2014 SAS Competition

Forecasting of Motorcycle Demand Using Calender

Variations, Hybrid Calender Variations-ANN and
Disagregation (Case Study in Jombang)
Muktar Redy Susilaa, Hanny Adiatib
a,b

Sepuluh Nopember Institute of Technology, Jalan Arief Rahman Hakim, Surabaya 60111, Indonesia
e-mail: [email protected], [email protected]

Abstract
The motorcycle market of Indonesia is the world's third-biggest market next to China and India, and the further
expanding demand is expected on the back of the steady economic basis in recent years [1]. In Indonesian market, three
type of motorcycle are dominated, there are Automatic (AT), Cub, and sport motorcycle. Forecasting to determine the
demand of motorcycle for each year is very important. Many kind of factors can effect the number of motorcycle
selling, therefore factors that affect it must be determined. This paper is released to present a review of the models
which are used to forecast of motorcycle selling for each month and each type in Jombang, Indonesia. In this research,
Calender Variations, Hybrid Calender VariationsANN, and Disagregation is used to find a best forecasting modeling.
The results show that using RMSE criteria, disaggregation is the best method to forecast motorcycle selling.
Keywords: motorcycle; Calender Variation, Hybrid Calender Variations-ANN, Disagregation.

1. Introduction
Motorcycle is one of the most popular vehicle in Indonesia. Increasing demand for motorcycles each
year is a complex problem, especially for the distributor of automobiles. PT. ABC is one of motorcycle
distributor company. The company has the problem with predicting the number of motorcycles that will be
needed by the population. Jombang is one area that is under the auspices of PT. ABC. Sales of motorcycles
consists of three types, namely AT (Automatic), Cub, and Sport.
Previous research on motorcycle sales forecasting carried out by Nursita [2]. In that study predicted
sales of motorcycles Honda Supra 125 cc with ARIMA method. Sitorus [3] predicted sales of Honda
motorcycles Automatic type, standard, and Sport using cointegration methods. The drawback of these
methods have not been able to capture the effects of variation in the calendar. According to Rey [4]
motorcycle sales in the Idhul Fitri increased.
Calender Variations (ARIMAX) Model is one of the forecasting methods are able to identify the
influence of calendar variations. Calender Variations-ANN hybrid is a combination between the ANN
Calender Variations. Calender Variations an approach for modeling linear, while the Neural Network is a
non-linear modelling approach. Disaggregation is a method of splitting the forecast sequence of high level
or called by "total" and then distributed to the lower level [5]. According to Lapide [6], the results of solving
high-level series prediction accuracy is more accurate.
Thus, in this study discussed monthly motorcycle sales forecasting and per type in Jombang who
allegedly influenced variation calendar approach and Hybrid Calender Variations-ANN. Besides
approaching Calender Variations and Hybrid Calender Variations-ANN, will be forecasting annual forecast
disaggregation approach to forecast monthly per type. Previously, annual forecasts obtained with causal
forecasting approach, one of the factors that affect consumer purchasing power that income communities

Susila & Adiati / SAS Competition (2015) 000000

[7]. So it is suspected that motorcycle sales are influenced by the number of productive population,
economic growth, GDP per capita, and the HDI. From such approaches, one will be selected the best
approach and is expected to help the PT. ABC to determine the amount of demand for motorcycles types
AT, Cub, and Sport in Jombang.
2. Material and Method
2.1. Data
Motorcycle sales data are obtained from PT. ABC. The data consist of motorcycle sales data and the
data that is expected to affect sales of motorcycles in Jombang. There are two types of sales data, the annual
motorcycle sales data for 2003 to 2012 and the data monthly motorcycle sales in January 2003 to March
2014. While the data is believed to affect the sales of motorcycles, namely the number of productive
population, GDP per capita, economic growth, and HDI 2003 to 2012 which is secondary data obtained
from the Badan Pusat Statistik (BPS) Indonesia.
2.2. Calender Variations or Autoregressive Integrated Moving Average with Exogenous Input (ARIMAX)
Model
Calender Variations (ARIMAX) models is ARIMA model with the addition of predictor variables [8].
In general, models ARIMAX is a time series with calendar variation effect. For Calender Variations
(ARIMAX) models with deterministic trends are as follows
() ( )
= + 1 1, + 2 2, + + , + ()
(1)

( )

where , is dummy variable for calendar variation.

2.3. Hybrid Calender Variations-ANN
Calender Variations an approach for modeling linear, while the Neural Network is a non-linear approach
to modeling. Mathematically incorporation of linear and non-linear models can be expressed by the
following equation [9]
= + +
(2)
where is linear component from Calender Variations model, as nonlinear component from Neural
Network and is residual from Hybrid ARIMAX-ANN models.
Therefore, residual from Calender Variations model become input for nonlinear modelling
= (1 , 2 , , ) +
(3)
Where f(.) is nonlinear function obtained from Neural Network and as a residual from Neural Network
models.
2.4. Disagregation Modelling
Disagregation is a method used to break into a series of high-level low-level series [5]. Disaggregation
basically divided into 2 groups: temporal and spatial models disagregation. The approach is based on the
proportion of forecast disaggregation is a method for splitting unit predictions of monthly and annual
forecasts for every type. To get a proportion of the monthly forecast sought total number of predictions or
forecasts yearly [10].
, = 1, + 2, + + 12,
(4)
Forecasting proportion for each month can be found from this equation
=

(5)

where , is monthly forecasting, , is amount of the mounthly forecast and is proportion jmounthly forecast.
Proportion for each type can be found
=

Where ,

is amount for each type and is j-forecast proportion .

(6)

Susila & Adiati / SAS Competition (2015) 000000

2.5. Model Selection

The best model is selected based on the minimum value of out-sample Root Mean Square Error
(RMSE) that is calculated using the following equation [10]:
1

RMSE = =1(+ ())

(7)

Where L is number of out-sample data, + is l-th out-sample, and is out of sample forecasting.
3. Results
3.1. Annual Motorcycle Sales Forecasting in Jombang
Motorcycle sales allegedly influenced by the number of productive population, economic growth,
GDP per capita, and the HDI. Based on regression analysis is obtained predictor variables that influenced
annual motorcycle sales is GDP per capita. Here's a regression analysis to motorcycles sales with GDP per
capita predictor variables.
Tabel 1 Regression Analysis for Motor Cycle Sales with GDP per Capita
Predictor

Coefisien

P-Value

Constan

10005

2.09

0.082

GDP

1985.6

3.8

0.009

So to predict annual motorcycle sales trend analysis for the GDP per capita following
= 2,574 + 1,16.

With these equations, obtained forecasts GDP per capita in 2013 amounted to 15.31 million and 16.47
million by 2014. With motorcycle sales regression equation to the GDP per capita predictor variables, the
obtained value of motorcycle sales in 2013 amounted to 35781 units and for the year 2014 amounted to
37554 units.
3.2. Calender Variations Model of motorcycle selling
3.2.1. Calender Variations Model for Automatic (AT)
According to ACF and PACF of data, the appropriate ARIMA model is ARIMA(1,0,0)(1,0,0)12,
ARIMA([1, 12], 0, 0). Therefore it can be obtained Calender Variations model with white noise residual
and have normal distribution. The first Calender Variations model is
= 44,23 + 531,871, + 480,932, + 570,373, + 324,144, + 433,385, + 446,546, + 517,647, + 898,2812, +
1
583,781 + 284,098 + 328,89+1 844,212, + (10,61)(1+0,5712 ) .

And the second Calender Variations model is

= 43,07 + 526,341, + 485,622, + 584,933, + 350,744, + 461,675, + 471,556, + 538,987, + 907,0012, +

1
599,421 + 319,56 + 366,63+1 856,052, + (10,49+0,4812 ) .

3.2.2. Calender Variations Model for Cub Selling

In cub selling type, the appropriate ARIMA model is ARIMA(2,0,4), ARIMA([2,4],0,0). Therefore it
can be obtained Calender Variations model with white noise residual and have normal distribution. The
first Calender Variations model is
= 1317,501, + 1185,802, + 1299,203, + 1253,304, + 1293,805, + 1306,106, + 1386,707, +
1600,508, + 1458,309, + 1319,6010, + 1171,9011, + 1430,3012, + 172,58 + 194,461, 14,742, +
(10,454 )

.
(1+0,532 )

And the second Calender Variations model is

= 1319,101, + 1179,102, + 1294,403, + 1248,804, + 1303,105, + 1387,106, + 1386,107, +

1605,508, + 1457,909, + 1318,3010, + 1174,3011, + 1425,512, + 183,13 + 197,761, 14,692, +
1
.
4)
(1+0,442
+0,32

Susila & Adiati / SAS Competition (2015) 000000

3.2.3. Calender Variations model for Sport Selling

For sport type selling , the appropriate ARIMA model is (1,0,0)12, ARIMA(0,0,1)12. Therefore it can
be obtained Calender Variations model with white noise residual and have normal distribution. The first
Calender Variations model is

= 205, 361, + 192,202, + 223, 543, + 231,064, + 230,075, + 227,616, + 266,817, + 252,668, +
217,879, + 199,1210, + 199,6711, + 272,6512, + 90,20+1 970,611, + 319,782, + 26,021, +
1
.
(1+0,7312 )

And the second Calender Variations Model is

= 206,921, + 184,522, + 219, 073, + 229,984, + 225,575, + 225,166, + 259,697, + 255,408, +

224,969, + 206,7810, + 200,4111, + 267,0712, + 64,46+1 1021,901, + 333,072, + 27,201, +
(1 12 ) .

3.3. Hybrid Calender Variations-ANN Model

3.3.1. Hybrid Calender Variations-ANN Model for Automatic (AT)
Minimum RMSE obtained from hidden layer with a number of five neuron for out of sample
forecasting model. The first Hybrid Calender Variations-ANN model is
= 44,23 + 531,871, + 480,932, + 570,373, + 324,144, + 433,385, + 446,546, + 517,647, +
1
898,2812, + 583,781 + 284,098 + 328,89+1 844,212, + (10,61)(1+0,5712 )
= 281,85 + 267,351 (. ) + 71,282 (. ) + 179,193 (. ) + 257,084 (. ) 221,215 (. )

where

1 (. ) = (1 + exp((0,08 24,601 0,1912 5,7413 )))1

2 (. ) = (1 + exp((0,01 0,481 12 + 0,0113 )) )1
3 (. ) = (1 + exp((0,04 + 0,201 1,4912 + 1,3213 )) )1
4 (. ) = (1 + exp((0,02 + 1,771 + 2,2912 + 1,0913 )) )1
5 (. ) = (1 + exp((0,01 5,651 0,9012 + 8,7413 )) )1 .

For the second model, four neuron in hidden layer make a minimum RMSE for out of sample
forecasting model. The second model of Hybrid Calender Variations-ANN is:
= 43,07 + 526,341, + 485,622, + 584,933, + 350,744, + 461,675, + 471,556, + 538,987, +
1
907,0012, + 599,421 + 319,56 + 366,63+1 856,052, + (10,49+0,4812 ) .
= 98,84 + 37,491 (. ) + 61,382 (. ) 0,643 (. ) + 130,114 (. )

where

1 (. ) = (1 + exp((0,41 + 66,981 34,4112 )) )1

2 (. ) = (1 + exp((0,32 43,941 27,5312 )) )1
3 (. ) = (1 + exp((0,77 + 10,971 + 82,5412 )) )1
4 (. ) = (1 + exp((0,03 + 81,901 128,7112 )) )1 .

3.3.2. Calender Variations-ANN Model for Cub Selling

Minimum RMSE obtained from hidden layer with a number of nine neuron for out of sample forecasting
model. The first Hybrid Calender Variations-ANN model is
= 1317,501, + 1185,802, + 1299,203, + 1253,304, + 1293,805, + 1306,106, + 1386,707, +
1600,508, + 1458,309, + 1319,6010, + 1171,9011, + 1430,3012, + 172,58 + 194,461,
(10,454 )

14,742, + (1+0,532 )

= 13,33 + 211,651 (. ) 62,142 (. ) 111,163 (. ) 99,844 (. ) 68,225 (. ) + 85,176 (. ) +

21,087 (. ) 147,238 (. ) 110,209 (. )

where

1 (. ) = (1 + exp((55,66 0,242 )) )1
2 (. ) = (1 + exp((102,54 2,042 )) )1
3 (. ) = (1 + exp((0,30 0,392 )) )1
4 (. ) = (1 + exp((59,62 + 10,992 )) )1
5 (. ) = (1 + exp((103,89 2,842 )) )1
6 (. ) = (1 + exp((264,63 3,772 )) )1
7 (. ) = (1 + exp((70,10 0,312 )) )1
8 (. ) = (1 + exp((165,96 + 14,032 )) )1
9 (. ) = (1 + exp((91,27 5,452 )) )1 .

For the second model, ten neuron in hidden layer make a minimum RMSE for out of sample
forecasting model. The second model of Hybrid Calender Variations-ANN is:

Susila & Adiati / SAS Competition (2015) 000000

= 1319,101, + 1179,102, + 1294,403, + 1248,804, + 1303,105, + 1387,106, + 1386,107, +

1605,508, + 1457,909, + 1318,3010, + 1174,3011, + 1425,512, + 183,13 + 197,761,
1
14,692, + (1+0,442

4)
+0,32

= 10.55 1,181 (. ) 274,662 (. ) + 17,393 (. ) + 178,444 (. ) 63,165 (. ) 65,156 (. ) +

19,817 (. ) + 216,648 (. ) 81,939 (. ) + 74,7010 (. )

where

1 (. ) = (1 + exp((0,02 + 6,422 + 3,294 )) )1

2 (. ) = (1 + exp((1,18 + 0,052 + 0,194 )) )1
3 (. ) = (1 + exp((0,23 5,132 + 9,974 )) )1
4 (. ) = (1 + exp((16,66 32,292 + 4,014 )) )1
5 (. ) = (1 + exp((0,18 8,392 0,294 )) )1
6 (. ) = (1 + exp((0,27 12,692 2,404 )) )1
7 (. ) = (1 + exp((0,14 3,152 + 6,064 )) )1
8 (. ) = (1 + exp((4,15 + 1,722 + 12,584 )) )1
9 (. ) = (1 + exp((11,00 41,782 32,304 )) )1
10 (. ) = (1 + exp((0,81 + 3,202 3,804 )) )1 .

3.3.3. Hybrid

Calender Variations-ANN Model for Sport Selling

Minimum RMSE obtained from hidden layer with a number of seven neuron for out of sample forecasting
model. The first Hybrid Calender Variations-ANN model is
= 205, 361, + 192,202, + 223, 543, + 231,064, + 230,075, + 227,616, + 266,817, + 252,668, +
217,879, + 199,1210, + 199,6711, + 272,6512, + 90,20+1 970,611, + 319,782, + 26,021, +
1

(1+0,7312 )

= 61,62 67,511 (. ) 84,492 (. ) 35,113 (. ) 43,784 (. ) + 53,145 (. ) + 78,046 (. ) + 53,657 (. )

where

1 (. ) = (1 + exp((44,20 + 0,5312 )) )1
2 (. ) = (1 + exp((5,78 + 18,6612 )) )1
3 (. ) = (1 + exp((28,30 + 16,3412 )) )1
4 (. ) = (1 + exp((595,35 23,5212 )) )1
5 (. ) = (1 + exp((15,81 + 13,6712 )) )1
6 (. ) = (1 + exp((8,36 + 10,2212 )) )1
7 (. ) = (1 + exp((28,20 3,6412 )) )1 .

3.4. Disagregation
In the previous discussion has obtained an annual forecasts using regression methods. To break up
the annual forecast, the monthly required proportion and the proportion of each type of motorcycle. Monthly
proportion used to break up the annual forecast monthly forecast. While the proportion of each type is used
to break down the forecast into a forecast annual three kinds of motorcycles namely Cub, AT, and Sport.
To obtain the proportion of monthly and per type of motorcycle used two approaches Calender Variations
and Hybrid Calender Variations-ANN. The results obtained forecast annual forecast for 2014 is 37554
units. Based on the monthly forecast for each type of motorcycle sales gained some models. Motorcycle
sales types AT and Cub obtained two Calender Variations and Hybrid models Calender Variations-ANN.
The sales of motorcycles types obtained two models Calender Variations Sport and one models Hybrid
Calender Variations -ANN. Therefore, breaking the annual forecast used a combination of the monthly
forecast.
3.5. The Comparison Between Calender Variations , Calender Variations-ANN, and Disagregation
The comparison of all model based on the RMSE value is shown in Table 1.
Tabel 2 RMSE Out of Sample
Kombinasi

Motorcycle
Type

RMSE
Calender
Variations

Hybrid
Calender
Variations ANN

Disagregation^

Disagregation ^^

AT

55253.15

68455.40

42842.48

57465.77

Cub

5544.40

10198.10

6647.44

11401.33

Sport

1126.79

758.76

1939.38

1878.58

Susila & Adiati / SAS Competition (2015) 000000

AT

55253.15

42064.92

Cub

5544.40

6796.97

Sport

1591.33

2885.48

AT

55253.15

68455.40

42702.11

57353.12

Cub

8151.53

13506.41

9091.61

14678.09

Sport

1126.79

758.76

1961.60

1906.85

AT

55253.15

41952.30

Cub

8151.53

9219.44

Sport

1591.33

2914.84

AT

61248.46

37852.95

48349.11

23308.63

Cub

5544.40

10198.10

3920.76*

8316.74

Sport

1126.79

758.76

1195.11

126.61

AT

61248.46

37852.95

48279.30

23258.24*

Cub

8151.53

13506.41

6941.37

11750.71

Sport

1126.79

758.76

1180.49

120.68*

AT

61248.46

48008.26

Cub

5544.40

4005.40

Sport

1591.33

AT

61248.46

47970.36

Cub

8151.53

6997.19

Sport

1126.79

918.12

921.99

Disagregation ^ : Yearly forecast disaggregation with monthly forecast proportion from Calender Variations model
Disagregation^^ : Yearly forecast disaggregation with monthly forecast proportion from Hybrid Calender Variations -ANN model

4. Conclusion and Future Work

Based on the analysis and discussion, it can be concluded that disagregation model yields better
forecast, compared to Calender Variations and Hybrid Calender Variations-ANN. For the prediction results
obtained by the method of disaggregation AT motorcycle sales will be high in December 2014, for the type
of high-Cub will be in August 2014, and for the type of Sport will be high in July 2014. So that the
management company must be aware of in these months, so that the company can anticipate the possibilities
that will happen.
To further study the forecasting methods used should be using disaggregation method. For the
disaggregation of annual forecasts can be used to forecast monthly proportions based on historical data and
forecast data proportions. For the proportion of forecast data can be used hybrid Calender Variations-ANN
approach.
References
[1] Badan Pusat Statistik (2012). Bermotor Menurut Jenisnya [On-line] diakses 13 Februari 2014, tersedia di http://www.bps.go.id.
[2] Nursita, L. (2010). Analisis Peramalan Penjualan Sepeda Motor di Mitra Pinasthika Mustika (MPM) Honda Motor dengan
Pendekatan ARIMA. Tugas Akhir, Jurusan Statistika, Institut Tehnologi Sepuluh Nopember.
[3] Sitorus, B. M., (2013). Analisis Peramalan Penjualan Sepeda Motor Honda Tipe Bebek, Matik, dan Sport di PT. Catur Putra Jaya
Menggunakan Metode Kointegrasi. Tugas Akhir, Jurusan Manajemen, Institut Tehnologi Pertanian Bogor.
[4] Rey (2013). Ramadhan Pecahkan Rekor, Honda Terjual 110.000 unit [On-line] diakses 1 Maret 2014, Tersedia di http:
//www.slideshare.net/ portalsurya/ epaper-surya-20-agustus-2013-25386477#.

Susila & Adiati / SAS Competition (2015) 000000

[5] Hidayah, E., (2011). Model Disagregasi Data Hujan Temporal dengan Pendekatan Bayesian Sebagai Input Pemodelan. Disertasi,
Jurusan Teknik Sipil, Institut Teknologi Sepuluh Nopember.
[6] Lapide, L. (2006). Top-Down & Bottom-Up Forecasting in S&OP. The Journal of Business Forecasting, Summer 2006.
[7] Anis, A., Aimon, A., dan Persaulian, B.,(2013). Analisis Konsumsi Masyarakat Indonesia. Jurnal Kajian Ekonomi, Vol. I, No. 02
[8] Lee, M. H., & Suhartono. (2010). Calendar Variation Model Based on ARIMAX for Forecasting Sales Data with Ramadhan Effect.
Proceedings of the Regional Conference on Statistical Sciences , 5.
[9] Zhang, G.P., (2003b). Time series forecasting using a Hybrid ARIMA and neural network model. Neurocomputing 50, 159-175.
[10] Wei, W.W., (2006). Time Series Analysis : Univariate and Multivariate Methods (2nd Ed.). New York: Pearson.

Appendix A. SAS Syntax

SAS Syntax for first Model of AT type

SAS Syntax for second Model of AT type

data motor;
input at I1 I2 I3 T M1M12 D1 D2 TD1 TD2;
datalines;
463
0 0 0 1 1 0
0 0
0
0
486
0 0 0 2 0 0
0 0
0
0

2082
0 0 0 59 0 0
1
1
59
59
2536
0 0 0 60 0 1
1
1
60
60

*
0 0 0 72 0 1
1
1
72
72
;
/*Proses Identifikasi*/
proc arima data=motor out=result1;
identify var=at crosscorr=(I1 I2 I3 T M1M12 D2)
nlag=60;
run;
/*Proses Estimasi*/
estimate input=(I1 I2 I3 T M1 M12 D2 )p=(1)(12)
noconstant;
run;
/*Pendeteksian outlier*/
outlier maxnum=3;
run;
/*uji asumsi normal*/
forecast lead=12 out=resi1;
run;
proc print data=resi1;
proc univariate data=resi1 normal;
var residual;
run;

data motor;
input at I1 I2 I3 T M1M12 D1 D2 TD1 TD2;
datalines;
463
0 0 0 1 1 0
0 0
0
0
486
0 0 0 2 0 0
0 0
0
0

2082
0 0 0 59 0 0
1
1
59
59
2536
0 0 0 60 0 1
1
1
60
60

*
0 0 0 72 0 1
1
1
72
72
;
/*Proses Identifikasi*/
proc arima data=motor out=result1;
identify var=at crosscorr=(I1 I2 I3 T M1M12 D2)
nlag=60;
run;
/*Proses Estimasi*/
estimate input=(I1 I2 I3 T M1 M12 D2 )p=(1,12)
noconstant;
run;
/*Pendeteksian outlier*/
outlier maxnum=3;
run;
/*uji asumsi normal*/
forecast lead=12 out=resi1;
run;
proc print data=resi1;
proc univariate data=resi1 normal;
var residual;
run;

Susila & Adiati / SAS Competition (2015) 000000

SAS Syntax for first Model of Cub type

SAS Syntax for second Model of AT type

data motor;
input cub I1 I2 I3 T M1M12 D1 D2 TD1 TD2;
datalines;
1425
0 0 0 1 1 0
0 0
0
0
1237
0 0 0 2 0 0
0 0
0
0

421
0 0 0 59 0 0
0 1
0
59
515
0 0 0 60 0 1
0
1
0
60

*
0 0 0 72 01
0
1
0
72
;
/*Proses Identifikasi*/
proc arima data=motor out=result1;
identify var=cub crosscorr=(I2 T M1M12D1 TD2)
nlag=60;
run;
/*Proses Estimasi*/
estimate input=(I2 T M1M12D1 TD2 ) p=(2) q=(4)
noconstant;
run;
/*Pendeteksian outlier*/
outlier maxnum=3;
run;
/*uji asumsi normal*/
forecast lead=12 out=resi1;
run;
proc print data=resi1;
proc univariate data=resi1 normal;
var residual;
run;

data motor;
input cub I1 I2 I3 T M1M12 D1 D2 TD1 TD2;
datalines;
1425
0 0 0 1 1 0
0 0
0
0
1237
0 0 0 2 0 0
0 0
0
0

421
0 0 0 59 0 0
0 1
0
59
515
0 0 0 60 0 1
0
1
0
60

*
0 0 0 72 01
0
1
0
72
;
/*Proses Identifikasi*/
proc arima data=motor out=result1;
identify var=cub crosscorr=(I2 T M1M12D1 TD2)
nlag=60;
run;
/*Proses Estimasi*/
estimate input=(I2 T M1M12D1 TD2 ) p=(2,4)
noconstant;
run;
/*Pendeteksian outlier*/
outlier maxnum=3;
run;
/*uji asumsi normal*/
forecast lead=12 out=resi1;
run;
proc print data=resi1;
proc univariate data=resi1 normal;
var residual;
run;

SAS Syntax for first Model of Sport type

SAS Syntax for second Model of Sport type

data motor;
input sport I1 I2 I3 T M1M12 D1 D2 TD1 TD2;
datalines;
194
0 0 0 1 1 0
0 0
0
0
173
0 0 0 2 0 0
0 0
0
0

541
0 0 0 59 0 0
0
1
0
59
624
0 0 0 60 0 1
0
1
0
60

*
0 0 0 59 0 1
0
1
0
72
;
/*Proses Identifikasi*/
proc arima data=motor out=result1;
identify var=sport crosscorr=(I3 M1M12 D1 D2 TD1)
nlag=60;
run;
/*Proses Estimasi*/
estimate input=(I3 M1M12 D1 D2 TD1 ) p=(12)
noconstant;
run;
/*Pendeteksian outlier*/
outlier maxnum=3;
run;
/*uji asumsi normal*/
forecast lead=12 out=resi1;
run;
proc print data=resi1;
proc univariate data=resi1 normal;
var residual;
run;

data motor;
input sport I1 I2 I3 T M1M12 D1 D2 TD1 TD2;
datalines;
194
0 0 0 1 1 0
0 0
0
0
173
0 0 0 2 0 0
0 0
0
0

541
0 0 0 59 0 0
0
1
0
59
624
0 0 0 60 0 1
0
1
0
60

*
0 0 0 59 0 1
0
1
0
72
;
/*Proses Identifikasi*/
proc arima data=motor out=result1;
identify var=sport crosscorr=(I3 M1M12 D1 D2 TD1)
nlag=60;
run;
/*Proses Estimasi*/
estimate input=(I3 M1M12 D1 D2 TD1 ) q=(12)
noconstant;
run;
/*Pendeteksian outlier*/
outlier maxnum=3;
run;
/*uji asumsi normal*/
forecast lead=12 out=resi1;
run;
proc print data=resi1;
proc univariate data=resi1 normal;
var residual;
run;