was trend

CH IX
Integrating us seasonal
average
-
w

- ↳ trend

AR IMA (P d
, , q) (fas

ut
m
N W
seasonal
mean not constant .
M part
.

Pregressive minionis Inno

I
m
visisd =

stationary -stationary
lag 1 +
d = 1

Nonseasonal
d
lang 2 2
+ =

↓ transform
layn + d= n

⑱ ri
# stationary .

mean constant Box-Jenkin models

w
nee
e

1
Assoc. Prof. Prapaisri Sudasna-na-Ayudthya
 Box-Jenkin Models
• a classical Time series models
called “ARIMA” or ARMA models.”
constant mean

• Deal with series of stationary time

series data.
nu

• Stationary means “constant m

mean
and variance.”
m

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 2

 integrate
M
P
ARIMA =>
ARMA
ne mem

I
↳ autoregressive
↓

moving average
5 dow wow t error
, nu
187
AR (1)
=y
I Noise MA(1)
z= 1 +
=

& ,
21 + fur time series Eur .
zf =

a+
-

8, a + 1
-

E + +1 =

& ,
z1 ; a +
=

0 MA(2)

If
=

A+
-

G1 At-1 -

Fzd+-2
AR (2)

z & &24+ a!
=

7 +
+

++ 1 ,
+ -
1

OR If =
&, 7
+ -
1
+

&27+ -
2
Types of Process Variability
~> mean
& variance din
• Stationary and uncorrelated − data vary around a fixed
mean in a stable or predictable manner
im Narwws
• Stationary and autocorrelated − successive observations
are dependent with tendency to move in long runs on
either side of mean shift down
owns .
Mean .

• Nonstationary − process drifts without any sense of a

stable or fixed mean · Li

I gap a
stationary stationary Nostationary gap even
:

wine

mean asso ~rwuN Sonia int vos

I-sap-jeap-]
visitin
Igapmins .

8
investis

I
↳ enters
/pattern S
wor n .

umwawws
-
Earwrs

auto correlation.

Chapter 4 Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU

Introduction to Statistical Quality Control, 5th Edition by Douglas C.
Montgomery.
3
 mosis mode

Model Building
.

y +

• Determine whether data are I

& miroduvis stationary or
nonstationary

st

stationary series or not by using

zen

“Time Series plot”.

-

(4+
+
-

1
-

diff -
e
unwomes 158 nonstationary
②
• If data are not “stationary”,
.

Y -

Y
+ -
1 + -

num

Differencing the series using vowing

low stationary
command “Differences”.
as wories (Diff ) Ownin downo
.
:

stationary .

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 4

 t y+ zt &t
=

1 yI

yz yz 71
The first differences of the time series
-

3 y3 y3 yz
-

values y1,y2,…,yn are " i i

N yn yn - yn-1

W zt = yt – yt-1 where t = 2,3,…,n

The second differences of the time series

values y1,y2,…,yn are

zt = (yt – yt-1) – (yt-1 – yt-2)

= yt – 2yt-1 + yt-2 for t = 3,4,…,n

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 5

 Time Series Plot of Yt
- + voni
n
20 trend a
trend , ww
nonstationary .

15 mean

10 drift

I
Yt

I lay
nu
5 diff :

d Y 4+
-
=
+
-

1
0

1 12 24 36 48 60
Index
72 84 96 108

3
120

Time Series Plot of C6 litt ricts

,

② 2

siterundaries ((6) 0
C6

=
lw stationary -1

-2

-3

-4
1 12 24 36 48 60 72 84 96 108 120
Index

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 6

③ Tritarian was und AR (p)

-
• Determine autoregressive (AR)
parameter for stationary data
using “Partial Autocorrelation
Function (PACF) plot”. Specify
significant Spike(s). spike ↓ .

Nin ARIMA (P 87 (P Als

⑦ M1(q)
, ,

• Determine moving average (MA)

↑
non seasonal seasonal

parameter for stationary data

using “Autocorrelation function
(ACF) plot”. Specify significant
• Spike(s).
Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 7
spilee Plot

AR12)
x
n
e

is T I die out
I
I -

I
I

DICE
I

↓sig-spines---- ARK ↳ 2

......I
:I
-

ACF
n
-

1 I

· 1
I
I

Or
......

PACF ,
MA(1)
 R ,
P
,
Li-5
mirover model ↑Chi-square
-

⑧
• Determine “ARIMA” model and test
by looking at residuals plot of ACF
and PACF.
u

AR(1) If At ze spike un
=
E+ =

1
+
, un

-
R =
die out 11t .

8
Af , At1-2
u
AR(2) Ef
-
=
+

* non-significant spikes indicate

appropriate model. Moreover, the
Chi-square test can be applied.
WSMTIUSINE

I
Hr
:

P-value > OR

Bauer svar-ner
He :
- . augins doin
: model Or is

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 8

 tha.
check si spike ASF
ser
is stationary vs

⑬ Autocorrelation Function forO C6

W
(with 5% significance limits for the autocorrelations)

1.0

0.8

0.6
sig spile=1
0.4 x
Autocorrelation

0.2
↳

0.0
die out ↑ VIG l
-0.2 ↳
-0.4

-0.6 MA(1)
-0.8

-1.0

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Lag

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 9

 PACK
Partial Autocorrelation Function for C6
(with 5% significance limits for the partial autocorrelations)

1.0

0.8

0.6 spile= 1 ARS17

Sig
-
Partial Autocorrelation

0.4

0.2

0.0

-0.2

-0.4
PACA
,
die=1 Ast

-0.6 ass ARIMA(1 , ⑧

,
1

-0.8 L
A d MA
-1.0 shell model stationary
- diff insur

Initabest
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Lag
Step 1-4 NIAN -

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 10

 stat time series > ARIMA

or

Promis in own as

um Grinding d
de We i cost
seasonal most
. 4 S

-> 0 P
IV - ⑧ D
->
0 Q

ARIMA (p ,
d , q7 (P , D G)s
,

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 11

 ①
ARIMA(1 , 1 17
Dit "
O
,
AR

MAG + P-values +
memoriainoon
, -

n
moun .

P-value max .

② ② min constant

·
↑

-x

win AR Our .

p-value :

max

a MIN12 2
do :

↳
:

Narmrdaine :
Na
VV Our
um

I
-
Plot In Mr
spike
-> do

....

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 12

 ⑧
PACF of Residuals for Yt
(with 5% significance limits for the partial autocorrelations)

1.0

0.8

0.6
No Significant spike WII1
Partial Autocorrelation

-> NAMIUNT :
0.4

0.2

I
0.0 MA(1)
-0.2

-0.4
Ino constant term)
-0.6

-0.8

-1.0

2 4 6 8 10 12 14 16 18 20 (with
22 5% 26
0
ACF of Residuals for Yt
24 significance
28 limits for the autocorrelations)
30
Lag
1.0

0.8
NWMIUSI NE =

0.6

0.4
Autocorrelation

0.2

0.0

-0.2

-0.4

-0.6

-0.8

-1.0

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Lag 13
Two useful types of Box-Jenkins models are
“autoregressive models and moving average
models.”

The nonseasonal autoregressive AR(1)

zt = φ1 zt −1 + at

For the nonseasonal MA(1)

zt = at − θ1at −1

Asst. Prof. Prapaisri Sudasna-na-Ayudthya, KU 14