Professional Documents
Culture Documents
net/publication/350962342
CITATIONS READS
0 724
4 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Thi-Nham Le on 21 April 2021.
To cite this article: Chia-Nan Wang, Min-Chun Yu, Nguyen-Nhu-Y Ho & Thi-Nham Le (2021):
An integrated forecasting model for the coffee bean supply chain, Applied Economics, DOI:
10.1080/00036846.2021.1887447
a
Department of Industrial Engineering and Management, National Kaohsiung University of Science and Technology, Kaohsiung, Taiwan;
b
Department of Business Administration, National Kaohsiung University of Science and Technology, Kaohsiung, Taiwan; cInternational School,
Vietnam National University Hanoi, Hanoi, Vietnam; dSoutheast Asian Languages and Cultures, National Chengchi University Taipei, Taiwan
ABSTRACT KEYWORDS
Coffee is the most traded commodity after petroleum. The Vietnamese coffee bean industry has Optimization; forecasting
raised concerns lately over an inefficient coffee value chain; bets on coffee price uncertainty are Model; supply Chain
increasing worldwide in the current. Accurate optimization of coffee bean prices helps manufac Management; coffee Bean
turers to control an unpredictable market and upgrade cooperativeness in sustainable agriculture.
The authors proposed a forecasting method to deal with demand volatility and uncertainty in
volumes and coffee bean prices. In this paper, we applied the forecasting nonlinear grey Bernoulli
model (NGBM) (1,1). NGBM (1,1), which is based on the parameter optimization algorithm, can
increase the precision of predictions. NGBM (1,1) was integrated with Fourier residual modification
model to forecast coffee bean price, which was a crucial factor in the Vietnamese coffee bean
supply chain. The price of coffee beans was calculated using a differential equation in an uncertain
system, along with actual data collected over the past six years. The results of this study demon
strate that an integrated forecasting model is an effective forecasting method. This research can
help companies to control risks that come with uncertain coffee prices and reduce risks in the
sustainable agriculture supply chain.
I. Introduction
variability in the agricultural production process
As with supply chain management for manufac is stochastic and not under farmers’ control. In
turers, agri-food supply chain management the meantime, improvements in modelling price
involves integration and collaboration among uncertainty in agricultural supply chains have
farmers, agricultural agents, manufacturers, distri been emphasized, and most of the effort invested
butors, and retailers to meet consumer needs and consisted of predicting future trends through effec
expectations for quantity, quality, and price. Agri- tive grey forecasting based on past and present
food supply chains are economic systems that data. One model has observed an extremely unfa
distribute benefits and apportion risks among par vourable trend in the future distribution of indi
ticipants. One of the food industry’s core compe genous Arabica coffee due to accelerated global
tencies is selecting appropriate suppliers, which climate change.
provide quality agricultural products for food- In the 1990s, the French brought coffee seeds to
processing activities. Vietnam, which eventually became the
Mussatto et al. (2011) and Feria-Morales (2002) world’s second-largest exporter of coffee, supplying
revealed that coffee is one of most popular bev approximately one-quarter of the United
erages in the world; and is also the second-most Kingdom’s coffee. Coffee contributes to Vietnam’s
traded commercial goods behind petroleum. The gross domestic product (GDP), creates jobs, and
coffee industry is expected to face severe challenges stimulates socio-economic development. There
in the decades to come because of uncertainty in was a disadvantage arrangement that uncertainty
commodity prices. In the agricultural supply chain, fluctuations in the market price of coffee beans are
management always wants to reduce risks, as there negative to the farmers’ benefit of growing coffee
is uncertainty in the fluctuation of prices in the beans. Thus, focusing on arrangements to promote
market. The extreme susceptibility to weather international trade in Vietnamese coffee beans
CONTACT Nguyen-Nhu-Y Ho hngny@vnu.edu.vn International School, Vietnam National University Hanoi, Hanoi Vietnam
© 2021 Informa UK Limited, trading as Taylor & Francis Group
2 C.-N. WANG ET AL.
through price forecasting is important in helping a result of these two weaknesses, Vietnamese cof
Vietnam maintain a good strategy in the global fee’s export price has always been lower than the
market for coffee. same product from other countries, and the price
has remained uncertain for a long time. According
to Asche, Jaffry, and Hartmann (2007) and Feng
The coffee industry in Vietnam
and Tan (2019), price always an important criter
Feria-Morales (2002) claimed that Vietnam, one of ion when selecting a vendor in a supply chain.
the top-10 coffee-producing countries, provides García-Martos, Caro, and Jesús Sánchez (2015)
high-quality green beans used as seeds for propa claimed that price forecasting was an interesting
gation. According to Franca et al. (2005), the cli problem in profit maximization strategy. Li et al.
mate, soil, cultivation practices, and irrigation (2018) indicated that the optimization of price
systems of specific locations play essential roles in uncertainty problems with small samples and
developing good quality coffee. insufficient information significantly improved
Agricultural commodity booms have challenged the performance models’ forecasting. In this
the coffee economies in Vietnam and other devel research, to forecast coffee beans’ future perfor
oped countries. Moving from a planned economy mance, we developed the NGBM (1,1) and adopted
to a market-oriented one, Vietnam has become one the Fourier series model to forecast coffee bean
of the fastest-growing economies in Asia from prices and make coffee bean selection less risk-
exporting coffee bean, just like Brazil in South averse.
America. Vietnamese coffee has risen to the top
of the world’s coffee manufacturers in world-
II. Literature review
record time, which has been affirmed by equally
fast switches policies to create a suitable market In agri-food supply chains, an organization’s beha
network. However, the Vietnamese coffee industry viour involves complex decision-making from the
faces problems of productivity, quality, and price, customer in complex networks consisting of small-
thus affecting the sustainable development of the and medium-sized enterprises. The agri-food sup
industry. Vietnam coffee is the main crop that ply chain became part of interactions with multi
brings the export value of about 3 USD billion national companies providing input to the supply
dollar per year. However, Vietnam only gets a low chain and engaging in its retail services. Leroy et al.
profit percentage because Vietnam always exports (2006) showed that high consumer demand
raw coffee bean. Following the Vneconomy website requires speciality coffee markets. Consumers
by coffee exports, in August 2018 Vietnamese cof paid the premium price required for special coffee
fee bean was estimated to reach 143 thousand tons that retains the original quality. Therefore, fore
with a value of 260 USD million dollar. casting further system improvement from pre
Accumulated coffee export in the first eight months viously observed data seems to be a practical
of 2018 is estimated at 2.5 USD billion dollar, up by alternative. With worldwide overproduction, this
15.5% in volume. However, this rate decrease by study considers coffee bean prices in the future.
2.5% in value over the same period in 2017. The Specific system development laws based on real
Vietnamese coffee industry still lacks reputable data must be discovered to obtain a reliable fore
brands and has a few advertising marketing activ cast. These laws would reduce the risk of market
ities. Therefore, Vietnam still has to export through price volatility in the future.
the third market, which means export prices face
uncertainty. Grey theory
The export price of Vietnamese coffee is deter The grey forecasting model (GM) originated by
mined based on the international coffee prices; Deng (1989) focuses on solving problems using
thus, the domestic coffee industry is fundamentally small sample sizes and insufficient information.
insecure and unstable. Furthermore, quality The conventional GM is based on knowledge of
improvement in Vietnam’s coffee beans was slow the least square method and first-order linear
due to the lack of standardized procedures. As ordinary differential equations. The GM is
APPLIED ECONOMICS 3
principally suited for handling situations in which studies applied NGBM (1, 1), which has provided
only limited data are convenient for forecasting, promising results, NGBM (1, 1) was sometimes
and the system follows the time series data. In unsuitable in some extraordinary situation exam
recent years, grey systems theory has been success ple: trending of business or the result in the future.
fully employed in the fields of agriculture, indus The NGBM was a novel grey forecasting model that
trial textiles, and industrial transportation. The GM integrates Bernoulli differential equations suitable
(1,N) is the main model in grey forecasting, for forecasting various developing trend curves in
Kayacan, Ulutas, and Kaynak (2010) in grey sys coffee prices. According to Hsu (2010), and Zhou
tems theory, GM (n,m) denotes a grey model, et al. (2009), in the forecasting model, the short-
where n is the order of the ordinary differential term forecasting is within 1–2 years, medium-term
equations and m is the number of Grey variables. is from 3 to 5 years, and the period of more than
Tien (2012) suggested that the various predictions 5 years is considered to be long time. In this paper,
of the existing model GM (1,n) can be an unex the NGBM is applied to medium- and long-term
pected and large sample, which is also a strength of coffee bean forecasting and an algorithm to opti
the basic grey methodology. The prediction accu mize the parameters of the NGBM is proposed.
racy can be handled in situations, which include Shaik and Walid (2011) showed that by optimiz
uncertain and insufficient information. The differ ing techniques effectively, the estimation of
ential equations of the GM (1, 1) model have time- NGBM, could enhance the accuracy of the predic
varying coefficients. tion model. Although those studies applied NGBM
Shrivastava and Panigrahi (2014) showed that (1, 1), which has provided promising results,
forecasting models are evaluated based on multiple NGBM (1, 1) was sometimes unsuitable in some
accuracy measures to reduce price risk. Coffee particular situations. The authors theorize that if
price forecasting is one of the overwhelming the data fluctuated wildly or included lots of noise,
threats to market participants, managers, and the data forecasted by NGBM (1,1) would also
farmers due to the high volatility of coffee bean fluctuate wildly. To deal with this problem, this
prices. Hovelaque, Duvaleix-Tréguer, and Cordier research adopts the improvements of NGBM
(2009) argued that the sustained competitiveness of (1, 1) and integrated Fourier series to build an
Vietnam’s coffee bean industry depends on its abil effective model with higher predictive accuracy.
ity to find stable and sustainable solutions to the This methodology, Providing strategy value to
imminent crisis in Vietnam’s coffee sector. This boost the distribution and consumption for opti
research aimed to integrate the combined non mization coffee bean value chain system and con
linear grey Bernoulli model – (NGBM) (1,1) and tributing more sustainability.
Fourier series to forecast coffee bean prices in the
future. Fourier series
In the model, non-linear relations González,
Nonlinear grey Bernoulli model (1,1) Jaramillo, and Carmona. (2008) adopted the
In 2008, Chen (2008) re-examined an example in Fourier series aiming to be used to forecast volati
Deng’s edition, and after applying the NGBM to lity values for which the series presents an apparent
the analytical results, demonstrated that it effec annual behaviour of the spectrum plot of the fluc
tively enhanced the precision of the methodology. tuation series with high-precision. The Fourier ser
The nonlinearity of natural phenomena increases ies was developed by Brigham and Brigham (1988)
NGBM’s precision. The traditional Grey Model to enhance forecasting accuracy when using highly
was easy to understand and simple to calculate, fluctuating data sets. According to Huang and Lee
with satisfactory accuracy. In case, if the data to (2011) a Fourier series where the fundamental fre
adjust the model have large variability, that acquire quency was assumed to be a modification model
will not get higher forecasting precision. This can refine the forecasting effectiveness of stochastic
investigation examines the improved grey model volatility data and time series. Fourier series are
and the knowledge of the elementary ordinary also adaptive models to refine the forecasting effec
nonlinear differential equations. Although those tiveness of random and uncontrolled data, which
4 C.-N. WANG ET AL.
describes the trends and fluctuations. The para researching the Fourier series, that the analytical
meter estimation of a Fourier series is formulated results demonstrated that applying a Fourier series
as one of the combinatorial optimization pro effectively enhanced the forecast’s accuracy. As
blems. A Fourier series was the model parameters previously mentioned, the mathematics in
by optimization techniques that can effectively a Fourier series becomes increasingly sophisticated,
intensify the model’s precision level. González, making the series difficult to understand and apply.
Jaramillo, and Carmona. (2008) suggested that The forecasting results in a wide amplitude fit with
the trend of a Fourier series, predicted with main a Fourier series. Thus, this study demonstrated
tenance planning and market research in the opti how to improve the Fourier series model by show
mized model, could enhance the prediction’s ing the error indices of fluctuation forecasting
accuracy. A Fourier series was also more accurate obtained by prior validation. In this paper, studies
at demand forecasting than using maintenance followed the isomorphism from the original equa
planning and market research to predict fluctua tion and were then combined with the methodol
tions. Fourier Series showed the effects of data ogy of elementary ordinary nonlinear differential
eventuality, or centre-symmetry situation could equations. To prove this novel approach works and
be adjusted to fit the result of one-time accumu is effective, this research has utilized the genetic
lated generating operation (1-AGO) of raw data. algorithm for the optimization problem and then
However, mathematics in Fourier Series precision constructed a Fourier-series-based NGBM. To
can be increased owing to the nonlinearity of improve the Fourier series model, we verified the
natural phenomena the authors adapted. The effectiveness of the proposed model. We proposed
non-linear Grey Bernoulli Model is used to adjust a research framework Figure 1 consisting of six
the model to acquire higher forecasting precision. steps, as follows:
Abate and Whitt (1992) found that many meth Step 1: Collect coffee bean price data. Historical
ods have been proposed to increase accuracy by coffee bean price data were collected from the
modelling the residuals of the original grey pre website Giacaphe (2019) from 2014 to 2019. The
diction. In fact, the residual diagnosis was forecasting techniques available have increased
a commonly used correction approach for time both in number and complexity. The purchasing
series prediction. One of those residual correction
approaches was to use the Fourier series. Besides,
Odan and Reis (2012), the linear forecasting Step 1: Collect coffee bean price data
methods were widely used because they were
easy to develop and implement, in addition to
being simple to understand and interpret.
However, in the real world, data have varied Step 2: Develop a Non-linear Grey
degrees of nonlinearity. The theory of combina Bernoulli Model - NGBM (1,1)
III. Methodology
Step 5: Verify the effectiveness of the
model
Forecasting techniques have developed rapidly,
thus becoming superior to conventional statistical
models. González (2008) determined, after Figure 1. Research framework.
APPLIED ECONOMICS 5
2 3 2 3
department in manufacturing paid attention to xð0Þ ð2Þ zð1Þ ð2Þ 1
forecasting future tendencies of some events, such 6 xð0Þ ð3Þ 7 6 zð1Þ ð3Þ 17
6 7 6 7
as investment in the market, which must be fore Y ¼ 6 . 7; B ¼ 6 .. 7 (1)
4 .. 5 4 . 5
casted to obtain higher profits and reduce the
investment risk for Vietnam’s coffee prices. xð0Þ ðnÞ zð1Þ ðnÞ 1
Step 2: Develop an NGBM (1,1) model that The least squares estimated sequence of the GM
improved the forecasting effectiveness of the sto (1,1) model in equation. (2) satisfies the condition
chastic fluctuation data, boosted an effective meth � 1
^a ¼ BT B BT Y. With all notations adopted from
odology for forecasting Vietnam coffee beans � 1
Theorem (3), if½a; b�T ¼ BT B BT Y, then dxdt þ
ð1Þ
price – designing the fluctuation of six-year fore
casting in cases involving samples with a large axð1Þ ¼ b was a whitenization equation of the GM
change in data forecasting. (1,1) model in equation (4). The time response
Step 3: Develop a Fourier series forecasting sequence of the GM (1,1) model in equation. (5)
model for coffee bean prices. Every simulation was given below:
result determined that the suggested model offered � �
ð1Þ ð0Þ b b
a more accurate forecast than several other types of ^x ðk þ 1Þ ¼ x ð1Þ e ak þ ; k
a a
forecasting models. Hence, this research improved
¼ 1; 2; . . . ; n (2)
the Fourier series based on NGBM (1,1) models to
refine the forecasting effectiveness of stochastic The parameters ð aÞ and b of the GM (1,1)
volatility data, thus providing an effective method model accumulated the development coefficient
for forecasting the price of Vietnam coffee beans and calculated by the ordinary least squares
and further improving the accuracy of forecasting method. The former reflected the development
that can be applied for cases involving sample data states of ^xð1Þ and ^xð0Þ . According to Cui et al.
with large fluctuations. (2013) the variables denoted by the system of inter
Step 4: Measure the integrated forecasting mod est external or pre-defined. The GM (1,1) model
el’s accuracy to judge its effectiveness at integrating was the system whitened by applying the inverse
the NGBM and Fourier series models. In this step, accumulated generating operation (IAGO) techni
we hypothesize that integrating the Fourier series que for parameter variance, rather than a first order
builds an effective model that increases its predic differential equation.
tive accuracy.
Step 5: Concerning the evaluation performance Nonlinear grey Bernoulli model – (NGBM (1,1)
of the forecasting volatility model, based on verify
ing the results of predicted values that match the According to Chen, Chen, and Chen (2008)
actual values, in this step, this study adopted mean designed the Nonlinear Grey Bernoulli Model –
absolute percentage error (MAPE) approaches to NGBM (1,1) model was as follow:
evaluate the forecasting performance model GM
(1,1), NGBM (1, 1), and Fourier series models. Step 1: Theorem: Let’s matrix Xð0Þ stand for the
Model characteristics included periodicity, ran nonnegative original history of time series data
� �
domness, and tendency. Xð0Þ ¼ xð0Þ ðti Þ , i ¼ 1; 2; . . . ; m; where xð0Þ ðti Þ
corresponds to the system output at a time ti and
m is the total number of the accurate data sequence.
Step 2: A new sequence Xð1Þ could be approxi
Grey forecasting model
mately rewritten as the following one-time accu
The progress of construction GM (1,1) model is mulated generating
� ð1Þ � operation (1-AGO), which is:
described below: ð1Þ
X ¼ x ðti Þ , where
Pk
Theorem: Let Xð0Þ ,X ð1Þ , and Zð1Þ be the original xð1Þ ðti Þ ¼ xð0Þ ðti ÞWhere t ¼ 1; 2; 3; . . . ; i
i¼1
sequence data except that X ð0Þ is non-negative. If Step 3: In NGBM (1,1) methodology the yields of
^a ¼ ða; bÞT (3) was a parameter sequence and the Grey differential equation showed:
6 C.-N. WANG ET AL.
h in
demonstrate the improvement in the accuracy of
xð0Þ ðtk Þ þ azðzÞ ðtk Þ ¼ b zð1Þ ðtk Þ (3)
NGBM (1,1), and serves as an adjustable parameter
The first-order accumulate progression satisfies detail below:
as follow difference equation:
Step 5: From the solution of equation (4) could be
dxð1Þ ðtk Þ h in obtained after the parameters a and b are predicted
þ axð1Þ ðtk Þ ¼ b xð1Þ ðtk Þ (4) value. Such as:
dtk
where Wheren�1; k ¼ 1; 2; 3; . . .
Step 6: The response function of inverse accu
zð1Þ ðtk Þ ¼ pxð1Þ ðtk Þ þ ð1 pÞxð1Þ ðtk 1 Þ; k mulated generating operation (I-AGO) to ^xð1Þ ðtk Þ,
¼ 2; 3; . . . ; m; (5)
the estimated the coffee bean price of ^xð0Þ ðtk Þ could
pwas denoted the value of production coefficient be forecasted as follows:
with a closed the background value of ½0; 1�, which
traditionally set balance as 0.5. To demonstrate the
expressions regarding parameters a and b in the
NGBM (1,1) model. In this case study, the power Development Fourier series modification by
n is used to be the adjustable parameter that the NGBM (1,1)
real numbers represent and serve as an adjustable Adapting González, Jaramillo, and Carmona.
parameter in terms of a matrix. Based on the pre (2008) separated the idea of Fourier series systems
vious research, if the parameter has n = 0.0381, and those of forecasting boxes as an important test.
NGBM (1,1) will get excellent flexibility.
Step 4: Chen, Chen, and Chen (2008) believed Theorem: Step 1: Assumed that x be the initial
that the accuracy forecasting model was directly values of m entries, and v is the predicted value
stochastic volatility by the parameter p in the back obtaining from NGBM (1,1). The residual series
ground value. The accuracy values of this para named ε which could be written as:
meter could be calculated by applying a specific ε ¼ feðkÞg,k ¼ 2; 3; . . . m
simultaneous optimization. However, in this where
research, equation (4), depending on the value of
parameters a and b could be calculated by adapting eðkÞ ¼ xðkÞ vðkÞk ¼ 2; 3; . . . m (12)
the ordinary least squares method (OLS):
Step 2: With the same definitions of a Fourier series
� �
a � 1 development by the residual sequence of NGBM
¼ B T B B T Ym (6) (1,1) expressed as follow the equation ^εðkÞ
b
approximately:
where
Xz �� � �� � ��
2 1 2πi 2πi
2 3 �n 3 ^εðkÞ ¼ að0Þ þ ai cos ðkÞ þ bi sin ðkÞ
xð0Þ ðt2 Þ zð1Þ ðt2 Þ zð1Þ ðt2 Þ 2 i 1
m 1 m 1
6 �n 7
6 xð0Þ ðt3 Þ 7 6 zð1Þ ðt3 Þ zð1Þ ðt3 Þ 7 (13)
6 7
Y¼6 . 7; B ¼ 6
6 .. 7
7
4 .. 5 4 . 5 k ¼ 1; 2; 3; . . . m
ð0Þ � n Where
x ðtm Þ ð1Þ
z ðtm Þ ð1Þ
z ðtm Þ
(7) ðm 1Þ
Z¼ 1 (14)
In equations (8) and (9), Zheng et al. (2011) 2
believed that model accuracy was directly influ Step 3: In this step, the minimum deployment
enced by the optimal solution of parameters p in frequency of the Fourier series, and Z only take an
the background power, given that the background integer value. Therefore, according to González,
value was expressed. Adopting the actual forecast Jaramillo, and Carmona. (2008), the residual series
ing example of coffee bean price rates to was rewritten as
APPLIED ECONOMICS 7
IV. An illustrative example of the Vietnamese different kinds of NGBM (1,1) models, along with
coffee supply chain the optimized Fourier series, showed the effective
ness of the proposed model in high fluctuation data
sets. This paper demonstrated the forecasting fra
Vietnamese coffee bean supply chain
mework for coffee bean prices applied to the real
In 2017, Vietnam was selected to become one of the case of Vietnam. The procedure for the terms of the
suppliers of Arabica coffee for the Starbucks coffee proposed prediction model has two stages. The first
company worldwide. Starbuck’s choice of Vietnam stage established the NGBM (1,1) to roughly pre
as a supplier proves that Vietnamese coffee will dict future data using a set of the most recent data.
retain a high price in the future, which is both The second stage used a Fourier series to refine the
a challenge and a chance for Vietnam’s coffee residual error in the NGBM (1,1) more details in
bean in the agribusiness supply chain. However, the next part.
the food industry’s sustained competitiveness will
depend on the ability to estimate price uncertain
ties in the future, demonstrating permanent and Grey forecasting model
sustainable solutions to address the imminent crisis The sequence of raw data Xð0Þ ¼ (1725.00, 1540.00,
in Vietnam’s coffee industry. This paper primarily 1637.00, 1637.00, 1651.00, 1683.00) simulates this
focuses on estimating unpredictable prices to help sequenceXð0Þ using the following three GM (1,1)
wholesale companies or agencies to control the models and comparing the accuracy of the
input material’s stability and risk allocation. simulation:
Vietnam grows Arabica coffee in the Lam Dong, From Eq. (2)xð0Þ ðkÞ þ azð1Þ ðkÞ ¼ b, compute the
Dien Bien, Son La, Quang Tri, and Dak Nong accumulation generation of Xð0Þ as follows:
provinces. Each region has its own unique coffee �
Xð1Þ ¼ xð1Þ ð1Þ; xð1Þ ð2Þ; xð1Þ ð3Þ; xð1Þ ð4Þ; xð1Þ ð5Þ ¼
flavour. In this study, we created a model and
applied it to a price-forecasting case study in the (3265.00, 4902.00, 6539.00, 8190.00, 9873.00).
most famous coffee province in Vietnam: Lam We check the quasi-smoothness from σ ð1Þ ðkÞ ¼
xð0Þ ðkÞ
Dong. xð1Þ ðk 1Þ
: It follows that σ ð1Þ ð2Þ = 0.69, σ ð1Þ ð3Þ= 0.37,
σ ð4Þ= 0.28, and σ ð1Þ ð5Þ= 0.22, σ ð1Þ ð6Þ= 0.18.
ð1Þ
Collect data of the coffee bean prices Therefore, k > 5, σ ð1Þ ðkÞ 2 0:1; 1� with σ ¼ 0:90,
Based on the data collected from the website that is, the law of quasi-exponentially and the con
Giacaphe (2019) in Table 1- uncertain coffee dition of quasi-smoothness is stratified. Thus, we
prices obtained for the six years from 2014 to can establish a GM (1,1) model for Xð1Þ . Using
2019 in Lam Dong, the research represents the adjacent neighbours of sequence, let Zð1Þ ¼
Vietnamese coffee beans. Procedures of the opti �
zð1Þ ð1Þ; zð1Þ ð2Þ; . . . ; zð1Þ ðnÞ be the sequence
mized NGBM (1,1) and Fourier series models
were established by minimizing an objective generated from X ð1Þ by the adjacent neighbour
function, with the constraints being0 � p � 1 means sequence Zð1Þ ¼
�
and n�1, to obtain the global optimization of xð1Þ ð1Þ; xð1Þ ð2Þ; xð1Þ ð3Þ; xð1Þ ð4Þ; xð1Þ ð5Þ ¼ (2495,
parameters p and n (just for NGBM (1,1) and 4083.5, 5720.5, 7365, 9031.5).
Fourier Series). In the coffee industry, forecast In addition, matrix B and constant vector YN are
ing the future tendency of some events, such as accumulated as follows:
investment in the market, must be forecasted to
obtain higher profits and reduce the investment 2 3 2 3
2595 1 1540
risk for Vietnam’s coffee prices. 6 4083:5 1 7 6 1637 7
6 5720:5 1 7 6 1637 7
B¼6 Y ¼
7 N 6 7
4 7364:5 1 5 4 1651 5
Validation of the forecasting model 1683
9031:5 1
In this section, three forecasting models were used
to compare the proposed model with several Using the least squares estimation, we obtain the
APPLIED ECONOMICS 9
sequence of parameters ^a ¼ ½a; b�T as follows By taking the above the formula settings the
� �
T
� 1 T 0:0182 optimal parameter value p ¼ 0:5and n = 0.0381,
^a ¼ B B B Y ¼ . We establish
1524:7043 which were calculated by adapting the ordinary
the following model dxdt
ð1Þ
0:0182xð1Þ ¼ 1524:7043 least squares method (OLS):
� �
and its time response from ^xðk þ 1Þ ¼ a � 1
� = BT B BT Ym
xð0Þ ð1Þ ba e aðkÞ þ ba ¼ 81693:8e 0:0102 b
83418:88333 Substituting different values of k into 2 3 2 3
1540 1725 4360
the equation: 6 1637 7 6 3265 6923 7
6 7
where Y ¼ 6 1637 7, B ¼ 6
6 4902 9541 7
7
k ¼ 1X ð1Þ ð1Þ ¼ 1725:00 6 7 4 5
k ¼ 2X ð1Þ ð2Þ ¼ 3446:54 4 1651 5 6539 12173
k ¼ 3X ð1Þ ð3Þ ¼ 5226:05 1683 8190 14846
k ¼ 4X ð1Þ ð4Þ ¼ 7011:07
k ¼ 4X ð1Þ ð4Þ ¼ 8774:14
k ¼ 5X ð1Þ ð5Þ ¼ 10523:81 Building the GNBM (1,1) form equations (8)
k ¼ 6X ð1Þ ð6Þ ¼ 11593:12
k ¼ 7X ð1Þ ð7Þ ¼ 13346:22
and (9), estimates of the parameters a and b were
k ¼ 8X ð1Þ ð8Þ ¼ 16948:95 as follows:
�
k ¼ 9X ð1Þ ð9Þ ¼ 18800:31 a ¼ 0:0172
k ¼ 10X ð1Þ ð10Þ ¼ 22606:11
b ¼ 3001961:3
Compute the simulated values of Xð0Þ using (5) From the solution of equation (4) could
the original series according to the be obtained after the parameters a and b
accumulated generating operation by using are predicted values. Such as:^xð1Þ ðti Þ =
^xð0Þ ðk þ 1Þ ¼ αð1Þ ^xð1Þ ðk þ 1Þ ^xð1Þ ðkÞ. Using the
same calculation and procedures, all forecasted (10)Wheren�1; k ¼ 1; 2; 3; . . .
values of all companies from 2020 to 2025 are (6) The estimated result ^xð0Þ ðti Þ from Fourier
found for the Vietnamese coffee evaluation shown Series modified using equation (10) as follows:
�
in Table 3. ð 0Þ ^xð0Þ ðtk Þ ¼ 1725:00
^x ðti Þ ¼
^xð0Þ ðtk Þ ¼ 3439:64 ^xð1Þ ðtk 1 Þ
Nonlinear grey Bernoulli model – (NGBM (1,1) (k ¼ 2; 3; . . . m)
The NGBM (1,1) model of the empirical process is Using the same calculation and procedures, all
described as follows forecasted values by the NBGM (1,1) of the coffee
(1) Denoted the original history of time series bean price from 2020 to 2025 were calculated for
data by the Vietnamese coffee evaluation shown in Table 3.
^εðkÞ ¼ 859:1093
Xz � � � � ��
2πi 2πi
(3)With the residual series can be estimated in þ 27:1971cos ðkÞ þ 31:6859 sin ðkÞ
i¼1
5 5
matrix ε ¼ PC form as
(5) Then the estimated ^v values to calculate
coffee bean price forecasting:
The Fourier series mathematics be calculated:
^v ¼ f1725; 1690:54; 1685:51; 1675:02; 1669:07;
1683:67g
Figure 2. Curves of actual versus simulated values using traditional and optimized GM (1,1), NGBM (1,1), and Fourier series.
After finding each model’s future price fore As shown in Figure 2, for the given fluctuating
casts, Figure 2 displays the validation of coffee curve, the NGBM (1,1) surpasses the performance
bean prices from 2014 to 2019 for each forecast expectations of the Fourier series model in the
ing model, such as GM (1,1), the NGBM (1,1), simulation results. Price forecasting was the best
and Fourier Series Model. We presented the method to generate new ideas for self-innovation
price forecasts for coffee beans grown in Lam in the coffee industry. In this research, we solved
Dong below: two major problems: first, combing the NGBM
Wang and Phan (2015) explained the remark (1,1) and Fourier residual modification models to
able higher accuracy of the NGBM (1,1) model improve the effectiveness of forecasting stochastic
compared with other GM (1,1) models, as exem volatility data. Second, providing an effective
plified in the fluctuating sequence X (0) = (1725, method for forecasting coffee bean prices in the
1540, 1637, 1637, 1651, 1683). In this section, Vietnam case study. Besides, improving the accu
we used coffee prices to compare the accuracy of racy of six-year forecasts in cases involving sample
the NGBM (1,1) with the Fourier series model data with large fluctuations. This price forecasting
to demonstrate the improvements in the Fourier method should help coffee companies ascend to
series model. Table 2 and Figure 3 show the greater heights by strengthening weak links in
results. According to Table 4, the NGBM (1,1) Vietnam’s agricultural supply chain.
model and Fourier series model has n = 0.0381,
p = 0.5. In this optimization method, the
V. Conclusions
Fourier series model’s average error decreased
from 3.926% to 2.077%. However, even though This paper developed an NGBM (1,1) and exam
the MAPE percentage was small, for the fluctu ined its validity in forecasting the price of coffee
ating curve given in Figure 2, the Fourier series grown in Lam Dong. Estimating this unpredictable
model surpasses the traditional Fourier series price would help companies manage their produc
model in the simulation results, being nearly tion in a timely manner and control input material
equal to the actual coffee bean price line. stability and risk allocation. Research results sug
Besides, Table 3 shows the MAPEs from 2012 to 2017 gest the combination of the NGBM (1,1) and
for coffee bean growers in Lam Dong, with an average Fourier series models to forecast the effect of cur
forecasting accuracy of 7.050% for both GMs (1,1), rent price movements on coffee prices in the future.
3.926% for the NGBM (1,1), and 2.077% for the Further analysis revealed the forecasting result of
Fourier series. The model precision improved obviously the uncertainty price’s Vietnamese coffee bean ana
when applied the Fourier series. These results strongly lysis. The MAPE of GM (1,1) and MAPE of NGBM
suggest that the forecasted accuracy was better than 90%. (1,1) having a percentage of MAPE at around 10%–
The performance of Fourier series forecasting earned the 20% shows that the NGBM (1,1) and GM (1,1) have
level of ‘excellent’ in MAPE. Thus, the Fourier series was adequate levels. However, with optimistic accuracy
an effective model for forecasting coffee bean prices in in the real work, the price uncertainty estimated by
this case study. Using a predictive mathematical model the Fourier series model with the MAPE has an
priced using three forecasting models would help econ ‘excellence’ level as well as with the actual value. In
omists estimate market pressures and build a more effi this paper, an applied and improved Fourier series
cient supply chain to control for volatility levels in the model, and the NGBM (1,1) were based on data
world coffee market. from 2014 to 2019 to forecast the uncertainty of
coffee prices. A mathematical model for integer
forecasting model solved prices of Lam Dong
growers. Moreover, this model could be applied
Fluctuating raw data sequence example to forecast different unit priced coffee in Vietnam
or in the world. Coffee bean price forecasts validat
The parameter evaluation of the fitted models was
ing the model could potentially improve forecast
developed by Wang and Phan (2015) and Chen
accuracy for a wide variety of macroeconomic out
(2008) below:
comes, such as risk reduction in the market.
12 C.-N. WANG ET AL.
Forecasting accuracy was determined using MAPE Feng, Z., and C. Tan. 2019. „Pricing, Green Degree and
calculations, and the results showed that the pro Coordination Decisions in a Green Supply Chain with
posed Fourier series model was more effective after Loss Aversion.„ Mathematics 7(3): 239. doi:10.3390/
math7030239
integration with NGBM (1,1). This research pro Feria-Morales, A. M. 2002. “Examining the Case of Green
poses a new and useful decision-making model that Coffee to Illustrate the Limitations of Grading Systems/
uses a real case study to show the proposed model’s expert Tasters in Sensory Evaluation for Quality Control.”
applicability and effectiveness. The integrated Food Quality Preference 13 (6): 355–367. doi:10.1016/
approach provides empirically meaningful and S0950-3293(02)00028-9.
helpful results for the sustainable development of Franca, A. S., L. S. Oliveira, J. C. Mendonça, and X. A. Silva.
2005. “Physical and Chemical Attributes of Defective Crude
the food supply chain industry in the future. In and Roasted Coffee Beans.” Food Chemistry 90 (1–2):
addition, further research could help develop 89–94. doi:10.1016/j.foodchem.2004.03.028.
a better perspective and an increasingly integrated García-Martos, C., E. Caro, and M. Jesús Sánchez. 2015.
approach for even more complex supply chains. “Electricity Price Forecasting Accounting for Renewable
Energies: Optimal Combined Forecasts.” Journal of the
Operational Research Society 66 (5): 871–884. doi:10.1057/
jors.2013.177.
Disclosure statement
Giacaphe (2019). Available online: https://giacaphe.com/gia-
No potential conflict of interest was reported by the authors. ca-phe-truc-tuyen/ (Accessed 8th December 2019).
González, R.-E., -M.-M. Jaramillo, and D.-F. Carmona. 2008.
“Monthly Electric Energy Demand Forecasting with Neural
Networks and Fourier Series.” Energy Conversion
ORCID Management 49 (11): 3135–3142. doi:10.1016/j.
Nguyen-Nhu-Y Ho http://orcid.org/0000-0002-7188-7927 enconman.2008.06.004.
Thi-Nham Le http://orcid.org/0000-0002-7319-8447 Ho, W., X. Xu, and P. K. Dey. 2010. “Multi-criteria
Decision Making Approaches for Supplier Evaluation
and Selection: A Literature Review.” European Journal
References of Operational Research 202 (1): 16–24. doi:10.1016/j.
ejor.2009.05.009.
Abate, J., and W. Whitt. 1992. “The Fourier-series Method for Hovelaque, V., S. Duvaleix-Tréguer, and J. Cordier. 2009.
Inverting Transforms of Probability Distributions.” “Effects of Constrained Supply and Price Contracts on
Queueing Systems 10 (1–2): 5–87. doi:10.1007/BF01158520. Agricultural Cooperatives.” European Journal of
Asche, F., S. Jaffry, and J. Hartmann. 2007. “Price Operational Research 199 (3): 769–780. doi:10.1016/j.
Transmission and Market Integration: Vertical and ejor.2008.08.005.
Horizontal Price Linkages for Salmon.” Applied Hsin, P.-H., and C.-I. Chen. 2015. “Application of Game Theory
Economics 39 (19): 2535–2545. doi:10.1080/ on Parameter Optimization of the Novel Two-stage Nash
00036840500486524. Nonlinear Grey Bernoulli Model.” Communications in
Brigham, E. O., and E. O. Brigham. 1988. “The Fast Fourier Nonlinear Science & Numerical Simulation 27 (1–3):
Transform and Its Applications, Englewood Cliffs 448 168–174. doi:10.1016/j.cnsns.2015.03.006.
Prentice Hall NJ.” ISBN- 13: 978–0133075052. Hsu, L.-C. 2010. “A Genetic Algorithm Based Nonlinear Grey
Chen, C.-I. 2008. “Application of the Novel Nonlinear Grey Bernoulli Model for Output Forecasting in Integrated
Bernoulli Model for Forecasting Unemployment Rate, Circuit Industry.” Expert Systems with Applications 37 (6):
Chaos.” Solitons & Fractals 37 (1): 278–287. doi:10.1016/j. 4318–4323. doi:10.1016/j.eswa.2009.11.068.
chaos.2006.08.024. Huang, Y.-L., and Y.-H. Lee. 2011. “Accurately Forecasting
Chen, C.-I., H. L. Chen, and S.-P. Chen. 2008. “Forecasting of Model for the Stochastic Volatility Data in Tourism
Foreign Exchange Rates of Taiwan’s Major Trading Demand.” Modern Economy 2 (5): 823. doi:10.4236/
Partners by Novel Nonlinear Grey Bernoulli Model me.2011.25091.
NGBM (1, 1).” Communications in Nonlinear Science Kayacan, E., B. Ulutas, and O. Kaynak. 2010. “Grey System
Numerical Simulation 37 (1): 278–287. doi:10.1016/j. Theory-based Models in Time Series Prediction.” Expert
cnsns.2006.08.008. Systems with Applications 37 (2): 1784–1789. doi:10.1016/j.
Cui, J., S. F. Liu, B. Zeng, and N. M. Xie. 2013. “A Novel Grey eswa.2009.07.064.
Forecasting Model and Its Optimization.” Applied Leroy, T., F. Ribeyre, B. Bertrand, P. Charmetant,
Mathematical Modelling 37 (6): 4399–4406. doi:10.1016/j. M. Dufour, C. Montagnon, P. MarracciniI, and D. Pot.
apm.2012.09.052. 2006. “Genetics of Coffee Quality.” Brazilian Journal of
Deng, J. 1989. “Introduction to Grey System Theory.” The Plant Physiology 18 (1): 229–242. doi:10.1590/S1677-
Journal of Grey System, Corpus ID: 9121622. 04202006000100016.
APPLIED ECONOMICS 13
Lewis, C. D. 1982. Industrial and Business Forecasting Methods: Engineering 4 (1): 37–56. doi:10.1080/
A Practical Guide to Exponential Smoothing and Curve Fitting. 19397038.2010.542836.
Butterworth-Heinemann. ISBN: 0408005599. Shrivastava, N. A., and B. K. Panigrahi. 2014. “A Hybrid
Li, S., X. Ma and C. Yang. 2018. „A Novel Structure-Adaptive wavelet-ELM Based Short Term Price Forecasting for
Intelligent Grey Forecasting Model with Full-Order Time Electricity Markets.” International Journal of Electrical
Power Terms and its Application.„Computers & Industrial Power Energy Systems 55: 41–50. doi:10.1016/j.
Engineering120: 53–67. doi:10.1016/j.cie.2018.04.016 ijepes.2013.08.023.
Maloni, M. J., and M. E. Brown. 2006. “Corporate Social Tien, T.-L. 2012. “A Research on the Grey Prediction Model
Responsibility in the Supply Chain: An Application in the GM (1, N).” Applied Mathematics Computation 218 (9):
Food Industry.” Journal of Business Ethics 68 (1): 35–52. 4903–4916. doi:10.1016/j.amc.2011.10.055.
doi:10.1007/s10551-006-9038-0. Wang, C. N., and V. T. Phan. 2015. “An Improved Nonlinear Grey
Mussatto, S. I., E. M. Machado, S. Martins, and Bernoulli Model Combined with Fourier Series.” Mathematical
J. A. Teixeira. 2011. “Production, Composition, and Problems in Engineering. doi:10.1155/2015/740272.
Application of Coffee and Its Industrial Residues.” Food Wang, Z. X., K. W. Hipel, Q. Wang and S. W. He. 2011. „An
Bioprocess Technology 4 (5): 661. doi:10.1007/s11947-011- Optimized NGBM (1, 1) Model for Forecasting the
0565-z. Qualified Discharge Rate of Industrial Wastewater in
Odan, F. K., and L. F. R. Reis. 2012. “Hybrid Water Demand China.„ Applied Mathematical Modelling 35 (12): 5524–
Forecasting Model Associating Artificial Neural Network 5532. doi:10.1016/j.apm.2011.05.022
with Fourier Series.” Journal of Water Resources Planning Wang, Z. X., K. W. Hipel, Q. Wang, and S. W. He. 2011.
Management 138 (3): 245–256. doi:10.1061/(ASCE) “An Optimized NGBM (1, 1) Model for Forecasting
WR.1943-5452.0000177. the Qualified Discharge Rate of Industrial Wastewater
Ou, S.-L. 2012. “Forecasting Agricultural Output with an in China.” Applied Mathematical Modelling 35 (12):
Improved Grey Forecasting Model Based on the Genetic 5524–5532. doi:10.1016/j.apm.2011.05.022.
Algorithm.” Computers Electronics in Agriculture 85: Zhou, J., R. Fang, Y. Li, Y. Zhang, and B. Peng. 2009.
33–39. doi:10.1016/j.compag.2012.03.007. “Parameter Optimization of Nonlinear Grey Bernoulli
Shaik, M., and A.-K. Walid. 2011. “Green Supplier Selection Model Using Particle Swarm Optimization.” Applied
Generic Framework: A Multi-attribute Utility Theory Mathematics Computation 207 (2): 292–299. doi:10.1016/
Approach.” International Journal of Sustainable j.amc.2008.10.045.