S

D NG M HNH ARIMA TRONG D

BO GI

Cao Ho Thi
Khoa Qun L Cng Nghip
i Hc Bch Khoa Tp.HCM
TM T T
Mc tiu ca nghin cu ny nhm gii thiu vic xy dng m hnh ca cc qu trnh
ngu nhin M Hnh ARIMA, v ng dng m hnh ny trong vic d bo. M hnh ny
gii thch s bin ng ca chui thi gian bng cch quan h vi cc gi tr qu kh v
tng c trng s cc nhiu ngu nhin hin hnh v cc nhiu ngu nhin c tr. M
hnh cng c ng dng mt cch minh ha nhm d bo gi c sng ti Thnh Ph H
Ch Minh.
ABSTRACT
The objective of this reseach is to introduce the construction of the model of stochastic
processes ARIMA model, and their use in forecasting. This model explains the
movement of the time series by relating it to the own past values and to the weighted sum
of current and lagged random disturbances. The model is also illustratively applied to
forecast the price of riverfish in HoChiMinh City.

I.

GI I THIU

Trong lnh vc Kinh T Lng, vic d bo thng da trn hai loi m hnh chnh l m
hnh nhn qu v m hnh chui thi gian. Trong m hnh nhn qu, k thut phn tch hi
qui c s dng thit lp mi quan h gia bin ph thuc v cc bin nguyn nhn.
Gi tr ca bin ph thuc s c d bo theo gi tr ca cc bin nguyn nhn. i vi
cc chui thi gian, m hnh ARIMA c s dng d bo cc gi tr trong tng lai.
Theo m hnh ny, gi tr d bo s ph thuc vo cc gi tr qu kh v tng c trng s
cc nhiu ngu nhin hin hnh v cc nhiu ngu nhin c tr.
Mc tiu ca nghin cu ny nhm gii thiu vic xy dng m hnh ca cc qu trnh
ngu nhin M Hnh ARIMA, v ng dng ca m hnh ny trong vic d bo. M hnh
cng c ng dng mt cch minh ha nhm d bo gi c sng ti Thnh Ph H Ch
Minh.
II.

M HNH ARIMA

Nhm mc ch gii thiu v m hnh T Hi Qui Kt Hp Trung Bnh Trt (ARIMA

AutoRegressive Integrated Moving Average), ni dung phn ny s trnh by tm lc
mt s c s l thuyt lin quan n tnh dng (stationary), tnh ma v (seasonality),
nguyn l Box-Jenkin; cng mt s nguyn tc nhn dng, xc nh cc thng s v v
cc kim nh v m hnh ARIMA.
Tnh d ng
Mt qu trnh ngu nhin Yt c xem l dng nu nh trung bnh v phng sai ca qu
trnh khng thay i theo thi gian v gi tr ca ng phng sai gia hai thi on ch
ph thuc vo khong cch hay tr v thi gian gia hai thi on ny ch khng ph
thuc vo thi im thc t m ng phng sai c tnh. C th:

Trung bnh:
E(Yt ) = = const

Phng sai: Var (Yt ) = 2 = const

ng phng sai:
Covar (Yt , Yt-k ) = gk
1

Tnh dng ca mt chui thi gian c th c nhn bit da trn th ca chui thi
gian, th ca hm t tng quan mu hay kim nh Dickey-Fuller.

Da trn th Yt = f(t), mt cch trc quan chui Yt c tnh dng nu nh th cho

thy trung bnh v phng sai ca qu trnh Yt khng thay i theo thi gian.

k
= SAC
o

Da vo hm t tng quan mu (SAC Sample Auto Correllation)

k =

k = E [( Y t Y )( Y t k Y ) =
o = E [( Y t Y ) 2 ] =

(Y

(Y

Y )2

Y ) (Y t k Y )
n

= Cov (Y t , Y t k )

= Var (Y t )

Nu SAC = f(t) ca chui thi gian gim nhanh v tt dn v 0 th chui c tnh dng.

Kim nh Dickey-Fuller (kim nh nghim n v) nhm xc nh xem chui thi gian

c phi l Bc Ngu Nhin (Random Walk; ngha l Yt = 1*Yt-1 + et) hay khng. Nu
chui l Bc Ngu Nhin th khng c tnh dng. Tuy nhin, Nu chui khng c tnh
dng th cha chc l Bc Ngu Nhin.

bin i chui khng dng thnh chui dng, thng thng nu ly sai phn mt ln
hoc hai ln th s c mt chui kt qu c tnh dng.

Chui gc: Yt

Chui sai phn bc 1: Wt = Yt Yt-1

Chui sai phn bc 2: Vt = Wt Wt-1

Tnh ma v
Tnh ma v l hnh vi c tnh chu k ca chui thi gian trn c s nm lch. Tnh ma
v c th c nhn ra da vo th SAC = f(t). Nu c sau m thi on th SAC li c
gi tr cao (ngha l th SAC c nh cao) th y l du hiu ca tnh ma v. Chui
thi gian c tn ti tnh ma v s khng c tnh dng. Phng php n gin nht kh
tnh ma v l ly sai phn th m. Nu Yt c tnh ma v vi chu k m thi on th chui
Z t = Yt Yt m s c kho st thay v chui Yt .
M hnh ARIMA
Theo Box- Jenkin mi qu trnh ngu nhin c tnh dng u c th biu din bng m
hnh T Hi Qui Kt Hp Trung Bnh Trt ARIMA.

M Hnh T Hi Qui Bc p - AR(p)

Trong m hnh t hi qui qu trnh ph thuc vo tng c trng s ca cc gi tr qu kh
v s hng nhiu ngu nhin

Yt = 1Yt 1 + 2Yt 2 + ... + p Yt p + + t

M Hnh Trung Bnh Trt Bc q MA(q)

Trong m hnh trung bnh trt, qu trnh c m t hon ton bng tng c trng s ca
cc ngu nhin hin hnh c tr

Yt = + t 1 t 1 2 t 2 ... q t q

M Hnh Hi Quy Kt Hp Trung Bnh Trt - ARIMA(p,d,q)

Phng trnh tng qut ca m hnh ARIMA l:

Yt = 1Yt 1 + ... + pYt p + + t 1 t 1 ... q t q

Nh n d ng m hnh
Nhn dng m hnh ARIMA(p,d,q) l tm cc gi tr thch hp ca p, d, q. Vi d l bc sai
phn ca chui thi gian c kho st, p l bc t hi qui v q l bc trung bnh trt.
Vic xc nh p v q s ph thuc vo cc th SPAC = f(t) v SAC = f(t). Vi SAC
c gii thiu trn v SPAC l T Tng Quan Ring Phn Mu (Sample Partial AutoCorrelation); ngha l tng quan gia Yt v Yt-p sau khi loi b tc ng ca cc Y
trung gian.

Chn m hnh AR(p) nu th SPAC c gi tr cao ti tr 1, 2, ..., p v gim nhiu

sau p v dng hm SAC gim dn.
Chn m hnh MA(q) nu th SAC c gi tr cao ti tr 1, 2, ..., q v gim nhiu
sau q v dng hm SPAC gim dn. Tm li,
Dng th SAC = f(t)
Gim dn
C nh q
Gim dn

Loi m hnh
AR(p)
MA(q)
ARMA(p, q)
cl

Dng th SPAC = f(t)

C nh p
Gim dn
Gim dn

ng cc thng s ca m hnh ARIMA(p, d, q)

Cc thng s fi v qj ca m hnh ARIMA s c xc nh theo phng php bnh

phng ti thiu (OLS-Ordinary Least Square) sao cho:
( Y t Y t ) 2 Min
Vi
t = (Yt Yt )
Kim tra chn on m hnh
Sau khi xc nh p, d, q v cc fi , qj; ngha l xc nh c phng trnh cho m hnh
ARIMA, iu cn phi lm l tin hnh kim nh xem s hng et ca m hnh c phi l
mt nhiu trng (white noise, nhiu ngu nhin thun ty) hay khng. y l yu cu ca
mt m hnh tt.
V mt l thuyt, et c to ra bi qu trnh nhiu trng nu:
E ( t ) = 0
2
+ t ~ N (0, )

Var ( t ) = 2 = const
+ k = Cov( t , t k ) = 0
Vic kim nh tnh nhiu trng s da trn th SAC ca chui et.

D bo
Da trn phng trnh ca m hnh ARIMA, tin hnh xc nh gi tr d bo im v
khong tin cy ca d bo.

D bo im:

Yt

Khong tin cy: Yt k ( t ) < Y t < Yt + k ( t )

Vi tin cy 95%, k =2.

III. S

D NG M HNH ARIMA TRONG D

BO GI

minh ha, nghin cu p dng m hnh ARIMA trong vic d bo gi c sng ti

thnh ph H Ch Minh. Nghin cu s dng chui gm 111 d liu thng t thng
1/1990 n thng 3/1999 v phn mm EVIEWS d bo gi tr thng 4/1999.
3

Cc d liu qu kh ca gi c sng c t tn l RFISH v chui sai phn bc 1 c

t tn l DRFISH. th RFISH = f(t) v DRFISH = f(t) c trnh by nh sau:
40000

12000

36000
8000
32000
28000

4000

24000
0
20000
16000

-4000

12000
-8000
8000
4000

-12000
90

91

92

93

94

95

96

97

98

90

91

92

RFISH

93

94

95

96

97

98

DRFISH

th RFISH cho thy chui RFISH khng c tnh dng. th DRFISH cho thy chui
DRFISH cng khng c tnh dng. Qua th trn v d liu ta nhn thy chui c tnh
ma v theo qu. Cc kim nh theo hm t tng quan mu hay kim nh DickeyFuller trong EVIEWS cng cho cho thy chui RFISH v DRFISH khng c tnh dng do
d liu c tnh ma v.
S dng phn mm EVIEW kh tnh ma v v tin hnh th nghim cho nhiu m
hnh ARIMA, cui cng ta c m hnh ti u c dng ARIMA(2,1,2) vi thi on kh
tnh ma v l m = 12. Kt qu v cc thng s fi v qj c trnh by trong bng sau:
Dependent Variable: D(RFISH)
Method: Least Squares
Date: 2/3/2002 Time: 18:17
Sample(adjusted): 1991:04 1999:03
Included observations: 96 after adjusting endpoints
Convergence achieved after 50 iterations
Backcast: 1990:02 1991:03
Variable

Coefficien Std. Error

t

t-Statistic

Prob.

C
AR(2)
SAR(12)
MA(2)
SMA(12)

-283.3601
0.413278
0.963121
-0.846851
-0.781433

1010.997
0.135466
0.044544
0.118603
0.078476

-0.280278
3.050799
21.62164
-7.140218
-9.957634

0.7799
0.0030
0.0000
0.0000
0.0000

R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat

0.614807
0.597875
2248.588
4.60E+08
-874.5842
1.718345

Mean dependent var

S.D. dependent var
Akaike info criterion
Schwarz criterion
F-statistic
Prob(F-statistic)

203.1250
3545.923
18.32467
18.45823
36.31124
0.000000

Sau khi xc nh c phng trnh cho m hnh ARIMA, cn phi tin hnh kim nh
tnh nhiu trng ca et . Kt qu kim nh da trn th SAC ca chui et. cho thy et c
tnh nhiu trng v c trnh by nh sau:

Date: 2/3/2002 Time: 18:20

Sample: 1991:04 1999:03
Included observations: 96
Q-statistic probabilities adjusted for
4 ARMA term(s)
Autocorrelation Partial Correlation

AC

. |*.
.*| .

|
|

. |*.
.*| .

|
|

.|.

.|.

.*| .

.*| .

.|.

.|.

.|.

.|.

1 0.108
2
0.060
3
0.002
4
0.107
5
0.046
6 0.003

.|.

.|.

.*| .

.*| .

.|.

.|.

.|.

.|.

10

.|.

.|.

11

. |*.
. |*.
.|.
.*| .

|
|
|
|

. |*.
.|.
.|.
.*| .

|
|
|
|

12
13
14
15

.*| .

.*| .

16

.|.
.|.

|
|

.|.
.*| .

|
|

17
18

.|.
.|.

|
|

.|.
.*| .

|
|

19
20

.*| .

.*| .

21

.*| .

.*| .

22

.*| .

.*| .

23

.*| .

**| .

24

. |**
. |*.

|
|

. |**
.|.

25
26

0.027
0.069
0.040
0.008
0.035
0.176
0.069
0.010
0.059
0.168
0.021
0.050
0.031
0.017
0.112
0.094
0.083
0.177
0.198
0.133

PAC Q-Stat Prob

0.108 1.1485
- 1.5093
0.072
0.013 1.5099
0.115
0.021
0.005
0.031
0.076
0.037
0.011
0.046
0.173
0.011
0.025
0.079
0.133
0.064
0.085
0.057
0.060
0.087
0.099
0.084
0.248
0.221
0.014

2.6913
2.9138 0.088
2.9147 0.233
2.9925 0.393
3.5015 0.478
3.6719 0.598
3.6789 0.720
3.8174 0.801
7.2986
7.8428
7.8547
8.2589

0.505
0.550
0.643
0.690

11.559 0.482
11.614 0.560
11.915 0.613
12.036 0.676
12.072 0.739
13.648 0.692
14.771 0.678
15.649 0.681
19.750 0.474
24.964 0.249
27.356 0.198
5

.|.

.*| .

27

.|.

.|.

28

.|.

.*| .

29

.|.

.|.

30

.|.

.*| .

31

.|.

.*| .

32

.|.

.|.

33

.*| .

.*| .

34

.|.

.|.

35

.|.

. |*.

36

0.053
0.039
0.024
0.017
0.005

0.062
0.013
0.070
0.023

27.733 0.226
27.944 0.262
28.023 0.307
28.065 0.355

- 28.069 0.407
0.075
- 28.084 0.460
0.010 0.059
- 0.061 28.084 0.513
0.002
- 28.878 0.524
0.072 0.075
- 29.128 0.563
0.040 0.028
0.045 0.117 29.450 0.596

Kt qu ca m hnh d bo c trnh by trong tp d liu RFISHF. th ca RFISH

v RFISHF c trnh by chung nh sau:
40000
36000
32000
28000
24000
20000
16000
12000
8000
4000
90

91

92

93

94

RFISH

95

96

97

98

RFISHF

Da trn phng trnh ca m hnh ARIMA, tin hnh xc nh gi tr d bo im v

khong tin cy ca d bo.
D bo im l Yt = 26267 v Khong tin cy 95% l [ 21742 , 30792 ]

Sau khi c kt qu d bo, ta em so vi gi tr thc vo thng 4/1999 l Yt = 26000 .

Gi tr ny nm trong khong tin cy 95% v xp x vi gi tr d bo im. Sai s d bo
l ( Yt -Yt)/ Yt *100 = (26267 26000)/26000 * 100 = 1,03%

KT LU N

Kt qu d bo cho thy th ca m hnh d bo RFISHF bm rt st th ca

chui d liu gc RFISH. iu ny chng t m hnh ARIMA(2,1,2) ny gii thch
c s s bin ng ca chui thi gian v gi c sng ti Thnh Ph H Ch Minh.
Gi tr d bo xp x vi gi tr trn thc t (sai s d bo nh) v khong tin cy 95%
cng cha gi tr thc . iu ny chng t tin cy ca m hnh d bo.
Ngoi v d minh ha trn, nghin cu cng p dng m hnh ARIMA d bo
cho hn 20 loi mt hng ti Thnh Ph H Ch Minh theo qui trnh tng t v cng t
c cc kt qu d bo vi tin cy cao. Tm li, M hnh ARIMA l mt m hnh
ng tin cy i vi d bo ngn hn.

TI LIU THAM KH O

Bowerman B.L., and OConnell R.T., 1993. Forecasting and Time Series. 3rd ed.,
Wadsworth, Inc.
Cao Ho Thi v Cc Cng S 1998. Bn Dch Kinh T Lng C S (Basic Econometrics
ca Gujarati D.N.). Chng Trnh FulBright v Ging Dy Kinh T ti Vit Nam.
EVIEWS, 2000. Quantitative Micro Software.
Pindyck R.S., and Rubinfeld D.L., 1991. Econometric Models and Economic Forecast. 3rd
ed., McGraw-Hill.
Ramanathan R., 2001. Introductory Econometrics with Applications. 5th ed., Harcourt
College Publishers

Lin h
a ch
Tel
Fax
Email

: Cao Ho Thi
: Khoa Qun L Cng Nghip, i Hc Bch Khoa Tp. HCM
268 L Thng Kit, Q.10, Tp. HCM
: 84 - 8 - 8650460
: 84 -8 - 8635058
: chthi@sim.hcmut.edu.vn