Price dependence in the coffee markets of Brazil,Ecuador and India: A copula approach
This study investigates the price dependence between coffee varieties, namely Arabica and Robusta, in Brazil, Ecuador and India at the farm gate level. For this reason, it utilizes monthly farm data on coffee between the two varieties and the copula methodology. The empirical results suggest that positive or negative shocks in the prices of the two coffee species in Brazil will not be transmitted, even though overall dependence is quite considerable. The picture is different in Ecuador and India where the overall dependence is relatively low but there is some degree of symmetric and asymmetric tail price dependence respectively.
Published on: Mar 4, 2016
Transcripts - Price dependence in the coffee markets of Brazil,Ecuador and India: A copula approach
UNIVERSITY OF IOANNINA, DEPARTMENT OF ECONOMICS
Ms ECONOMIC ANALYSIS
PRICE DEPENDENCE IN THE COFFEE MARKETS OF BRAZIL,
INDIA AND ECUADOR: A COPULA APPROACH
SUPERVISORS: Panagiotou Dimitrios
The present work is dedicated to my soul mate Evaggelia Moraiti!
This study investigates the price dependence between coffee varieties,
namely Arabica and Robusta, in Brazil, Ecuador and India at the farm
gate level. For this reason, it utilizes monthly farm data on coffee
between the two varieties and the copula methodology. The empirical
results suggest that positive or negative shocks in the prices of the two
coffee species in Brazil will not be transmitted, even though overall
dependence is quite considerable. The picture is different in Ecuador and
India where the overall dependence is relatively low but there is some
degree of symmetric and asymmetric tail price dependence respectively.
TABLE OF CONTENTS
CHAPTER 1: An overview of coffee industry…………………………..4
CHAPTER 2: Literature review………………………………………..22
CHAPTER 3: Methodology analysis…………………………………...32
CHAPTER 4: Empirical analysis………………………………………38
CHAPTER 1: An overview of coffee industry
Coffee is a popular beverage and an important commodity, which has become
a daily routine and coffee shops are now a common social meeting ground. Since
coffee shops are springing up at every half block in one’s local downtown, it may not
be surprising that coffee has become the second largest traded commodity next to oil
Therefore, the coffee industry has recently received increased attention from
economic researchers. Since the crisis period of the early 1990s, coffee has been on
the leading edge of economic, social, and environmental development schemes that
now reach many major industries.
The procedure of coffee production
Coffee production is the industrial process of converting the raw fruit of the
coffee plant into the finished coffee. The coffee tree is a tropical evergreen shrub
(genus Coffea) and grows between the Tropics of Cancer and Capricorn. The two
most commercially important species grown are varieties of Coffea arabica (Arabicas)
and Coffea canephora (Robustas).
The average Arabica plant is a large bush with dark-green oval leaves. The
fruits, or cherries, are rounded and mature in 7 to 9 months; they usually contain two
flat seeds, the coffee beans. When only one bean develops it is called a peaberry.
Robusta is a robust shrub or small tree that grows up to 10 metres high. The
fruits are rounded and take up to 11 months to mature; the seeds are oval in shape and
smaller than Arabica seeds.
Ideal average temperatures range between 15 to 24ºC for Arabica coffee and
24 to 30ºC for Robusta, which can flourish in hotter, harsher conditions. Coffee needs
an annual rainfall of 1500 to 3000 mm, with Arabica needing less than other species.
Whereas Robusta coffee can be grown between sea-level and about 800 metres,
Arabica does best at higher altitudes and is often grown in hilly areas.
A coffee plant usually starts to produce flowers three to five years after it is
planted and it is from these flowers that the fruits of the plant (widely known as coffee
cherries) appear, with the first useful harvest possible around five years after planting.
The cherries ripen around eight months after the emergence of the flower, by
changing color from green to red, and it is at this time that they should be harvested.
As coffee is often grown in mountainous areas, widespread use of mechanical
harvesters is not possible and the ripe coffee cherries are usually picked by hand. The
main exception is Brazil, where the relatively flat landscape and immense size of the
coffee fields allow for machinery use.
Coffee trees yield an average of 2 to 4 kilos of cherries and a good picker can
harvest 45 to 90 kilos of coffee cherry per day; this will produce nine to 18 kilos of
coffee beans. In most coffee-growing countries, there is one major harvest a year;
though in countries like Colombia, where there are two flowerings a year, there is a
main and secondary crop, the main one April to June and a smaller one in November
Coffee is harvested in two ways: 1. Strip Picked: all the cherries are stripped
off of the branch at one time, either by machine or by hand. 2. Selectively Picked:
only the ripe cherries are harvested and they are picked by hand. Pickers check the
trees every 8 to 10 days and individually pick only the fully ripe cherries. This method
is labour intensive and more costly. Selective picking is primarily used for the finer
Coffee is produced in more than 60 countries and provides a livelihood for
some 25 million coffee-growing families around the world. Over the last 50 years,
there has been steady growth in world production, interspersed with periodic falls.
The average growth rate since 1963 was 2.4%, with 2.8% annual growth in the
market-controlled period, and 2% since 1990. In crop year 2012-2013, world coffee
production reached 145.1 million bags, the largest on record. With the exception of
Africa, all coffee-growing regions recorded a steady growth in their production over
the time period.
Figure 1: World coffee production by region (1963/64 – 2012/13)
More specifically, coffee is mostly produced and exported by developing
countries in equatorial and subequatorial regions of the globe. Beans from different
countries or regions can usually be distinguished by differences in flavor, aroma,
body, and acidity. These taste characteristics are dependent not only on the coffee’s
growing region, but also on genetic subspecies and processing.
Figure 2: World coffee production by varieties
There are two principal types of coffee: Robusta and Arabica. Robusta is
grown at lower altitudes, typically in West and Central Africa and South East Asia,
has a higher caffeine content and is generally considered to have a harsher taste. In
contrast, Arabica is lower in caffeine content and is commonly thought to have a
more refined taste. Arabica coffee is grown in Latin America, Central and East Africa
and in India, and at generally higher altitudes than Robusta.
Figure 3: Geography of world coffee production y variety
Source: Ubilava, 2012
Total exports by exporting countries increased steadily during the last 50 years
despite some interruptions in the upward trend, notably between 1976 and 1978, as
well as in 1987/88 and 1994/95. Up until 1989, the coffee market was regulated by a
series of International Coffee Agreements (ICA) which were intended to manage
supply and maintain price stability. This system subsequently collapsed, and since
1990 the coffee market has been subject to the free market forces of supply and
Figure 4: Volume and value of exports by all exporting countries (1964-2012)
Compared to the average for the period 1990 to 2011, the following exporting
countries recorded significant earnings: Brazil (US$3.1 billion), Colombia (US$1.7
billion), Vietnam (US$888 million), Indonesia (US$573 million), Guatemala
(US$525 million), Mexico (US$502 million) and Honduras (US$422 million).
During the free market period, namely the period after 1990, importing
countries earned an annual average of US$5.5 billion from re-exports of 24.2 million
bags of coffee, compared to US$962 million for the period 1965 to 1989. Germany
earned the substantial sum of US$1.4 billion for average re-exports of 6.6 million
bags a year, accounting for 27% of the total value of re-exports by all importing
countries during the free market period. Belgium ranked second with US$542 million,
representing 9.9% of the total value of re-exports by all importing countries, followed
by the USA and Italy.
Overall, it should be noted that re-exporting activities in importing countries
have shown consistent growth over the last 50 years, with re-export value gaining
considerable momentum since 2000. These re-exports comprise all three forms of
coffee, namely green coffee, roasted coffee and soluble coffee. Nevertheless, re-
exports by some importing countries are clearly dominated by a particular form of
coffee (Pendergrast, 1999). In Belgium and Germany green coffee is the predominant
form of coffee re-exported, while roasted coffee is the predominant form of coffee re-
exported by Italy, Poland, Sweden and the United States. For the other importing
countries soluble coffee is the main form of coffee re-exported, more particularly by
Japan, Spain and the United Kingdom.
The coffee industry
The coffee industry currently has a supply chain which is the sequence of
activities and processes required to bring the coffee from its raw state to the finished
goods sold to the consumer. This chain is often complex, and varies in different
countries but typically includes the following categories: producers, middlemen,
processors, government agencies, exporters and importers, dealers, roasters and
More specifically, producers (or growers) usually work on a very small plot of
land of just one or two hectares. Many of them do some primary processing, such as
drying or hulling, themselves. Processors are individual farmers who have the
equipment to process coffee.
Middlemen (or intermediaries) may be involved in many aspects of the supply
chain. They may buy coffee at any stage between coffee cherries and green beans,
they may do some of the primary processing, or they may collect together sufficient
quantities of coffee from many individual farmers to transport or sell to a processor,
another intermediary, or to a dealer. Notice that there may be as many as five
intermediary links in the chain.
Afterwards, there are the government agencies. In some countries the
government controls the coffee trade, perhaps by buying the coffee from processors at
a fixed price and selling it in auctions for export. Exporters: they buy from
co-operatives or auctions and then sell to dealers. Their expert know-ledge of the local
area and producers generally enables them to guarantee the quality of the shipment.
While middlemen exporters purchase coffee directly from small farmers
(Ubilava, 2012; ICO, 2014), large coffee estates and plantations often export their
own harvests or have direct arrangements with a transnational coffee processing or
distributing company. Under either arrangement, large producers can sell at prices set
by the New York Coffee Exchange. Green coffee is then purchased by importers from
exporters or large plantation owners. Importers hold inventory of large container
loads, which they sell gradually through numerous small orders. They have capital
resources to obtain quality coffee from around the world, capital normal roasters do
not have. Roasters' heavy reliance on importers gives the importers great influence
over the types of coffee that are sold to consumers (Pendergrast, 1999).
Dealers or brokers supply the coffee beans to the roasters in the right
quantities, at the right time, at a price acceptable to buyer and seller. Finally, roasters
are experts to turn the green coffee beans into products people enjoy drinking. The
company also adds value to the product through marketing, branding and packaging
Coffee reaches the consumers through cafes and specialty stores selling
coffee, of which, approximately, 30% are chains, and through supermarkets and
traditional retail chains. Supermarkets and traditional retail chains hold about 60% of
market share and are the primary channel for both specialty coffee and non-specialty
coffee. Retailers (or sellers) of coffee products range from large supermarkets, to
hotel and catering organizations, to small independent retailers.
On the whole, the world coffee consumption is thriving. Although many
traditional markets are growing only modestly, with consumption appearing to have
reached saturation point in many countries, there still exist several dynamic niche
opportunities for producers to benefit from, such as specialty and certified coffees.
Furthermore, over the last 50 years, the perception of coffee as a healthy product has
changed significantly, with a corresponding increase in the amount of scientific
research into the positive health properties of coffee.
Consumption in emerging markets and exporting countries has been growing
rapidly and shows strong potential for further growth. Exporting countries whose
economic prospects are favourable to increased coffee consumption include Brazil,
Indonesia, India and Mexico. Living standards in some coffee-exporting countries
continue to improve, creating a strong potential for growth in domestic coffee
consumption given an expanding middle class.
Emerging markets are found in newly industrialized countries which have
experienced considerable economic and social development. As a group, these
emerging markets have recorded considerable growth in coffee consumption, with an
increase from 10.2 million bags in 1990 to 27.9 million in 2012, representing an
average annual growth rate of 4.7%.
Figure 5: World coffee consumption by type of market (1964-2012)
The ICO and the price indicators for coffee
The coffee prices which are used in empirical studies are mainly indicator
prices and come from the International Coffee Organization (ICO) database.
Before continuing to the construction of the price indicators, let us explain a
few things for the ICO and its mission. The International Coffee Organization is the
main intergovernmental organization for coffee, bringing together exporting and
importing governments in order to tackle the challenges facing the world coffee sector
through international co-operation. Note that its member governments represent 95%
of the world coffee production and 83% of the world consumption. The ICO’s
mission is to strengthen the global coffee sector and promote its sustainable expansion
in a market-based environment for the betterment of all participants in the coffee
industry. The ICO was set up in London in 1963 under the auspices of the United
Nations because of the great economic importance of coffee. It administers the
International Coffee Agreement (ICA), an important instrument for development
co-operation. The latest agreement, the ICA 2007, entered into force on 02/02/2011.
The price indicators for coffee have been defined in the International Coffee
Agreement. More particularly, indicator prices concern spot prices of four main
groups consisting of similar types of coffee. This subdivision results in the following
groups: Unwashed Arabicas (mainly Brazilian coffee), Colombian Milds (mainly
coffee from Colombia), other Milds (mainly coffee from other Latin American
countries) and Robustas (mainly African coffee).
The International Coffee Organization, which is the executive organization for
the International Coffee Agreement, computes the indicator prices for these groups of
coffee types, according to its regulations. Therefore, these indicator prices are based
on the daily spot prices of the relevant types of coffee originating from the trade in the
New York market. It is worth-mentioning that, although all the prices concern coffee,
these coffee types are different. They have tastes which are not equally appreciated,
and they are also processed in different ways into different final products, such as
roasted and soluble coffee.
The price which determines when export quota become effective is the
“composite indicator price” (CIP), which is computed as an average of the indicator
prices. The composite indicator prices have been defined in different ways during the
history of the ICA. For example, the composite indicator price of 1968 (CIP-‘68) is an
average of the indicator prices of all the four coffee types, whereas the one of 1976
(CIP-‘76) is the average of only the indicator price of Other Milds and Robustas. Such
a change was made because the prices of Other Milds and Robustas were considered
to represent the world market prices of coffee more accurately at that time. Brazil and
Colombia often made use of discounted prices and 'special deals' with roasters in
Relationship between fundamentals and prices
The main factors determining coffee prices are production, consumption and
stock movements, as well as any exogenous elements which may change the impact
of these factors on price formation. Correlations between these fundamental market
factors and coffee prices are explored below. Figure 6 shows developments in supply,
production, stocks and consumption compared to the ICO1
composite indicator price
during the period 1964 to 2012.
ICO stands for International Coffee Organization.
Figure 6: Relationship between market fundamentals and coffee prices
Global coffee supply is made up of world production and stocks, comprising
opening stocks in exporting countries and inventories held in importing countries. As
is the case of most agricultural commodities, coffee is subject to considerable
variations in production that are attributable to agricultural and climatic conditions.
Too little or too much rainfall, for instance, can affect the volume of production from
one crop year to the next. Increases in the costs of fertilizers and labour can also limit
their use, leading to reduced production. Correlation coefficients between production
and coffee prices are not very significant, however there is a negative correlation
observed between world stocks and prices. In other words, low stock levels generally
entail high prices (Pendergrast, 1999).
It is aforementioned that up until 1989, the coffee market was regulated by a
series of International Coffee Agreements which were intended to manage supply and
maintain price stability. This system subsequently collapsed, and since 1990 the
coffee market has been subject to the free market forces of supply and demand.
More specifically, prices levels during the regulated market period (1965 to
1989) were relatively high since both upward and downward trends were corrected
through the application of export quotas. Under this system, export quotas were in
effect during the periods between 1963 and September 1972, from October 1980 to
February 1986 and from November 1987 to July 1989.
The free market period beginning in 1990 had two sub-periods of markedly
low price levels: 1989 to 1993 and 1999 to 2004. The latter sub-period recorded the
longest period of low prices ever recorded, known as the coffee crisis, with severe
negative consequences on the coffee economies of exporting countries. Prices
recovered strongly after 2004, reaching a 34-year high in mid-2011. However, there
has subsequently been a severe deterioration in prices while costs of coffee production
inputs, particularly fertilizers and labour, continue to rise.
Figure 7: ICO composite indicator price (monthly averages: 1965-2013)
However, price volatility can be a major source of uncertainty and
vulnerability for coffee producers. Investment and planning decisions have to be made
based on the information available at the time, but unforeseen changes in prices can
undermine these choices, often with detrimental consequences (Donnet et al., 2008).
Coffee prices are determined day-to-day on the world commodity markets in
London and New York. However, there are circumstances in which farmers can
receive more than the market price, for example if the quality of their coffee is high, if
they undertake some or all of the processing stages which someone else would
otherwise be paid to do, or if they can sell direct to a manufacturer rather than to
Farmers can also reduce their costs if they are able to share processing and
transportation facilities with other farmers. Coffee farmers may sell their coffee in a
number of ways, such as selling to the next link in the traditional supply chain, selling
to government agencies in countries where the coffee trade is government controlled,
although this is becoming less common, or, selling direct to a manufacturer.
Over the last 50 years, world coffee prices have varied enormously, from less
than 50 US cents/lb to over 300 cents. In the free market period since 1990,
small-holder farmers in many countries have been more exposed to fluctuations in
coffee prices, as the internal regulatory mechanisms in producing countries were
predominantly dismantled. These price fluctuations have increased rural poverty as it
became difficult for small producers to efficiently plan their resource allocations. As a
result, risk management strategies are becoming increasingly recommended to
producers in developing countries; however the scope and applicability of these
instruments can vary significantly.
Price volatility is just one side of the economic sustainability equation.
Changes in production costs over time can severely affect a producer’s ability to make
a sustainable living from their coffee crop. The main components of production costs
for coffee producers are labour, fertilizers and phytosanitary products such as
pesticides (Donnet et al., 2008).
Labour costs are one of the major limiting factors for the development of
coffee production. In many countries, one of the explanatory reasons for declining
production levels is the ageing agricultural population, and the lack of youthful
workers to replace it, due largely to urban migration from rural areas. Coffee is a
labour-intensive crop, with little mechanization in many producing countries, and
global urban wages are generally increasing.
Spot and future markets
Coffee is an important source of export revenue for many Less Developed
Countries (LDCs) and the underlying asset for the largest futures markets in soft
. World Bank and IMF policy has often directed developing countries to
invest in the production of cash crops and the exports of many developing nations are
often concentrated upon a limited number of primary commodities (Milas et al., 2004;
Fry et al., 2010). Significant amounts of trade occur on the associated futures markets,
making them important subjects of research in their own right. However, as discussed
in Morgan et al. (1999), coffee sector participants in LDCs may have restricted access
to such markets. The question of how futures’ market trading activity affects the
volatility of coffee spot prices is therefore important in both economic and welfare
terms. Of particular interest is how speculative behaviour in futures markets affects
the associated spot markets. Also of interest is the wider question of the
interdependence of spot and futures markets, with a range of theoretical and empirical
work suggesting that causality may also run from spot markets to futures markets (Fry
et al., 2010).
Futures markets present an opportunity for participants to hedge their exposure
to variations in the value of the underlying asset. In this regard they provide a
valuable economic function. Transparent price discovery and price dissemination are
key features of successful futures markets (Hall et al., 2006). Even speculation in
futures markets may thus provide value in driving market efficiency and depth.
However, there remains the possibility that futures’ market trading activity
contributes to volatility in the price of the underlying asset (Yang et al., 2005). These
effects may be further compounded if some of the participants in the spot market,
particularly those in LDCs, do not have easy access to futures markets and the
hedging opportunities that they present.
The soft commodities are made up of cocoa, coffee, cotton, corn, wheat, soybean, fruit, orange juice
and sugar. The term generally refers to commodities that are grown rather than mined.
However, the interplay between spot and futures markets may exhibit richer
behavior and may not be restricted to a unidirectional flow of information from
futures markets to spot markets (Fry et al., 2010).
Fair Trade Coffee
From the aforementioned paragraphs, it becomes obvious that coffee is a very
popular product, with a high consumption, and it can be profitable for both the
producers and the sellers. However, the existence of so many middlemen standing
between the producer and the consumer, cannot ensure that coffee growers receive a
fair reward for their labour.
Fair Trade has its origins in an initiative started in Netherlands by a
church-based NGO in 1988 in response to low coffee prices. The stated aim of the
initiative was to ensure growers were provided “sufficient wages”. The NGO created a
fair trade label for their products, Max Havelaar, after a fictional Dutch character who
opposed the exploitation of coffee pickers in Dutch colonies (Dragusanu et al., 2013).
Over the next half decade, Max Havelaar was replicated in other European
countries and North America, and similar organizations, such as TransFair, emerged.
In 1997, the various labeling initiatives formed an umbrella association Fair Trade
Labelling Organization International (FLO) along with three other organizations
(including TransFair). The Fair Trade Certification mark was launched in 2002.
The stated goal of Fair Trade is to improve the living conditions of farmers in
developing countries. In practice, this is accomplished through two primary
mechanisms: a guaranteed minimum price for coffee sold and a price premium that is
paid. Both are set by Fair Trade Labeling Organization.
The minimum price is meant to cover the average costs of sustainable
production, and acts as a price floor that reduces the risk faced by coffee growers. Fair
Trade buyers must pay producers at least the minimum price when the world price is
lower, and must pay the higher price when world price is above the FT minimum
price (Dragusanu et al., 2013).
According to the World Fair Trade Organization and the other three major Fair
Trade organizations (Fairtrade Labeling Organizations International, Network of
European Worldshops and European Fair Trade Association), the definition of fair
trade is “a trading partnership, based on dialogue, transparency and respect, that
seeks greater equity in international trade”. The stated goal is to offer better trading
conditions to marginalized producers and workers. Fair trade organizations, along
with the backing of consumers, campaign for change in the rules and practice of
conventional international trade. However, fairtrade certification is not free; there is
an application fee, initial certification fee, membership dues, annual audit fees and
more. Certification can cost thousands of Euros for a single plantation (Wilson and
As it has become obvious, coffee is a very important product for international
trade. Moreover, it is the second traded product next to oil and coffee consumption
has been increasing over the past 50 years (Pendergrast, 1999). However, these
increases in the consumption side have not been accompanied by corresponding
increases in raw coffee prices and improvements of coffee producers’ livelihood.
After the International Coffee Association (ICA) dissolved in 1989, coffee production
increased substantially, mainly due to mass production by Brazil and the entry of new
coffee producers in Asia and Africa. Moreover, the world coffee prices fell by 50%
(Giannakas and Omidvar, 2015).
The general trend seemed to be the following: while farm-gate prices have
been declining during the most of the past 25 years, the prices in consuming countries
have been soaring and so have the profits of middlemen. These trends have reduced
further the producer welfare and have pushed many coffee producers into poverty
(Giannakas and Omidvar, 2015). Therefore, it became obvious the necessity for a Fair
Trade method in the coffee industry.
The guaranteed premium for coffee sold as Fair Trade must be set aside and
invested in projects that improve the quality of life for producers and their
communities. The specifics of how the premium is used must to be decided upon in a
democratic manner by the producers themselves. Potential projects that could be
funded with the Fair Trade premium include the building of schools and health
clinics, offering instruction courses for members of the community, provision of
educational scholarships, investments in community infrastructure, improvements in
water treatment systems, conversion to organic production techniques, etc. Since
2011, five cents of the premium must be invested towards improving the quality and
productivity of coffee (Dragusanu et al., 2013).
As it has already been discussed, for coffee to be sold under the Fair Trade
mark, all actors in the supply chain, including importers and exporters, must obtain
Fair Trade certification. On the production side, the certification is open to small
farmer organizations and cooperatives that have a democratic structure, as well as
commercial farms and other companies that employ hired labor (Fair Trade
The certification entails meeting specific standards that are set and maintained
by Fair Trade Labeling Organization (FLO). An independent certification company
FLO-CERT, which split from Fair Trade Labelling Organization International in
2004, is in charge of inspecting and certifying producers (Fair Trade Foundation,
These Fair Trade compliance criteria focus on the social, economic and
environmental development of the community. In terms of social development, the
producer organization must have a democratic structure and transparent
administration in place, and must not discriminate against its members. To satisfy the
economic development criteria, organizations need to be able to effectively export
their product and administer the premium in a transparent and democratic manner.
The environmental development criteria are meant to ensure that the members work
towards including environmental practices as an integral part of farm management, by
minimizing or eliminating the use of certain fertilizer and pesticides and replacing
them with natural, biological methods, as well as adopting practices that ensure the
health and safety of the cooperative members and the entire community (Fair Trade
In the case of commercial plantations that employ a large number workers, the
Fair Trade standards entail that hired workers are not children or forced workers, and
are free to bargain collectively. Hired workers must be paid at least minimum wage in
the respective region, and must be given a safe, healthy, and equitable environment
(Fair Trade Foundation, 2012).
The world coffee market has undergone a significant transformation over the
last half century, as we described above. However, world production grew steadily
over the last 50 years despite climatic shocks. But it will be difficult to maintain this
trend mainly on account of the continued rise in production costs as well as problems
related to pests and diseases which could affect this steady growth in production.
Moreover, climate change could also have a negative impact on production in many
countries unless urgent research is carried out on adaptation measures.
On the other hand, prospects for growth in world coffee demand remain
promising, mainly in emerging markets and exporting countries, in addition to the
expansion of niche markets in traditional consuming countries. This growth in
consumption should help to maintain a tight balance between supply and demand. In
addition, the development of a processing industry in exporting countries could enable
them to increase added value in the coffee sector.
It is worth noting that the barriers to increased coffee consumption in many
exporting countries and emerging markets are related mainly to cultural and economic
factors, as per capita GDP is generally low and coffee is considered as a luxury good
in many countries. As incomes rise, consumption of coffee will likely become more
prevalent and the vast potential of these emerging markets and exporting countries
will be increasingly fulfilled. Finally, sustainable development of the coffee economy
needs to be based on actions designed to promote a consistent balance between supply
and demand that is remunerative to growers.
CHAPTER 2: Literature review
In the following pages, we discuss the literature review concerning two major
subjects. In the first part, we discuss the application of copula in agricultural
economics. This is an interesting point in our study because it can shed light on the
importance and the suitability of this methodological tool to our empirical research. In
the second part, we focus on the empirical studies concerning the price dependence of
coffee. From this literature review, we aim to find relevant studies in order to
highlight their advantages and disadvantages and show our intuition and novelty in
this branch of literature.
Copula in agricultural economics
As we have already mentioned in the previous parts of the present dissertation,
copulas are a way of formalizing dependence structures of random vectors. Although
they have been known about for a long time, it seems that they have been
rediscovered recently in applied sciences, such as biostatistics and biology.
The use of copulas in the economics literature is more recent and most
empirical applications are found within the financial economics field (Parra and
Koodi, 2006; Patton, 2004). Copulas allow flexible characterization of dependence
between random variables, being especially useful if no obvious choice for the
multivariate density function exists.
Financial analysts began to take a strong interest in copulas as a result of the
financial crisis of 2008. Hence, they began to ask whether stock returns are more
highly correlated during financial crises than in normal times, thus rendering stock
portfolios riskier than predicted by conventional asset pricing models. The questions
being addressed by financial analysts are analogous to those that must be addressed in
index insurance design: in both cases, one is concerned with the degree of dependence
exhibited by two or more random variables at the extremes of their distribution.
Copulas provide a formal framework for addressing questions such as these.
But financial economics is not the unique field of economics, where copulas
can be applied. The agricultural sector usually faces a combination of risks rarely
found in other enterprises. Two of the major risks affecting agriculture are climatic
and natural risks that influence agricultural yields, and market risks that may lead to
agricultural price fluctuations. Recent dismantling of public commodity price
stabilization mechanisms leading to increased dependence of prices on global markets
may have increased price risk (Gilbert and Morgan, 2010; Anton and Kimura, 2011).
Since the 2008 global, financial crisis, changes in both food price levels and
volatility became more the norm than the exception. According to the European
Commission (2011), the food price volatility is likely to persist in the upcoming years.
This statement has renovated the interest of how the changing conditions occurring in
one market of a commodity can be transmitted by the mechanism of prices to the
other commodity market.
Emmanouilides et al. (2014) apply the copulas in order to study the price
dependence in the three principal EU olive oil markets, Spain, Italy and Greece. As
we have already mentioned, recent dismantling of public commodity price
stabilization mechanisms, leading to increased dependence of prices on global
markets may have increased price risk (Gilbert and Morgan, 2010; Anton and Kimura,
2011). Besides, the analysis of spatial price interrelationships enable researchers to
assess whether geographically separated markets are segmented, meaning
regionalized, or globalized, meaning integrated. At the EU level, the European
Commission has pursued vigorously the goal of the integration of national markets
over the last 30 years through a number of economic initiatives, policies, and
programmes. In this context, one of the objectives of their empirical study has been to
investigate price interrelationships in the principal EU olive oil markets. This
objective has been pursued using copulas.
The empirical data are monthly olive oil prices and they consider the prices of
two quality differentiated olive oil grades; extra virgin and lampante. According to the
relevant empirical results, over the period 2002 to 2012 there has been a variety of
degrees and intensities of price co-movement in the three geographically separated
markets. Depending on the market pair and the olive oil grade considered, Kendall’s
tau indicated that the tendency of prices to co-move was not particularly high. This
result can be, to a certain extent, attributed to the presence of transaction costs in
international trade. These tend to differ among countries because they use non
tradable inputs, which may create a wedge.
Therefore, measures of overall price dependence are not expected to take very
large values. In two out of the six market pairs studied, tail dependence was
symmetric, while for the other four pairs tail dependence was asymmetric; prices
boomed together but they did not crash together. This was especially true for prices in
Italy and Spain, which are by far the most important players in the EU olive oil
market. The relatively low overall degree of dependence, and, more importantly, the
evidence in favor of asymmetric price co-movements, suggests that the three principal
EU olive oil markets cannot be thought of as one great pool. Last but not least, it is
important to note that the Gaussian copula family turned out to be irrelevant for all six
price pairs analyzed. This implies that empirical investigations relying on the
assumption of multivariate normality for the price processes in the major EU olive oil
markets could probably suffer from misspecification.
Moreover, Emmanouilides and Fousekis (2014a,b) extended their empirical
research with copula to other fields of agricultural economics, too. More specifically,
they study the vertical price transmission in the US pork industry (Emmanouilides and
Fousekis, 2014a) as well as the vertical price dependence structures in the beef supply
chain in the USA (Emmanouilides and Fousekis, 2014b). These papers belong to the
widely-defined range of agricultural economy, because livestock is another important
parameter of agriculture and the primal economic sector of production.
In the first paper (Emmanouilides and Fousekis, 2014a), they investigate the
degree and the structure of price co-movement in the US pork industry. They choose
study this issue by using copulas, because this is a flexible statistical approach in
order to analyze price dependence and co-movement, especially during extreme
market upswings and downswings. Notice that in well-functioning (or integrated)
markets price shocks in any market level are transmitted to other market levels;
primary producers benefit from price increases at the wholesale and the retail levels
and final consumers benefit from cost reductions upstream.
In this research, they employ bivariate copulas and they use monthly prices at
three market levels (farm, wholesale, and retail) over the period 1970 to 2012 and
their mail empirical results indicate that the appropriate copula models for the US
pork industry differ across time periods and across markets. For instance, the
existence of asymmetric price transmission and low degrees of price dependence
(especially in the second half of the sample for the pair wholesale-retail) raise
concerns about the efficiency and the distribution of benefits among the stakeholders
of the pork supply chain in the US. Therefore, they conclude on the need for further
research on this topic.
In the second paper (Emmanouilides and Fousekis, 2014b), the empirical
question remains the same, though the authors change the product under study. More
specifically, they examine with copula the degree and the structure of price
dependence along the beef supply chain (farm, wholesale and retail) in the US
economy. For the needs of the study, they use monthly rates of price changes over the
period 2000-2013. The analysis considers two pairs of markets, namely the pair
farm-wholesale and the pair wholesale-retail. According to their empirical results, it
turned out that price co-movement for the first pair is relatively stronger and it is
described with the Gumbel-Clayton copula, while that for the second pair is rather
weak and it is described by the Gumbel copula. The empirical findings point to the
existence of price transmission asymmetry, which is much more important for the pair
On conclusion, we notice that Emmanouilides, Fousekis and Grigoriadis apply
the methodological tool of copula on a wide range of empirical questions in the
agricultural economics. They apply the copula in both horizontal and vertical price
More specifically, they investigate the price dependence and structure in
the three principal EU olive oil markets, namely Spain, Italy and Greece, which is a
horizontal analysis of price dependence, as well as in the study of US agricultural
economy in the sector of livestock, namely pork and beef, which consists a vertical
analysis of price dependence. In the first case, we are interested in the price
dependence of a product between two or more countries, whereas in the second case,
we are interested in the price dependence and structure of a product in its supply chain
within a country.
More recently, Panagiotou and Stavrakoudis (2015) employ copulas in
order to examine the degree and the structure of price dependence between different
cuts in the US pork industry. On contrary to the study of Emmanouilides and Fousekis
(2014a) who apply the copula for the pork US industry at three market levels (farm,
In the empirical analysis of the present dissertation, we focus on the price dependence of coffee in a
horizontal basis and then we compare the copulas in order to derive the corresponding economic
intuition and policy implications for the coffee market.
wholesale, and retail) over the period 1970 to 2012, Panagiotou and Stavrakoudis
(2015) apply the copula only at the retail level. For the needs of their study, they
exploit monthly retail data of pork cuts. The relevant retail prices are the prices from
centre cut-bone in chops, boneless chops, and other pork chops. The empirical results
suggest that for all pairs, retail prices are not likely neither to boom nor to crash
together, even though overall dependence is quite considerable for two of the three
pairs considered in their study. This can be an indication that given identifiable
differences between specific pork cuts, sellers at the retail level don’t adopt different
pricing strategies when market conditions change. Finally, the authors find no
evidence of asymmetric price responses between the pairs of pork cuts considered in
On the whole, concerning the livestock and particularly the pork industry,
there are plenty of relevant papers using copula. For instance, Qiu and Goodwin
(2012) estimate asymmetric price transmission in the hog supply chain using copulas,
which is the case of vertical transmission of price dependence and changes. Their
results indicate evidence of symmetry and asymmetry. On the one hand, for the farm
to wholesale and wholesale to retail marketing chains, shocks in one market would
transfer to the other market. On the other hand, for the farm to retail case, retail prices
do not respond to price reduction in farm level prices.
Empirical research in coffee price dependence
It has already been mentioned that coffee is an important commodity and has
become the second largest traded commodity next to oil (Pendergrast, 1999).
Therefore, the coffee industry has attracted the attention of economic researchers.
However, according to our knowledge, there is no any empirical research employing
copulas in order to study price dependence in coffee.
In the following pages, we present the relevant empirical studies and their
main findings according to price dependence in coffee. Actually, the majority of
papers deals with the fair trade on coffee and its impacts on coffee growers or the
coffee price and the exporting performance rather than the price dependence in a more
clear question. Since low-income and developing countries depend mostly on just a
few commodities for the bulk share of their export earnings, commodity price
fluctuations directly affect the incidence of poverty, as the vast majority of the poor
depend on primary commodities for their livelihoods. Finally, many researchers
investigate the relationship between oil products and coffee production as well as the
interrelationship between domestic and international coffee prices.
We cite Dragusanu and Nunn (2014) who estimate the effects of fair trade
certification on coffee producers in Costa Rica. For these reasons, they examine a
panel of all coffee producers between 1999 and 2010 and conclude that fair trade
certification is associated with higher export prices, while they find no evidence that
certification is associated with more sales (either domestic or for export) or with
higher domestic prices.
On the other hand, Branchi et al. (1999) intend to investigate the role of price
policies in explaining international differences in coffee exports. As it is well-known
from the relevant economic theory, national policies cannot be taken as completely
independent of the level of the international price. However, it is likely that
differences in the domestic price policy might have had a recognizable impact on the
long-run production and export trends of the various producers. In fact, variations in
the real exchange rate are expected to influence coffee exports via their impact on the
average competitiveness of each country, while the level of protection captures the
effect of the taxation policy on the domestic supply of coffee.
Therefore, the authors study the impact of price on coffee production and
exports in a selected group of developing countries, with particular focus on a
subgroup of Sub-Saharan countries (SSA). Due to the dependency of coffee producers
on the vagaries of the international market, direct crop taxation and exchange rate
policies in these countries are found to be only partially endogenous. The long-run
impact of policies on producers’ behavior is then tested by means of a cross-country
linear regression model. About one third of cross-country variability in planted areas
is found to be attributable to exchange rate and, to a lesser extent, taxation policies.
However, price policies do not appear to exert any significant impact on yields. Their
results show that, in the case of coffee, the weight of domestic price policies in
determining the performance of traditional export sectors is relevant, but should not
be overrated. Exogenous events, like the vagaries of world markets, on one hand, and
of weather, on the other, still play a major role in influencing the evolution of coffee
exports, as they did in the past.
Quite different is the scope of the following papers. In particular, the purpose
of ICO (2015) is to analyze the nature of the relationship between the price of coffee
and oil products as well as the movement of the US dollar. Oil plays an important role
in modern agriculture. Along with fuel for transport and agricultural machinery,
plastic materials, nitrogen fertilizers and pesticides require a high consumption of
hydrocarbons. Fertilizers are effectively substances, which, through soil enrichment,
provide the plants with nutritive food supplements, which boost their growth and
productivity. Moreover, the development of the oil industry has increased the use of
fertilizers. Natural deposits of phosphate and potassium contribute equally to the
development of fertilizers. A number of coffee exporting countries use fertilizers in
order to improve productivity. Three main nutritive elements are used in coffee
growing: nitrogen fertilizers, potassic fertilizers, and phosphate fertilizers.
Thus, the research which was held by ICO (2015) has a twofold target: on the
one hand, to conduct a comparative analysis between the price of coffee and the price
of oil products, and on the other hand, to analyze the relationship between coffee
prices and the US dollar exchange rate in relation to the national currencies of
selected coffee exporting countries. The methodology of the study is based on the use
of statistical tests, apart from copula, and linear regression analysis in order to
determine the relationship between average monthly prices of coffee and oil products,
and also the relationship between coffee prices and the exchange rate between the US
dollar and national currencies of the selected exporting countries. They cover the
period from January 1990 to December 2014, which corresponds to the era of free
trade following the abandonment of the regulated quota market system that was in
place the previous decades. The reference price of coffee are the prices on the New
York and London futures markets, specifically, the average of the second and third
positions. Indeed, futures prices react very quickly to new information relating to
physical prices. On the other hand, prices in physical markets react with delay, for
transactions take time to become effective.
Apart from the aforementioned dependence from oil products, agricultural
production suffers from increased price volatility, too. Since 2000, the 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.
Although supply and demand fundamentals played a significant role in the food crisis
More recently, Hernandez et al. (2015) examine the degree of
interdependence and volatility transmission between international and domestic price
returns to assess whether Ethiopian coffee prices have become more integrated to
world prices in recent years. Their study becomes interesting particularly after the
implementation of the Ethiopian Commodity Exchange (ECX) in the late 2008. For
the needs of the analysis, the authors focus on monthly prices of the five major coffee
varieties in Ethiopia and implement a GARCH approach to derive dynamic
conditional correlations and evaluate volatility transmission from international to
The estimation results highlight the lack of a higher integration between local
and world markets in recent years, measured through interdependencies in prices.
More specifically, they indicate that not all coffee regions have become more
interrelated with the world markets after the implementation of the ECX. On the one
hand, despite the general increase in producer coffee prices after 2008, only few
regions show a higher interdependence between local and international price returns.
On the other hand, volatility spillovers from international to domestic prices have also
not increased in the past years. Finally, according to their empirical results, the ECX
does not seem to have produced a major shift in the dynamics of domestic coffee
price returns, at least not in the following months after its implementation.
Another empirical contribution examining the price transmission between
international and local markets and the effect to the producer price is the one of Xile
Li and Sayed Saghaian (2013). They employ a vector error correction model in order
to measure the interdependence between producer and world prices. Their empirical
results show that there is some degree of co-integration between the producer and
world prices in Vietnam. Moreover, the short-run price transmission is asymmetric for
Robusta. They conclude that the quality and the government intervention affect the
results of the price link.
On the other hand, Ramirez et al. (2014) apply a non-structural approach to
analyze coffee price returns behavior over time. Particularly, they analyze the
non-linearities in the return of Colombian Arabica coffee price. In their study, they try
to understand, among others, the dynamics of the price of Colombian Arabica coffee
over time, so as to design a derivative market to hedge the exposure to price
fluctuations. For the needs of the empirical study, they apply GARCH models.
According to their results, there are some interesting policy recommendations. For
instance, they conclude on the increasing competition in the international coffee
markets, where the numbers of producers, players and wholesalers are limited.
Copula had not been widely used in agricultural economics, although
nowadays this methodology is getting familiar to researchers in fields of agricultural
economics too. However, to our knowledge, copulas have not yet been applied in
studies for price dependence in coffee.
In this chapter of the present dissertation, we presented the relevant literature
on two categories. In the first category, we focused on the use of copula in the
agricultural economics. Therefore, we discussed the price dependence patterns on
olive oil, oranges and apples on the European market, as well as on livestock, as pork
and beef in the US economy. The main result can be the recognition of copulas as a
useful tool in empirical studies for the economy and with many applications in the
financial economics, agricultural economics, etc.
In the second category, we focus on the empirical studies for coffee. The
reason lies on our empirical question for price dependence in coffee. Thus, we will to
discuss the preceding papers on the structure and pattern of coffee prices and the
possible dependence on them. We should notice that the relevant literature focuses
mainly on the Fair Trade Certifications for coffee, the dependence from oil prices, the
importance of coffee as the major exporting product especially for poor and/or
developing countries as well as the relationship between domestic and international
coffee prices. Unfortunately, we found only a few papers on price dependence. As a
result, the following empirical study of the present dissertation aims to fill a two-fold
gap in the literature. On the one hand, we aspire to contribute to the empirical analysis
for the price dependence of coffee, where we locate a gap and that the researchers
have not dedicated the appropriate interest for that issue. On the other hand, we differ
from the preceding literature, because to the best of our knowledge, we are the first
who employ copulas in studying the dependence pattern of coffee prices. By utilizing
this methodological tool, we are able to shed light on this issue through a more
Consequently, we derive the conclusion that the methodology of copula fits
many fields of study and can highlight different points of explanations and economic
intuition. According to our main question, it is useful in agricultural economics and it
starts to be used by many researchers for a variety of empirical issues. This may solve
some of the empirical questions that remain in the literature.
CHAPTER 3: Methodology analysis
Copula procedure in modeling dependence
Copula dates back to Sklar’s (1959) theorem, according to which a
multivariate distribution of a vector of random variables is fully specified by the
individual marginal densities and a joining function known as copula.
More specifically, univariate distributions of economic time-series are usually
found to be characterized by excess kurtosis, skewness and non-normality. According
to Patton (2006), it has also been found that related price series may show asymmetric
dependence, which is an indicator of multivariate non-normality.
Based on the Sklar’s (1959) theorem, an n-dimensional joint distribution
characterizing dependence of n economic variables can be decomposed into n
univariate distributions and a copula function. The latter fully describes the
dependence structure between the variables under study. In other words, copula
functions allow a flexible dependence structure between the n random variables and
are specially suited when no obvious choice for the multivariate density exists (Serra
and Gil, 2012).
Here, we will present the methodological tool of copula for a two-dimensional
and bivariate case4
Let 1 2( , )X X X be a random vector with distribution function 1 2( , )H x x ,
where ( )i iH x are the marginal distribution functions of iX (i=1,2). Utilizing the
Sklar’s (1959) theorem, there exists a copula function
: 0,1 0,1C , such that for
1 2( , )x x R it holds that
1 2 1 1 2 2( , ) ( ( ), ( ))H x x C H x H x (1)
As long as the marginal distributions are continuous, there is a unique copula
function 1 1
1 2 1 21 2( , ) ( ( ), ( ))C u u H H u H u
, where 1
( )iiH u
are the inverse distribution
functions and they are quantiles (probabilities) of the uniform distribution
function 0,1U .
For the sake of simplicity, we consider the bivariate case. However, the analysis can be extended to a
p-variate case with p>2. In general, a copula is a multivariate distribution function on [0, 1]p
Hence, the joint density function associated with C is
1 2 1 21 2
1 2 1 1 2 2 1 1 2 21 2
( ( ), ( )) ( , )
( , )
( ( )) ( ( )) ( ) ( )
h H u H uC h x x
c u u
u u h H u h H u h x h x
where h is the joint density function of H, and hi is the marginal density function of Hi
(recall that i=1,2).
We should remind that copula is a function that ties together two marginal
probability functions, as we have shown above.
Rewriting the previous relationship, (2), we have that
1 2 1 1 2 2 1 1 2 2( , ) ( ( ), ( )) ( ) ( )h x x c H x H x h x h x (3)
In general, a joint probability density function of two random variables (X1
and X2) contains information on the marginal behavior of these variables and,
moreover, on the dependence between them. In 1 1 2 2( ( ), ( ))c F x F x each random
variable is fed on its own distribution function. As a result, all information contained
in the marginal distribution functions is swept away and what is left in c and C is pure
joint information about X1 and X2.
Hence, form relationship (3) we derive that the copula function fully
characterizes the co-movement (dependence structure) of the random variables by
capturing the information missing from the marginal distributions to complete the
joint distribution (Meucci, 2011, Emmanouilides et al, 2014).
The inverse of Sklar’s (1959) theorem holds and it means that given H1 and H2
and any copula function C, the function H of relationship (1) defines a valid joint
distribution function with margins H1 and H2.
The next step is to define the appropriate measures for dependence. A standard
rank-based measure of functional dependence is Kendall’s tau, defined as
N N N
P Q P
N N N
where N is the total number of observations, and PN and QN are the number of
concordant and discordant pairs, respectively. Also we deploy here and the theoretical
version of Kendall’s tau, defined as
1 2 1 2
4 ( , ) ( , ) 1C u u dC u u
Usually, economists, managers and policy makers are more interested in the
dependence at the tails (the lowest, λL, and the highest, λU, ranks). The tail dependence
concerns with the study of co-movement between extreme values. The relevant
dependence coefficients, , 0,1U L , are defined as
1 2 ( , )
lim ( | ) lim
u C u u
prob U u U u
( , )
lim ( | ) limL
C u u
prob U u U u
In order to have upper or lower tail dependence, λU and λL need to be positive.
Otherwise, there is upper or lower tail independence. Therefore, the two
aforementioned measures of tail dependence provide information about the likelihood
for the two random variables to boom and to crash together, respectively. Note that
since λU and λL are expressed via copula function, certain properties of copulas apply
to tail coefficient as well (Emmanouilides et al, 2014; Panagiotou and Stavarkoudis,
The most commonly used families of bivariate copulas are the Gaussian and
t-Copula belong to the elliptical family of copulas, while Clayton, Gumbel, Frank,
Gumbel-Clayton and Joe-Clayton belong to the Archimedean family.
Concerning the elliptical family, the Gaussian involves a single dependence
parameter, ρ (the linear correlation coefficient corresponding to the bivariate normal
distribution). The t-copula involves two parameters, the correlation coefficient ρ and
the degrees of freedom (denoted as v). When v≥30 the t-copula collapses to a
The Clayton, the Gumbel, the Frank, the Gumbel-Clayton, and the
Joe-Clayton are members of the family of Archimedean copulas. The first three
contain a single dependence parameter (denoted as θ) while the last two contain two
dependence parameters (denoted as θ1 and θ2).
The selection among the aforementioned seven alternative copula families,
presented above, is carried out in the following way. In a first step, the Akaike
(AIC) and the Bayesian Information Criterion6
applied to each family:
In the case where the two criteria are in agreement, the selection process is
terminated with the appropriate (the best fitting) model being the one that gives
the lower AIC (and BIC) value.
In the case of disagreement, the goodness of fit of the two competing models is
further assessed using the rank-based versions of the Kolmogorov-Smirnov (KS)
and the Cramer-von Mises (CvM) tests.
Apart from the above criteria, for the selection of the appropriate copula model we
can use the Clarke test(2007). For the purpose of our study we will stand on this
The next step is to present the empirical version of the copula function.
Let the pair of ranks (R11, R21), …, (R1i, R2i), …, (R1N, R2N), where R1i stands
for the rank of observation X1i in the random vector X1 and R2i stands for the rank of
observation X2i in the random vector X2.
The copula data are derived by normalizing the ranks by a factor
The empirical copula is defined as
( , )
l u u
C u u
with domain (0,1)2
, and l(.) is an indicator function. The method of maximum
pseudo-likelihood maximizes the rank-based, log-likelihood function
( ) ln ,
where cθ is the probability density function of a copula with parameter vector θ
(Genest et al, 1995; Genest and Favre, 2007).
The values of the AIC is computed as 1 2
2 ln ( , ) 2
c u u k
, where k is the number of the
estimated dependence parameters.
The values of the BIC is computed as 1 2
2 ln ( , ) ln( )
c u u N k
, where k is the number of
the estimated dependence parameters.
Consistency test for dependence
It is quite useful before estimate a given parametric model using our data, to
test for constancy of the empirical copulas. For this reason we employ the tests of
Busetti and Harvey (2011) which are based on the sample τ-biquantics, for a bivariate
Let a bivariate time series Y=(Y1,Y2). Let also the empirical copula CT(τ,τ)
which gives the proportion of cases both series in pair are less than or equal to
particular quatiles, q1(τ) and q2(τ) respectively. The τ-biquantic is defined as
1 1 2 2 1 1 2 2( ( ), ( )) ( , ) ( ( )) ( ( ))TBIQ y q x q C H y q y q
Where H(.) is an indicator function. The null hypothesis (H0) of constancy, interpreted
as the quantile τ of the bivariate empirical copulas is constant while the alternative
(H1) that the quantile τ of the bivariate empirical copulas exposed to breaks. The BH
test statistic is
( ( ))
( , )(1 ( , ))
T C C
It’s worth mentioning that for the following empirical analysis all estimations
and tests have been carried out using the CDVine package in R (Schepmeier &
Appendix: Clarke test
The Clarke test calculates the distance between two copula models. Let C1 and
C2 two parametric copula models with estimated parameter vectors 1 and 2
respectively. Let also the log differences of their likelihoods dt=ln(C1(ut1 , ut2;
1)/C2(ut1, ut2; 2)), for t = 1,2,…,T. The null hypothesis (H0) defines that the two
models are statically identical. The Clarke test is
B= I(d>0), where I(.) is the
indicator function.. The null is rejected for values of B that are significantly different
(smaller or larger) than T/2.
When the competing copula models are more than two, then a bivariate copula
model is compared with all alternative bivariate copula models under consideration at
a specific significant level. A score +1 (-1) is assigned to a model that is superior
(inferior) to another, no score is assigned to identical models. The best model is the
one that achieves the highest total score (Belgorodski, 2010; Brechmann and
CHAPTER 4: Empirical analysis
Monthly data on coffee prices for two types of coffee, Arabica and
Robusta at the farm gate level were used in this study. The data refer to the
period 2002:1 to 2014:12 and have been obtained from the International
. In order to achieve as many observations as possible,
we select only three producing countries, namely Brazil, India and Ecuador
The reason for that is our intention to study the degree and the structure of
price dependence between Arabica and Robusta and conclude if there is any
substitution effect among them. We suggest that if this is the case, implies
that the prices of the two commodities have a long-run relationship and
shocks occurring in one market will evoke responses in other market. In
other words as the price of one commodity increases following for example
an international shock, demand may shift towards another commodity
resulting higher prices on these as well. Therefore, we choose these three
major producing countries as representatives of the whole coffee market.
However, these are not the first three major producing countries.
According to the ICO, the first producing country is Brazil, followed by
Vietnam, Colombia and Indonesia. For the needs of our empirical
investigation, we need as many as possible observations for both the Arabica
and Robusta coffee. Taking into account these two parameters, we end up
with Brazil, India and Ecuador, because they satisfy both criteria. As a result,
we make the assumption that these three producing countries represent the
global market in a satisfying way, because they produce both species and
provide a wide range of data on prices.
The data for coffee prices are available at the site of International Coffee Organization (ICO)
On the following pages, we analyze and describe our data, in order to have
some basic early results, before proceeding on the econometric analysis. Therefore,
we present some useful descriptive statistics and diagrams, in order to highlight the
details of our data.
Note that the variables are the following: ara.BR stands for Arabica in Brazil,
ara.IN stands for Arabica in India, ara.EC stands for Arabica in Ecuador, rob.BR
stands for Robusta in Brazil, rob.IN stands for Robusta in India and rob.EC stands for
Robusta in Ecuador.
Figure 1: Time-series evolution of coffee prices (raw data)
Figure 1 presents the time-series plot for the raw data of prices. The blue line
corresponds to the prices of Arabica and the red line corresponds to the prices of
Robusta. In all cases, the price of Arabica is higher than the price of Robusta, but the
prices appear to be moving together. However, in 2012 there is a greater and
considerable increase in the prices of Arabica. This pattern seems to be repeated in
2014, though we do not have enough data in order to study further another similar
In Table 1, we present some useful descriptive statistics, in order to highlight
the details of our data considering normality, density, skewness, kurtosis and other
Table 1: Descriptive statistics
Brazil India Ecuador
Arabica Robusta Arabica Robusta Arabica Robusta
156 156 156 156 142 142
number of null 0 0 0 0 0 0
0 0 0 0 14 14
min 24.79 15.46 36.37 16.41 22.68 7.89
max 248.98 125.11 243.38 131.44 246.52 106.82
median 97.36 73.25 113.27 74.53 113.40 67.43
mean 104.60 68.40 118.05 71.97 115.57 60.85
CI of the mean
7.97 4.34 8.51 4.84 10.13 4.59
variance 2540.56 753.05 2897.08 937.24 3726.51 765.90
50.40 27.44 53.82 30.61 61.05 27.67
skewness 0.81 -0.25 0.36 -0.24 0.25 -0.36
kurtosis 0.62 -0.95 -0.45 -1.12 -0.61 -1.05
0.94 0.96 0.96 0.94 0.95 0.94
Note: The prices of coffee varieties are expressed in us cents/lb
In Table 1 are presented some very important and interesting descriptive
statistics. In the first three lines, we verify the fact that these three countries are those
with the most complete dataset. Notice that in only Ecuador we miss 14 observations.
Then we present the minimum and maximum value for these variables, the median
and the mean. Furthermore, we estimate the 95% confidence interval of the mean.
Confidence intervals only assess sampling error in relation to the parameter of interest
(the mean) and show us the range within which 95% or 99% or 99.9% of observations
could be expected to lie. The next two statistics are quite important: variance and
standard deviation. In fact, the standard deviation is simply the square root of the
variance. These are measurements of the spread between observations in a data set.
The variance measures how far each number in the set is from the mean. However,
the usefulness of variance is limited since the units are squared and not the same as
the original data. Therefore, standard deviation is more useful.
The following statistics deal with skewness and kurtosis. Skewness is a
measure of symmetry, or more precisely, the lack of symmetry. If the bulk of the data
is at the left and the right tail is longer, we say that the distribution is skewed right or
positively skewed; if the peak is toward the right and the left tail is longer, we say that
the distribution is skewed left or negatively skewed. More specifically, if skewness is
positive, the data are positively skewed or skewed right, meaning that the right tail of
the distribution is longer than the left. If skewness is negative, the data are negatively
skewed or skewed left, meaning that the left tail is longer. In addition, we can get a
general impression of skewness by drawing a histogram, as we see at the following
If a distribution is symmetric, the next question is about the central peak: is it
high and sharp, or short and broad? We can get some idea of this from the histogram,
but a numerical measure is more precise. The height and sharpness of the peak
relative to the rest of the data are measured by a number called kurtosis. Higher values
indicate a higher, sharper peak; lower values indicate a lower, less distinct peak. The
values for asymmetry and kurtosis between -2 and +2 are considered acceptable in
order to prove normal univariate distribution (George and Mallery, 2010). Recall that
the mean and standard deviation have the same units as the original data, and the
variance has the square of those units. However, the kurtosis has no units: it’s a pure
Finally, we take into account the normality of the variables, using the
Shapiro-Wilk test for normality. In statistics, normality tests are used to determine if a
data set is well-modeled by a normal distribution and to compute how likely it is for a
random variable underlying the data set to be normally distributed. Tests of univariate
normality include D’Agostino’s K-squared test, the Jarque-Bera test, the
Anderson-Darling test, the Cramér–von Mises criterion, the Lilliefors test for
normality (itself an adaptation of the Kolmogorov-Smirnov test), the Shapiro-Wilk
test and the Pearson’s chi-squared test.
In case of the Shapiro-Wilk test, the null-hypothesis is that the population is
normally distributed. Thus, if the p-value is less than the chosen level, then the null
hypothesis is rejected and there is evidence that the data are not from a normally
distributed population. In other words, the data are not normal. On the contrary, if the
p-value is greater than the chosen level, then the null hypothesis that the data came
from a normally distributed population cannot be rejected. However, since the test is
biased by sample size, the test may be statistically significant from a normal
distribution in any large samples. Thus a Q–Q plot is required for verification in
addition to the test, as we see in the following figures of the present dissertation.
Looking at the values of this test, we conclude that the population is not normally
Another interesting topic is the existence or not of correlation. We examine
this subject with the Pearson’s correlation between the two coffee species in a
particular country each time. The correlation for Brazil is 0.86 (p-value=0.000),
therefore we reject the null hypothesis that the true correlation is equal to zero, in
favor of the alternative one. In case of India, the results from the estimation are
similar: the correlation is 0.87 (p-value=0.000) and we conclude that the true
correlation between Arabica and Robusta in India does not equal zero. Finally, the
same holds for Ecuador. Its correlations is 0.94 and the prices of these two species are
Figure 2a illustrates the histograms of the raw data for Arabica and Figure 2b
the respective histograms for Robusta. Similarly, Figures 3a and 3b illustrate the
histograms of the ranks for Arabica and Robusta, respectively. Recall from the
aforementioned analysis of the methodology of copula that ranks are necessary for the
estimation of copula.
This is a very helpful statistical tool. In general, a histogram is a bar graph of
raw data that creates a picture of the data distribution. The bars represent the
frequency of occurrence by classes of data. A histogram shows basic information
about the data set, such as central location, width of spread, and shape. The
histogram’s shape and statistical information help us decide how to improve the
system. If the system is stable, you can make predictions about the future performance
of the system. After improvement action has been carried out, we may continue
collecting data and making histograms to see if the theory has worked.
Some of the aforementioned descriptive statistics, such kurtosis and skewness,
can help us interpret the histogram and can show us whether the data distribution is
normal or not. In particular, the histogram should be flat for a uniform sample, but the
visual perception varies depending on whether the histogram has, for example, 15, 10,
5, or 3 bins.
Figure 2a: Histograms of raw data for Arabica
Figure 2b: Histograms of raw data for Robusta
As if has been mentioned, the flatter the histogram, the more uniform the
sample. Therefore, Ecuador presents the more uniform picture compared with the
other two countries.
Figure 3a: Histograms of ranks for Arabica
Figure 3b: Histograms of ranks for Robusta
Histograms are the usual tool for representing medium sized data distributions
graphically, but they suffer from several defects. The kernel density estimate is an
alternative computer-intensive method, which involves smoothing the data while
retaining the overall structure. It is a good method of reconstructing an unknown
population from a random sample of data, overcomes the problems of histograms.
Diagrams 4a and 4b illustrate the kernel density and the kernel polygon for
each country and for both coffee species. The estimation through the kernel density is
a non-parametric way to estimate the probability density function of a random
variable. Kernel density estimation is a fundamental data smoothing problem where
inferences about the population are made based on a finite data sample.
Figure 4a: Kernel densities and polygons of Arabica
Figure 4b: Kernel densities and polygons of Robusta
Finally, we should notice the similarities between the histograms and the
kernel densities, since the kernel density estimates are closely related to histograms.
However, they can be endowed with properties, such as smoothness or continuity, by
using a suitable kernel.
Figures 5a and 5b illustrate the Quantile-Quantile or Q-Q plots for Arabica and
Robusta. The Q-Q plot is an exploratory graphical tool used to check the validity of a
distributional assumption for a given data set. In general, the basic idea is to compute
the theoretically expected value for each data point based on the distribution in
question. If the data indeed follow the assumed distribution, then the points on the
Q-Q plot will fall approximately on a straight line. This is quite true for the data under
study, as we can see from Figures 5a and 5b, which illustrate the Q-Q plots of Arabica
and Robusta, respectively, in Brazil, India and Ecuador.
Figure 5a: Normal Q-Q plot of Arabica
Figure 5b: Normal Q-Q plot of Robusta
Empirical Results and Discussion
For the empirical analysis, we employ the semi-parametric approach proposed
by Chen and Fan (2006), as aforementioned in a previous chapter of the present
dissertation, which involves three steps. In the first step, an Autoregressive Moving
Average - Generalized Autoregressive Conditional Heteroskedasticity
model is fit to each series of the rates of price change of the
relevant coffee species. In the next step, the obtained standardized residuals (filtered
data) are used to calculate the respective empirical distribution functions, creating in
this way the desired copula data. In other words, in the second stage, the original
time-series are transformed into the so-called copula data, namely data with
approximately uniform marginal distributions on [0,1], by using their respective
empirical distribution functions. In the last step, we estimate the copula models by
applying the maximum likelihood (ML) estimator to the copula data (Canonical ML).
Following the semi-parametric approach we can estimate the copula and the margins
using different methods. We have to say that the Canonical ML estimator is
consistent but less efficient relative to the parametric one. Thus, the empirical
distributions of the copula parameters and the dependence measures (such as
Kendall’s tau and tail coefficients) should be approximated using resampling methods
so as to calculate their standard errors.
As shown by the above we employ an ARMA (2, 1) - GARCH (1, 1) 9
on the rates of price change to obtain the filtered data. Table 2 presents the p-values
from the application of the Ljung-Box and the ARCH Lagrange Multiplier (LM) tests
to the filtered data among various lag lengths. The results show no autocorrelation and
ARMA-GARCH models are commonly used to obtain filtered data in empirical investigations of
dependence among stochastic process with copulas (e.g. Rémillard, 2010; Czado, Schepmeier, Min, 2012;
Aloui, Hammoudeh and Nguyen, 2013; Zinmer, 2013, Emmanouilides and Fousekis, 2014; Panagiotou and
The selection of ARMA (2, 1)-GARCH (1, 1) was based upon the goodness of fit tests against other
alternative ARMA-GARCH models. The ARMA (2, 1)-GRACH (1, 1) had the highest p-values.
Table 2: p-values of the test for autocorrelation and ARCH effects
No. of lags
1 6 12
No. of lags
1 6 12
ara.BR 0.8266 0.5568 0.6494 0.3693 0.7797 0.9320
ara.EC 0.9987 0.8677 0.4372 0.6446 0.6088 0.5949
ara.IN 0.7250 0.9835 0.6702 0.4045 0.6795 0.8411
rob.BR 0.9994 0.7188 0.7162 0.7745 0.6595 0.6677
rob.EC 0.9231 0.3247 0.5958 0.5960 0.9361 0.9251
rob.IN 0.8932 0.9922 0.9079 0.6408 0.9123 0.9801
A copula defines the probability that the observations from two time series (in our
case the filtered data) are below a given quantile. It is proposed that time- varying
tests constructed from indicator variables be used to test against the hypothesis that a
copula is changing over time. Therefore, before the selection of the appropriate copula
model, we apply the Busseti-Harvey test for constancy of the copula, for 1%, 5% and
10% significance levels. Table 3 presents the results for the Busseti-Harvey test for
the above three quantiles of the bivariate empirical copulas. As we can see for 1 per
cent statistical significance there are no breaks and/or gradual but persistent shifts in
the respective empirical copulas.
Table 3. Tests for the constancy of the qualities of the empirical copula
Empirical values of the nτ(ΒQ) statistics
τ = 0.25 τ = 0.5 τ = 0.75
ara.BR-rob.BR 0.241 0.731 0.577
ara.EC-rob.EC 0.523 0.312 0.600
ara.IN-rob.IN 0.338 0.601 0.300
Note: The critical values are 0.743, 0.461 and 0.347, at the 1%, 5% and 10% levels, respectively.
The selection among the different models was made based on Clarke test. We
compute a total score for each bivariate copula family under consideration. For each
possible pair of copula families, the Clarke test decide which of the two families fits
the given data best and assigns a score. The best model is the one that achieves the
highest total score. In case that the Clarke criteria can not discriminate between two or
more copulas, we apply the Cramér–von Mises goodness of fit test for the copulas
that appear to fit the data best and we choose the one that has the maximum p-value.
Table 4 and Table 5 give us the scores of Clarke test and the p-values of CvM
goodness of fit test.
Table 4: Total scores of the ten models considered.
Country test 1 2 3 4 5 6 7 8 9 10
Clarke -4 2 -9 3 5 -7 5 1 -4 8
Selected family: 10(BB8)
Clarke -6 7 -6 1 3 -9 7 -1 4 0
selected family: 7 (BB1),
Clarke -7 9 -3 1 -4 -1 3 1 4 -3
selected family: 2
Table 5. P-Values of the Cramér–von Mises test
India 0.948 0.971
Note: The p-values have been obtained by the bootstrap using 1000 replications
In case of Brazil, we select the Joe-Frank (BB8) copula. This copula has two
parameters to be estimated (θ1 and θ2) but exhibits no tail dependence10
. In case of
India, we can see that both P-values of CvM test are very high for the two competing
copulas, so we choose the one that has the highest P- value. With this in mind we
select the Clayton-Gumbell (BB1) that exhibits possible asymmetric tail dependence.
This copula has also two dependence parameters. On the contrary, in case of Ecuador
Clarke criteria indicate the selection of the Student-t copula, which is a bivariate
elliptical copula. This copula has two parameters: one correlation parameter, θ, and
the relevant degrees of freedom, v, and exhibits tail dependence.
10 For more information on the copula functions, the parameters, Kendall’s tau, tail dependence
and, generally, the relevant families, you may advice the Table 1A on the Appendix section.
Figure 6 presents the scatterplots of the copula data within a country for the
two different coffee species, Arabica and Robusta, respectively.
Figure 6: Scatterplots of the copulated data
The first scatterplot represents Brazil, the next represents India and the last
one represents Ecuador. Looking at those three scatterplots, we can say that they
indicate a positive association between the prices of the two coffee species within a
country, because the majority of the relative rank pairs lie along the positive
diagonals. However, this is more true in case of Brazil (the first scatterplot), while it is
less obvious in case of Ecuador (the last scatterplot). However, comparing the
scatterplots among the three countries, we can derive the conclusion that the prices
between the two species, namely Arabica and Robusta, are more associated in Brazil,
as they are less dispersed and more concentrated. On the contrary, the least connected
prices between Arabica and Robusta are present in Ecuador, as they are more
dispersed in the corresponding scatterplot.
Table 6 presents the selected copulas with their corresponding parameters, and
the estimates for Kendall’s tau (τ, as well as for the lower and upper tail dependence
coefficients, λL and λU, respectively). The standard errors of the copula parameters
and dependence measures have been obtained using bootstrap method (Appendix D)
It is worth-mentioning at this point some notion and symbols used in the
presentation of the results of our empirical investigation. More specifically, note that
in all our tables *, ** and *** denote statistical significance in 1%, 5% and 10%,
Table 6: Results from copula parameter estimation for Arabica and Robusta within each country
Selected Copula Parameters Kendall’s τ λL λU
Note: *** denotes statistically significant at 1% level or less
Estimation results were obtained after 1000 bootstrap repetitions.
Beginning with Brazil, the Clarke criterion suggests the selection of the
Joe-Frank(BB8) copula. According to Table 6, the tail dependence in that case is (λL,
λU) => (0, 0), meaning that there is no tail dependence; therefore the prices do not
boom or crash together. In other words, this empirical finding suggests that price
shocks in Arabica (Robusta) are not associated with price shocks (crashes) in Robusta
(Arabica). The estimation of this copula shows statistically significant parameters θ1
and θ2. Moreover, the statistically significant value for Kendall’s tau equals to 0.44
indicating that although the concordant pairs well exceed the discordant ones, the
overall strength of the relationship between the rates of price change is quite
considerable. Recall that the strength of the relationship between two variables can be
measured by Kendall’s tau, which is computed from ranks and is more appropriate as
a measure of concordance than a correlation coefficient.
In case of India, Clarke criterion suggests the selection of the Clayton-Gumbel
(BB1) copula. According to Table 6, there exists asymmetric tail dependence, which
is statistically significant. The value of λU implies that with a probability 0.27 a price
boom in Arabica (Robusta) will be associated with a price boom in Robusta
(Arabica). The value of λL implies that with probability 0.43 a price crash in Arabica
(Robusta) will be associated with a price crash in Robusta (Arabica). The difference
between the two probabilities offer an indication that price shocks during market
downswings are likely to be transmitted with higher intensity compared with price
shocks during market upswings. The statistical significant value of Kendall’s tau is
0.35(considerably lower compared to that for Brazil).
Finally, the Clarke criterion suggests the selection of the Student-t copula, in
case of Ecuador indicating symmetric tail dependence. The selected copula families
for Brazil and India belong to the bivariate Archimedean families, this one belongs to
the bivariate elliptical families. The degrees of freedom ( v=3.56) are well bellow 30
suggesting a very strong departure from normality. Kendall’s tau is 0.09 (very small)
indicating limited degrees of dependence over the entire joint distribution. The
coefficients of tail dependence (co–movement at the extremes) are statistical
significant at any reasonable level and give the probability of 0.13 that the two price
changes are together at the upper or at the lower joint tails.
. According to the results, we can conclude that despite the considerable co-
movement of coffee prices, the Joe-Frank (BB8) copula for the pair Arabica-Robusta
in Brazil indicates that close enough to the tails of the joint distribution of price
changes, extreme events occur independently in each margin. We can see that there is
no empirical evidence that price shocks to Arabica (Robusta) will be transmitted to
Robusta (Arabica). The interpretation of this finding maybe suggests that the two
varieties are not perfect substitutes, so the quality differentiation between the two
species plays an important role on the eyes of middlemen (or processors).
Regarding India, the finding of an asymmetric Clayton-Gumbel (BB1) copula
for the pair Arabica-Robusta implies that price shocks (booms) in Arabica will be
transmitted with lower intensity in Robusta compared to negative price shocks
(crashes). In other words, a plausible explanation of this empirical finding is that the
two coffee species may be imperfect substitutes, so quality differentiation exists to
some extend point.
Finally for Ecuador, the empirical evidence of a t-copula for the pair Arabica-
Robusta implies that extreme positive and extreme negative changes in Arabica prices
are equally likely to be passed in the Robusta. We have to mention that the degree of
this price transmission is very small which indicates that although these two coffee
species have substitute relationship, there are many other factors that restrain the
degree of the price symmetry co-movement.
Last but not least, we present the following contour diagrams. The first picture
represents the empirical copula, while the second one represents the theoretical one
(of the corresponding copula family).
In general, graphs provide a way of visualizing functions of two variables.
One important way of visualizing such functions is by drawing their contour
diagrams, i.e. a curve along which the function has a constant value. In the following
pictures, we can see the difference between the two copulas, empirical and theoretical.
Figure 7a: Contour diagram for Brazil
Figure 7b: Contour diagram for India
Figure 7c: Contour diagram for Ecuador
The subject of price a-symmetry has been an important topic in agricultural
economics, since it may be an indicator of market inefficiency and it may call for
policy intervention. In our empirical research, the main question is the identification
of the degree and the structure of price dependence between the two coffee species,
namely Arabica and Robusta, within the countries of Brazil, India and Ecuador.
More specific, the present dissertation has a two-fold contribution to the
existing literature. Our motivation lies on the aforementioned statement about price
co-movement. First, we apply the copula methodology which provides us with more
insights regarding this question of research. Second, we choose the agricultural
product of coffee, that has not been examined yet thoroughly concerning the possible
existence of price asymmetry in the coffee market.
According to the first novelty, a modern and quite helpful methodological tool
is the application of copulas, which is a way of formalizing dependence structures of
random vectors. More specifically, they allow flexible characterization of dependence
between random variables, being especially useful if no obvious choice for the
multivariate density function exists. Although they have been known about for a long
time in research, it seems that they have been rediscovered recently in applied
sciences, such as agricultural economics.
On the other hand, it is worth mentioning that coffee is the second largest
traded commodity next to oil (Pendergrast, 1999). Therefore, the coffee industry has
recently received increased attention from economic researchers. Since the crisis
period of the early 1990s, coffee has been on the leading edge of economic, social,
and environmental development schemes that now reach many major industries.
Our empirical investigation can be thought as a small, yet, significant
contribution to the relevant literature on that subject. The specific contribution lies on
the methodology in coffee, as well as in the coffee. Up to our knowledge, there is no
empirical research for price dependence using copulas in the coffee market. For that
reason, we examined three major coffee producing countries, Brazil, India and
Ecuador. These countries produce significant amounts of both Arabica and Robusta
coffee species. Data were ranged from 2002:1 up to 2014:12.
Our empirical results show the existence of co-movement between the prices
of Arabica and Robusta within the countries of Brazil, India and Ecuador. For the case
of India and Ecuador, the selected copula families indicate that a price boom (crash)
in the one of coffee species will be matched with a price boom (crash) to the other
coffee species and vise versa. For the case of Brazil we see no evidence for price co-
movement on the extreme events occurring in the coffee market.
Appendix A. Relationships between copula parameters and
Table 1A: Copula functions, parameters, Kendall’s tau, tail dependence
Copulas Parameters Kendall’s tau
Gaussian ( 1,1)
2 1 ,
4 ( )
1 log( )(1 )t t t dt
(BB1) 1 20, 1
1 2 2
2 ,2 2
(BB6) 1 21, 1
1 ( log(1 1 ) )(1
*(1 (1 ) ))
(BB7) 1 20, 0
1 ( (1 (1 ) )
(1 (1 ) ) 1
2 ,2 2
1 21, 0,1
1 2 2
(1 ) 14
1 ( log
(1 ) 1
*(1 )(1 (1 ) ))
t t dt
Source: Brechmann and Schepsmeier (2013)
Appendix B. The CvM test
The CvM test compares the distance between CT and CT,θ where the latter is an
estimator of Cθ. The null hypothesis(H0) specified as C = Cθ, which means that there is
no statistical difference between the empirical and the estimated theoretical copula.
The distance between them is defined as ,( )T T TD T C C . The CvM statistic is
( ) ( )T T TCvM D u dC u which is consistent.
Appendix C. Estimation of the standard errors for copula
parameters and dependence coefficients of Kendall’s tau and tail
measures λU, λL.
Patton (2013) proposes an i.i.d. bootstrap method to calculate the standard errors for
copula parameters and dependence coefficients. More specific this method is ideally
when the copula model has been estimated by applying the semi parametric procedure
and when the copula is being stationary over time.
Ahmed, O. and Serra, T. (2015) Economic analysis of the introduction of
agricultural revenue insurance contracts in Spain using statistical copulas,
Agricultural Economics, 46, pp.69-79.
Anton, J. and Kimura, S. (2011) Risk management in agriculture in Spain,
OECD Food, Agriculture and Fisheries Working Papers. No.43, OECD Publishing.
Branchi, M., Gabriele, A. and V. Spiezio (1999) Traditional agricultural
exports, external dependence and domestic price policies: African coffee exports in a
comparative perspective, UNCTAD Discussion Papers.
Brechmann, E.C. and Schepsmeier, U. (2013) Modeling dependence with c-
and d- vine copulas: The r-package cdvine, Journal of Statistical Software, 52,
Busetti, F. and Harvey, A. (2011) When is a copula constant? A test for
changing relationships, Journal of Financial Econometrics, 9 (1), pp.106-131.
Chen, X. and Fan, Y. (2006) Estimation of copula-based semiparametric time
series models, Journal of Econometrics, 130, pp.307-335.
Donnet, M.L., Weatherspoon, D.D. anf Hoehn, J.P. (2008) Price determinants
in top-quality e-auctioned specialty coffees, Agricultural Economics, 38, pp.267-276.
Dragusanu, R. and N. Nunn N (2014) The Impacts of Fair Trade Certification:
Evidence From Coffee Producers in Costa Rica. Working Paper, available at:
Emmanouilides, C.J. and Fousekis, P. (2014a) Vertical Price Transmission in
the US Pork Industry: Evidence from Copula Models, Agricultural Economics
Review, forthcoming (indexed in SCOPUS).