Forecasting Model Selection of Red Chilli Price at Retail Level

Ketut Sukiyono1) Miftahul Janah2),

1)
Department of Agricultural Socio-Economics, Faculty of Agriculture, University of Bengkulu
2)
Department of Magister Agribusiness, Faculty of Agriculture, University of Bengkulu

Abstract. This study aims to select the best forecasting model of red chilli prices at the retail
level. The data used are monthly data, from 2011 - 2017. Five forecasting models are aplied and
estimated including Moving Average, Single Exponential Smoothing, Double Exponential
Smoothing, ARIMA, and Decomposition. The best model is selected based on the smallest
MAPE, MSE and MAD values. The estimation results show that the best forecasting model is
ARIMA at ordo (1,1,9).

Keywords: Curly Red Chilli, Retail, Forecasting.

1. Introduction
As an important commodity for the Indonesian economy, chilli often experiences sharp price
increases resulting in a high contribution to inflation. Chilli is included in the group of vegetable
plants which have sharp price fluctuations. The Ministry of Agriculture (2016) reported that in
the period of 2010-2014, chilli prices at a producer level have raised by 17.70% from IDR
16,343 in 2010 to IDR 19,237 in 2014. Many factors contributed to those phenomena; among
them is consumption per capita, the rainy season, Ramadan, and the New Year (Farid et al.,
2012). In addition, the Central Statistics Agency (BPS) noted that the increase in prices of
foodstuffs such as shallots and chillies had an impact on inflation in March 2016 which was
0.19%. Although in the past few periods, chilli with other food commodity groups has also
contributed to deflation in Indonesia (Bank Indonesia, 2018).
The uncertainty of chilli prices and their impact on the Indonesian economy result in very
important information on chilli prices in the future. This information can only be provided with a
forecasting process and this requires an appropriate forecasting method. Forecasting is estimating
something in the future based on past data that is scientifically analysed, especially using
statistical methods. Forecasting, according to Gapersz (2005) is an activity of a business function
that predicts sales and use of products so that products can be made in the right amount. Many
forecasting methods can be applied to forecast chilli price, such as Moving Average, Exponential
Smoothing, Decomposition and ARIMA (Autoregressive Integrated Moving Average). The
problem is whether the forecasting model is the most appropriate and accurate with chilli price
data at the retail level. The choice of this model can be done by estimating all available
forecasting models and selecting the best model with the available statistical rules. This method
has been used by Sukiyono & Rosdiana (2018) for rice prices, Novanda, et al (2018) for coffee
prices, and Sukiyono, et al (2018) for cocoa prices.
Departing from the discussion above, this study aims to determine the chilli price forecasting
model at the best retail level. For this purpose, the MA, Exponential Smoothing, Decomposition,
and ARIMA models are estimated and selected using the MAPE, MAD, and MSE indicators.

1
2. Methodology
2.1. Description of Statistics on Red Chilli Prices
The data used in this study is monthly time series data of the red Chilli at retail level in 2011-
2017 (84 observations). Data is obtained from the Central Statistics Agency and the Ministry of
Trade.
Time Series Plot of Red Chili Prices
70000

60000

50000
Red Chili Prices

40000

30000

20000

10000
1 8 16 24 32 40 48 56 64 72 80
Index

Figure 1. Red Chilli Prices Plot

Based on the results of historical data regarding the price of red Chilli at the retail level, prices
fluctuate and are not stable. The price instability can be seen in Figure 1 where there is a
tendency for the price trend to increase over a certain period of time, namely in December,
January, March. In addition, data declined in May, June, July. So that it can be concluded that
the price of red Chilli undergoes fluctuating changes and has a trend component and seasonal
variations. Price increases occur from October to January. According to Sukiyono & Rosdiana
(2018) the appropriate method for forecasting price data that has fluctuated in several months is
the Decomposition method. However, this judgment should be compared with other models
including Moving Average, Exponential Smoothing, and ARIMA.
The following is a summary of statistics from the retail price of red Chilli data from January
2011 to December 2017 can be seen in Table 1. Table 1 presents that the average value of red
Chilli at the retail level from 2011 to 2017 is 27,978 (84 observations), while for data deviations
(standard deviation) to the average is 9455.

Table 1 Descriptive Statistics of Red Chilli Prices at Retail Level

Mean Standard Deviation Minimum Maximum
Red Chilli Price (IDR) 27.978 9455 12.718 65.945

2.2. Model Analysis

Simple Moving Average Model
The moving average method uses actual data for new prices to generate forecast values for future
price (Padang, (2013) and Gaspersz (2005). Wayne (2004) stated this method is often used on
quarterly or monthly data to help observe components of a time series. The greater the order of
moving averages, the greater the smoothing effect. Research by Siregar & Riksakomara (2016)

2
are some examples of research that use this model for forecasting silver product demand and
stock prices respectively. Statistically, Moving Average model can be writes as follows:
P  Pt 1  Pt 2  ...  Pt  N 1
Pt 1  t
N
where Pt 1 is the forecasted price for the period t+1, Pt is the price in the period t, and N is the
period of the moving average.
Exponential Smoothing
Exponential smoothing methods are the most widely used forecasting methods. This method was
initially developed by Robert G. Brown for the US Navy during World War II (Gass & Harris,
2000) and further developed by Holt (1957) for forecasting the inventory control systems.
Exponential smoothing is forecasting method that weights the observed time series unequally.
Two types of exponential smoothing models are widely used, i.e., simple exponential smoothing
(SES) and Double Exponential Smoothing (DES). SES can be used for time series that contain
neither no-trend nor seasonality (Ruiter, 2017).
Formula Single Exponential Smoothing is

Pˆt 1  Pt  1   Pˆt

where Pˆt 1 is the forecast price at the time t+1; Pt is the actual price at t; observed; P̂t is the
forecast Pt ; and 0    1 is the smoothing parameter.

Holt (2004) explained that Double Exponential Smoothing (DES) computes a trend equation
through the data using a special weighting function that places the greatest emphasis on the most
recent time periods. The forecasting equation changes from period to period.
The forecasting algorithm makes use of the following formulas:


Pˆt  Pt  1    Pˆt 1  bt 1 
t  t t 1 
b   Pˆ  Pˆ  1   b t 1

Ft m  Pˆt  bt m

Where, m is the number of periods to be predicted; Ft  m is forecasting prices in the next m

period; bt is a smoothing trend in period t; and  and  are smoothing parameter.
In the double exponential smoothing method one of the most decisive parameters is smoothing
constants α and γ which are both between 0 and 1. Furthermore, the trial and error method will
be used to select smoothing constants α and γ which produce the smallest forecasting error value
for both settlement type double exponential smoothing from Brown and Holt.
Decomposition
Decomposition methods are used both for short and long term predictions. Makridakis, et al
(1993) explains that the decomposition method basically is to decompose series data into several
patterns; namely trend, cyclical and seasonal, and identify each of these components separately.

3
This is done, according to Makridakis et al, to help improve forecasting accuracy and help
understand better data behaviour. The Decomposition Model has two components namely
Additive and Multiplicative which observe in predicting trends, seasonal and forecasting cycles
(Kendek, Prang & Paendong, 2014). Mathematically, the additive decomposition model can be
written as:
Pt  Tt  St  Ct  I t
Whereas the multiplicative decomposition model can be written:
Pt  Tt  St  Ct  I t
where Pt is red chilli price in period t; Tt is trend component in period t; St is seasonal
component in period t; C t is cyclic component in period t; I t irregular component in period t; and
t is time period.
ARIMA (Autoregressive Integrated Average)
ARIMA model is applied to forecast a red chilli in a response time series as a linear combination
of its own past prices and past errors (Weiss, 2000). This model procedure provides a
comprehensive set of tools for univariate time series model identification, parameter estimation,
and forecasting. Many research applied ARIMA model to forecast price in the future, such as
Zhang, (2011). Juang et al., (2017); and Anokye et al., (2018). Box & Jenkins (1976) explained
that ARIMA is divided into three stages, namely, identifying, estimating/fitting, and forecasting.
In addition, according to Ekananda (2014), an ARIMA model is characterized by the notation
ARIMA (p, d, q), where p, d, and q denote orders of Auto-Regression (AR), Integration
(differencing) and Moving Average (MA), respectively. Thus, ARIMA model is explained
briefly as follows:
Autoregressive Model (AR)
The general form of the autoregressive model with order p (AR (p)) or the ARIMA model
(p,0,0) is stated as follows:
Pt  0  1Pt 1  ...    Pt  p  et
where Pt denotes the red chilli price at at t,  0 denotes constant;  p denotes autoregresive
parameter to p; and et is error value at t.
Moving Average (MA)
The moving average order q (MA (q)) or ARIMA (0,0,q) is stated as follows:
Pt  0  1et 1  ...  q et  q  eq
0 denotes a constant; q denotes moving average parameter to q; and et  q is an error value
of t-q.
Process Autoregressive Moving Average (ARMA)
The general model for the process of mixing AR (p) pure and MA (q) is pure, for example
ARIMA (p,0,q) is confirmed as follows:
Pt  0  1Pt 1  ...    Pt  p  1et 1  2et  2  ...  q et q

4
 Process Autoregressive Integrated Moving Average (ARIMA)
If non-stationary is added to the ARMA process mixture, then fulfilling the general
ARIMA model (p,d,q) is fulfilled. The equation for the simple case of ARIMA (p,1,q) is
writtenas follows:

 
Pt  0  1  1 Pt 1  ...  1    Pt  p  1et 1   2 et  2  ...  q et  q

2.3. Forecasting Accuracy Measures

There are 3 indicators used for selecting the best forecasting model, namely, Mean Squared Error
(MSE); Mean Absolute Deviation (MAD); and Mean Absolute Percentage Error (MAPE).
According to Karim, Awala & Akhter (2010), the smaller measurement values show more
accurate forecasts since it produces minimum forecasting error.
Mean Squared Error (MSE)
This MSE value is obtained through the difference between the actual value and the forecasting
value squared divided by the number of time series forecasting. The MSE value is used when the
residual magnitude is evenly distributed throughout the observation (Jarett, 1991).

MSE 
1 n

 Pt  P̂t
n t 1
 2

Mean Absolute Percentage Error (MAPE)

MAPE is the most useful measure to compare the accuracy of forecasts between different items
or products and commonly used in quantitative methods of forecasting (Makridakis;Wheelwright
& Hyndman 1998). The MAPE value is used to examine the extent to which the forecasting
method is biased.

1 n et
MAPE    100%
n t  1 Pt

1 n Pt  P̂t 
 
n t  1 Pt
 100%

MAD (Mean Absolute Deviation)

MAD is the average absolute error during a certain period regardless of whether the forecasting
results are bigger or smaller than the reality. Mathematically, MAD is formulated as follows:
(Nasution & Prasetyawan. 2008)
n
 Pt  P̂t
t 1
MAD 
n

5
3. Results and Discussion
Estimated Forecast Model Results

Moving Average Method

Estimates for the MA model are carried out using price movements starting from order 1 with
observations of 84 periods. The model estimation results are presented in Table 2 and Figure 2.

Table 2. Moving Average Model Result

MA (1) MA (2) MA (3) MA (4)
Forecast price (IDR/kg) 34,269 32,841 32,192 36,124
Accuracy Indicators
a. MAPE 16 20 22 24
b. MAD 4637 5548 6021 6457
c. MSD or MSE 52,437,885 64,724,284 75,317,501 83,245,779

MA(1) MA(2) MA(3)

Figure 2. Results of Estimating the Forecast Model with the Moving Average Method

In Table 2 and Figure 2, MA (1) has the smallest MAPE, MAD, and MSE values compared to
MA(2), MA(3), and MA(4). These results conclude that MA(1) is the most accurate model for
predicting the price of red chilli at the retail level in the furute. With this model, the next period
value of the red chill price is IDR 34,269.
Single Exponential Smoothing Method
A Single Exponential Smoothing Method is a forecasting technique developed from simple
moving averages (Biri, Langi & Paendong, 2013). This method is carried out with several levels
of α, starting from 0.1 to 0.9. The results of the estimation of the Single Exponential Smoothing
method are shown in Table 3.

Table 3 Comparison of the results of forecasting the single exponential smoothing method
Α MAPE MAD MSE
0,1 22 5,907 69,607,292
0,2 22 5,835 67,422,931
0,3 21 5,678 65,454,232
0,4 21 5,546 62,991,400
0,5 20 5,415 60,384,169
0,6 19 5,282 57,977,616
0,7 18 5,117 56,011,201
0,8 18 4,982 54,628,498

6
 0,9 17 4,861 53,912,425

Based on Table 3, the smallest value of MAPE, MAD and MSE is obtained at the value of α =
0.9. Thus, the SES model is P̂t 1  0.9 Pt  0.1P̂t and based on this model, the price of red chilli
for the next period is equal to IDR 33,977.
Metode Double Exponential Smoothing
The Double Exponential method is a method used when data shows a trend. The working
principle of the Double Exponential Smoothing Holt method is to smooth the trend value with
parameters from the original data. For the Holt method smooth the trend value with different
parameters from the parameters used in the original series. This parameter smoothing the double
exponential smoothing from Holt with a value between 0 and 1, where 0 <α <1.

In Holt's analysis, it is found that the value of  is 0.2, and the value of  is 0.1 to 0.9. Table 4
presents the results of the analysis for the Double Exponential Smoothing method with various
values of  .
Table 4. Double Exponential Smoothing Result with the value of  = 0.2
 MAPE MAD MSE
0,1 24 6,231 75,067,727
0,2 25 6,584 83,400,842
0,3 27 7,028 92,857,308
0,4 29 7,555 103,617,602
0,5 30 8,046 114,841,014
0,6 32 8,460 125,598,529
0,7 34 8,898 135,960,796
0,8 35 9,226 146,726,474
0,9 36 9,531 158,394,417

To see the best selection of models and the best accuracy, can be seen from the smallest MAPE
and MSE values. In Table 4. The smallest MAPE value is at y = 1, α = 0.2, namely MAPE = 24
and MSE value = 75067727.MODEL????

Decomposition Method
The Decomposition Model is a method that identifies separate components of the basic pattern
that characterize data, especially time series, economic and business data [15]. The results of this
Decomposition method are using both Additive Decomposition and Multiplicative
Decomposition. The results of the analysis can be seen in Figure 3.

Multiplicative Additive

7
 Figure 2. Results of Estimated Model Prediction with Decomposition Method

In the picture above it can be seen that the predictive data of multiplicative and additive
decomposition prices with the actual data are not coincide. This means that for this method it is
not appropriate to predict the price of red Chilli. However, even so, the model can be massaged
from several assessment indicators, namely the values of MSD, MAPE and MAD.
Table 5. Comparison of Multiplicative and Additive Decomposition Methods
Decomposition MAPE MAD MSD
Multiplicative 22 5,813 64,572,977
Additive 21 5,808 64,540,448

Based on Table 5, for the decomposition method the results tend to be almost the same for the
permits in the amount of IDR 36.671 and IDR 36.670. But in this case, if chosen, which is good
for fitting forecasting, a better additive decomposition method. This is based on the criteria for
smaller MSE indigo than the other methods.

ARIMA
For the use of the ARIMA method, one of the main requirements before doing the next data
processing is that the data must first be stationary, both to the variety and the average. To
determine the dependence of observation on the time series model used, ACF (Autocorrelation
Function) and PACF (Partial Autocorrelation Function) on the data so that the correlation value
between observations is known to be a limitation of the time series model in predicting the price
of red Chilli. The estimation results using the ARIMA method can be seen in Figure 5. Data has
experienced stationary on Lag 1 difference, both ACF and PACF.

ACF PACF
Figure 2. Results of Estimating the Forecast Model with the ARIMA Method

After estimation by ensuring data stationarity, then the price of red Chilli can be forecasted. This
is after differencing on lag 1. For several models in ARIMA used are ARIMA (0,1,1) and
ARIMA (1,1,0). The following are the estimation results of the ARIMA model (0,1,1) and
(1,1,0):

8
Table 6. ARIMA Estimation Results
Ordo MAPE MAD MSE
1,1,5 15,047 4,110,255 6,499,988
1,1,6 15,199 4,139,229 6,612,798
1,1,7 15,271 4,218,777 6,786,387
1,1,8 15,362 4,335,571 6,884,533
1,1,9 14,700 4,066,461 6,334,551
2,1,1 15,489 4,194,841 6,649,005
2,1,2 15,979 4,304,786 6,778,801
2,1,3 15,736 4,248,225 6,778,371
2,1,4 15,322 4,177,631 6,756,981

Tabel 7. Estimated Model Parameters of Red Chilli Prices

No. Data Coefficient SE-Coefficient t-statistic Probability
1. Red Chili Price
AR (1) -0,255 0,509 -0,501 0,618

MA (1) 0,145 21,646 0.007 0,995

MA (2) 0,400 18,490 0,022 0,983
MA (3) 0,393 9,805 0,040 0,968
MA (4) 0,424 1,282 0,331 0,741
MA (5) 0,090 7,956 0,011 0,991
MA (6) 0,112 9,909 0,011 0,991
MA (7) 0,064 12,350 0,005 0,996
MA (8) -0,304 13,757 -0,022 0,982
MA (9) -0,331 7,180 -0,046 0,963

Selection of the Best Method

The selection of the best method is done by comparing the values of MAPE, MSD, and MAD
from each model that has been done before is the best level of accuracy. The following are the
results of the comparison used to determine the best model selection can be seen in Table 7.

Table 7. Comparison of Inter-Model Goodness Criteria

Forecasting Model MAPE MAD MSE
MA (1) 16 4,637 52,437,885
Single Exponential Smoothing 17 4,861 53,912,425
Double Exponential Smoothing 24 6,231 75,067,727
Dekomposisi Additif 21 5,808 64,540,448
ARIMA 14 4,066 6,334,551

9
The results of the selection of MAPE, MAD and MSD criteria in Table 7 indicate that a good
method for forecasting retail red pepper prices is the Moving Average (MA) method, with the
forecast price for the price of red Chilli in the next time period of IDR 34,269.

Conclusion
The results of the research that has been done to get the best model, the Moving Average method
is a good method for predicting red Chilli prices at the retail level. Furthermore, if you use the
Single Exponential Smoothing model that is used, that is at α = 0.9. Whereas if using the Double
Exponential Smoothing Holt method, the model used is the value of α = 0.2 y = 0.1. The
selection of the best model is based on the criteria for the MAPE, MAD and smallest MSE
values. In addition, the ARIMA (1,1,0) and ARIMA (0,1,1) models can also be used as good
model choices for forecasting models that have α <5%.

References
Anokye R., Acheampong E., Owusu I., & Obang E. I. 2018. Time series analysis of malaria in
Kumasi: Using ARIMA models to forecast future incidence Cogent social sciences 4,
1:1-13. https://www.tandfonline.com/doi/abs/10.1080/23311886.2018.1461544
Bank Indonesia. 2018. Inflasi Juni 2018 Terjaga Didukung Harga Pangan yang Terkendali.
Retrived from https://www.bi.go.id/id/moneter/...inflasi/.../Analisis%20Inflasi%20-
%20Juni18.pdf
Biri, Romy, Yohanes A.R. Langi dan Marline S. Paedong. 2013. Penggunaan Metode Smoothing
Eksponensial dalam Meramalkan Pergerakan Inflasi Kota Palu. Jurnal Ilmiah Sains. 13.
(1) : 68-73. https://ejournal.unsrat.ac.id/index.php/JIS/article/viewFile/2035/1621.

Box, G.E.P. & Jenkins, G.M. (1976). Time Series Analysis: Forecasting and Control, Revised
Edition, San Francisco: Holden Day.

Ekananda, Mahyus. 2014. Analisis Data Time Series. Jakarta: Mitra Wacana Media.

Farid, M. dan N.A. Subekti. 2012. Tinjauan terhadap produksi,konsumsi, distribusi dan dinamika
harga cabe di Indonesia. Buletin Ilmiah Litbang Perdagangan 6(2): 211-233. Retrieved
from http://jurnal.kemendag.go.id/index.php/bilp/article/view/132

Gaspersz, Vincent, 2005. Production Planning and Inventory Control, Gramedia Pustaka Utama,
Jakarta.

Gass, S.I., & Harris, C.M. (Eds.). 2000. Encyclopedia of Operations Research and Management
Science (Centennial edition), Dordrecht, The Netherlands: Kluwer.

Holt, C. C., 1957. Forecasting trends and seasonals by exponentially weighted averages, O.N.R.
Memorandum 52/1957, Carnegie Institute of Technology.

10
Holt, C. C., 2004. Forecasting Seasonals and Trends by Exponentially Weighted Moving
Averages. Int. J. Forecasting, 20(1):5-10.
https://doi.org/10.1016/j.ijforecast.2003.09.015.

Jarett, J.1991. Business Forecasting Methods, Cambridge, MA: Basil Blackwell.

Juang W C Huang S J Huang F D Cheng P W Wann S R. 2017 Application of Time series

Analysis in Modelling and Forecasting Emergency Departement Visits in A Medical
Centre in Southern Taiwan BMJ Open 7:1-7.
https://www.ncbi.nlm.nih.gov/pubmed/29196487

Karim. R.; A. Awala & M. Akher. 2010. Forecasting of Wheat production in Bangladesh.
Bangladesh J. Agric. Res. 35(1): 17 – 28.
https://www.banglajol.info/index.php/BJAR/article/view/5863/4598

Kementerian Pertanian. 2016. Outlook Cabai Merah. Pusat Data dan Informasi Pertanian.
Kementerian Pertanian. Jakarta.

Kendek, Olvi J.; Jantje D. Prang; dan Marline Paendong. 2014. Prediksi Jumlah Pengunjung
Perpustakaan Universitas Sam Ratulangi Manado Menggunakan Metode Dekomposisi.
JdC, 3 (1) : 73-80. https://ejournal.unsrat.ac.id/index.php/decartesian/article/view/4000

Makridakis, S.; S.C. Wheelwright, & R. J. Hyndman, 1998. Forecasting Methods and
Applications, New York, Wiley.
www.forecastingprinciples.com/files/MakridakisbyFaria.pdf
Nasution, A. H., Prasetyawan. 2008. Perencanaan dan Pengendalian Produksi. Yogyakarta :
Graha Ilmu.
Noeryanti, Ely Oktafiani dan Fera Andriyani. 2012. Aplikasi Pemulusan Eksponensial dari
Brown dan dari Holt untuk Data yang Memuat Trend. Prosiding Seminar Nasional
Aplikasi Sains dan Teknologi (SNAST) Periode III. : 447-455. Retrieved from
repository.akprind.ac.id/sites/files/.../2012/noeryanti_15454.pdf
Novanda, Ridha Rizki., Eko Sumartono, Putri Suci Asriani, Ellys Yuliarti, Ketut Sukiyono,
Basuki Sigit Priyono, Irnad, Reswita, Melly Suryanty and Vera Octalia. 2018. A
Comparison of Various Forecasting Techniques for Coffee Prices. Journal of Physics:
Conference Series. 1114. https://iopscience.iop.org/article/10.1088/1742-
6596/1114/1/012119/pdf.
Padang, E., Gim Tarigan & Ujian Sinulingga. 2013. Peramalan JumlahPenumpang Kereta Api
Medan- Rantau Prapat dengan Metode Pemulusan Eksponensial Holt-Winters. Saintia
Matematika, 2: 161-174. Retrieved from
https://media.neliti.com/media/publications/221287-peramalan-jumlah-penumpang-
kereta-api-me.pdf.
Ruiter, Michelle de. 2017. Using Exponential Smoothing Methods for Modelling and
Forecasting Short-Term Electricity Demand. Erasmus School of Economics Bachelor
Econometrics and Operations Research. http://hdl.handle.net/2105/38564

11
Sukiyono, Ketut., Musriyadi Nabiu, Bambang Sumantri, R.R. Novanda., Nyayu Neti Arianti,
Sriyoto., M. Zulkarnain Yuliarso., Redy Badrudin, M.Mustopa Romdhon., dan H.
Mustamam. 2018. Selecting an Accurate Cacao Price Forecasting Model. . Journal of
Physics: Conference Series. 1114. https://iopscience.iop.org/article/10.1088/1742-
6596/1114/1/012116/pdf .
Sukiyono, Ketut dan Rosdiana. 2018. Pendugaan Model Peramalan Harga Beras pada Tingkat
Grosir. Agrisep Vol. 17 No. 1 : 23-30. Retrived from
https://ejournal.unib.ac.id/index.php/agrisep/article/view/4503
Wayne L, Winston. 2004. Operation Research: Application and Algorithm, 4th Ed,Thomson
Brooks/Cole.
Weiss, E. 2000. Forecasting commodity prices using ARIMA, Technical Analysis of Stocks &
Commodities, Vol. 18, No.1, pp.18-19, 2000.
Yuniastari, Ni Luh Ayu Kartika & IGP Wirarama Wedashwara Wirawan. 2004. Peramalan
Permintaan Produk Perak Menggunakan Metode Simple Moving Average Dan
Exponential Smoothing. Jurnal Sistem Dan Informatika vol. 9(1): 97-106. Retrieved
from https://jsi.stikom-bali.ac.id/index.php/jsi/article/view/41
Siregar, Ruben A., & Edwin Riksakomara. 2016. Pembangunan Aplikasi Berbasis WebUntuk
Peramalan Harga Saham Dengan Metode Moving Average, Exponential Smoothing, Dan
Artificial Neural Network. Jurnal Teknik ITS 5(2): 347-352. Retrieved from
http://ejurnal.its.ac.id/index.php/teknik/article/view/17070/2927

12