0000) **
7.4242 2.7702 4.4671 49.8746
RBD Test RBD( 5)
(0.1909) (0.7353) (0.4842) (0.0000) **
9.7338 9.2869 6.7983 4.5154
RBD( 10)
(0.4641) (0.5051) (0.7443) (0.9211)
Note: * shows the 5% level of significance and ** shows significance at 1%. Q-Statistics is the Ljung-Box statistics based on standardized residual
and square of standardized residual up to lag 50 with H0: no serial correlation. LM-ARCH (n) Lagrange multiplier test for ARCH effect up to order
n, its H0: series is not subject to ARCH effect. JB (Jarque Bera) test H0: series is normal. RBD is the Residual based diagnostics.

The results imply that in case of Pakistan the volatility of wheat prices is transmitted into rice market and causes
volatility in its prices, whereas change in rice prices do not affect the prices of wheat. Additionally, it suggests that
wheat and rice are not substitute goods for each other in case of Pakistan. Furthermore, the utility from the
consumption of rice is assumed to be unaffected by consumption of wheat (Swamy and Binswanger, 1983; Cornelisse
et al., 1987; Kurosaki & Fafchamps, 2002; Atkin, 2013). This is evident that the due to difference in consumption
habit in Pakistan thus it hard to make a strong case for price transmission across substitutes in either direction.
(Cornelisse et al., 1987; Atkin, 2013).

19
Spillover Effect between Food and Agricultural Commodities…

Various studies have shown that the demand for poultry is increasing more than beef, which is a close substitute
(Capps, 1986; McCorriston et al., 2004; Showell et al., 2012). The change in consumption behavior is not entirely due
to the changes in relative prices or income but because of captive supplies, market information, meat packing
concentration and it is also due to the large variety of chicken products available in the market (Capps, 1986; Showell
et al., 2012). According to the low of demand and supply low prices is a result of large supplies and which relative to
current demand conditions. Further, when the price of meat increase, customer start buying closest substitutes like
poultry and lamb. The existence of substitution effect between two commodities is observed through price
transmission, but if the market is competitive6, then shifting into substitutes would have no effect on price transmission
(McCorriston et al., 2004).
Spillover effect of poultry on beef is also investigated and the best fitted specifications chosen are ARMA (2, 0),
t-GARCH (1, 1). Coefficient for residual and residual square are insignificant for poultry. Other parameters of mean
and variance are all significant. Results show that series is negatively skewed and kurtosis is high (leptokurtic). Jarque
Bera test provides evidence of non-normal distribution. Q-statistics shows no serial correlation up to lag 20, Q 2 –
statistics shows no volatility clustering. LM-ARCH test exhibits no ARCH effect in beef series. No evidences of
heteroscedasticity in residuals were found that shows that the GARCH (1, 1) model is a good fit. It conclude that in
Pakistan the price of poultry do not show price and volatility transmission across the prices of beef.
Specification for the spillover effect of beef on poultry is ARMA (0, 3), t-GARCH (1, 1) where, parameters of
mean and variance equation are significant. For the residual and residual square of beef, the coefficient are
insignificant showing that the return series of poultry is not affected by beef. Additionally, volatility of poultry market
is not transmitted across beef market (see Table 7 & 8). Sum of α1 + β1 for poultry return series is 0.99861 showing
positive constrain (α1 + β1 ≥ 0) and satisfies the stationary condition as less than 1 (α1 + β1 < 1). The high value of α,
β implies more volatile behavior due to shock. Return series is negatively skewed and value of kurtosis is small, so
the distribution is platykurtic. Further, value of the Jarque-Bera is still high showing that the distribution is not normal.
Diagnostic test shows little serial correlation in standardized residuals for lag 10, 20, 50 and no/little volatility
clustering was found in square standardized residuals. After applying GARCH (1, 1) model poultry series show no
ARCH effect.
The above analysis shows that in Pakistan there is significantly no price transmission between beef and poultry,
further the negative sign of coefficients for both beef and poultry indicates that these commodities are substitutes
based on price, which is similar to the finding of Goodwin and Holt (1999). Goodwin (2006) is of the view that
analysis based on only prices is not enough to evaluate the price transmission between substitutes and a deep analysis
of the market and its institutions is mandatory. Additionally, the utility from the consumption of one commodity is
assumed to be different from the consumption of other substitute commodity (Swamy and Binswanger, 1983;
Cornelisse et al., 1987; Kurosaki & Fafchamps, 2002; Atkin, 2013). This is obvious that due to difference in
consumption habits, the utility received from beef is different from the utility received from poultry for an individual
in case of Pakistan (Cornelisse et al., 1987; Atkin, 2013).

5. Conclusion

In this study, the volatility of food and agricultural commodities is investigated through GARCH models. For this
purpose monthly data on twelve food and agricultural commodities and factors effecting volatility is taken from April,
1983 to April, 2013 for a total of 360 monthly observations.
Like other financial time series data, agricultural and food prices are also characterized with heteroscedasticity.
Therefore, to deal with the problem of heteroscedasticity, this research study used GARCH-type model introduced by
Engle (1982) to investigate the type and the rationale of volatility regarding the price of agricultural commodities over
time in Pakistan.
Food and agricultural commodities price return series are stochastic in nature with no fixed pattern and historical
evidences are therefore, not enough to predict the future prices. However, if the GARCH models are correctly
specified, its implications are very strong and convenient. Therefore, conditional GARCH models is applied on time
series date of prices in food and agricultural commodities and on explanatory variables.
The graphical analysis shows that the series selected for spillover analysis are not normally distributed, and ARCH
effect is visible in each variable. Volatility spillover effect for substitute goods was estimated for two pair of
commodities i.e. rice and wheat, beef and poultry. Again, GARCH (1, 1), EGARCH (1, 1) and GJR-GARCH (1, 1)
models are estimated and best fitted model is selected on the bases of diagnostic tests. Analysis of this study shows
prices of rice do not affect the mean prices of rice and vice versa. Results conclude that volatility of rice do not affect

6
See McCorriston et al. (2004) for details.
 A. Ismail

the volatility of wheat series, but the volatility of wheat significantly affect the volatility of rice price series, but the
positively of the rice coefficient shows that these commodities are not substitutes. Whereas, price of poultry do not
affect the mean prices of beef and vice versa. Additionally, volatility of poultry market is not transmitted to beef
market and vice versa.
This study motivates for further research in volatility spillover effect for more substitute for crude oil like natural
gas, which is used as an alternate fuel in vehicles and its impact on the food and agricultural prices should also be
investigated. Further volatility in the ghee prices can also be analyzed as a substitute for edible oils.
Second, the crucial behavior in return prices of poultry is to be deeply investigated through seasonal GARCH
models. Additionally, the volatility of inputs for poultry industry needs to be analyzed, which will help to understand
the cause of highly volatile prices in poultry sector of Pakistan.

Reference

Abbott, P. C., Hurt, C., and Tyner, W. E. (2008), “What's driving food prices?”, Farm Foundation Issue Report.
Agnolucci, P. (2009), “Volatility in crude oil futures: A comparison of the predictive ability of GARCH and implied
volatility models”, Energy Economics, 31, 316–321.
Alom, F., Ward, B., and Hu, B. (2011), “Spillover effects of World oil prices on food prices: evidence for Asia and
Pacific countries”, In Proceedings of the 52nd Annual Conference New Zealand Association of Economists, 29.
Aloui, C. (2007), “Price and volatility spillovers between exchange rates and stock indexes for the pre- and post-euro
period.” Quantitative Finance, 7(6): 669-85.
Apergis, N., and Rezitis, A. (2003), “Agricultural price volatility spillover effects: the case of Greece”, European
Review of Agricultural Economics, 30(3), 389-406.
Arouri, M., Jouini, J., Nguyen, D.K. (2011), “Volatility spillovers between oil prices and stock sector returns:
implications for portfolio management.” Journal of International Money and Finance, 30, 1387-1405.
Atkin, D. (2013), “Trade, tastes, and nutrition in India”, The America Economic Review, 103(5), 1629-1663.
Babula, R. A., and Somwaru, A. (1992), “Dynamic impacts of a shock in crude oil price on agricultural chemical and
fertilizer prices”, Agribusiness, 8(3), 243-252.
Baffes, J. (2007), “Oil spills on other commodities”, Resources Policy, 32(3), 126-134.
Balcombe K. (2009), “The nature and determinants of volatility in agricultural prices: An empirical study from 1962-
2008”, Proceedings from FAO Workshop. November, 2009.
Bantilan, M.C.S., and Davis, J.S. 1991, “Across-commodity spillover effects of research and opportunity costs in a
multiproduct production environment”, ACIAR/ISNAR Project Paper Series no. 30. Canberra, Australia:
Australian Centre for International Agricultural Research and The Hague: International Service for National
Agricultural Research.
Bollerslev T. (1986), “Generalized autoregressive conditional heteroscedasticity”, Journal of Econometrics. 31. p.
307-327.
Bozdogan, H. (2000), “Akaike's information criterion and recent developments in information complexity.” Journal
of Mathematical Psychology, 44(1), 62-91.
Broyden, C. G. (1970), “The convergence of a class of double-rank minimization algorithms 1 general
considerations”, IMA Journal of Applied Mathematics, 6(1), 76-90.
Bubak, V., Kocenda and F. Zikes. (2011), “Volatility transmission in emerging european foreign exchange markets.”
Journal of Banking and Finance, 35(11): 2829-41.
Buguk, C., Hudson, D., and Hanson, T. (2003), “Price volatility spillover in agricultural markets: An examination of
US catfish markets”, Journal of Agricultural and Resource Economics, 86-99.
Capps O. (1986), “Changes in domestic demand for food: impacts on Southern agriculture”, Southern Journal of
Agricultural Economics, 18(01).
Chatrath, A., Adrangi, B., Dhanda, K. (2002), “Are commodity prices chaotic?” Agricultural Economics, 27, 123–
137.
Choudhry, T. (1996), “Stock market volatility and the crash of 1987: evidence from six emerging markets.” Journal
of International Money and Finance, 15(6), 969-981.
Cooke, B. and Robles, M. (2009), “Recent food price movements: a time series analysis”, IFPRI Discussion Paper.
Cornelisse, P. A., Kuijpers, B., and Salam, A. (1987), “A Policy Model of the Wheat and Rice Economy of
Pakistan”, The Pakistan Development Review, 385-400.
Dahl, C.M., Iglesias, E.M. (2009), "Volatility spill-overs in commodity spot prices: new empirical results”, Economic
Modelling, 26, 601-607.
Dehn, Gilbert, L., and P.Varangis. (2005), “Agricultural commodity price volatility”,

21
Spillover Effect between Food and Agricultural Commodities…

Dehn, J. (2000), “Commodity price uncertainty in developing countries”, World Bank Policy Research Working
Paper, (2426).
Du, X., Yu, C.L., Hayes and D.J. (2011), “Speculation and volatility spillover in the crude oil and agricultural
commodity markets: a Bayesian analysis”, Energy Economics, 33 (3), 497–503
Elmarzougui, E., and Larue, B. (2013), “On the evolving relationship between corn and oil prices”, Agribusiness.
Engle, R (1982), “Autoregressive Conditional Heteroskedasticity with estimates of the variance of United Kingdom
inflation”, Econometrica 5 987-1008
Engle, R. F., Lilien, D. M., and Robins, R. P. (1987), “Estimating time varying risk premia in the term structure: the
ARCH-M model”, Econometrica: Journal of the Econometric Society, 391-407.
Ghaith, Z. and Awad, I. (2011), “Examining the long term relationship between crude oil and food commodity prices:
Co-integration and causality”, International Journal of Economics and Management Sciences, 1(5), 62-72.
Gilbert, C. L. (2006), “Trends and volatility in agricultural commodity prices”, In D. A. Sarris, & D. Hallam,
Agricultural Comodity Market and Trade: New Approaches to Analyzing Market Structure and Instability. Edward
Elgar Publishing limited.
Gilbert, C.L. and Morgan, C.W., (2010a), “Has food price volatility risen?” Pdf Available at:
<http://ideas.repec.org/p/trn/utwpde/1002.html> Accessed on April 2010.
Glosten, L. R., Ravi J. and David E. R. (1993). “On the relationship between the expected value and the volatility of
nominal excess returns on stocks”, Journal of finance. 48:5, pp. 1779-801
Goodwin, B. K. (2006, April). Spatial and vertical price transmission in meat markets. In Market Integration and
Vertical and Spatial Price Transmission in Agricultural Markets Workshop. University of Kentucky, Lexington,
April (Vol. 21).
Goodwin, B. K. and M. T. Holt. (1999), “Asymmetric adjustment and price transmission in the U.S. beef sector,”
American Journal of Agricultural Economics, 79: 630-637.
Hamao, Y., Masulis, R. W., and Ng, V. (1990), “Correlations in price changes and volatility across international stock
markets”, Review of Financial studies, 3(2), 281-307.
Huchet-Bourdon, M. (2011), “Agricultural commodity price volatility:An overview”, OECD Food, Agriculture and
Fisheries Working Papers No.52 .
Khiyavi, P. K., Moghaddasi, R., Eskandarpur, B., & Mousavi, N. (2012), “Spillover effects of agricultural products
price volatilities in Iran (Case Study: Poultry Market)”, Journal of Basic and Applied Scientific Research, 2(8),
7906-7914.
Kim, J. and H. Doucouliagos. (2008), “Realized volatility and correlation in grain futures markets: Testing for
spillover effects.” Review of Futures Markets, 17(3): 275-300.
Ljung, G. and Box, G. (1978), “On a measure of lack of fit in time series models.” Biometrika, 65(2), 297-303.
Malliaris, A., Urrutia, J. (1996), “Linkages between agricultural commodity futures contracts.” Journal of Futures
Markets, 16(5), 595-609.
McCorriston, S., Morgan, W., and Rayner, T. (2004). Price transmission in imperfectly competitive vertical markets.
University of Nottingham, School of Economics.
McGuirk, A. and Mundlak, Y. (1991). “Incentives and constraints in the transformation of Punjab agriculture.” Intl
Food Policy Res Inst, 87.
Mitchell, D. (2008), “A note on rising food price”, Policy Research Working Paper .
Nelson, D. B. (1991), “Conditional heteroskedasticity in asset returns: A new approach”, Econometrica 59: 347-370.
Rezitis, A. (2003), “Mean and volatility spillover effects in Greek producer-consumer meat prices”, Applied
Economics Letters, 10(6), 381-384.
Roache, S. K. (2010), “What explains the rise in food price volatility?”, IMF Working Paper , pp. 1-29.
Serra, T., Zilberman and J. Gil. (2011), “Price volatility in ethanol markets.” European Review of Agricultural
Economics, 38(2): 259–280.
Showell, B. A., Williams, J. R., Duvall, M., Howe, J. C., Patterson, K. Y., Roseland, J. M., & Holden, J. M. (2012).
USDA Table of Cooking Yields for Meat and Poultry.
Skintzi, V. and A. Refenes. (2006), “Volatility spillovers and dynamic correlation in european bond markets.” Journal
of International Financial Markets, Institutions and Money, 16(1): 23-40.
Sumner, D. A. (2009), “Recent commodity price movements in historical perspective”, American Journal of
Agricultural Economics. , 91 (5), 1250-1256.
Swamy, G., and H.P. Binswanger (1983), “Flexible consumer demand systems and linear estimation: food in India,”
American Journal of Agricultural Economics, 65, 675-684.
Tse, Y. K. (2002), “Residual‐based diagnostics for conditional heteroscedasticity models”, The Econometrics
Journal, 5(2), 358-374.
 A. Ismail

Kurosaki, T. and Fafchamps, M. (2002), “Insurance market efficiency and crop choices in Pakistan”, Journal of
Development Economics, 67(2), 419-453.
Wu, F., Z. Guan and R. Myers. (2010), “Volatility spillover effects and cross hedging in corn and crude oil features.”
The Journal of Futures Markets, 31(11): 1052–1075.
Zhang, Z., Lohr, L., Escalante, C and Wetzstein, M. (2010), “Ethanol, corn, and soybean price relations in a volatile
vehicle-fuels market”, Energies, 2 (2), 320-339

23