A partnership with academia

Building knowledge for trade and development

Vi Digital Library - Text Preview

Unraveling the Underlying Causes of Price Volatility in World Coffee and Cocoa Commodity Markets

Discussion paper by Maurice, Noemie Eliana, Davis, Junior / UNCTAD, 2011

Download original document (English)

In recent years, Commodity Dependent Developing Countries (CDDCs) have faced multiple global food, energy and climate crises, compounded by the recent financial and economic crises, which have increased their vulnerability to excessive price volatility in commodity markets. Moreover, structural vulnerabilities in most CDDCs render their economies more vulnerable to increased commodity market turbulence than developed countries, given their comparatively lower income and high dependence on commodity exports. Therefore, this paper empirically examines the patterns and underlying causes of excessive price volatility for two major soft commodities of critical importance to many of the poorest CDDCs: coffee and cocoa. It aims to identify interactions, similarities and causalities between coffee and cocoa prices on the one hand and, oil and futures prices on the other hand. Our analysis of coffee and cocoa historical prices shows that, coffee price volatility has uneven or differing reactions depending on the nature of the market shock. Oil price spillover effects on coffee and cocoa markets are also assessed using cointegration and causality models. Long-run causality is found between oil prices, and coffee and cocoa prices but, only cocoa has an equilibrium relationship with oil in the longterm. Given the results, this study proposes some policy recommendations for managing price risk and addressing regulation in cocoa and coffee exporting countries


1


Unraveling the
underlying causes
of price volatility in
world coffee and
cocoa commodity
markets
UNCTAD Special Unit on
Commodities working paper series
on commodities and development

September 2011


Noemie Eliana Maurice and Junior
Davis


Discussion paper 1


The papers in this series are the preliminary results of
research undertaken by SUC staff members and do not
necessarily reflect the views of UNCTAD. They are
published in this form to stimulate discussion and
comment on work that is generally still in progress.





2


Unravelling the underlying causes of price volatility in world coffee


and cocoa commodity markets




Noemie Eliana Maurice1 and Junior Davis2






Abstract:
In recent years, Commodity Dependent Developing Countries (CDDCs) have faced
multiple global food, energy and climate crises, compounded by the recent financial
and economic crises, which have increased their vulnerability to excessive price
volatility in commodity markets. Moreover, structural vulnerabilities in most CDDCs
render their economies more vulnerable to increased commodity market turbulence
than developed countries, given their comparatively lower income and high
dependence on commodity exports. Therefore, this paper empirically examines the
patterns and underlying causes of excessive price volatility for two major soft
commodities of critical importance to many of the poorest CDDCs: coffee and cocoa.
It aims to identify interactions, similarities and causalities between coffee and cocoa
prices on the one hand and, oil and futures prices on the other hand. Our analysis of
coffee and cocoa historical prices shows that, coffee price volatility has uneven or
differing reactions depending on the nature of the market shock. Oil price spillover
effects on coffee and cocoa markets are also assessed using cointegration and
causality models. Long-run causality is found between oil prices, and coffee and
cocoa prices but, only cocoa has an equilibrium relationship with oil in the long-
term. Given the results, this study proposes some policy recommendations for
managing price risk and addressing regulation in cocoa and coffee exporting
countries.

Keywords: Commodity price volatility, financialization, error correction modelling,
cointegration theory, commodity dependent developing countries, least developed
countries.

JEL classification: E30; F24; O11



1 Erasmus School of Economics, Erasmus University, Rotterdam, The Netherlands. E-mail:
noemiee.maurice@gmail.com
2 UNCTAD, Special Unit on Commodities United Nations. E-mail: junior.davis @unctad.org. The
authors would like to thank Ms. Laurent (SUC) for assistance with statistical data collection.





3


Contents


1 Introduction 4


2 Overview of the world coffee and cocoa markets 5
2.1 Commodity price volatility 7


3 Modelling coffee and cocoa price volatility 10


4 Impact of oil spillover effects and speculation on coffee and cocoa prices 16
4.1 Cross commodity causality: Oil vs. Coffee and Cocoa 17


4.2 Cointegration models and results: the effect of speculation 19


5 Policy recommendations and conclusions 21


6 Terms & Acronyms 28


7 Annexes 29




Figures
Figure 1 Coefficients of variation for commodities in the short- and long run 9
Figure 2 Percentage variations in Prices, consumption and production of (A) coffee and (B)
cocoa 16


Tables
Table 1 Specification for commodity prices 11
Table 2 Granger-causality tests results 18





4




1 Introduction

Since 2000, Commodity Dependent Developing Countries (CDDCs) have faced
multiple global food, energy and climate crises, compounded by the recent financial
and economic crises which have increased their vulnerability to excessive price
volatility3 in commodity markets. Moreover, structural vulnerabilities in most
CDDCs render their economies more vulnerable to increased commodity market
turbulence than developed countries, given their comparatively lower income and
high dependence on commodity exports. The World Bank estimates that 119 million
more people have been pushed into hunger as a result of the 2008 food crisis. There
are now an estimated 1.02 billion malnourished people worldwide (World Bank
2009).

Meanwhile, the FAO estimates that more than 75 million people were driven into
hunger between 2006 and 2010 (FAO 2011). LDCs and CDDCs were particularly
harmed by this crisis. Indeed, in most of the LDCs, consumed food is not processed
or at least, less than in the developed countries. Therefore, following the 2007-2008
food prices crisis, the affordability of food products became more worrying in
developing countries than in the developed ones. Another reason why the LDCs
were particularly affected by the food crisis is because they spend a larger share of
their income on food. Some low income countries spend up to 70-80 per cent of
their income on food (UNCTAD 2009).

Although supply and demand fundamentals played a significant role in the food
crisis outbreak, many other factors contributed to the economic turmoil. For
example, large increases in oil prices contributed to rising production costs and
drove food prices higher. Besides, the World Bank estimates that weakness of the
dollar accounted for 15% of the food price increases between 2002 and 2008
(Mitchell 2008). Additionally, over the last decade, weather events such as drought
in Russia, freezes in Brazil and, heavy rains in Canada and in Australia caused major



3 Volatility is a statistical measure of the tendency of an asset's price to vary over time. It is usually
captured in the standard deviation or variance.





5


disruptions in the agricultural commodities production i.e. grains, and tropical
foodstuff. Global warming also proves partly responsible for livestock and crops'
diseases thereby, threatening food security and exacerbating food supply problems.
Price fluctuations are inherent in agricultural markets – partly due to the supply-
demand dynamics and the unpredictability of weather patterns and harvest yields.

There are debates as to the extent to which activity in futures trades and over the
counter markets (OTC) for agricultural commodities impact on this volatility.
Whatever the cause, extreme volatility in food prices deters producers from making
the necessary investments for increasing productivity and production: this is one of
the underlying causes of continued worldwide food insecurity.

This study intends to explore the gravity of the commodity trade and development
problematique vis-à-vis high food, energy prices and volatile markets for the world’s
most vulnerable CDDCs. It aims to empirically explore underlying price behavior
and volatility in the coffee and cocoa markets, and also to identify interactions,
similarities and causalities between coffee and cocoa prices on the one hand and, oil
and futures prices on the other hand. This study will first provide an overview of the
world coffee and cocoa markets. Next, we introduce the data employed for use in the
empirical analyses. The Generalized Autoregressive Conditional Heteroskedasticity
(GARCH) models for Arabica, Robusta and cocoa are then estimated and interpreted.
We then empirically consider the price-effects of both energy and financial products
using Granger-causality and cointegration methods to explore potential long-term
trend similarities. Last, we consider the empirical results to formulate a few policy
recommendations aimed at reducing risks associated with price volatility in CDDCs.

2 Overview of the world coffee and cocoa markets

Coffee and cocoa are both tropical commodities mainly produced in CDDCs and that
have experienced extreme variability in their prices over the last 40 years. In fact,
coffee and cocoa price variations have proven very large compared to grains or
meat. This study will differentiate between Arabica and Robusta coffee as they are
different varieties of coffee and traded on separate exchange markets. Coffee and
cocoa have similar long-run price trends (Graph 2- Annexes). Moreover, to the
production of both commodities is mostly located in LDCs and developing countries





6


in Africa, South America and South Asia (Annexes- Doc1). Thus, coffee and cocoa
price volatility is of acute economic importance for CDDCs whereas the tea trade for
instance has no major impact on its main producers' (China and India) trade
balances. As coffee and cocoa are the two major export crops of the Sub-Saharan
African region (SSA), they represent a major source of income for many LDCs or
developing countries that have strong commodity-export dependence. For instance,
cocoa crop exports provide a livelihood for 25 per cent of the Cote d'Ivoire's
population (FAO 2006) while, the share of coffee in total exports represents 79 per
cent in Burundi and 64 per cent in Ethiopia (FAO 2006). For coffee and cocoa
exporting CDDCs, price volatility is a major cause of concern while it is a relatively
minor concern for most importing countries. For the former, significant fluctuations
in world prices may have dramatic effects both at the national and producer levels
as extreme volatility in prices deters producers from making the necessary
investments for increasing productivity and production. For most importing
countries, changes in coffee or cocoa prices would probably only result in relatively
minor changes in consumption habits.

Involving over fifty producing countries, of which thirty are importers, coffee is one
of the most widely traded commodities. Coffee is a perennial crop that is an
agricultural commodity produced from the same root structure for two or more
years. It is also noteworthy that coffee is a seasonal crop; seasons vary from country
to country which makes supply for the most part unpredictable. For many
developing country governments, and the private sector coffee production, trade
and consumption is a critical contributor to socio-economic development.

The International Coffee Organization (ICO) is the main intergovernmental
organization in charge of collecting and sharing information on coffee and of
establishing international cooperation in the coffee sector. In 1882, with its entry
into the Coffee Exchange of New York (later part of the Coffee, Sugar and Cocoa
Exchange), coffee prices became more volatile. The Intercontinental Exchange (ICE)
which is part of the New York Board of Trade (NYBOT) governs the world Arabica
price through Futures U.S. Coffee "C" contracts while Robusta coffee has been traded
for over twenty years on the London International Financial Futures Exchange
(LIFFE).





7



Cocoa, although produced and exported in smaller volumes, has many similarities
with coffee. Ninety per cent of the cocoa producing countries also produce coffee
(Annexes - Doc 1). While primarily consumed in Organization for Economic
Cooperation and Development (OECD) countries, cocoa is exclusively produced in
developing countries; which makes cocoa price volatility an important issue for
CDDCs. Cocoa harvests and thus productivity levels are highly dependent on
prevalent weather conditions. The mandates of the International Cocoa
Organization (ICCO) focus on enhancing the economic, social and environmental
sustainability of the world cocoa economy. Since 1925, cocoa has been traded on the
New York Cocoa Exchange before joining the Coffee, Cocoa and Sugar Exchange and
later the ICE, as part of NYBOT. Cocoa futures contracts is primarily traded and
denominated in UK pounds.

2.1 Commodity price volatility

Commodity prices have shown considerable volatility over the past decade.4 The
price boom between 2002 and 2008 was the most pronounced in several decades –
in magnitude, duration and breadth. Moreover, the price decline following the onset
of the recent global crisis in mid-2008 stands out both for its sharpness and for the
number of commodities affected. Since mid-2009, and especially since the summer
of 2010, global commodity prices have been rising again rapidly (excepting some
temporary setbacks in the second quarter of 2011).

There are many explanations for the apparent volatility in commodity markets,
including the so-called financialization of commodities as an asset class. The high
prices across a broad range of commodities -- and the potential diversification
benefits of a wide array of investment opportunities -- has attracted speculative
investors (e.g. hedge funds, commodity index and exchange-traded-funds) into
commodity markets. Between 2003 and 2008, speculative investment in commodity
indexes was estimated to have increased from $15 billion to around $200 billion.



4 Price volatility is a measure of price variation from one period to the next.





8


The issue of commodity price development and the financialization of commodity
trading is discussed further in Annex 1.

Long-term comparisons show that recent price volatility is not unprecedented for
individual commodities.5 For example, oil price volatility in 2008, while remarkable,
remained well below its spike of the early 1970s. Examining the short-term constant
prices provides a better insight with regard to recent food price developments. The
chart below presents the coefficients of variation (CV) for various food commodities
and oil.


µ
σ


=CV
(1)



The CV (1) connects the standard deviation (σ ) to the mean ( µ ) so that the context
of the mean of the data is considered allowing for cross-commodity comparisons. CV
is a basic measure of price dispersion; it serves to compare the degree of variability
from one data series to another.
Figure 1 shows the long-term volatility of commodities prices using monthly
nominal prices for 6 commodities over the period 1960-2011 and indicates that the
recent price fluctuations are not extraordinary for specific commodities (Calvo-
Gonzales, Shankar and Trezzi, 2010). The volatility of coffee prices was similar to
that of most agricultural products in the past. Petroleum and sugar prices were the
most volatile during the period 1960-1989. Indeed, price volatility in 2000-2011
fell with respect to its long-term average for 85% of the commodities, and rose for
15% of the commodities.

However, it should be noted that the volatility estimates in this paper do not take
into account trends which could be important in the context of a commodity super
cycle,6 as for example in the case of real metals prices. Moreover, the high speed and



5 Jacks DS, O’Rourke KH and Williamson JG (2011). Commodity Price Volatility and World Market
Integration since 1700. Review of Economics and Statistics, (forthcoming) and Calvo-Gonzales O,
Shankar R and Trezzi R (2010). Are Commodity Prices More Volatile Now? A Long-Run Perspective.
Policy Research Working Paper No. 5460, World Bank, Washington, DC, October.
6 John T. Cuddington and Daniel Jerrett (2008). Super Cycles in Real Metals Prices? IMF Staff Papers,
Vol. 55, No. 4. International Monetary Fund.





9


amplitude of recent price swings for a broad range of commodities clearly
distinguishes them from earlier ones (Baffes and Haniotis, 2010).7 More specifically,
the magnitude of the most recent upswing of food and metals prices was above the
historical average, while the magnitude of the price rebound for oil was similar to
historical averages, but occurred at a higher speed.

The volatility of coffee prices was similar to that of most agricultural products over
the past 50 years. Petroleum and sugar prices were the most volatile during the
period 1960-2010. However, it should be noted that the volatility estimates below
do not take into account trends which could be important in the context of a
commodity super cycle, as for example in the case of real metals prices (Cuddington
and Jerret, 2008).

Figure 1 Coefficients of variation for commodities in the short- and long run



Source: Unctadstat (2011).

The coefficient of variation is very sensitive to outliers hence; for example, the large
amplitude of price swings that occurred during the 1979 financial crisis8 for a broad



7 Baffes J and Haniotis T (2010). Placing the 2006/08 commodity price boom into perspective. Policy
Research Working Paper No. 5371, World Bank, Washington, DC, July.
8 The financial crisis of 1979-1981 had many similarities to the recent global financial crisis of 2009-
2010. For example, the US dollar was falling, inflation in the USA was approaching 13% and a high
level of unemployment at 13% was exacerbated by a concomitant energy crisis in 1979 which let to





10


range of commodities biases the indicator. Although the CV does not reach its 1980
historical record, most of the commodities' volatility has significantly risen over the
last decade. We explore these issues empirically in sections 3 and 4 of the paper.


3 Modelling coffee and cocoa price volatility

In this paper on coffee and cocoa price volatility and GARCH-type models, the
sample size consists of 249 observations. We use logarithmic transformations of
monthly constant prices of Arabica and Robusta from January 1990 to September
2010 (12 months*20 years+9 months= 249 months)9. For the second part of the
study, we use the logarithms of monthly current prices for Arabica, Robusta, cocoa
and oil. Daily futures prices of Arabica, Robusta and cocoa were collected from
Bloomberg. Monthly averages were computed in order to conduct a causality
analysis. Cocoa futures prices are extracted from the London International Financial
Futures and Options Exchange (LIFFE) and therefore are converted from UK (£)
pounds sterling to US dollars using the monthly average of the Bank of England’s
spot exchange rate statistics.

Table 1 lists the commodity price series, sources and units of measurement utilized
in this paper. The deflator that is used to compute constant prices from current
price ( 100*/tan MUVCurrenttCons = ) is the UN Unit Value Index of Manufactured
(MUV) goods exports.



rapidly escalating energy food prices. On commodity markets, precious metals again became a safe
haven for investors with gold reaching $850 and silver $50 an ounce.
9 The 1990-2010 period corresponds to the free market period on commodity markets.





11



Table 1 Specification for commodity prices


Source: ICO, ICCO Bloomberg, the World Bank

Food price variations are often large and unpredictable. Greater price
unpredictability and uncertainty about future developments, often leads to higher
price risks being borne by producers, exporters, importers and stock holders who
are then very likely to review their investment decisions. To reduce disruptions in
both coffee and cocoa markets will require an empirically accurate measure of
volatility that takes into account specifications relative to each commodity and
allows the prediction of future price developments. ARCH and GARCH processes
defined as "mean zero, serially uncorrelated processes with non-constant variances
that are conditioned on past information" (Aradhyula and Ho, 1988) are useful
economic analysis tools with strong forecasting accuracy.

GARCH models use past prices to model and forecast conditional variances. They
also allow a wide range of possible specifications to both model volatility and
examine volatility persistence and asymmetry in coffee prices over time. Any
GARCH model assumes that prices have a time-varying (non-constant) variance
which means that in some periods, markets are more volatile than in others. The
objective of this section of the paper is to characterize the conditional variance of


Commodities Period (mm/yyyy) Price Specifications Source Unit
Arabica (A) 01/1990 - 09/2010 Monthly average ICO USc/kg
constant prices
Robusta (R) 01/1990 - 09/2010 Monthly average ICO USc/kg
constant prices
Cocoa (C) 01/1990 - 09/2010 Monthly average ICCO USc/kg
constant prices
Arabica (A) 01/1990 - 04/2011 Monthly average ICO USc/kg
current prices
Robusta (R) 01/1990 - 04/2011 Monthly average ICO USc/kg
current prices
Cocoa (C) 01/1990 - 04/2011 Monthly average ICCO USc/kg
current prices
Petroleum Crude 01/1990 - 04/2011 Monthly average prices Bloomberg $/bbl
Of Brent, Dubai and World Bank
West Texas
(A) futures prices 01/1990 - 04/2011 Daily current prices Bloomberg US$/lb
(R) futures prices 11/1991 - 01/2009 Daily current prices Bloomberg US$/MT
(C) futures prices 01/1990 - 04/2011 Daily current prices Bloomberg GBP/MT





12


price series of Arabica, Robusta and cocoa. Let us assume that the Arabica prices
series AtP 10 are generated by the autoregressive process:


t


p


i


A
ti


A
t PcP εφ ++= ∑


=




1
1 (4.1)



While the conditional variance is presented in a GARCH (1, 1) model with a constant,
past information about volatility ( 2 1−tε ) and past forecast variance ( 2 1−th ):


),0(~1 ttt hN−Ωε
2


1
2


1
2


−−


++= ttt hh βαεδ (4.2)

The conditional variance 2th of the information set available at time t-1


1−Ωt considers varying confidence intervals of volatility. Table 4 (Annexes) contains
univariate GARCH (1, 1) parameters for the mean and the variance equations of
both coffees and cocoa. The preferred regression has the AR order p and the moving
average (MA) order q that minimize the Schwarz information criterion (SIC). In
addition, regressions are estimated using a range of {1; 5} for p and {0; 5} for q and
the combination of p and q with the lowest SIC is the preferred model. The Arabica
results show that AR(1) is the specification that maximizes the quality of the fit.
Robusta on the other hand is best approximated with the model ARMA(1,1) and,
both the AR and the MA coefficients are significantly different from 0. Finally cocoa
is better approximated by an AR(1) model. All the coefficients in table 4 of the
Annexes are significant and the regressions show a high adjusted R-squared,
meaning that the estimated parameters of the conditional mean have a strong
explanatory power of historical price movements. Given the high adjusted R-
squared, it would seem that GARCH models perform well at modelling conditional
variance. Nonetheless, this is no guarantee that the GARCH process is a statistically
valid improvement over the AR(MA) process (Aradhyula and Holt, 1988). It is thus



10 RtP stands for Robusta price and CtP for cocoa price





13


relevant to test the GARCH hypothesis that the conditional variances are in fact, not
constant using the following hypothesis:


0,0:0 == βαH
00:1 ≠≠ βα orH



A Wald test of the joint significance of α and β is conducted for the three
commodities in Table 5 (Annexes). The statistics used in a Wald test is the Chi-
squared; if the p-value of the chi-squared exceeds the significance level (0.05) the
null hypothesis of stationarity in the volatility cannot be rejected. Results indicate
that p-values of the Chi-squared distributions of Arabica, Robusta and Cocoa are all
equal to 0, thus, we reject the null hypothesis of stationarity in the conditional
forecast variances; GARCH is an improvement over the AR process for the three
tropical commodities.

From our GARCH analysis, it is possible to infer that shocks in prices are reflected in
volatility, but one might also consider how changes in variability evolve when
shocks are positive or negative. Such a distinction may be modelled with
econometric tools and, by adding precision to the model, provide a better
forecasting tool. Understanding volatility in response to positive or negative shocks
is crucial for CDDC producers so they can predict future volatility in commodity
prices with more accuracy and thus, improve the estimation of future revenue
streams. We do this by introducing symmetry or leverage effects in the variance to
GARCH models. The Exponential Generalized Autoregressive Conditional
Heteroskedasticity (EGARCH) is the most widely used of these models to estimate
the logarithm of conditional variance in order to determine whether or not the
observed volatility reacts asymmetrically to good and bad news. Good news in the
case of a commodity might be favourable weather forecasts for coffee and cocoa
crops or policies that promote agricultural development and growth; whilst bad
news may for example be a natural disaster or calamitous weather event (hurricane,
tornado, flooding etc) or sharp rises in oil prices for instance. Nelson (1991) and
Schwert (1989) maintain that stock volatility is higher during recessions and
financial crisis. In order to assess this for cocoa and coffee we estimate the following
EGARCH:





14




)log()log( 2 12
1


1
22


1


1
1


2








− +++= t
t


t


t


t
t hhh


h βεπεπδ (4.3)



In this model the effects of residuals is exponential and not quadratic. The
asymmetry is measured by the coefficient 2π ; if it is negative and significant, as for
many financial assets, there is positive asymmetry and negative price shocks have a
stronger impact on price volatility than positive shocks. The impact of positive
shocks (good news) is measured by 2 121 )( −+ thππ whereas the impact of negative


shocks is captured by 2 121 )( −− thππ . The hypothesis tested by the EGARCH model
is the following:


0: 20 =πH


0: 20 ≠πH



The results in Table 7 (see Annex) show the EGARCH preferred regressions for
cocoa, Arabica and Robusta with regard to the SIC. Results show that none of the
asymmetric 2π coefficients is negative and, only 2π for cocoa is approximately equal
to zero ( 2π =0.035) meaning that, positive and negative shocks have approximately
the same impact on its volatility. In addition, the GARCH (1, 1) model has a smaller
SIC than the EGARCH model and thus, cocoa volatility is better approximated with
the asymmetry specification. On the other hand, the asymmetry coefficients for
arabica and robusta are large and significant: for arabica, 422.02 =π , and for
Robusta 351.02 =π and, both p-values are equal to zero. The SIC indicates that the
EGARCH describes the volatility in world coffee prices better than the GARCH (1, 1).
Positive shocks have a more prominent effect on the observed volatility than
negative shocks.

An empirical examination of the varying volatility of coffees and cocoa allows us to
estimate the best fit for the modelling of these three commodities. For cocoa, prices
follow an autoregressive process of order one AR(1) and its conditional variance is a
GARCH (1,1) process. Arabica and robusta prices follow an ARMA model of order





15


p=4 q=2 for arabica and p=1 q=1 for robusta. Both coffees conditional variances are
better estimated with the EGARCH model.

Although the price correlations between the three commodities is very high (0.8 in
the long-run) (see Annexes- Table 2), specificities in terms of their price volatility
are less obvious and requires more complex models.

Volatility, expressed by the conditional variance of the price series, is modelled with
different features for arabica, robusta and cocoa, and suggests that there may be
persistence in volatilities and that price series are best estimated with a varying
variance. We find different results for each of the three tropical commodities. The
price model AR(1) is used for the cocoa price series, robusta's prices are modelled
with ARMA(1,1) process and, Arabica prices follow a ARMA(4,2) process. The
conditional variance definition follows an EGARCH process with similar coefficients
and a positive and significant 2π for both coffees, which suggests that, their
volatility is more affected by positive shocks in prices than by negative price shocks.
Moreover, a large increase in oil prices (listed as a negative shock) will have a lower
impact on coffee price variability than a steep decline in oil prices (positive shock)
of a similar magnitude. Cocoa, on the other hand does not show any asymmetric
pattern in its varying volatility. Thus, in a world of high oil prices, coffee price
volatility is not as excessive as in a context of low oil prices; whilst cocoa price
volatility is largely unchanged.





16



4 Impact of oil spillover effects and speculation on coffee and cocoa


prices

This section addresses two of the main underlying causes of coffee and cocoa price
volatilities. Logically, changes in commodity prices result from changes in their
fundamentals namely, supply and demand. Graph 6 (see Annexes) shows that for
non-essential goods, variation in fundamentals do not necessarily reflect the extent
of the price surges that have occurred over the past 20 years (see Figure 2).

Figure 2 Percentage variations in Prices, consumption and production of (A)
coffee and (B) cocoa



Source: ICO, ICCO accessed July 2011.

One of the reasons for the detachment between production and price in commodity
markets may be explained by the Separation theorem according to which "when a
future market exists, the optimum production of the firm does not depend upon the
(subjective) distribution of the random price nor upon the firm's attitude toward
risk" (Broll and Zilcha, 1992). Thus whenever a futures market is available, the price
and production of the commodity may grow independently. Therefore, we do not
dwell upon an empirical analysis of the fundamentals for coffee and cocoa, but
rather focus on two external drivers of these commodity prices namely, the energy
sector represented by crude oil prices and the financial sector which is reflected by
futures prices. In this section, all the commodity prices are denominated in current
dollar prices as only current prices are traded in the financial markets. However,
constant dollar prices provide a better fit for estimating historical volatility.


B.


-40%


-20%


0%


20%


40%


60%


80%


1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010


Price Demand Supply


A.


-40%


-20%


0%


20%


40%


60%


80%


100%


120%


140%


1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010


Price Demand Supply





17


Barnard (1983) highlighted the potential for fuels to be disruptive to agricultural
commodity prices. Activities such as: planting, the application of fertilizer,
harvesting, storage and transportation require an important amount of diverse
fuels; the most usual being crude oil, coal, gas, and more recently biofuels. Also, it
has been argued that the prices of both coffees and cocoa are influenced by oil prices
(Baffes J. 2007), and that current prices have been volatile in recent years hence
providing traders with significant “trend-following opportunities” (ICE 2011). We
utilize Granger-causality tests to assess the long-term causality links between oil
and commodities prices while cointegration methods are used to assess the long-
run relationship between cash and futures prices of cocoa and coffee.

4.1 Cross commodity causality: Oil vs. Coffee and Cocoa

In sub-Saharan Africa, cocoa is mainly grown by smallholder farmers (≤ 1 hectare)
and often on a subsistence basis (ITC, 2001). Larger cocoa plantations exist in Brazil,
Ecuador and Malaysia. Although cocoa is particularly sensitive to weather
conditions and diseases that may negatively affect production, relatively little
fertilizer is utilized (FAO 2006). On the other hand, coffee production is increasingly
mechanized and uses various chemical fertilizers (e.g. nitrogen, potassium etc.)
which are by-products of the petroleum industry. Here, we only consider the
indirect effect of fertilizers prices on coffee and cocoa prices through the oil price.
Fuels are also required for storage and transportation thus directly enhancing the
potential transmission effect of oil prices on coffee and cocoa prices. Graph 8
(Annexes), shows that coffee and cocoa price changes were often preceded by
variations in the oil price of a similar magnitude over the past fifty years. Therefore,
we aim to determine whether causality between oil prices and, coffee and cocoa
prices holds in the long-run considering the time-horizon: 1990-2010 and then,
whether a similar trend between oil and, cocoa and coffee is empirically observed.
First, we conduct Granger causality tests11 for crude oil, Arabica, Robusta, and cocoa
using large lag lengths in order to account for a long adjustment period of the
commodities prices to variations in the oil price, the results of which are presented
in Table 2.

11 'x is a Granger cause of y if present y can be predicted with better accuracy by using past values of x
rather than by not doing so, other information being identical' (Charemza and Deadman 1992).





18



Table 2 Granger-causality tests results

Null Hypothesis Lags included Observations F-statistic Prob.
LN_OIL does not Æ LN_ARABICA 48 208 1.901 0.003
LN_ARABICA does not Æ LN_OIL 1.152 0.270
LN_OIL does not Æ LN_COCOA 36 220 1.736 0.012
LN_COCOA does not Æ LN_OIL 1.025 0.441
LN_OIL does not Æ
LN_ROBUSTA 51 205 1.694 0.012
LN_ROBUSTA does not Æ
LN_OIL 1.091 0.349
Source: Annexes - Table 1

Table 5.1 shows that we cannot reject the hypothesis that the oil price Granger-
causes Arabica, Robusta and cocoa price variability at the 5 percent level (p-values:
Prob. > 0.05). However, the oil price is never Granger-caused by Arabica, Robusta or
cocoa prices at the 5 percent level. It is important to highlight that the oil-
commodity causality conclusions are dependent on the number of lags included. The
results show that oil price spillover effects on Arabica and Robusta take
approximately 4 years while it takes only 3 years for cocoa; which seems consistent
with observations outlined in Graph 8 (Annexes).

The concept of cointegration enables us to further determine the possible
relationship between the variables. Now that a long-run causality link has been
established between oil and beverages, we use cointegration tests to ascertain the
long-run relationship between these variables. Empirically, two I(1) cointegrated
series are defined, therefore if a linear combination of both is stationary I(0); an
adjustment between these two variables prevents errors becoming larger in the
long-term. Also, it s important to ensure current coffee-, cocoa-, and oil prices follow
an I(1) process. The results of our ADL tests reveals the presence of unit roots in
levels (p-values > 0.05) but not in first differences (p-values < 0.05) hence, prices of
the studied commodities are I(1) (Annexes- Table 8). Granger cointegration tests
are run by; first, estimating the equation (5.1), generating the residuals series tû and
then, estimating an ADL unit root test on those residuals by means of equation (5.2).
Cointegration of the series implies that the ADL unit root test of the residuals tû is
stationary.





19



attat uOilcC ,, . ++= η (5.1)


atC , : Current price at time t of a : { tA , tR , tCocoa }


at


p


j
ajtajatat uuu ,


1
,,,1, ˆˆˆ εαβ +∆+=∆ ∑


=


−−


(5.2)



The results of equation (5.1) are presented in Table 9 of the Annexes. The reported
adjusted R-squared provides a first hint regarding the cointegration of the variables.
In the first regression, it indicates that variations in cocoa, Arabica and Robusta
prices respectively explain 45%, 10% and 2% of the variations in oil prices. Test
results indicate that, only cocoa prices are cointegrated with oil prices at the 5%
level. Cointegration between oil prices and coffees prices (Arabica and Robusta) is
weakly rejected at the 10% level. This suggests that although coffee production uses
more technological and petro-chemical fertilizer inputs than cocoa, there is no
linear relationship between coffee and oil whereas, such a relationship is observed
for cocoa and oil. In fact, cocoa and oil price series may trend together in the long-
run. In summary, although long-run causality from the oil sector to the beverage
commodity sector is a valid assumption, only cocoa shares the same long-term trend
as oil. Besides, a short-run analysis confirms the consistency of the long-run
equilibrium relationship between cocoa and oil prices. As most coffee and cocoa
exporting countries are oil importing price-takers, there is limited policy space for
them to reduce their vulnerability to oil price fluctuations, whatever the
implications for their commodity exports.

4.2 Cointegration models and results: the effect of speculation

The global economic crises since 2008-2009 may have altered the nature of the
relationship between futures and cash prices of some agricultural commodities. The
2000 deregulation of financial instruments (futures) encouraged speculators to
massively trade commodities in which they had no business interest; and therefore,
contributed to the price surges in food and energy sectors, destabilizing businesses
and producer incomes (Ash et al., 2010, Gilbert and Morgan 2010). In fact, since
1990 cash coffee and cocoa prices and futures prices have tended to move in a
similar direction, irrespective of increased speculation. It could therefore be argued





20


that futures markets are quite efficient; as futures prices and cash prices are
convergent and it is also likely that both variables are cointegrated. After verifying
that futures prices are I(1) (see Annexes- Table 12), we conducted Granger
cointegration tests and obtained the following results (see Annexes - Tables 12 and
13) for the equations (5.3) and (5.4):


atatat uFC ,,, . ++= χϕ (5.3)


atC , : Cash price at time t for commodity a : { tA , tR , tCocoa }


atF , : Future price at time t for commodity a : { tA , tR , tCocoa }


at


p


j
ajtajatat uuu ,


1
,,,1, ˆˆˆ επγ +∆+=∆ ∑


=


−−


(5.4)



If the two price series are I(1) and the linear combination of them is I(0), the
variables are said to be cointegrated and thus, bivariate models may be specified to
take into account the linear relationship between the two series in the short-run.
ADL test results in table 5.2 attest to the rejection of the null hypothesis of a unit
root in the residuals at the 1% level (Prob. <0.05), thereby futures series and their
corresponding cash prices series are cointegrated. The cointegration order (1, 1)
and the cointegrating vector [1, - χ̂ ] corresponding to: [1, 0.98] for Arabica, [1, 1.02]
for Robusta and [1, 0.925] for cocoa may be positively accepted. Engle and Granger
(1987) have demonstrated that all cointegration series have an error correction
representation. Positively accepted cointegration suggests that an error correction
model (ECM) maybe estimated to assess short-term price adjustments. We estimate
the error correction mechanism with an unrestricted OLS in equation (5.5):


atatatatat FCFC ,,1,12,10, ).( εχααα +−+∆+=∆ −− (5.5)

We replace χ by its previously computed OLS estimate χ̂ so that atC ,∆ , atF ,∆
and ).ˆ(


,1,1 atat FC −− − χ are all )0(I (Charemza and Deadman, 1991) and the error is
corrected ( at ,ε ~ )0(I ). Given the results (Table 15 – Annexes), we assume that 1ˆ =χ
hence, the Engle Granger (5) equation is simplified as follow:





21




atatatatat FCFC ,,1,12,10, )( εααα +−+∆+=∆ −− (5.6)

The Arabica model (Table 16 - Annexes) suggests that the predictive power of the
model is very high; especially for Arabica and Robusta. Indeed adjusted R-squared
for Arabica, Robusta and cocoa models are respectively 0.95, 0.90 and 0.70.

Despite the low frequency of monthly data, it is possible to estimate the speed of
adjustment between futures and cash prices. An ECM provides a good
representation of short-run adjustments between cash and futures markets for
Arabica, Robusta and cocoa. Short-run adjustments are consistent with the long-run
relationship equilibrium existing between cash and futures series suggesting that
the speed of adjustment is very fast, and futures cocoa and coffee markets are
efficient.


5 Policy recommendations and conclusions

Price fluctuations are inherent in agricultural markets – partly due to the supply-
demand dynamics and the unpredictability of weather patterns and harvest yields.
There are debates as to the extent to which activity in futures trades and over the
counter markets (OTC) for agricultural commodities impact on this volatility.
Whatever the cause, extreme volatility in food prices deters producers from making
the necessary investments for increasing productivity and production: this is one of
the underlying causes of continued worldwide food insecurity. Indeed, recent
weather catastrophes, oil price surges, inflation, declining value of the U.S. dollar
and, growing financialization on futures exchange markets have greatly led to the
unpredictability of food prices and market developments. Several international
organizations have investigated policy responses in order to mitigate the risks
associated with high prices and volatility in global food markets. A policy
recommendation put forward by the G2012 suggests strengthening the long term



12 Policy reports elaborated by FAO, IFAD, IMF, OECD, UNCTAD, WFP, the World Bank, the WTO,
IFPRI, and the UN HLTF (2011).





22


productivity, sustainability and resilience of the CDDCs agricultural sector, through
enhanced public investment and national food security programs. Increasing
transparency in food and futures markets and, eliminating domestic trade policies
would also reduce trade distortions and markets instabilities (Staatz and Weber,
2011 and, Limao and Panagariya, 2003).

This paper examined volatility, oil, and futures spillover effects on three major
tropical commodities: Arabica, Robusta and Cocoa. Volatility developments and
implications were analyzed from the supply-side that is, exporting LDCs and CDDCs.
In this case, large price decreases are simultaneously reflected in the trade balance
and in the longer term has a detrimental effect on growth. On the other hand,
sudden price hikes may encourage producers to increase production and adjust
their investment decisions, which may trigger even more instability in the markets.
The results of the presented GARCH models provide an accurate assessment of
commodity price volatility. The conditional variances are found variant over time
due to volatility clustering13, thus reverting to the mean rather than remaining
constant or moving in monotonic fashion over time, which justifies the use of a
GARCH model. Further analysis reveals uneven effects in Arabica and Robusta price
volatilities, which, are more affected by positive shocks than negative shocks. A good
harvest in coffee crops will trigger more volatility in its price than a bad harvest.
However, cocoa volatility reacts symmetrically to the market shocks whether
positive or negative. Cocoa price volatility is evident, regardless of whether there is
a good or poor harvest.

This paper investigates causality links between the crude oil price and, both coffees
and cocoa prices in the long-run. It appears that variations in coffee and cocoa prices
follow oil price variations with, respectively 4 and 3-year intervals. Nevertheless,
the hypothesis of a long-run equilibrium relationship only holds between oil and
cocoa prices meaning that, structural changes in the oil price will be directly
reflected in cocoa prices. Baffes (2007) shows that the average elasticity for cocoa;
was high and significant while the average coffee elasticity was particularly low; in



13 In contrast to the often-assumed log-normal distribution of asset price returns, it is often observed
that periods of high price volatility follow periods of low volatility and vice versa.





23


short a 100 per cent variation in the oil price causes a 49 per cent shift in cocoa
prices, but does not cause a significant variation in coffee prices. In summary, oil
price developments have no significant effect on coffee price variability in the short-
run. On the other hand, policy-makers should closely monitor oil price surges as
they appear to strongly influence cocoa prices and their volatility in both the short
and long-run.

We also examined the relationship between Arabica, Robusta and cocoa cash prices
and their corresponding futures prices. The deregulation of financial and physical
instruments in 2000, along with the introduction of new electronic trading
opportunities in 2007 has raised concerns about efficiency in the coffee and cocoa
futures markets. However, in this study, the observed cointegration between cash
and futures series between 1990 and 2010 suggests that both ICE and LIFFE futures
markets are (statistically) unbiased and therefore, serve as price discovery channels
for coffee and cocoa sector participants. The very short adjustment period
noticeable between futures and cash prices suggests that, hedging strategies
mitigate price risk only if they are an immediate reaction to market activity.
Nonetheless, the lack of statistical bias of futures markets does not necessarily imply
a full-hedging of price risk (Broll and Zilcha 1992).

In fact, the Separation theorem states that unbiased futures estimators of the spot
prices do not imply that price risk is entirely avoided. Recent studies have shown
that major speculative activity has increased price risk for cash market participants,
particularly commercial traders (Schaffnit-Chatterjee, 2011 and, Schutter, 2010). As
a consequence of increasing speculative activity, small farmers growing cocoa and
coffee in developing countries are even more exposed to price risk, especially as few
alternatives to manage price risk are available to them. Gabre-Madhin (2010) and,
Fortenbery and Zapata (2004) have proposed the creation of local commodity
exchanges which are more accessible to commercial hedgers (for example; the
Ethiopia Commodity Exchange which reduces the incentives of speculators by
imposing mandatory delivery and higher margins. Such initiatives may largely
reduce price risk and thus, promote economic stability in many CDDCs.





24


Commodity producers in developed countries are increasingly relying on hedging to
mitigate exposure to price volatility. However, the extent of hedging in developing
countries remains limited. A few countries have used market-based instruments to
mitigate the income risks.14

The main reason for the low use of financial instruments is the lack of familiarity on
the part of both private sector operators (especially farmers and exporters) and, in a
few instances, the lack of interest from government officials. Using financial
instruments in hedging requires technical and managerial expertise and an
institutional framework that ensures adequate reporting, recording, monitoring and
evaluating mechanisms. Furthermore, it is also necessary to establish internal
control procedures that avoid and protect against speculative transactions.15

Market-based instruments can play a fundamental role in building tailor-made
facilities to address commodity price instability. However, it is doubtful whether the
futures markets are as suitable for addressing problems emanating from price
variability as they are for reducing uncertainty in revenue flows. This
notwithstanding, futures markets do allow Governments to eliminate uncertainty
associated with variability.

Apart from emergency measures designed to assist the most vulnerable and the
long-term measures designed to tackle excessive commodity price volatility on the
supply side, there is a need to consider how the functioning of commodity
derivatives markets could be improved in a way that would enable those trading
venues to better fulfill their role of providing reliable price signals to commodity
producers and consumers.

In light of the vital role of information flows in commodity price developments, a set
of four policy responses to improve market functioning should be considered: First,



14 For example, Mexico hedged, via options, all of its oil sales for 2009 in 2008 at a strike price of US$
70 a barrel when the oil price was US$ 100 a barrel.14 The cost of purchasing options at US$ 1.5
billion enabled the programme to make a savings of more than US$ 5 billion..
15


Claasens S (1992). How can developing countries hedge their bets? Finance and Development. September
1992.





25


greater transparency in physical markets would enable the provision of more timely
and accurate information about commodities, such as spare capacity and global
stock holdings for oil, and for agricultural commodities, areas under plantation,
expected harvests, stocks and short-term demand forecast. This would allow
commercial market participants to more easily assess current and future
fundamental supply and demand relationships.

Second, a better flow of and access to information in commodity derivatives
markets, especially regarding position-taking by different categories of market
participants, would further improve market transparency. In particular, measures
designed to ensure reporting requirements for trading on European exchanges
similar to those enforced in US exchanges would considerably improve
transparency of trading and discourage regulatory migration.

Third, tighter regulation of financial market participants, such as through
establishing position limits, could contain financial investors’ impacts on commodity
markets. For example, a rule could be applied to physical traders, prohibiting them
from taking financial positions and betting on outcomes that they are able to
influence due to their strong economic position in physical markets. This calls for
finding the right balance between being adopting overly restrictive regulation,
which would impair the price discovery and risk transfer functions of commodity
exchanges, and overly lax regulation, which equally impairs the basic functions of
the exchanges.

Finally, there appears to be support for the contention that the behaviour of
financial investors in following investments that align to their own preferences help
explain movements in coffee and cocoa prices that the fundamentals alone are
unable to account for. The rises in coffee and cocoa prices attracts more speculation
from parties with no interests in owning the actual commodity but are investing
solely on the basis of expected price changes on futures markets. As a result, the
behaviour of financial investors/speculators continues to push prices above the
equilibrium price of the commodity. In the very short-run (e.g. in daily price
formation), a declining dollar seems likely to stimulate speculation in commodity
markets rise in prices. We also find that growing speculation appears to link





26


financial variables with coffee and cocoa prices during some periods. Although
speculation was particularly high over the past four years, the equilibrium between
financial and commodity variables holds (i.e. is linked) in the long-term.


References



Aradhyula and Ho (1988). GARCH time-series models: an application to retail livestock prices.
Ash, M. et al. (2010). Regulation of commodities futures markets in the reconciliation of the


financial reform bill. Letter to the Reconciliation Committee of the US Congress.
Baffes J. (2007). Oil spills on other commodities. World Bank.
Broll and Zilcha (1992). Optimal hedging by firms with multiple sources of risky revenues.
Charemza Wojciech W. and Deadman Derek F. (1992). Econometrics Practice, General to


Specific Modelling, Cointegration and Vector Autoregression.
Combes Jean-Louis and Guillaumont Patrick (2002). Commodity Price Volatility,


Vulnerability and Development. Development Policy Review.
FAO, IFAD, IMF, OECD, UNCTAD, WFP, the World Bank, the WTO, IFPRI and the UN HLTF


(2011). Price Volatility in Food and Agricultural Markets: Policy Responses. Draft of
Policy report


FAO (2011). Hunger Statistics. FAOSTAT.
FAO (2011). Why has Africa become a net food importer? Explaining African agricultural


and food trade deficits.
FAO (2011). Who's Feeling the Effects of Rising Food Prices?
FAO (2008). Number of Hungry People Rises to 963 Million.
FAO (2006). Fertilizer use by crop. Fertilizer and Plant Nutrition Bulletin
Fortenbery and Zapata (2004). Developed Speculation and Under Developed Markets- The


Role of Futures Trading on Export Prices in Less Developed Countries.
Gabre-Madhin (2010). Structured Commodity trade and Price Volatility: Are Commodity


Exchanges the Solution, or the Problem? Presentation to FAO.
Gilbert C.L. and Morgan C. W. (2010). Commodities Market Speculation: The Risk to Food


Security and Agriculture. Review. Institute for Agriculture and Trade Policy.
IATP (2008). Commodities Market Speculation: The Risk to Food Security and Agriculture.


Trade and Global Governance Program.
International Coffee Organization (ICO 2005). Coffee price volatility. Working paper.
ICE Futures U.S. (ICE 2011). Cocoa, Coffee "C".
International Trade Centre (ITC 2001). Cocoa, A guide to trade practices. Product and


market development.





27


International Trade Centre (ITC 2002). Coffee, An exporter's guide. Product and market
development.


Kaufman Frederick (2010). THE FOOD BUBBLE, How Wall Street starved millions and got
away with it. Report.


Kebede Yohannes (1992). Causality and efficiency in the coffee futures markets. MPRA.
Limao and Panagariya (2003). Why is there an Anti-trade Bias in Trade Policy?
Mitchell Donald (2008). A note on rising food prices. Policy Research Working Paper Series


from the World Bank.
Nardella M. (2007). Price efficiency and speculative trading in cocoa futures markets.
Nelson Daniel B. (1991). Conditional Heteroskedasticity in Asset Returns: A New Approach.
OECD (2006). Agricultural market impacts of future growth in the production of biofuels.


Directorate for food, agriculture and fisheries; committee for agriculture.
Penaranda F. and Micola A. R. (2009). The emergence of biofuels and the co-movement


between crude oil and agricultural prices.
Rapsomanikis George (2011). Price Transmission and volatility spillovers in food markets;


Safeguarding food security in volatile global markets. Trade and Markets Division
(FAO)


Schaffnit-Chatterjee Claire (2011). Where are food prices heading? Short term drivers,
trends and implications. Deutsche Bank Research


Schutter Olivier (2010). Food Commodities Speculation and Food Price Crises, Regulation to
reduce the risks of price volatility United Nations Special Rapporteur on the Right to
Food.


Staatz and Weber (2011). Evaluating the impact on Market Performance of Investments in
Market Information Systems: Methodological challenges.


Swaray R. B. The University of York (2002). Volatility of primary commodity prices: some
evidence from agricultural exports in Sub-Saharan Africa.


Timmer C. Peter (2009). Did Speculation Affect World Rice Prices?
United Nations Conference on Trade and Development (UNCTAD 2010). The Least


Developed Countries Report 2010.
United Nations Conference on Trade and Development (UNCTAD 2009). The financialization


of Commodity Market. Trade and Development Report 2009.
Wahl Peter (2008). Food Speculation: The Main Factor of the Price Bubble in 2008. Briefing


Paper. World Economy, Ecology & Development
World Bank (2009). Global Economic Prospects; commodities at the crossroad. Report to the


12th session of the Human Rights Council
WTO (2004). Selected Issues in trade and trade policies. World Trade Report 2004.





28


6 Terms & Acronyms

ADL Augmented Dickey Fuller
CDDCs Commodity-dependent developing countries
CFA Communaute Financiere Africaine
FAO Food and Agriculture Organization of the United Nations
IATP Institute for Agricultural Trade Policy
ICA International Coffee Agreements
ICCO International Cocoa Organization
ICO International Coffee Organization
IFAD International Fund for Agricultural Development
IFPRI International Food Policy Research Institute
IMF International Monetary Fund
ICE Intercontinental Exchange
ITC International Trade Centre
LDCs Least Developed Countries (*)
LIFFE London International Financial Futures and Options Exchange
MDG Millennium Development Goals
NYBOT New York Board of Trade
ODA Official Development Assistance
OECD Organization of Economic Cooperation and Development
SIC Schwarz Information Criterion
UN-CTAD United Nations Conference on Trade and Development
UN-HLTF United National High-Level Task Force on the Global Food Security Crisis
WFP United Nations World Food Programme
WTO United Nations World Trade Organization
Notes: *LDCs: forty-eight countries designated by the UN using three criteria: “low-income”, “human
assets weakness”, “economic vulnerability”: Afghanistan, Angola, Bangladesh, Benin, Bhutan, Burkina
Faso, Burundi, Cambodia, the Central African Republic, Chad, the Comoros, the Democratic Republic
of Congo, Djibouti, Equatorial Guinea, Eritrea, Ethiopia, the Gambia, Guinea, Guinea-Bissau, Haiti,
Kiribati, the Lao People’s Democratic Republic, Lesotho, Liberia, Madagascar, Malawi, Mali,
Mauritania, Mozambique, Myanmar, Nepal, Niger, Rwanda, Samoa, Sao Tome and Principe, Senegal,
Sierra Leone, the Solomon Islands, Somalia, Sudan, Timor-Leste, Togo, Tuvalu, Uganda, the United
Republic of Tanzania, Vanuatu, Yemen and Zambia.





29


7 Annexes

Annex contents


List. 1 Beverage exporting countries

Graph 1 Coffees and Cocoa: monthly prices & volatility (short term)
Graph 2 Monthly price-volatilities of beverage commodities in the long run (1960-


1990)
Graph 3 Returns of Beverage annual prices
Graph 4 Percent variations in: Coffee and Cocoa prices, consumption and production
Graph 5 Evolution of Arabica, Robusta, Cocoa, and Oil current prices
Graph 6 Percent Variation in Cocoa- Arabica- Robusta prices vs. Oil prices

Table 1 Correlations in current & constant prices
Table 2 Descriptive Statistics of Arabica, Robusta and Cocoa (in log)
Table 3 GARCH (1, 1) tests results
Table 4 Wald Test: Test of the GARCH hypothesis
Table 5 EGARCH: tests results for Cocoa, Arabica and Robusta
Table 6 Unit root in level and first-difference for Arabica Robusta Cocoa and Oil
Table 7 Ordinary Least Squares equations
Table 8 Cointegration: ADL test on residuals
Table 9 Unit root in level and first-difference for Arabica Robusta Cocoa futures prices
Table 10 Ordinary Least Squares equations
Table 11 Cointegration: ADL test on residuals
Table 12 Wald Test
Table 13 OLS Error Correction Model





30



List 1. Beverage exporting countries


Cocoa exporting countries Coffee exporting countries Tea exporting countries
Brazil
Cameroon
Côte d'Ivoire
Dominican Republic
Ecuador
Gabon
Ghana
Malaysia
Nicaragua
Nigeria
Papua New Guinea
Sierra Leone
Togo
Trinidad and Tobago
Venezuela

















Angola
Brazil
Burundi
Central African Republic
Colombia
Costa Rica
Cote d'Yvoire
Cuba
Ecuador
El Salvador
Ethiopia
Gabon
Ghana
Guatemala
Honduras
India
Indonesia
Kenya
Liberia
Mexico
Nicaragua
Panama
Papua New Guinea
Philippines
Sierra Leone
Tanzania
Thailand
Timor-Leste
Togo
Uganda
Vietnam
Yemen


China
India
Indonesia
Vietnam
Turkey
Sri lanka
Kenya
Japan
Argentina
Iran
Bangladesh
Malawi
Uganda



















Source: FAO (2011)





31



Table 1 Correlations in current & constant prices




SHORT RUN



Current









Constant








LONG RUN


1960-2010 Cocoa Arabica Robusta
Cocoa -
Arabica 0.908 -
Robusta 0.418 0.921 -





1968-1990 Cocoa Arabica Robusta
256 obs. Cocoa -
Arabica 0.84 -
Robusta 0.90 0.96 -


1990-2011 Cocoa Arabica Robusta
256 obs. Cocoa -
Arabica 0.60 -
Robusta 0.36 0.77 -


1990-2010 Cocoa Arabica Robusta
249 obs. Cocoa -
Arabica 0.29 -
Robusta 0.09 0.76 -





32



Table 2 Descriptive Statistics of Arabica, Robusta and Cocoa (in log)



ln(Rt) ln(At) ln(Ct)
Mean 4.746 5.293 4.891
Median 4.755 5.299 4.847
Maximum 5.881 6.274 5.580
Minimum 3.969 4.579 4.427
Std. Dev. 0.391 0.321 0.264
Skewness 0.226 0.383 0.575
Kurtosis 2.768 2.828 2.841

Standard deviation 0.082 0.061 0.054

Sum 1181.668 1317.876 1217.934
Sum Sq. Dev. 37.918 25.523 17.332

Observations 249 249 249






33


Graph 1 Coffees and Cocoa: monthly prices and volatility (short term)




Source: Authors estimates.


4.4


4.8


5.2


5.6


6.0


6.4


1991 1993 1995 1997 1999 2001 2003 2005 2007 2009


LN_ARABICA


-.2


-.1


.0


.1


.2


.3


.4


1991 1993 1995 1997 1999 2001 2003 2005 2007 2009


Differenced LN_ARABICA


3.6


4.0


4.4


4.8


5.2


5.6


6.0


1991 1993 1995 1997 1999 2001 2003 2005 2007 2009


LN_ROBUSTA


-.3


-.2


-.1


.0


.1


.2


.3


.4


1991 1993 1995 1997 1999 2001 2003 2005 2007 2009


Differenced LN_ROBUSTA


4.4


4.6


4.8


5.0


5.2


5.4


5.6


1991 1993 1995 1997 1999 2001 2003 2005 2007 2009


LN_COCOA


-.2


-.1


.0


.1


.2


.3


1991 1993 1995 1997 1999 2001 2003 2005 2007 2009


Differenced LN_COCOA





34



Graph 2 Monthly Price volatilities of beverage commodities in the long-run


(1960-1990)


0


100


200


300


400


500


600


700


800


1963 1967 1971 1975 1979 1983 1987 1991 1995 1999 2003 2007


COCOA ARABICA
ROBUSTA TEA





Graph 3 Returns of Beverage annual prices



Source: Authors estimates.


-.6


-.4


-.2


.0


.2


.4


.6


.8


1964 1969 1974 1979 1984 1989 1994 1999 2004 2009


Differenced ARABICA


-0.6


-0.4


-0.2


0.0


0.2


0.4


0.6


0.8


1.0


1964 1969 1974 1979 1984 1989 1994 1999 2004 2009


Differenced ROBUSTA


-.4


-.2


.0


.2


.4


.6


.8


1964 1969 1974 1979 1984 1989 1994 1999 2004 2009


Differenced COCOA


-.6


-.4


-.2


.0


.2


.4


.6


1964 1969 1974 1979 1984 1989 1994 1999 2004 2009


Differenced TEA





35



Table 3 GARCH (1, 1) tests results



Cocoa: AR (1)


ttt pcCocoa εφ ++= −11
Arabica: AR (1)


ttt pcA εφ ++= −11
Robusta: ARMA (1,1)


tttt pcR εεγφ +++= −− 1111
Conditional variance 2


1
2


1
2


−−


++= ttt hh βαεδ


Cocoa Arabica Robusta
ARMA c 4.940 5.260 4.610


(0.158) (0.132) (0.206)
φ 0.976 0.969 0.972
(0.011) (0.015) (0.014)
γ 0.241


(0.075)
GARCH δ 0.001 0.002 0.002


(0.000) (0.001) (0.001)
α 0.247 0.178 0.144
(0.080) (0.067) (0.067)
β 0.622 0.505 0.525
(0.121) (0.210) (0.244)
α+β 0.870 0.682 0.669


Schwarz -2.742 -2.264 -2.418
Adjusted R^2 0.947 0.940 0.968



Table 4 Wald Test: Test of the GARCH hypothesis



Wald Test: 0,0:0 == βαH
Test Statistic Value df Probability


F-statistic 53.76003 (2, 243) 0.000
Equation: COCOA_GARCH Chi-square 107.5201 2 0.000 REJECT


F-statistic 31.58837 (2, 243) 0.000
Equation: ARABICA_GARCH Chi-square 63.17674 2 0.000 REJECT


F-statistic 15.88593 (2, 242) 0.000
Equation: ROBUSTA_GARCH Chi-square 31.77186 2 0.000 REJECT





36



Table 5 EGARCH: tests results for Cocoa, Arabica and Robusta



Cocoa: AR (1)


ttt pcCocoa εφ ++= −11
Arabica: ARMA (4, 2)


tttttttt ppppcA εεγεγφφφφ +++++++= −−−−−− 221144332211
Robusta; ARMA (1, 1)


tttt pcR εεγφ +++= −− 1111


EGARCH: )log()log( 2 12
1


1
22


1


1
1


2








− +++= t
t


t


t


t
t hhh


h βεπεπδ




* Note: Only Cocoa 2π coefficient is significantly equal to 0.







Coefficient Cocoa Arabica Robusta
ARMA c 4.911 5.410 4.747


0.139 0.285 0.258
AR 1φ 0.974 1.248 0.980


0.010 0.075 0.010
2φ - -1.048 -
- 0.096 -
3φ - 1.037 -
- 0.080 -
4φ - -0.269 -


- 0.069 -
MA 1γ - -0.088 0.223


- 0.029 0.067
2γ - 0.931 -


- 0.032 -
EGARCH δ -2.073 -3.178 -2.308


0.710 0.574 0.777
1π 0.542 -0.036 0.015
0.135 0.141 0.146
2π 0.035* 0.422 0.351
0.090 0.104 0.086


β 0.712 0.402 0.579
0.117 0.110 0.138
SIC -2.721 -2.280 -2.466





37


Graph 4 Per cent variations in Prices, consumption and production of coffee and
cocoa



Cocoa variations


-.4


-.2


.0


.2


.4


.6


.8


1991 1993 1995 1997 1999 2001 2003 2005 2007 2009


Cocoa consumption
Cocoa Price
Cocoa production



Coffee variations


-.4


-.2


.0


.2


.4


.6


.8


1991 1993 1995 1997 1999 2001 2003 2005 2007 2009


Coffee consumption
Coffee price
Coffee production



Source: ICO, ICCO





38



Graph 5 Current prices of: Arabica, Robusta, Cocoa, and Oil




0


1


2


3


4


5


6


7


1964 1969 1974 1979 1984 1989 1994 1999 2004 2009


OIL ARABICA
ROBUSTA COCOA









39


Graph 6 Per cent Variation in Cocoa- Arabica- Robusta prices vs. Oil prices



Arabica vs. oil


-0.8


-0.4


0.0


0.4


0.8


1.2


1.6


1964 1969 1974 1979 1984 1989 1994 1999 2004 2009


Differenced OIL
Differenced ARABICA


Robusta vs. oil


-0.8


-0.4


0.0


0.4


0.8


1.2


1.6


1964 1969 1974 1979 1984 1989 1994 1999 2004 2009


Differenced OIL
Differenced ROBUSTA


Cocoa vs. oil


-0.8


-0.4


0.0


0.4


0.8


1.2


1.6


1964 1969 1974 1979 1984 1989 1994 1999 2004 2009


Differenced OIL
Differenced COCOA





40



Table 6 Unit root in level and first-difference for Arabica Robusta Cocoa and Oil







Table 7 Ordinary Least Squares equation





Table 8 Cointegration: ADL test on residuals


Arabica Cocoa Robusta
Lag length 1 0 1
t-stat Prob. t-stat Prob. t-stat Prob.
ADL statistic -1.614 0.1003 -2.2436 0.0242 -1.569 0.1096


1% -2.574 -2.574 -2.574
5% -1.942 -1.942 -1.942


Critical values: 10% -1.616 -1.616 -1.616


Arabica Cocoa Robusta Oil
Unit root in first-differences
Lag length 1 0 1 0
t-stat Prob. t-stat Prob. t-stat Prob. t-stat Prob.


ADL statistic -12.80 0.00 -14.094 0.000 -11.790 0.000 -11.486 0.000
Unit root in levels
Lag Length 1 0 1 1
t-stat Prob. t-stat Prob. t-stat Prob. t-stat Prob.


ADL statistic 0.746 0.87 1.1 0.93 0.408 0.801 0.784 0.882
Critical values: 1% -2.574 -2.574 -2.574 -2.574
5% -1.942 -1.942 -1.942 -1.942
10% -1.616 -1.616 -1.616 -1.616


Method: Least Squares

Dependent Variable:
LN_COCOA LN_ARABICA LN_ROBUSTA
Variable Coef. Std. Error Coef. Std. Error Coef. Std. Error


η (LN_OIL) 0.368 0.025 0.211 0.037 0.105 0.044
C 3.796 0.087 4.735 0.129 4.539 0.153
Adjusted R-squared 0.453 0.112 0.018





41


Table 9 Unit root tests for Arabica Robusta Cocoa futures prices





Table 10 Ordinary Least Squares equations



* denotes insignificance at a 5% level


Futures Arabica "C" Futures Cocoa Futures Robusta
Unit root in first-differences
Lag length 1 0 1


t-statistic Prob.
t-
statistic Prob. t-statistic Prob.


ADL statistic -13.451 0.000 -12.819 0.000 -11.19 0.000

Unit root in levels
Lag length 1 0 1


t-statistic Prob.
t-
statistic Prob. t-statistic Prob.


ADL statistic 0.675 0.861 0.728 0.871 0.24 0.755
Critical values: 1% -2.574 -2.574 -2.574
5% -1.942 -1.942 -1.942
10% -1.616 -1.616 -1.616


Dependent Var.:
LN_COCOA LN_ARABICA LN_ROBUSTA
Variable Coefficient Std. Error Coefficient Std. Error Coefficient Std. Error



χ
ϕ


0.981
0.0647


0.006
0.0318


1.0213
-0.069*


0.01
0.055


0.925
0.446


0.0058
0.0278


Adjusted R-
squared 0.989 0.976 0.982





42



Table 11 Cointegration: ADL test on residuals


Arabica futures Cocoa futures Robusta futures
t-statistic Prob. t-statistic Prob. t-statistic Prob.


-2.789 0.0054 -9.139 0.000 -2.803 0.0052
1% -2.574 -2.574 -2.574
5% -1.942 -1.942 -1.942


ADL
statistic

Critical
values:


10% -1.616 -1.616 -1.616





Table 12 Wald Test: 1ˆ =χ
Wald Test
Test Statistic Value df Probability
Arabica t-statistic 2.12 254 0.035
F-statistic 4.50 (1, 254) 0.035
Chi-square 4.50 1 0.034
Cocoa t-statistic -3.05 254 0.003
F-statistic 9.31 (1, 254) 0.003
Chi-square 9.31 1 0.002
Robusta t-statistic -13.04 205 0.000
F-statistic 169.97 (1, 205) 0.000
Chi-square 169.97 1 0.000








43



Table 13 OLS Error Correction Model



atatatatat FCFC ,,1,12,10, )( εααα +−+∆+=∆ −−






Dependent Variable atC ,∆ : Variable Coefficient Std. Error t-Statistic Prob.


0α -0.001 0.002 -0.729 0.466
Arabicaa : 1α 0.907 0.013 69.790 0.000


2α -0.030 0.018 -1.724 0.086
adjusted 2R 0.951


Cocoaa :






-0.001

0.800


0.003

0.032


-0.226

24.993



0.821

0.000


2α 0.034 0.033 1.018 0.310
adjusted 2R 0.716


0α 0.005 0.003 1.445 0.150
Robustaa : 1α 0.843 0.021 40.622 0.000


2α -0.059 0.032 -1.844 0.067
adjusted 2R 0.892




Login