2
Modeling Long-Term Oil Demand in the Agricultural Sector
About KAPSARC
KAPSARC is an advisory think tank within global energy economics and sustainability
providing advisory services to entities and authorities in the Saudi energy sector
to advance Saudi Arabia's energy sector and inform global policies through
evidence-based advice and applied research.
This publication is also available in Arabic.
Legal Notice
© Copyright 2024 King Abdullah Petroleum Studies and Research Center (“KAPSARC”).
This Document (and any information, data or materials contained therein) (the
“Document”) shall not be used without the proper attribution to KAPSARC. The
Document shall not be reproduced, in whole or in part, without the written permission
of KAPSARC. KAPSARC makes no warranty, representation or undertaking whether
expressed or implied, nor does it assume any legal liability, whether direct or indirect,
or responsibility for the accuracy, completeness, or usefulness of any information that
is contained in the Document. Nothing in the Document constitutes or shall be implied to
constitute advice, recommendation or option. The views and opinions expressed in this
publication are those of the authors and do not necessarily reect the ofcial views or
position of KAPSARC.
3
Modeling Long-Term Oil Demand in the Agricultural Sector
The agricultural sector and global food security are facing an increasing number of risks due to
climate change, the increasing population size, rising energy and agricultural demands, competing
demands for land for biofuel production, and the degradation of soil quality. Between 2001 and 2018,
the annual consumption of food and agricultural products increased rapidly by approximately 48%, and
there was twofold population growth (WEF 2021). This increase in agricultural demand has also stimulated
the demand for energy products, as the agricultural sector relies heavily on specic fuels for heating,
machinery, and other activities. Given the consistent increasing trend of the global population, the world will
be faced with the need to feed approximately 10 billion people by 2050, or approximately 50% more food
than in 2010. This methodology paper aims to provide a description of the global long-term outlook of oil
consumption in the agricultural sector. The analysis disaggregates oil demand outlook by primary fuel type
and geographical location. That is, the modeling process disaggregates global oil consumption into eight
regions and different primary fuels that each region consistently consumes in its agricultural sector. Given
the importance of the challenges faced by the global agriculture and food security sector, this analysis
highlights the need for data-driven approaches that allow for the long-term management of energy demand
in this sector. This paper also emphasizes the importance of agricultural energy demand forecasting and its
role in shaping future energy policies.
Keywords: Energy demand; Agriculture; Forecast; Energy Models.
Abstract
4
Modeling Long-Term Oil Demand in the Agricultural Sector
Through the Green Revolution of the 1950s
and 1960s, also known as the Third
Agricultural Revolution, the yield of global
agriculture increased dramatically, preventing the
spread of famine and malnutrition (Moore 2010).
Nevertheless, the world’s population has grown
by more than 5 billion people since the Green
Revolution began, necessitating continued growth
in agricultural production. The agricultural sector
is also at the heart of the food production process
(i.e., it includes cereals, fruits, vegetables, meat,
sh, poultry, milk, etc.) and economically valuable
nonfood plant goods (e.g., tobacco, jute, and hemp).
The sector also involves other processes, such
as fertilizer production, postharvest processing,
and food transportation. Between 1960 and 2015,
agricultural production increased more than
threefold, partly due to the productivity-enhancing
technologies of the Green Revolution and the
considerable expansion in the use of land, water,
and other natural resources for agricultural activities.
Moreover, during the same period, there was the
notable industrialization and globalization of food
and agriculture. The evolving agrarian transition
from farmer agriculture to extensive commercial
agriculture has been redesigning many regions’
production systems.
Furthermore, some forecasts suggest that global
agricultural production will need to at least double
by 2050 to satisfy the projected food demand
resulting from both population growth and the shift
toward richer diets (Alexandratos and Bruinsma
2012). The agricultural sector also remains the key
sector for developing and diversifying the economy
in many developing countries. Furthermore, the
enhancement of productivity and upgrading of
agricultural production systems are the key drivers
of poverty alleviation in many regions worldwide,
with energy playing a critical role in this regard.
Indeed, harnessing power for modern, sustainable
agricultural production and processing systems is
an essential factor in moving beyond subsistence
farming to food security, adding value in rural areas,
and expanding into new agrarian markets.
The world’s agricultural systems are consumers
of large amounts of energy and contributors to
climate change. Agriculture consumes 15–30%
of the world’s primary energy and emits 25–34%
of its total greenhouse gas (GHG) emissions
(Schramski, Woodson, and Brown 2020). In
the U.S., for example, the agricultural sector is
responsible for approximately 1.9% of total primary
energy consumption (Department of Agriculture
(USDA)). Diesel, electricity, natural gas, gasoline,
and liqueed petroleum gas (LPG) are the most
common energy sources for agricultural activities
in the U.S. It is also critical to mention that in the
period from 2012 to 2015, there was an increase in
energy intensity in the U.S. agricultural sector, with
its energy consumption increasing by over 10% and
its output growing by only 6% (USDA).
Typically, energy consumption varies at every phase
of the agri-food chain, and agriculture’s energy use
is segmented into two categories: direct and indirect
energy. Direct energy use in agriculture reects
the energy that is spent directly on agricultural
activities, including those activities farm machinery,
vehicles, trucks, equipment, drying operations,
business operations, and marketing. Indirect
energy consumption in agriculture includes the
chemical inputs used in the production of fertilizers,
pesticides and other chemical products used in
agriculture. Energy in the agricultural sector is used
primarily for groundwater pumping and agricultural
machinery, such as threshers and tractors. Starting
with fertilizer production to process and transport
food to market, the industrial food system relies
1. Energy Demand in the Agricultural
Sector
5
Modeling Long-Term Oil Demand in the Agricultural SectorModeling Long-Term Oil Demand in the Agricultural Sector 5
heavily on fossil fuels to produce basic commodity
monocultures (see Figure 1). For instance,
approximately 80% of the agricultural sector’s
energy needs are met by fossil fuels.
Although energy consumption in agriculture
generally tends to be at a low level compared to
that in industry or transportation, the sector is of
signicant importance to economic activity and
employment, especially in developing countries.
Additionally, considering the extra energy demand
of agricultural production (Sharma et al. 2021) and
its environmental footprint (Pellegrini and Fernández
2018; Rosa et al. 2021), the energy demand and
environmental costs of such additional production
must be managed (World Bank 2023; WRI 2018).
Figure 1. Framework for Energy Use in Agriculture and the Food Chain.
PRODUCTION
• On-farm
• mechanization
• Reduction in human
• labour requirements
• Increased operational
• efficiencies
STORAGE &
HANDLING
• Cold storage
• Moisture control
• Mechanized
• sorting/packaging
TRANSPORT &
LOGISTICS
• Warehouse
• Road, rail
• and maritime
• transport
END-USER
• Cooking
• Transport
• Household
• appliances
INPUTS
• Seed
• Irrigation/
• pumping
• Livestock feed
• Fertilizer
TRANSPORT
• Farm to collection
• centre
VALUE ADDED
PROCESSING
• Drying
• Grinding
• Milling
• etc.
MARKETING &
DISTRIBUTION
• Packaging
• Retail
• (supermarkets)
• Refrigeration
Food (energy) losses
Food
Source: Adapted from FAO/USAID, 2015
• Collection centre to
• proceeing facility/market
1. Energy Demand in the Agricultural Sector
6
Modeling Long-Term Oil Demand in the Agricultural Sector
Forecasting agricultural energy demand
is a crucial aspect of managing energy
resources, as it helps predict future energy
usage patterns in the agricultural sector, which,
in turn, can assist with planning and optimizing
energy production and distribution. In this context,
there is growing interest in the literature on energy
demand modeling in the agricultural sector (Gezer,
Acaroglu, and Haciseferogullari 2003; Taki et al.
2018; Bolandnazar et al. 2020). However, how to
adequately forecast energy demand continues to
be a challenge due to the lack of global agricultural
data, especially in developing countries. Next,
we summarize some key studies and their
methodological approaches to estimate the levels
of energy consumption in the agricultural sectors of
various countries and regions globally.
Fei and Boqiang (2017), using an econometric
method, investigated China’s agricultural energy-
saving potential. First, the authors performed
a cointegration analysis and constructed an
error-correction model to investigate the long-
run equilibrium relationship between agricultural
energy demand and its driving factors, such as
energy price, mechanical power, agrarian output,
agricultural industrial structure, and tax expenditure,
over the period 1980–2012. Then, stability and
t effect tests and Monte Carlo simulation were
implemented to validate the soundness of the
prediction model. Furthermore, the scenario analysis
method was used to predict energy-saving potential
under various scenarios within 2020 and 2025.
Another study was carried out by Rokicki et
al. (2021), which aimed to identify and outline
the current situation and evolution of energy
consumption in agriculture in European Union (EU)
countries. The study included an examination of
the degree of concentration of energy consumption
in agriculture in EU countries, pointing out the
directions of their evolution and the types of energy
used and identifying the connection between energy
consumption and the changes in the economic and
agricultural situation in EU countries. Using data
from Eurostat, the above study covered the period
2005–2018.
Shakibai and Koochekzadeh (2009) used a
multilayer perceptron neural network model and
autoregressive integrated moving average (ARIMA)
process to model and predict agricultural energy
demand in Iran. Data for the period 1976–2001 were
used for modeling, and information for the period
2002–2007 was used to assess the predictive
power of the used techniques. More recently, in
the same context, Farajian et al. (2018) employed
the Box–Jenkins approach to model agricultural
demand for four major energy sources, namely,
gasoline, kerosene, diesel, and electricity, in Iran
over the period from 1988 to 2014. The results were
used to check the model’s suitability and generate
information about the state of energy demand
in the Iranian agricultural sector over the period
2015–2026.
Recently, Bolandnazar et al. (2020) used different
modeling approaches to predict potato energy
demand in Iran. The primary methods used
included multiple linear regressions, Cobb–Douglas
functions, support vector machines, and radial basis
functions. The above authors showed that the radial
basis function model yielded the most signicant
prediction performance among the different
methods.
Finally, a study aimed at understanding the
relationship between agricultural energy
consumption and production from a global
perspective was conducted by Chen et al. (2020),
investigating the decoupling status between the
energy consumption and economic growth of the
farming industry in 89 countries with available
data from 2000 to 2016. The study disaggregated
2. Literature Review
7
Modeling Long-Term Oil Demand in the Agricultural SectorModeling Long-Term Oil Demand in the Agricultural Sector 7
agricultural energy use in the 89 countries into the
effects of a driving factor (i.e., agricultural economic
output) and three constraining factors (i.e., labor
intensity, agricultural land, and energy intensity, in
descending order).
The above literature review shows that projections
of agricultural oil demand at the regional level are
relatively rare. While there several studies have
focused on specic fuels, countries or regions,
there is a lack of research and modeling using
econometric methods from a global perspective.
Thus, this paper aims to ll this gap by providing
insights into the long-term global energy demand
disaggregated by eight regions and the primary fuels
consumed by the agricultural sector in each region
until 2050. These regions and fuels are described in
the below section.
2. Literature Review
8
Modeling Long-Term Oil Demand in the Agricultural Sector
This section demonstrates the methodology
of collecting data and constructing long-term
outlook models and lists the regions and
main different fuels for which we construct outlook
models. This section also lists different variables
that play critical roles in shaping oil demand in the
agricultural sector.
3.1. Data
As mentioned above, the aim of this study is
to construct different models to forecast global
primary fuel demand disaggregated by eight
regions: Organisation for Economic Co-operation
and Development (OECD) America, OECD
Europe, OECD Asia Oceania, Latin America and
the Caribbean, Middle East and Africa, China,
non-OECD Asia (excluding China), and Eurasia.
Each region has its own individual fuel models
depending on the historical fuel consumption level
in the agricultural sector. In addition to historical fuel
consumption, long-term forecast models incorporate
nominal gross domestic product (GDP), GDP per
capita, real GDP, total population, urban population,
Brent oil price, and rural population as explanatory
variables. The last two variables are the shares of
urban and rural populations of the total population
for each region, in percentages. Nominal GDP, per
capita GDP, and rural population data are from the
World Bank National Accounts database (World
Bank 2022). Total population and urban population
data are from the United Nations Population Division
(UN 2022). The historical data extend for 48 years,
from 1971 to 2019.
It is critical to emphasize that agricultural fuel
consumption data include data on all agricultural,
hunting, and forestry activities. These activities
include fuel consumption for traction, power, and
heating (IEA 2023). After examining the different fuel
consumption levels of the eight regions, we nd that
the agricultural sector consumes four main fuels:
1) LPG, 2) kerosene, 3) gas and diesel oil, and
4) fuel oil (IEA 2022). Notably, kerosene
consumption does not include that used in
aircraft transport. In addition, gas and diesel oil
consumption excludes biofuels (IEA 2022). The
dependent variables, explanatory variables, and
geographic distribution are as follows:
Dependent Variables
- LPG demand (LPG)
- Other kerosene demand (KERO)
- Gas/diesel oil excluding biofuel demand
(Gas)
- Fuel oil demand (Fuel)
Explanatory Variables
- Nominal GDP at current prices, denoted
as GDPcu
- GDP per capita (constant 2015 US$),
denoted as GDPpc
- Real GDP (constant 2015 US$), denoted
as GDPre
- Total population, denoted as popul
- Urban population (% of the total population),
denoted as urban
- Rural population (% of the total population),
denoted as rural
- Brent oil price, denoted as Brent
Regions
- OECD America
- OECD Europe
- OECD Asia Oceania
- Latin America and the Caribbean
- Middle East and Africa
3. Data and Methodology
9
Modeling Long-Term Oil Demand in the Agricultural SectorModeling Long-Term Oil Demand in the Agricultural Sector 9
- China
- Other Asia
- Eurasia
3.2. Methodology
The methodological approach emphasizes mainly
the ARIMA with exogenous factors (ARIMAX), which
has the ability to identify underlying patterns in time-
series data and quantify the inuence of external
factors. This approach offers the capacity to identify
the effects of high-impact variations that are both
external and internal in nature (Andrews et al. 2013).
The ARIMAX is an extended version of the ARIMA
model. ARIMA, also known as the Box–Jenkins
method (Box and Jenkins 1990), requires historical
time-series information of the underlying variable. The
ARIMAX is quite similar to the ARIMA model, except
that the former includes relevant explanatory factors.
Although the incorporation of exogenous factors
increases the complexity of the model-building
process, the model can accommodate the impact of
external drivers (e.g., the economic situation).
Although the ARIMAX works like a multivariate
regression model, it leverages the autocorrelation
that may be present in the residuals of the regression
to improve forecasting accuracy. More specically,
the ARIMAX is a multiple linear regression model
including autoregressive (AR) and/or moving average
(MA) terms. The AR terms denote that the factor
of interest is regressed on its own value, while the
MA terms specify that regression error is a linear
combination of the error values that occurred in the
past. Finally, the I-part (for “integrated”) stands for
the differentiation order to transform the time series
into a stationary one (if needed).
The general specications of ARIMAX models are
expressed as follows:
ARX: AR with exogenous factors:
Yt=∅
(
L
)
Yt+
β
Xt+
ε
t. (1)
MAX: MA with exogenous factors:
Y
t
=
ϕ
(
L
)
ε
t
+
β
X
t
. (2)
ARMAX: ARMA with exogenous factors:
∅
(
L
)
Yt=
ϕ
(
L
)
ε
t+
β
Xt, (3)
where Xt depicts the explanatory factors,
b
depicts
the model coefcients, ∅(L)Yt depicts the AR model,
and
ϕ
(L)
e
t is the MA model.
The ARIMAX model is described by the following
parameters:
p, the number of AR terms;
d, the number of nonseasonal differences
needed for stationarity;
q, the number of lagged forecast errors in the
prediction equation; and
n, the number of explanatory factors.
Mathematically, the model may be specied as
follows:
ARIMAX
(
p
,
d
,
q
,
n
):
Yt=
ω
+
i
=1
p
∑
ϕ
iYt−
1
−
j
=1
q
∑
θ
j
ε
t−j+
l
=1
n
∑
δ
lXl
,
(4)
where
ω
represents the model’s constant term,
ϕ
i, i = 1,..., p;
θ
j, j = 1,..., q; and
δ
l, l = 1,..., n represent
the model parameters, Yt and Yt–1 for i = 1,..., p are
the predicted values of the outcome variable,
e
t–j for
j = 1,..., q represent the regression error, and XI for
I = 1,..., n represent the explanatory factors.
3. Data and Methodology
10
Modeling Long-Term Oil Demand in the Agricultural Sector
To assess the validity of the ARIMAX model, it is
necessary to verify six conditions. Among these six
conditions, two of them (designated as Hypotheses 1
and 2) relate to the residuals generated by the
model, and the remaining four of them (designated
as Hypotheses 3 through 6) pertain to the exogenous
factors that make up the model (Andrews et al. 2013):
H1: The stationarity of the series indicates that the
mean and variance of the residuals are stable over
time.
H2: There is an absence of serial correlation.
H3: The coefcient estimates of the explanatory
variables must be signicantly different from zero.
H4: There is an absence of reverse causality.
H5: The coefcient sign of each signicant external
factor must be appropriate.
H6: There is an absence of multicollinearity.
Figure 2 displays the four steps used to construct
and assess the validity of the ARIMAX prediction
model, namely, identication, estimation, model
assessment, and forecasting.
Figure 2. Key Steps of ARIMAX Forecasting.
Estimation of Model Coefficients
Estimation parameters of the model
identified in the previous stage
Identification of Model Orders
Check for the stationary in the series and identify
the initial orders of model parameters
Diagnosis of the Fitted Model
Evaluate the performance of the
estimated model, using appropriate
metrics, to verify if it adequately
represents the dynamics of the data.
Forecast
Generate forecasts for future time periods using
the estimate model.
Step 4
Step 3
Step 2
Step 1
3. Data and Methodology
11
Modeling Long-Term Oil Demand in the Agricultural Sector
The main goal of this paper is to construct
regional long-term oil demand forecasts for
the agricultural sector. However, it might be
useful to showcase the primary results to provide
a full picture of the model output. This section
shows the validation output as well as the long-term
forecasts of the eight regional models for each fuel
that when combined, constitute global demand. In
other words, we show the forecasts of the regional
demand for each fuel that form global demand.
4.1. Regression Results
As mentioned earlier, we develop 32 models to
forecast four fuels in eight different regions using
the ARIMAX modeling technique. The regression
outputs of OECD America, for example, are shown
in Table 1. The remaining regression outputs are
listed in Appendix 2. For simplicity, we show only the
main regressors.
4. Results
Table 1. Regression results of OECD America.
Dep. variables Ln(LPG) Ln(KERO) Ln(Gas) Ln(Fuel)
Regressors
Ln(GDPREj)0.306586***
Ln(GDPREj) t–1 5.21092**
Ln(GDPR)t–1
Ln(popj)−137.197*** −324.884***
Ln(pop)t–1 157.011*** 563.722**
Ln(pop)t–2 −251.074**
Ln(Brentj)0.458963***
Ln(Brent)t–2 0.344116
Ln(LPG)t–1
Ln(KERO)t–1 1.02859***
Ln(Gas)t–1
Ln(Fuel)t–1 0.474648***
C−389.292*** −1.56710*** 0.316147 85.0915
Adj. R20.994 0.986 0.745149 0.950335
No. of
observations
47 40 40 40
Normality test:
Chi2(2)
3.7336 [0.1546] 4.6602 [0.0973] 5.4038 [0.0671] 0.88058 [0.6438]
Hetero test F(14,30) = 1.9951 [0.0552] F(2,35) = 1.1078 [0.3416] F(2,37) = 0.85025 [0.4355] F(14,25) = 1.1271 [0.3835]
RESET23 test F(2,34) = 0.67196 [0.5174] F(2,34) = 0.36647 [0.6959] F(2,36) = 2.9937 [0.0627]
Notes:
* signicant at 10%.
** signicant at 5%.
*** signicant at 1%.
12
Modeling Long-Term Oil Demand in the Agricultural Sector
4.2. Regional Fuel Forecasts
After constructing the regression equations for the
four fuels of the eight regions, we show the long-
term fuel forecasts of these regions. As discussed
earlier, the goal of this paper is to predict the
long-term growth of oil demand in the agricultural
sector according to different primary fuels. Thirty-
year forecast horizons are estimated for each of
the 32 cases (4 fuels × 8 regions). This subsection
discusses the forecast of each fuel and shows how
it is expected to evolve in each region. Notably,
forecasts depend mainly on future GDP, population,
and oil prices and, thus, might not capture any
potential endogenous factors, such as future
technological changes.
LPG is a common fuel in agriculture and farming
activities and has many applications in agriculture,
such as providing heat for crop and vegetable
growth, poultry production, and water heating.
Farmers nd LPG attractive due to its portability,
accessibility, and convenience of use (Exceptional
Energy in Action 2023). LPG forecasts show a
signicant demand increase in OECD America
and a slight demand increase in Latin America and
Other Asia, while the other regions do not show a
signicant long-term change (Figure 3).
The forecasted growth in LPG consumption can also
be explained by the expected growth of both global
population and GDP, as illustrated in Figure 4. That
is, the growth in population and GDP stimulates
increased demand for agricultural products and,
hence, for agricultural activities. For instance, the U.S.
accounted for 30% of the global production of corn
in 2019, and Brazil accounted for approximately 40%
of the global production of sugarcane. In addition,
Figure 3. Regional LPG Forecasts
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
80,000
90,000
1971
1976
1981
1986
1991
1996
2001
2006
2011
2016
2021
2026
2031
2036
2041
2046
Ktoe
America
Europe
Asia Oceania
Latin America
Middle East
China
Other Asia
Eurasia
4. Results
13
Modeling Long-Term Oil Demand in the Agricultural SectorModeling Long-Term Oil Demand in the Agricultural Sector 13
Indonesia and Malaysia (which fall into the Other Asia
category) are collectively responsible for approximately
85% of global palm oil production (Food and
Agriculture Organization (FAO)). These countries are
expected not only to continue leading the production
of the abovementioned products but also to increase
their shares as global demand increases.
Global demand for kerosene is expected to
exponentially increase throughout the forecasted
horizon, as shown in Figure 5. This growth in demand
is solely driven by Europe, while we expect to see
diminishing demand levels in the other regions.
Kerosene consumption, as mentioned earlier, does
not include that consumed in aircraft transport.
If we look at the historical demand for kerosene,
then we can see that Europe witnessed a sharp
increase in demand in 2016, which might indicate
a switch from different fuels to kerosene or an
emerging technological change in a specic sector
that stimulated the demand for kerosene. In addition
to LPG, kerosene is also an alternative fuel for
agricultural activities such as heating and even used
to be the primary fuel for some agricultural machinery.
Gas/diesel oil includes different heavy gas oils,
which include mainly transportation diesel. Diesel
is widely used in different sectors and applications,
including passenger cars, trucks, and marine areas.
In addition, the agricultural sector relies heavily on
diesel equipment. Diesel is one of the most cost-
effective fuels for engines and machinery in terms
of efciency, power, and durability and is consumed
throughout almost the whole agricultural supply
chain, especially in countries where technological
advancement is at a relatively high level. The global
forecasts show that most regions are expected to
maintain their demand for gas and diesel oil in the
long term, while Asia Oceania’s demand is expected
to grow exponentially, as shown in Figure 6.
Such growth in diesel demand is expected, as
Australia, Japan, and South Korea are among the
Figure 4. Global Population and Real GDP
0
2
4
6
8
10
12
0
20
40
60
80
100
120
140
160
180
1971
1975
1979
1983
1987
1991
1995
1999
2003
2007
2011
2015
2019
2023
2027
2031
2035
2039
2043
2047
Population (Billions)
Real GDP (Trillions)
Real GDP
Population
4. Results
14
Modeling Long-Term Oil Demand in the Agricultural Sector
Figure 5. Regional Kerosene Forecasts
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
1971
1976
1981
1986
1991
1996
2001
2006
2011
2016
2021
2026
2031
2036
2041
2046
Ktoe
America
Europe
Asia Oceania
Latin America
Middle East
China
Other Asia
Eurasia
Figure 6. Regional Gas/Diesel Oil Forecasts
0
50,000
100,000
150,000
200,000
250,000
300,000
350,000
400,000
450,000
500,000
1971
1976
1981
1986
1991
1996
2001
2006
2011
2016
2021
2026
2031
2036
2041
2046
Ktoe
America
Europe
Asia Oceania
Latin America
Middle East
China
Other Asia
Eurasia
4. Results
15
Modeling Long-Term Oil Demand in the Agricultural SectorModeling Long-Term Oil Demand in the Agricultural Sector 15
Figure 7. Regional Fuel Oil Demand Forecasts
0
20,000
40,000
60,000
80,000
100,000
120,000
1971
1976
1981
1986
1991
1996
2001
2006
2011
2016
2021
2026
2031
2036
2041
2046
Ktoe
America
Europe
Asia Oceania
Latin America
Middle East
China
Other Asia
Eurasia
4. Results
top consumers of diesel in the agricultural sector
(NM). Europe, in contrast, is expected to gradually
move away from gas/diesel products, which may
reasonably explain its expected growth in demand
for kerosene. In other words, Europe is expected
to replace diesel with kerosene in its agricultural
activities in the long term.
Global fuel oil demand is also expected to witness
growth in the long term (Figure 7). However, this
demand is expected to be driven mainly by the
Other Asia region, where India and Pakistan are
among the largest players. In 2020, for example,
India consumed approximately 192 thousand metric
tons of fuel oils for agricultural activities.
16
Modeling Long-Term Oil Demand in the Agricultural Sector
The primary purpose of this analysis is to provide a global view of future energy demand in the
agricultural sector. The projection of long-term energy demand for agriculture helps determine the
capacity needed to support the growing global demand for food. Such projections are also crucial
for analyzing the scope and composition of energy-supply expansion projects involving alternative energy
sources. Furthermore, these projections are broadly useful for understanding the agricultural sector’s energy
needs to achieve economic and ecological sustainability and evaluate the effectiveness of relevant policy
decisions (biofuel subsidies, regulations, labeling, etc.).
In addition, this study highlights the lack of specic studies focusing on agricultural energy demand
prediction, despite the extensive literature focusing on the agricultural sector within the context of food
security and climate change. This study also highlights that while there is substantial literature on the
agricultural sector within the context of food security and climate change, specic studies focusing on the
forecasting of energy demand from agriculture are somewhat limited. Two potential explanations for this
scarcity are considered. First, the level of direct energy demand of the agricultural sector is relatively low
compared to other sectors of the economy. Second, most studies and outlooks designed to predict energy
demand integrate agricultural consumption with that of other sectors, such as industry and construction.
In summary, this paper constructs a set of ARIMAX models for forecasting the global oil demand
disaggregated by region and fuel type. Eight regions are accounted for in the global demand: OECD
America, OECD Europe, OECD Asia Oceania, Latin America and the Caribbean, Middle East and Africa,
China, non-OECD Asia (excluding China), and Eurasia. After examining oil demand in the agricultural
sector, we nd that the sector consumes four main types of fuels: liqueed natural gas (LNG), kerosene,
gas/diesel oil, and fuel oil. Thus, a total of 32 models are constructed (8 regions × 4 fuels) and utilized
to forecast the fuel demand levels with a 30-year horizon. We can see from the results that in general,
the demands for all four fuels are expected to grow exponentially in the long term. Moreover, some
regions are expected to increase their demand for a particular fuel and deviate from other fuels, while few
regions are expected to maintain their demand for some fuels, for example, gas/diesel oil in Europe and
OECD America.
5. Conclusions
17
Modeling Long-Term Oil Demand in the Agricultural Sector
Alexandratos, N. and J. Bruinsma. 2012. World
Agriculture Towards 2030/2050: The 2012 Revision.
Rome: The Food and Agriculture Organization of the
United Nations.
Andrews, Bruce H., Matthew D. Dean, Robert Swain,
and Caroline Cole. 2013. “Building ARIMA and ARIMAX
Models for Predicting Long-Term Disability Benet
Application Rates in the Public/Private Sectors.” Society
of Actuaries (August): 1–54.
Beltran-Peña, Areidy, Lorenzo Rosa, and Paolo
D’Odorico. 2020. “Global Food Self-Sufciency in
the 21st Century Under Sustainable Intensication of
Agriculture.” Environmental Research Letters 15 (9):
095004. DOI 10.1088/1748-9326/ab9388
Bolandnazar, Elham, Abbas Rohani, and Morteza Taki.
2020. “Energy Consumption Forecasting in Agriculture by
Articial Intelligence and Mathematical Models.” Energy
Sources, Part A: Recovery, Utilization, and Environmental
Effects 42 (13): 1618–1632. DOI: https://doi.org/10.1080/1
5567036.2019.1604872
BP. 2020. Annual Energy Outlook. London: BP.
Chen, Xi, Chenyang Shuai, Yu Zhang, and Ya Wu.
2020. “Decomposition of Energy Consumption
and Its Decoupling With Economic Growth in the
Global Agricultural Industry.” Environmental Impact
Assessment Review 81 (March): 106364. DOI:10.1016/j.
eiar.2019.106364
Farajian, Leila, Reza Moghaddasi, and Safdar Hosseini.
2018. “Agricultural Energy Demand Modeling in Iran:
Approaching to a More Sustainable Situation.” Energy
Reports 4 (November): 260–265. DOI: 10.1016/j.
egyr.2018.03.002
Fei, Rilong, and Boqiang Lin. 2017. “Estimates of
Energy Demand and Energy Saving Potential in China’s
Agricultural Sector.” Energy 135 (September 15): 865–
875. DOI: 10.1016/ j.energy.2017.06.173
Gezer, Ibrahim, Mustafa Acaroǧlu, and Haydar
Haciseferoǧullari. 2003. “Use of Energy and Labour
in Apricot Agriculture in Turkey.” Biomass and
Bioenergy 24, no. 3 (March): 215–219. DOI: 10.1016/
S0961-9534(02)00116-2
Moore, Jason W. 2010. “The End of the Road?
Agricultural Revolutions in the Capitalist World‐Ecology,
1450 –2010.” Journal of Agrarian Change 10, no. 3 (July):
389–413. DOI: 10.1111/ j.1471- 0 36 6. 2010.0 0276 .x
Pellegrini, Pedro, and Roberto J. Fernández. 2018. “Crop
Intensication, Land Use, and On-Farm Energy-Use
Efciency During the Worldwide Spread of the Green
Revolution.” Proceedings of the National Academy of
Sciences 115, no. 10 (March): 2335–2340. DOI: 10.1073/
pnas.1717072115
Rokicki, Tomasz, Aleksandra Perkowska, Bogdan
Klepacki, Piotr Bórawski, Aneta Bełdycka-
Bórawska, and Konrad Michalski. 2021. “Changes
in Energy Consumption in Agriculture in the EU
Countries.” Energies 14 (6): 1570. DOI: 10.3390/
en14061570
Rosa, Lorenzo, Maria Cristina Rulli, Saleem Ali, Davide
Danilo Chiarelli, Jampel Dell’Angelo, Nathaniel D.
Mueller, Arnim Scheidel, Giuseppina Siciliano, and Paolo
D’Odorico. 2021. “Energy Implications of the 21st Century
Agrarian Transition.” Nature Communications 12, no. 1
(July 16): 1–9. DOI: 10.1038/s41467-021-22581-7
Schramski, John R., C. Brock Woodson, and James
H. Brown. 2020. “Energy Use and the Sustainability of
Intensifying Food Production.” Nature Sustainability 3,
no. 4 (April): 257–259. DOI: 10.1038/s41893-020-0503-z
Shakibai, Alireza, and Somayeh Koochekzadeh.
2009. “Modeling and Predicting Agricultural Energy
Consumption in Iran.” American-Eurasian Journal
of Agricultural and Environmental Science 5, no. 3
(January): 308–312.
Sharma, Gagan Deep, Muhammad Ibrahim Shah,
Umer Shahzad, Mansi Jain, and Ritika Chopra.
2021. “Exploring the Nexus Between Agriculture and
Greenhouse Gas Emissions in BIMSTEC Region: The
Role of Renewable Energy and Human Capital as
Moderators.” Journal of Environmental Management 297
(November): 113316. DOI: 10.1016/ j.jenvman. 2021.113 316
Taki, Morteza, Abbas Rohani, Farshad Soheili-Fard,
and Abbas Abdeshahi. 2018. “Assessment of Energy
Consumption and Modeling of Output Energy for Wheat
Production by Neural Network (MLP and RBF) and
Gaussian Process Regression (GPR) Models.” Journal of
Cleaner Production 172 (January 20): 3028–3041. DOI:
10.1016/j.jclepro. 2017.11.107
References
18
Modeling Long-Term Oil Demand in the Agricultural Sector
References
World Bank. 2022. “Population, Total.” Accessed July
2023. https://data.worldbank.org/indicator/SP.POP.TOTL.
World Resources Institute (WRI). 2018. World Resources
Report: Creating a Sustainable Food Future. Washington,
D.C.: World Resources Institute.
19
Modeling Long-Term Oil Demand in the Agricultural Sector
Region Denitions
i. OECD America:
Canada
Chile
Guam (no energy balance)
Mexico
Puerto Rico (no energy balance)
United States of America
United States Virgin Islands (no energy balance)
ii. OECD Europe:
Austria
Belgium
Czech Republic (no energy balance)
Denmark
Estonia
Finland
France
Germany
Greece
Hungary
Iceland
Ireland
Italy
Luxembourg
Netherlands
Norway
Poland
Portugal
Slovakia
Appendix 1
20
Modeling Long-Term Oil Demand in the Agricultural Sector
Slovenia
Spain
Sweden
Switzerland
Turkey
United Kingdom
iii. OECD Asia Oceania
Australia
Japan
New Zealand
OECD Asia Oceania, Other (no energy balance)
Republic of Korea
iv. Latin America and the Caribbean
Anguilla
Antigua and Barbuda
Argentina
Aruba
Bahamas
Barbados
Belize
Bermuda
Bolivia (Plurinational State of)
Brazil
British Virgin Islands
Cayman Islands
Colombia
Costa Rica
Cuba
Dominica
Appendix 1
21
Modeling Long-Term Oil Demand in the Agricultural Sector
Appendix 1
Dominican Republic
El Salvador
French Guiana
Grenada
Guadalupe
Guatemala
Guyana
Haiti
Honduras
Jamaica
Martinique
Montserrat
Netherlands Antilles
Nicaragua
Panama
Paraguay
Peru
St. Kitts and Nevis
St. Lucia
St. Pierre et Miquelon
St. Vincent and the Grenadines
Suriname
Trinidad and Tobago
Turks and Caicos Islands
Uruguay
Ecuador
Equatorial Guinea
Venezuela
v. Middle East and Africa
Algeria
22
Modeling Long-Term Oil Demand in the Agricultural Sector
Angola
Republic of the Congo
Gabon
IR Iran
Iraq
Kuwait
Libya
Nigeria
Qatar
Saudi Arabia
United Arab Emirates
Bahrain
Benin
Botswana
Burkina Faso
Burundi
Cameroon
Cape Verde
Central African Republic
Chad
Comoros
Côte d’Ivoire
Democratic Republic of the Congo
Djibouti
Egypt
Eritrea
Ethiopia
Gambia
Ghana
Guinea
Guinea-Bissau
Appendix 1
23
Modeling Long-Term Oil Demand in the Agricultural Sector
Jordan
Kenya
Lebanon
Lesotho
Liberia
Madagascar
Malawi
Mali
Mauritania
Mauritius
Mayotte
Morocco
Mozambique
Namibia
Niger
Oman
Réunion
Rwanda
Sao Tome and Principe
Senegal
Seychelles
Sierra Leone
Somalia
South Africa
South Sudan
Sudan
Swaziland
Syrian Arab Republic
Togo
Tunisia
Uganda
Appendix 1
24
Modeling Long-Term Oil Demand in the Agricultural Sector
United Republic of Tanzania
Western Sahara
Yemen
Zambia
Zimbabwe
vi. China
People’s Republic of China
vii. Other Asia
Afghanistan
American Samoa
Bangladesh
Bhutan
Brunei Darussalam
Cambodia
China, Hong Kong SAR
China, Macao SAR
Cook Islands
Democratic People’s Republic of Korea
Fiji
French Polynesia
India
Indonesia
Kiribati
Lao People’s Democratic Republic
Malaysia
Maldives
Micronesia (Federated States of)
Mongolia
Myanmar
Appendix 1
25
Modeling Long-Term Oil Demand in the Agricultural Sector
Nauru
Nepal
New Caledonia
Niue
Pakistan
Papua New Guinea
Philippines
Samoa
Singapore
Solomon Islands
Sri Lanka
Thailand
Timor-Leste
Tonga
Vanuatu
Viet Nam
viii. Other Eurasia
Albania
Armenia
Azerbaijan
Belarus
Bosnia and Herzegovina
Bulgaria
Croatia
Cyprus
Georgia
Gibraltar
Kazakhstan
Kyrgyzstan
Latvia
Appendix 1
26
Modeling Long-Term Oil Demand in the Agricultural Sector
Lithuania
Malta
Montenegro
Republic of Moldova
Russia
Romania
Serbia
Tajikistan
The Former Yugoslav Republic of Macedonia
Turkmenistan
Ukraine
Uzbekistan
Appendix 1
27
Modeling Long-Term Oil Demand in the Agricultural Sector
Appendix 2
Table A2.1. Regression results for OECD Europe.
Dep. variables Ln(LPG) Ln(KERO) Ln(Gas) Ln(Fuel)
Regressors
Ln(GDPREj)2.60362 0.817115
Ln(GDPREj)t–1 3.01981
Ln(GDPR)t–1
Ln(popj)− 4 3.7119 −17.6 8 61
Ln(pop)t–1 36.3423 15.0991
Ln(pop)t–2
Ln(Brentj)−0.0359199 − 0.174 814
Ln(Brent)t–1 − 0.14 90 82
Ln(LPG)t–1
Ln(KERO)t–1 0.785961
Ln(Gas)t–1 0.414465
Ln(Fuel)t–1 0.547019
C75.3026 −90.5990 −323.015 358.496
Adj. R20.972114 0.98941 0.897312 0.996054
No. of
observations
47 47 47 47
Normality test,
Chi2(2)
0.44503 [0.8005] 0.84147 [0.6566] 1.1814 [0.5539] 2.1052 [0.3490]
Hetero test F(10,36) = 1.2930 [0.2711] F(13,31) = 0.88917 [0.5722] F(10,35) = 2.4257 [0.0258] F(11,35) = 0.71045 [0.7204]
RESET23 test F(2,38) = 2.2558 [0.1186] F(2,33) = 1.1851 [0.3184] F(2,38) = 0.080631 [0.9227] F(2,37) = 3.6599 [0.0355]
28
Modeling Long-Term Oil Demand in the Agricultural Sector
Table A2.2. Regression results for OECD Asia Oceania.
Dep. variables Ln(LPG) Ln(KERO) Ln(Gas) Ln(Fuel)
Regressors
Ln(GDPREj)
Ln(GDPREj)t–2 0.53501
Ln(GDPR)t–1
Ln(popj)−46.9285
Ln(pop)t–1 44.7839
Ln(pop)t–2
Ln(Brentj)−0.227627 −0.0811527
Ln(Brent)t–1 0.150512 0.0583978 − 0.12 805
Ln(LPG)t–1 0.741843
Ln(KERO)t–1 0.978521
Ln(Gas)t–1 0.957646
Ln(Fuel)t–1 0.960628
C−14.7968 0.426828 41.6024 0.589687
Adj. R20.987057 0.916833 0.98285 0.955708
No. of observations 47 47 47 47
Normality test: Chi2(2) 1.391 [0.4988] 5.9399 [0.0513] 1.9909 [0.3696] 1.5988 [0.4496]
Hetero test 0.94215 [0.4496] 0.21993 [0.9681] 0.43938 [0.9162] 0.88671 [0.4804]
RESET23 test 2.5607 [0.0899] 0.10527 [0.9003] 1.8683 [0.1687] 1.5276 [0.2292]
Appendix 2
29
Modeling Long-Term Oil Demand in the Agricultural Sector
Appendix 2
Table A2.3. Regression results for Latin America.
Dep. variables Ln(LPG) Ln(KERO) Ln(Gas) Ln(Fuel)
Regressors
Ln(GDPREj)4.79132
Ln(GDPREj) t–2 −5.35509
Ln(GDPR)t–1
Ln(popj)− 5 07. 512
Ln(pop)t–1 501.684
Ln(pop)t–2
Ln(Brentj)−0.765143
Ln(Brent)t–1
Ln(LPG)t–1 0.332006
Ln(KERO)t–2 0.155851
Ln(Gas)t–1 0.931207
Ln(Fuel)t–1 0.670904
C124.655 19.7918 0.640112 1.55699
Adj. R20.964872 0.912081 0.990661 0.436464
No. of observations 47 47 47 46
Normality test: Chi2(2) 1.7621 [0.4143] 0.64590 [0.7240] 3.9875 [0.1362] 3.9 037 [0.1420]
Hetero test 1.9295 [0.0996] 1.2684 [0.2810] 0.78938 [0.4605] 2.4998 [0.0940]
RESET23 test 4.1089 [0.0236] 0.53269 [0.5916] 1.1628 [0.3222] 0.81199 [0.4508]
30
Modeling Long-Term Oil Demand in the Agricultural Sector
Table A2.4. Regression results for the Middle East.
Dep. variables Ln(LPG) Ln(KERO) Ln(Gas) Ln(Fuel)
Regressors
Ln(GDPREj)−1.09218 −3.01494 1.48364
Ln(GDPREj)t–2 −1.44701
Ln(GDPR)t–1
Ln(popj)93.2267 −88.8367 0.0730009
Ln(pop)t–1 − 9 0.1469 90.8895
Ln(pop)t–2
Ln(Brentj)
Ln(Brent)t–1 −0.0619985 −0.224096
Ln(LPG)t–1 0.379794
Ln(KERO)t–2
Ln(Gas)t–1 1.06442
Ln(Gas)t–2 −0.313223
Ln(Fuel)t–1 0.669346
C−32.0079 52.8271 1.05671
Adj. R20.994082 0.935943 0.910058 0.93872
No. of observations 28 28 28 28
Normality test, Chi2(2) 0.029551 [0.9853] 1.4849 [0.4760] 2.7287 [0.2555] 1.2654 [0.5312]
Hetero test 0.52431 [0.8461] 0.53769 [0.8284] 0.52753 [0.8213] 0.80318 [0.5368]
RESET23 test 1.0978 [0.3561] 0.57701 [0.5707] 2.3159 [0.1233] 2.3540 [0.1185]
Appendix 2
31
Modeling Long-Term Oil Demand in the Agricultural Sector
Table A2.5. Regression results for China.
Dep. variables Ln(LPG) Ln(KERO) Ln(Gas) Ln(Fuel)
Regressors
Ln(GDPREj)−0.0461426
Ln(GDPREj)t–2
Ln(GDPR)t–1 −3.55197 0.0790086 0.0370960
Ln(popj)94.5434 78.6934
Ln(pop)t–1 −93.3042 −78.832
Ln(pop)t–2
Ln(Brentj)0.510098
Ln(Brent)t–1 −0.0951232
Ln(LPG)t–1
Ln(KERO)t–2
Ln(Gas)t–1 0.563222 0.563222
Ln(Gas)t–2 0.069022
Ln(Fuel)t–1 0.598281
C85.7508 1.80921 1.54365 2.01594
Adj. R20.821252 0.949934 0.92372 0.836008
No. of observations 48 48 48 48
Normality test: Chi2(2) 4.9603 [0.0837] 5.3186 [0.07] 5.5182 [0.0751] 1.3058 [0.5205]
Hetero test 0.50725 [0.8739] 0.60788 [0.782] 0.63825 [0.7201] 1.3141 [0.2495]
RESET23 test 0.50321 [0.6084] 2.1504 [0.1308] 0.8732 [0.4654] 1.515 [0.2335]
Appendix 2
32
Modeling Long-Term Oil Demand in the Agricultural Sector
Table A2.6. Regression results for other Asia.
Dep. variables Ln(LPG) Ln(KERO) Ln(Gas) Ln(Fuel)
Regressors
Ln(GDPREj)10.2991
Ln(GDPREj) t–2 6.76184
Ln(GDPR)t–1
Ln(popj)499.424 − 2 07.718
Ln(pop)t–1 − 517. 3 05 440.015 −200.337
Ln(pop)t–2 −17.6262 208.140
Ln(Brentj)
Ln(Brent)t–1 −0.0642976
Ln(LPG)t–1 0.258165
Ln(KERO)t–2
Ln(Gas)t–1 0.775029
Ln(Gas)t–2 − 0.227172
Ln(Fuel)t–1 −230.659 0.666781
Ln(Fuel)t–2 −0.235070
C18 4.110 80.6663 −31.0276 −163.476
Adj.R20.92782 0.635928 0.996263 0.893252
No. of observations 47 35 47 47
Normality test: Chi2(2) 1.4661 [0.4804] 0.28754 [0.8661] 0.18217 [0.9129] 1.1352 [0.5669]
Hetero test 1.8353 [0.0850] 1.4426 [0.2353] 2.0728 [0.0485] 0.99417 [0.4787]
RESET23 test 2.4823 [0.0970] 0.73297 [0.4895] 1.5979 [0.2160 1.3338 [0.2762]
Appendix 2
33
Modeling Long-Term Oil Demand in the Agricultural Sector
Table A2.7. Regression results for Eurasia.
Dep. variables Ln(LPG) Ln(KERO) Ln(Gas) Ln(Fuel)
Regressors
Ln(GDPREj)0.917216
Ln(GDPREj) t–2 1.77912 −4.63751 2.03810
Ln(GDPR)t–1 −1.96904 −1.119 62
Ln(popj)−278.236 371.256 − 5 3. 311
Ln(pop)t–1 501.047 −342.996 124.738
Ln(pop)t–2 −222.851 −71.869 −30.7066
Ln(Brentj)0.0597688 −0.397552
Ln(Brent)t–1 0.350974
Ln(Brent)t–2 −0.409705 −1.06778
Ln(LPG)t–1 −0.395647
Ln(LPG)t–2 0.145594
Ln(KERO)t–2 −0.347902
Ln(Gas)t–1 0.748255
Ln(Gas)t–2 0.219646
Ln(Fuel)t–1 0.293646
Ln(Fuel)t–2
C10.9714 −425.638 14.3082 556.275
Adj. R20.929509 0.976679 0.993406 0.941484
No. of observations 47 47 47 37
Normality test: Chi2(2) 0.78473 [0.6755] 0.22023 [0.8957] 0.45079 [0.7982] 0.87429 [0.6459]
Hetero test 1.6976 [0.1039] 1.8996 [0.0678] 1.3837 [0.2194] 1.0503 [0.4358]
RESET23 test 15.397 [0.0000] 0.71438 [0.4961] 1.19 06 [0.3164] 1.9757 [0.1575]
Appendix 2
34
Modeling Long-Term Oil Demand in the Agricultural Sector
Notes
35
Modeling Long-Term Oil Demand in the Agricultural Sector
Notes
36
Modeling Long-Term Oil Demand in the Agricultural Sector
About the Authors
Fateh Belaïd
Fateh Belaïd was a full professor of economics at Lille Catholic University and
director of the Smart & Sustainable Cities research unit. Fateh has also held
various positions at the French Scientic and Technical Center for Building
and led multiple collaborative projects for the French Ministry of Ecological
Transition and the European Commission. He is an energy and environmental
economist drawing from the elds of applied microeconomics, energy
modeling, and econometrics.
He has published widely on household energy consumption, energy-saving
behaviors, individual preference and investment in energy efciency, energy
poverty, renewables, and energy policy. He received a habilitation for
supervising doctoral research from Orléans University, a Ph.D. in Economics,
an M.S. in Applied Economics & Decision Theory from Littoral University, and
an engineering degree in statistics.
His work has been published in journals including Ecological Economics, The
Energy Journal, Energy Economics, Economic Surveys, Energy Policy, and
Environmental Management
Mohammad Aldubyan
Mohammad is a research lead in KAPSARC’s Climate & Sustainability
program. His research focuses on energy efciency and energy demand in
buildings. He is currently leading the Residential Energy Model (REEM), which
simulates residential energy demand and estimates the impact of energy
efciency programs on Saudi Arabia’s housing sector. He also leads the
long-term KAPSARC Oil Market Outlook (KOMO) in buildings and agriculture
sectors.
Mohammad holds an M.Sc. in Renewable and Clean Energy from the
University of Dayton, Ohio and an M.Sc. in Economics from Purdue University,
West Lafayette.
37
Modeling Long-Term Oil Demand in the Agricultural Sector
About the Project
The aim of this project is to construct a model to forecast the direct oil demand driven by buildings
and agriculture sectors. The forecast takes in consideration the growth in buildings and agriculture
sectors and its contribution to the overall oil demand by accounting for some factors that primarily
impact energy demand such as national economy growth, energy efciency level, etc. It also
considers some technical and economic characteristics of both demand- and supply-side. The main
goal of conducting this forecast is to feed KAPSARC long-term KOMO model for better long-term
oil demand outlook.
This project is expected to provide an annual long-term forecast of oil demand associated with
buildings and agriculture sectors. A detailed forecast of oil demand is to be explained by many
factors such as the growth of each building type relative to the others, overall energy efciency
level, and geographical distribution, to name few.
38
Modeling Long-Term Oil Demand in the Agricultural Sector
www.kapsarc.org
