See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/317886554

Time Series Forecasting Using Arima Methodology with Application on

Census Data in Iraq

Article in Science Journal of University of Zakho · December 2016

DOI: 10.25271/2016.4.2.116

CITATION READS

1 2,954

1 author:

Qais Abdulqader
Duhok Polytechnic University
5 PUBLICATIONS 21 CITATIONS

SEE PROFILE

All content following this page was uploaded by Qais Abdulqader on 14 September 2019.

The user has requested enhancement of the downloaded file.

Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

Time Series Forecasting Using Arima Methodology with Application on Census

Data in Iraq

Qais Mustafa Abdulqader

Dept. of Hospital Management, Zakho Technical Institute, Dohuk Polytechnic University, Zakho, Iraq.
(Accepted for publication: April 24, 2016)

Abstract:
In this paper, the methodology of Box-Jenkins of Autoregressive Integrated Moving Average (ARIMA)
has been used for applying and forecasting the census in Iraq by taking (61) observations of the annually
census from 1950 to 2010. Several adequate models of time series have been built and some of the
performance criteria have been used for the purpose of comparison between models. Results of the
analysis showed that the ARI(2,2) model is adequate to be used to forecast the annually census data of
Iraq. During the period 2011 to 2020, there will be (33.58%) increase in the population, and the
population of Iraq in 2020 would be (41358200) persons.
Keywords: Box-Jenkins, ARIMA Models, Time Series Forecasting, Census.
the electric power. The results of the analysis

T
1. Introduction:
ime series analysis is the process of using showed that The best model was ARIMA(1,0,2)
statistical techniques to model and than ARIMA(1,0,1) model and AR(1) from
explain a time-dependent series of data performance of predict methods.
points. Time series forecasting is the process of Recently, the ARIMA methodology has
using a model to generate predictions (forecasts) been used in population forecasting studies.
for future events based on known past events. (Wan et. al., 2013), have used the ARIMA
Box-Jenkins Forecasting methodology is a methodology for forecasting prison populations
univariate version method and it is a self- using sentencing and arrest data .The results
projecting time series forecasting method. It has from the analysis showed that although
popularized and become widely known by modeling suggests an uptrend in the remand
George E. and Gwilym M. Jenkins in 1970 prisoner population, this should be more than
(George et. al, 2008). offset by a decrease in the sentenced prisoner
Many applications have been done in this population over the next months. (Pang and
area. (Zakria and Mohammad, 2009), have used McElroy, 2014), used ARIMA methodology for
ARIMA models for forecasting the population of forecasting fertility and mortality by
Pakistan. They showed that the estimated model race/ethnicity and gender. Results of the analysis
ARI(1,2) are close to other researcher’s finding are produced using fertility and mortality data
as well as non-government organizations for dating from 1989 to 2009. For total rates, it is
future planning and projects. (Mutar and Ilias, determined that a model without drift produces
2010), have made a comparative forecasting more tenable forecasts in comparison to the
work between ARIMA methodology and neural occasionally implausible results from the model
network method. They showed that ARIMA with drift. (Brajesh and Shekhar, 2015), used
methodology has given more appropriate statistical models to forecast the population of
forecasts than those given by feed forward accidental mortality in India. The results of the
artificial neural network. (Tuama, 2012), used study showed that on validation of models,
ARIMA methodology to forecast numbers of the ARIMA performed better than the damped trend
patients malignant tumors in Anbar province. exponential smoothing (DTES). This will be
The results from the analysis showed that the help for policy maker to control such type of
proper and suitable model is integrated incidence in future.
autoregressive model of order two ARI(2,1). The underlying goal in this paper is to use
(Sarpong, 2013), applied ARIMA models for ARIMA methodology so as find an appropriate
modeling and forecasting maternal mortality. formula when building a model from time series
The results of the study showed that the
data of the population census in Iraq, so that the
ARIMA(1, 0, 2) model is adequate for
forecasting quarterly maternal mortality ratios at residuals are as small as possible and exhibit no
the hospital. (Ghafil, 2013), used the Box- pattern.
Jenkins models for forecasting the production of

258
Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

2. Methodology stationary if it has a fixed mean, fixed constant

Box-Jenkins analysis methodology refers to a variance, and a constant auto-covariance
systematic method of identifying, fitting, structure. If the last condition is not satisfied,
checking, and using integrated autoregressive, then the series is said to be stationary in the
moving average (ARIMA) time series models. weak sense, or second order (Yafee and McGee,
The method is appropriate for time series of 1999).
medium to long length (at least 50 observations). When preparing the data and testing for non-
The general Box-Jenkins ARIMA (P,I,Q) model stationarity If the autocorrelation starts high and
for w is written as (George et. al., 2008): decline slowly, then the series is non-stationary,
w = ∅ w + ∅ w + ⋯+ ∅ w +a and the Box-Jenkins methodology recommend
−θ a −θ a −⋯ differencing one or more times to get
−θ a (1) stationarity. Also we should keep in mind that
Where ∅ and θ are unknown parameters and the variance of the errors of the underlying
the a are independent and identically distributed model must be invariant (i.e. constant). This
normal errors with zero mean, P is the number of means that the variance for each subgroup of
lagged value of w , it represents the order of data is the same and does not depend on the
autoregressive (AR) dimensions, I is the number level or the point in time. If this is violated then
of times w is differed, and Q is the number of one can solve this by stabilizing the variance
lagged values of the error terms representing the through taking a suitable and kind of
order of moving average (MA) dimension of the transformation by using Box-Cox test. We refer
model. The term integrated means that to obtain the reader to (Box and Cox, 1982) for more
a forecast for w from this model it is necessary details.
to integrate the forecast w . ARIMA The identification phase for choosing and
methodology may be involves in three phases: building the appropriate (P,Q) values of the
identification, estimation and testing, and ARMA model for the stationary series w is
application. carried out on the grounds of its characteristics,
2.1. Identification the Model that is, the mean, the autocorrelation
It should be borne in mind that preliminary function(ACF), and the partial autocorrelation
identification model commits us to nothing
function (PACF).Table1 represents the
except tentative consideration of a class of
ARIMA models that will later be efficiently autocorrelation patterns of ARMA processes.
fitted and checked (George et. al., 2008). The We refer the reader to (Brockwell and Davis,
Box-Jenkins assumes that the time series is 2002) for more details.
stationary. A series is said to be strictly
Table (1): Autocorrelation patterns of ARMA processes
Process ACF PACF
AR(P) Infinite: exponential Finite: cut off at lag p
and/or sine-cosine
wave decay
MA(Q) Finite: cut off at lag p Infinite: exponential and/or sine-cosine
wave decay
ARMA(P,Q) Infinite: exponential Infinite: exponential and/or sine-cosine
and/or sine-cosine wave decay
wave decay
to the further past observations (Ngo and Bros,
2.2. Estimation and Testing the Model
Not only does the Box-Jenkins model have to 2013).There are many approaches of estimation
be stationary, it also has to be invertible. to fitting Box-Jenkins models such as (George
Invertible means recent observations are more et. al., 2008, Shumway and Stoffer, 2011):
heavily weighted than more remote 1- Ordinary least square method
observations; the parameters used in the model 2- Maximum likelihood method
decline from the most recent observations down 3- Non-linear estimation method

259
Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

4- Moments method HQC =

To select the best model from several 2Ln(RMSE) +
adequate models we should use suitable criteria ( ( ))
(5)
that deal with measures of accuracy and also
This criterion uses a different penalty for the
with measures of goodness of fit of a model. In
number of estimated parameters.
this paper we depend on some criteria such as
A fitted model must be examined carefully to
(Ayalew et. al., 2012, Makridakis et. al., 1998,
check for possible model inadequacy. If the
Polhemus, 2011):
model is adequate, then the residual series
1- Root Mean Square Error(RMSE)
should behave as a white noise (or independent
∑ at 2
when their distributions are normal). The ACF
RMSE = (2)
n−c and PACF of the residuals can be used to check
Where t is the time period, n is the total the closeness of at to a white noise. The
number of observations and c is the number of procedure is by studying the autocorrelation
parameters in the model. The RMSE has the plots of the residuals to see if further structure
advantage of being easier to handle (large correlation values) can be found. If all the
mathematically. autocorrelations and partial autocorrelations are
2- Mean Absolute Error(MAE) small, the model is considered adequate and
∑ |a t | forecasts are generated.If some of the
MAE = (3)
n autocorrelations are large, the values of P and/or
The selected model is the one with the
Q are adjusted and the model is re-estimated
smallest Mean Absolute Error. The MAE has the
(Tsay, 2002).
advantage of being more interpretable and is
It is also possible to test the joint hypothesis
easier to explain to non-specialists.
that all m of the rk correlation coefficients are
3- Akaike Information Criterion(AIC)
simultaneously equal to zero using the
AIC =
Portmanteau test developed by Box and Pierce in
2Ln(RMSE) + (4) 1970,and the formula can be represented as
Where (RMSE) is the Root Mean Square (Brooks and Tsolacos, 2010):
Error during the estimated period, C is the Q=
number of estimated coefficients in the fitted n∑ r (6)
model. Notice that the AIC is a function of the Where n is a sample size, m is a
variance of the model residuals, penalized by the maximum lag length. The correlation
number of estimated parameters. In general, the coefficients are squared so that the positive and
model will be selected that minimizes the mean negative coefficients do not cancel each other
squared error without using too many out. Since the sum of the squares of independent
coefficients. standard normal variates is itself a χ2 variate with
4- Hannan-Quinn Criterion(HQC) degrees of freedom equal to the number of

260
Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

squares in the sum, it can be stated that the and to make forecasts and the users of the
Portmanteau test is asymptotically distributed as forecasts will be evaluating the pros and cons of
2
a χ m under the null hypothesis that all m the model as time progresses. A forecasting
autocorrelation coefficients are zero. As for any assignment is not complete when the model has
joint hypothesis test, only one autocorrelation been fitted to the known data. The performance
coefficient needs to be statistically significant of the model can only be properly evaluated
for the test to result in a rejection. after the data for the forecast period have
2.3. Application and Enforment become available (Makridakis et. al., 1998). The
Once a model has been selected carefully and Box - Jenkins methodology can be represented
judiciously and its parameters estimated and summarized through figure1.
appropriately, then it can be used for application

Figure (1): Schematic representation of the Box-Jenkins methodology

3. The Data
Censuses provide population numbers, household or family size and composition, and information
on sex and age distribution. They often include other demographic, employment, disability, fertility,
migration, education, economic and health-related topics as well. Table2 and also figure2 shows the
variable used in the analysis which is the annually data of census in Iraq (in thousands) and represents
a sample size (61) observations from 1950 to 2010.The first (51) observations were used for
estimation and the last (10) observations were used for forecasting. The source of the data present in
the web page of the United Nations Statistics Division.

261
Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

Table (2): The annually data of census in Iraq (in thousands) during the period 1950-2000
Years Population Years Population Years Population
1950 5719 1975 11685 2001 24517
1951 5902 1976 12068 2002 25238
1952 6065 1977 12461 2003 25960
1953 6216 1978 12860 3004 26674
1954 6360 1979 13258 2005 27377
1955 6503 1980 13653 2006 28064
1956 6647 1981 14045 2007 28741
1957 6795 1982 14436 2008 29430
1958 6951 1983 14823 2009 30163
1959 7116 1984 15203 2010 30962
1960 7290 1985 15576

1961 7475 1986 15941

1962 7674 1987 16302

1963 7889 1988 16673

1964 8122 1990 17074

1965 8376 1991 17518

1966 8651 1992 18010

1967 8947 1993 18547

1978 9261 1994 19124

1970 9586 1995 19732

1971 9918 1996 20363

1971 10256 1997 21017

1972 10600 1998 21694

1973 10951 1999 22387

1974 11312 2000 23091

Figure (2): Census of Iraq during the period 1950-2000

262
Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

Figure2 shows upward increasing trend and suggests that the given time series is non-stationary.
Figure3 presents the plots of the autocorrelation function (ACF) and partial autocorrelation function
(PACF) respectively.The values of the ACF are gradually declining from a first - order autocorrelation
coefficient to the end. The computed Portmanteau test of Box-Pierce with seventeen lags takes a value
of 261.468 (p-value = 0.00), which is highly significant, confirming the autocorrelation pattern. The
partial autocorrelation function shows a large peak at lag 1 with a rapid decline thereafter, which is
indicative of a highly persistent autoregressive structure in the series.

Figure (3): ACF and PACF of census of Iraq during the period 1950-2000 from left to right.

Box-Cox transformation gave the values of = 0.0 and its interval was (-0.745, 0.741) which
contains the value zero. This recommended that the log transformation is appropriate choice to make
our series stationary in variance before to take the difference of the series. After applying the log
transformation on the original series and checking again the ACFs and PACFs, we concluded that the
series need to be difference twice so as to be stationary in the mean. Figures4 and 5 represents the
transformed series after 2nd differencing, ACF and PACF after 2nd differencing respectively.

Figure (4): Log of census in Iraq after 2nd differencing 1950-2000

263
Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

Figure (5): ACF and PACF of log of census in Iraq after 2nd differencing 1950-2000 from left to right.

After getting stationarity, we proceed to fit an ARMA model to the log of the second difference of
the census of Iraq series. We apply the two measures of accuracy: RMSE and MAE with the two
measures of the goodness of fit of a model AIC and HQC mentioned in theoretical part to select the
appropriate model order. Table2 shows different combinations of ARIMA specifications and the
estimated criteria values.

Table (3): ARIMA model comparison using criteria values

Models RMSE MAE AIC HQC
ARI (2,2) 3.34155 2.50926 2.49130 2.52025
ARIMA(2,2,1) 3.38335 2.51019 2.55538 2.59880
ARIMA(2,2,2) 3.40277 2.49847 2.60604 2.66394
ARIMA(1,2,2) 3.85454 2.71061 2.81615 2.85958
ARIMA(1,2,1) 4.46132 3.18464 3.06932 3.09827
From table 3 interestingly, both measures of the goodness of fit AIC and HQC selected ARI(2,2)
having the smallest values comparing with the others. Despite the fact that AIC often tends to select
higher order ARMAs, in this paper there is a consensus across the two criteria. Also, the accuracy
measure RMSE selected the same model and the measure MAE selected the ARIMA(2,2,2), but its
value was very close to the ARI(2,2). In general and depending on measures of accuracy, and
measures of the goodness of fit of a model, the suitable and appropriate model is ARI(2,2). The
estimated ARI(2,2) is presented in table4.

Table (4): Estimation of ARI(2,2)

Parameters Estimates Standard Error t-ratio P-value

AR(1) 1.703 0.067 25.557 0.000

AR(2) -0.795 0.062 -12.735 0.000

After estimating the ARI (2, 2) model, we have to check for randomness. Figure6 shows the ACF
and PACF of residuals using ARI (2, 2) on log of census of Iraq.

264
Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

Figure (6): ACF and PACF of residuals using ARI(2,2) on log of census of Iraq from left to right.

Through looking at the figure6, we conclude After we checked and selected the best model
that none of the autocorrelations coefficients of which represented an ARI(2,2) of log of the
ACF and PACF are statistically significant, annually census data of Iraq, it’s time to use the
implying that the time series may well be model for application and make forecasts,
completely random (white noise). Also, we did a because the main purpose of modeling a time
test for randomness of residuals using a series is to make forecasts which are then are
Portmanteau test (or Box-Pierce test), which has used directly for making decisions. We validate
been mentioned in the theoretical part using the the forecast by splitting the data in two parts:
equation (6). The value of the test statistics was one part of the data (i;e, the first 51
equal to (8.52774) and the P-value was observations) we used it for modeling and the
(0.860063). Since the P-value for this test is other part of the data (i;e, the last10
greater than or equal to 0.05, we cannot reject observations) is used for forecasting. Table5
the hypothesis that the series is random (white shows the forecasting for the next twenty years
noise) at the 95% or higher confidence level. of the annually census (thousands) of Iraq.
Table (5): Forecast values of the annually census (thousands) of Iraq using ARI(2,2) model

Lower limit Upper limit

Year True value Forecast
95% 95%

2001 24517 24513.9 24500.9 24527.0

2002 25238 25228.7 25177.3 25280.2

2003 25960 25946.0 25817.6 26075.0

2004 26674 26668.1 26412.3 26926.3

2005 27377 27398.3 26955.6 27848.2

2006 28064 28140.7 27445.5 28853.4

265
Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

2007 28741 28899.5 27883.4 29952.6

2008 29430 29679.0 28273.4 31154.5

2009 30163 30482.9 28621.2 32465.7

2010 30962 31314.3 28933.8 33890.6

2011 32175.5 29218.0 35432.4

2012 33068.2 29480.3 37092.8

2013 33993.3 29726.2 38872.9

2014 34951.1 29960.0 40773.6

2015 35941.2 30184.5 42795.9

2016 36963.3 30401.4 44941.5

2017 38016.6 30611.3 47213.3

2018 39100.5 30813.9 49615.5

2019 40214.4 31008.2 52153.9

2020 41358.2 31192.8 54836.2

Looking at the table5 and using the model true values during the period 2001 to 2010 was
obtained ARI (2,2), we forecast the annually (30.08%), and depending on forecasting values
census data of Iraq from 2001 to 2020 and was (31.57%) and the two ratios are close to
compared it to the first ten observed and real each other. Hence, we can say that, ARI(2,2)
values from 2001 to 2010, with the statistical model is adequate to be used to forecast annually
software Statgraphics Centurion XVI. We can census data of Iraq, and during the period 2011
see that in 2001, the predicted value (24513.9) is to 2020, there will be (33.58%) increase in the
very close to the true value (24517) recorded and population, and the population of Iraq in 2020
published by the United Nations Statistics would be (41358200) persons. Figure7 presents
Division. Also, this observed value fall inside the forecasts for the log of the annually census
the confidence interval, and so on for the remain data of Iraq from the period 2001 to 2020 using
values. There is something else to be mentioned, ARI (2,2) model.
the increase of population of Iraq depending on

266
Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

Figure (7): Forecasts for the log of the annually census of Iraq from 2001 to 2020 using ARI (2,2)
model
4. Conclusions Brockwell P. J. and R. A. Davis R. A. (2002).
Introduction to time series and forecasting.
In this paper, we have built and used a
(2nd ed.). Springer.
systematic methodology of Box-Jenkins ARIMA
Brooks C. and Tsolacos S. (2010). Real Estate
forecasting for annually census data of Iraq.
Modeling and Forecasting. Cambridge
Indeed, we concluded that
university press.
1- The suitable model for forecasting the census
George E., Jenkins G. M. and Reinsel G. C.
of Iraq is ARI(2,2).
(2008). Time series analysis: Forecasting and
2- The ratio of increase of population of Iraq of
control, (4th ed.). John Wiley & Sons, INC.
true values and forecasted values during the
Ghafil A. A. (2013). Using Box-Jenkins
period 2001 to 2010 was close to each other.
(ARIMA) models for forecasting the
3- During the period 2011 to 2020, there will be
production of electric power. Journal of
(33.58%) increase in the population, and the
Karbala University, 11, 196-207.
population of Iraq in 2020 would be (41358200)
Makridakis S., Wheelwright S. C. and R. J.
persons.
Hyndman R. J. (1998). Forecasting: Methods
5. Recommendations
and applications. (3rd ed.). John Wiley &
1- We may use this model for forecasting the
Sons, INC.
census of Iraq for future.
Mutar D. R. and I. I. Ilias I. I. (2010). Analysis
2- We recommend comparing the results of
and modeling time series of water flow into
ARIMA methodology for forecasting the census
Mosul city: A comparative study. Iraqi
of Iraq with other methods like wavelet
Journal of Statistical Sciences.10, 1-32.
transforms or neural network method to see the
Ngo T. H. D. and Bros W. (2013). The Box-
differences.
Jenkins Methodology for Time Series
References:
Models. Journal of statistics and data
Ayalew S., Babu M. C. and Raw L.K. (2012). analysis, 454, 1-13.
Comparison of New Approach Criteria for Pang O. and McElroy T. (2014). Forecasting
Estimating the Order of Autoregressive Fertility and Mortality by Race/Ethnicity and
Process. Journal of Mathematics, 1, 10-20. Gender. Center for Statistical Research &
Box G.E.P. and Cox D.R. (1982). An analysis of Methodology, 3, 1-55.
transformations, revisited, rebutted. Journal Polhemus N. W. (2011).Time Series Analysis
of American Statistical Association, 77, 209- Using Statgraphics Centurion. StatPoint
210. Technologies, INC.
Brajesh and Shekhar C. (2015). Accidental Sarpong S. A. (2013). Modeling and Forecasting
mortality in India: Statistical models for Maternal Mortality; an Application of
forecasting. International Journal of ARIMA Models. International Journal of
Humanities and Social Science Invention, 4, Applied Science and Technology, 3, 19-28.
35-45.

267
 Journal of University of Zakho, Vol. 4(A), No.2, Pp 258-268, 2016 ISSN: 2410-7549

Shumway R. H. and Stoffer D. S. (2011). Time populations using sentencing and arrest data.
series analysis and its applications. (3rd ed.). Crime and Justice Bulletin, 174, 1-12.
Springer. Yaffee R. and McGee M. (1999). Introduction to
Tsay R. S. (2002). Analysis of Financial Time time series analysis and forecasting: With
Series. John Wiley & Sons, INC. applications of SAS and SPSS. Academic
Tuama S. A.(2012). Using analysis of time Press.
series to forecast numbers of the patients Zakria M. and Muhammad F. (2009).
Malignant Tumors in Anbar province. Al- Forecasting the population of Pakistan using
Anbar University Journal of Economics and ARIMA models. Pakistan Journal of
Administration Sciences, 4, 371-393. Agricultural Science, 46, 214-223.
Wan W-Y, Moffatt S., Xie Z., Corben S. and
Weatherburn D. (2013). Forecasting prison

@@@ARIMA@bàbäŠbØ@bäbåï÷ŠbØ@l@ômbØ@æŽî−Œ@bä‹Ùïåïj“Žïq
@@ŽôÔaÈ@æŽîín“ïäa†@a‹ŽïàˆŠó@æŽïîbma†@Šó@ß@ç‹ÙŽïuójŽïu@ßó †
@@ZŽôåïÜíÙŽïÜ@bïmŠíØ
@l@ ómbè@ ARIMA ŠbØìaìóm@ æŽîìó÷@ ÛíÅÝÅÜ@ æŽîŠójÄbä@ ì@ ôØóîŽí‚@ bäìíjÑï“ä@ bî@ @ åÙåïu@ MØíi@ bàbäŠbØ@ a†@ ŽôåïÜíØóÄ@ ŽôÄ@ †@@@
@æŽîín“ïäa†@a‹ŽïàˆŠó@ˆ@bäa†@H61I@båm‹ Šòì@l@õˆ@ìó÷@ŽôÔaÈ@æŽîín“ïäa†@a‹ŽïàˆŠó@l@ç‹Ø@ôåïj“Žïqì@ç‹ÙŽïuójŽïu@bàòŠóà@l@çbåï÷ŠbØ
@Žôäbåï÷@ê¡@æŽîŠóÅïq@Ûò‡åè@ì@ç‹ØbÄb÷@óåmbè@Šíu@ìaŠíu@ì@õb−í @æŽîìó÷@ômbØ@æŽî−Œ@æŽïÝŽî†íà@Ûò‡åè@N@a†@1950-2010@Žõìbà@†@ŽôÔaÈ
@@@@@Na†@þŽî†íà@båîói@Äbä†@Žôä‹Ø†ŠìaŠói@bàòŠóà@l@çbåï÷ŠbØ@l@óåmbè
@æŽïî@óäýb@a‹ŽïàˆŠó@bä‹Ø@ôåïj“Žïqì@Žôäbåï÷ŠbØ@l@íi@ˆ@óîb−í @ì@óîóè@a‹Žïm@bî@ARI(2,2)@þŽî†íà@íØ@‹ØŠbî†@Žôä‹ØóÄíÝ’@æŽïàb−ó÷@@@
@ŽôÔaÈ@æŽîín“ïäa†@aŠbàˆ@ì@(33.58%)@a‰ŽîŠ@l@a‡äbóØ@aŠbàˆ@†@oi@a‡îóq@çìíiò‡ŽîŒ@Žõ†@a†@2011-2020@Žõìbà@†@NŽôÔaÈ@ôäaín“ïäa†
@@N‘óØ@(41358200)@oi@Žõ†@Žõ@(2020)@ýb@ß
@@
@@
@@
@@@ARIMA@óïvéåà@ãa‡ƒnbi@óïåàÜa@ÞþÜbi@üjånÜa@
@@×a‹ÉÜa@À@ðäbÙÜa@†a‡ÉnÜa@pbäbïi@ôÝÈ@ÖïjnÜa@Êà
@@Z ó–þ©a
@ÖïjnÜa@Þua@æà@ ARIMA@óïÝàbÙnÜa@óØ‹zn¾a@Ãbìÿaì@ðma‰Üa@Ša‡®flÜ@åÙåïu@ M@ Øíi@óïvéåà@ãa‡ƒna@@szjÜa@a‰è@À@@@
@öbåi@@N1950 – 2010@óïåàÜa@òÑÜa@ßþ‚@ñíåÜa@ðäbÙÜa@†a‡ÉnÜa@æà@ò‡èb“à@(61)@‰‚bi@Ú܈ì@×a‹ÉÜa@À@ðäbÙÜa@†a‡ÉnÜbi@üjånÜaì
@@N@xˆbáåÜa@μi@óäŠbÕ¾a@‹ÍÜ@öa†ÿa@ïîbÕà@Éi@ãa‡ƒna@ì@óïåàÜa@ÞþÜa@æà@óÑÝn¬ì@óïÐbØ@xˆb¹
@ßþ‚N×a‹ÉÜa@ À@ ñíåÜa@ ðäbÙÜa@ †a‡ÉnÜbi@ üjånÜa@ À@ êàa‡ƒnfi@ â÷þàì@ ÀbØ@ íè@ ARI(2,2)@ xˆíáåÜa@ çbi@ ÞïÝznÜa@ w÷bnä@ p‹éÄa
@(2020)@ ãbÈ@ À@ ×a‹ÉÜa@ çbÙ@ †‡Èì@ (33.58%)@ b芇Ô@ ójåi@ çbÙÜa@ †‡È@ À@ ò†bîŒ@ Ûbåè@ çíÙï@ 2011-2020@ óïåàÜa@ òÑÜa
@@@N˜ƒ’@(41358200)@|j—ï

268

View publication stats