|The cou rse idroduces Ectimatina auocorrelartion

|
|modellinq and anal yts time function ond fiting ARINA modd
Lerles dada 8)Fore casting
9)ting ProCRr in freguen
CouRSE DUTLINE domoinY
Introduction and characteristiar af
time. sories General
Componerl time uerie ,uimple |Remark:
descriptive techn iques and he uniguenecc ime serier
meacurpmert rend and ueapnal. doeta is h Hh data ic orrelae
3.Auto covariance, auto corre lation and and can be ucedto orecast the
corrp logram. data
Hture valuer he given
4Rerid onalyri, Ceries.
-ler rancomnecr
Jample artocorrelation junction tar INT RODUC TLDN.
Jurning point test |Timecerie i a collpclion
- Kendallr tet lobservati ons made cequentially
5Pro babilty tet for t imp eries in ime.
- Aationarty: is a relaiDnchip betuepn a
Casuality: |varioble« one uhich mud be
Inverdibility Hime.
hte nore.
-Rondom walk.
6Linear filtersi moving average ,
Exampletime
lime
orie
Serie Occur in many ielde
.
aut oreqresrive,ARMA ARIMA models.ranging rom eco nomjcs 4o engineering
 Hhe ueri, fadors eg trends and
- Share sometimer be eanl
reasonaldy can
prices onuCcea Ye days sepn from cuch plot
Currency
exchange. Ho-entthe pl onabler one to
EpOrt {alc in GUcretivo momh look or
un ucual obrofvalion 'e
incomp in Cuccersiye oullier:
Averag
month
Company prete in Cuccksrire a) ExplaogteoExploratiun
or
khen obcervall onc are taken on
) yearu
Phycical dime ceIBL,
|more variables, H may be poati ble
Many time serie octur
phy<ical science
in the
in
Ho urp he vari ion in on
CPrier to explain the
pt ime
varioti on in
i
particulary fer
meleorrlog4 ,marine ccienceS and andher time reries
geo phyia: Thic lead to a beHer un dortand
lock
Rainfall on Cuc¢ertire hot genoroled
Anrtemp. in cuccausiYe
ng
dayhous, Hhe 4h Procpd to Trends

qi ven time ceries

day or months Taing
a)oduial dime eriel )ledictigc;
Analyrif cale 1qurer in Given on y obierved ime veri pu,
SuccRsSiwp wepke an imporanlone may war to gore carl thp
problem jn commer ce uure val ver the erier.
Hicten impordan 40 10recat This is o very imporlart lask in
future cales inordgrto plan sales ore casti nq and inthe
production anolu eDnomc ond indurtria
H may alo be inerert to examineime . ere
the relai onchip btwn sales and
olher ime series Such ay advert. 4)Contropupesei
expendture.
One mg Setk to control 1he
4) bemagcophic timp serias;
This Aime Jerier ocur in sytem either generoting by
the cudy |Uuarning Gignpkr
POpulrtion ihreugh a cenuur
1ture n
D
eVpms or Omining wh wud
carried ovl ofter every ioy
|happen inp or paramete hp uneti

4he Pm are changed.

|heciver ime cener analusis
Reccution; 5) Statitial
model buildingi
Nhen presente d with a time ceries
Several jint dependert variabler
Hhe girt tep dhe analycic are Cosidpred to comp up wih
correlalipn
in K

|uaia lly to plf the dara aHis a atktical model. Auto

Approachc do dime serie analycik

)Clasical opproach;
Thic 1crered do q analyc in

|
he time domain.
dercriber he evolution

a proces hroygh time.

Time The main tool onalo kthe
ki can qive a picture o what the autororrelati on qunction.
do lek lke 1rom ohich a |a)AnaluindheIRguEnSydoman
|darcription s made. This deRS Cribes how variationu in
To dtain a cimple deseripiVe a reries can be accounted
EOure the maln prPerip bu variattDnr at dErprt 1regutrci
 Ihe main dool
1s Ahe vpecrol
qunchon derty during harvectvg
a) Manmade force oq Fod ivrtie
AnalaY in dhe time
- The saler n
a cupermarkel
domain will
|eical mehod 4or high in

aml4inqdime decodkery be

)
cCries fanges
rom relaively
Horward ciragh
decriplive melbodu t
Aed ierpnti al ophicla Cyclic vaialivn
echnique 1hece are vari oti ons in time
e conider dhe cinple
darcriphi ve Veries olh period accillation tme
|Aechniquer
nol eq uol to one year
Examples include, rt
COMPONEN TS oF TIME

v0r1afing
TradHionol m-lhodl
SERIES. - baily tempe rcture.s in Mombace
peig

eries Conict analyina hme are high around midday

dpcompuing he Bucinecr cycles are orp lhan m

Yariationt in a cpipC U Cem

into, one ear,
-Irends
The includes Proc perrty, recesrin,
egional varition.
depresciprn b recoven
Cyclic varialion.
Irregular glutuotins Random fluctuaions
|Ater Arend, ceacona lity and
Irend:
Th the qFneral dendoncy
is
lber yehe ect hove beno
dota |remnoved 1rom ti me serigs hen
to increase or decrease over time
|uhat remane r due to randem
H is ar areuut thofe orer uhich lucucrt i pnr.
have erher

|
conetant or change very These flutuotion are due to forces
gradua or ner a Larae Periatos beyond hismsn
ime rnntrol.
Eg; Eathquakes, crikes, epidemics
Apart from upward and dow nwa rd c
Hen dencies come tim e cerles have
ralue lueluing around -a tornetarl |Problens n dume serier
|which deer chorge uorth tme . Dldentiy the compones
Anal4fi
at work.
a)\sclate, study, analyse and
measure he
components
in
dependertly
SlatitieatHeehmque for naty
MeacuremeIrs drend in time
Ceriec data

|Analycit 4eries baving -rend

|depende on whelher one w ant to
Time
me ure rend or to rpmovetrend
in order lo analuse lecal luctuatione
Seasonal variadion: H ako depend Dn whelher here
These are periodic qnd req ular
eauonolily or n.
movc mens in a
dime series wdh aherv are a n umber
peribd
The
lass han
varialion
one year
may be tribuled to-
| mearuring irend
Graphic method
;
|he folleuo ingi Method
0 Hdlural fot6pr eq. produdion frincipler
curve
leasl
foting
qure
ome cDModrhes melhodr
depend on Cean Methd
Drices
agricuural preucts -Medhod
1iHec
eni -averqqer
drop dur inq planting and
increare
 araahieWehod
he ime series X4 can be
aqainrt t to plotted
obtain a 4ree hand
Smooth Durve
hi
enabler one do hap an idea
abouthe 9e neral 1rend time
mori hing Ahe curve eliminarler
|bohveaDnaliy and random
IH ic fecls
eaY do apply but very
H
cubjective
doec nl
meaure trend Slhous trend

edhed
The SRm)-Qvergges
D bivde
procedure a fl louoc,
the whole data cet into two
parts wnd tiMe
Compule the arrthmedic m eqn qor Advartag es'
Pach pard and plot there two
OVeroges 0goind he mid-peimtr
e rerpethve periods toverpd
0jeive
biskdvantaqe;
-
and comprehensive

- Ancumer a linear trend

by each part
a) Jornhe wo
poaint to obtain the Melbcd aCuve
qitting
reguired trend whch can be ued |A drddtional
to eutmale uture value

Evample
fru
wqy dealing wrlh
-seasona l data which cprlain
01
lorl early
asimpleubdiox do the
dr
data.
Conaider the 1pllooinq
€et gres 1Dr lhere are variou upar
Curver
compony X Hhat can be uspd fo describe
Ahe drend in a qiven dorta s
SalesYear et

52 1976 |The melhods ain

a curvp that minimses
coming oth 4
79 I477 Thp
70 1978 |doviations 1rm.the data poimtr
can-1o Hhe ine
69
1979
1&D
Part
) bRd id.
Polynomials;
94 1981 X0t{ linear function.
: dp ta.{ +0H° Quadraic nc.
2).
CUr Ve
79 1983 yeneral
q7
|986
1984
|9 85
Pat Me an:3333
log 4 :log a t log b

3)Groth curves
1987
Eslime cala in 1q19, |988 and
a+bc Modhed exp curye
Log
1989.
X Gompertz Cc ve

oan 0G6€ Mdq79

Log loga c' lg b
FAt Bc
1922i487
Uetn 93-3333 Mid1145
or all 4hese curvas, the fitted
|4undion provider the meaure
Ahe trend and
he
|an ertimote
The recidu als
are
|OhPr valion and
he
lutuaion
he
rasidual
provide
locol

duetencer bhen
. Qa, - au -a

correrponding valuec| for a

| he ted Curve.
pol ynom.al degree lhe
Hir duacondlan and the
cerond and higher dt are Z ero

ExoMple,
ve Let Ux: aot a,x+ aax°
Ux - - Ux Ux

How Ao chose a curvei

No clearly cut ancwer to hn

eer exicte
0- (AU.)
0ooroximahionu at he =A la,+201X +a.)
pe LurvP b be uPd con be made -0a
|by ue adheore m baied 1nte U (A U)
dif4 erentes. - A (20s)
Let ,, Ka
be the cbserved
valuer arandom yarinble at hmel

1ne4 X X-X |ngeneral, we can conclude that

1e second
cferpnce
dere nce x A(X-
=
(X
x)
\apolynomial
Uy
degroen

-
Ak+- A
X+a-X,-
:a-3X1
In
Condant r>n
similarman per,he qro wth

Hqher order
be derined.
dferenceur
t| Curve can be aPpromled
can smilorly Conclan the
)X
Nodyed
rains
by he

exponertial
,
X-1

-9
|)lag Gompert z
-5
54
-13 |3) () Lgieic curye
35

Elample,
Let Ux= ao t
Method e

Xa
Leatquace
ttngastaighd line
= atbt )

.
a,
Then S-2(-a- b)
: Qota,x ta, - Q, -Q,x. ds -aŽ (x4-a -b)
U - (U.) as-at(u-a-4)=o
Ue - Ux t:
 {1 - A + 8 2°

n24"-(21)
Similarly
Ea9ople,
b=ntK - The 4able belaw rhows the 1gure4
n13- ()?
ppoducrinihouupnd
Thace Year19c31965 964 1961 1968|11¢9972
which
qre ihe normal eualionr
a and b qre derived romrcd|l| Sa| 94| 5 98
hence;
X -a tb4 hich linear i) EA a trajght line by melhod
lead Gquarer and lobulate
HlB:The ame procedure can be he trend value
uKed to tit he and degree u) Eiminte he Arend.
trend

S-(X- 96-a,t-a1)
) Whe
What
produ
) Ectimate the
eti
Componentr
Seriar are left?
Ahe monthy increare
on

pr oduction
+he time

in
in the

197d
-2> 4-0-qt- 9.)-0[
Lei he Arend eqn. be
AS -2 2tlXa, -91-a,)-0... X. -a+ +e
i

Tho nor Mal eqn are

The normal equation are

2 -noo t a,2faa 24
Since n -
4hot the
lodd)

middle
, celec t Suh
""
2t-a. Ztta.24? + a1 ear X|{ |x|0tl
0

2t4qoZ42 0 +0t' 1963 -QG818- 5 tuwo

There 1465 88 4
76s2.88 even,

44 -94 85 07 we
middle

1961 85 87-120
|1468 Sq- +5
ondolpnecualecue:
X4 : ab'te |1969 98
celed
poirte

492 to
Log legatt legtt lege be
I49 orther

Fronm eqn (0 o-+

Al9 eqn l) 1qa - a tb

B =log b
El log e 1343= ha t/245 b t

2b
Led z 2 (9 - - B) s/2213

The nor mal, pq uatiDnu are j

:-2 [4-A-B+)0

a- 86 1
 G13-1arb
I99 at 5\b divided

chall cn der
on 4tting
e the curve
nelhd
and we
3
GI3- 1a
1390:1at35
b
-
celkcted
x,, ,a corrpo
pots

om vaue
to 3
4.,
7$0
356
35Ç b
350
|eguidsl
ta and 3 cuch hal

b2 1910
G13-7Q+2 19t0 |Subettuting 4,deta in l0, w
get
X,=a tbe
Xa:41bet
Xa a 1 be J
= 812G 219t
X,
Auming addi4ivp mpdel time
Ceripc wp mi nate lhp trend valuer
eli 2
{rom cubtracti ng then -from he
9iven yalueu;

=I:5,5.12, 8 93,-2.2Gi 55,6 30, =bclete.

Nle The resulting values shou
fluctuolons
Period
whch change
mere thon one ypar ,1hereore
within a bele)
whod rgmauns are cyclic variahpa
a- a

() X.af fed
buthe qradien he line nnce
+ve ,{he yeorly role in cret Sabtting3
-
prodution 19.
K3- Ka

Value
X4 :81 correrponding
26 4 2 19l3)
to I90;
Subtitun
fquot ron
bhcinany
wP qP
93 63

Assignment
The follewinq talle hous million

mguep-Couugr
1429Lgkg |49lq59 I769 1979199 '
5 |25,|m92L9l2cll43 9
Pop
)Y: ab' t e' the trend
Log Log clg b
ond tabulate Let A: Log a

) value
Estimte he populati on in 1999.
log
B: Lg
Bc
b

EiHing grodh ueS i) now

is a modfied cxpone mial
-a+bc. curve ard he constanls A,B
Leac uare
mehod n applicabieand C an be elimrled b4
Lsince dhe number Pramrlpr-obe hhe m elhod 3-elected point
evalualed itqrerier hon he no. foally 1he coneant a,b and
v ar iableg 9iven c are 4hen obluned ar foll out
Special te chinioue hove been ea, e:b.
 beo
Findi Ci00A:1s,9,10, 1, 12
nding
reciproc

K
Complte a Syear Moving g
I00

e
K
100

I00I00
I04
C= e 2+
Eqn
104
(iv) is agaim in4he orm
a modyi ed curve ond so
exponenlial I0D o0
Hhe previouc method can be uted to
|evaluate he paa MPlerr n vol ved.
19 0D
|Mehodeun9averog
|Moing vg eittn misaseries ople
| fucresrive ave10ger or arihmetic |Arsuming
|edrend ba4-yr cy cle, calculate
means of m Aerms Glaring wth he he method movim
1jnti second ,1hird dernmu etc
11 moving g the meon y1he-finl
avq
oo
4olagdta
ronm he
ae11a211149s19
m ers
the qcrage -16e econd
m erm elartinq fron the seco nd 551511 586CI2.
Her m to (ma1erD +Yua A- Certeed.

i f

middle value
m
morin g
odd, ie
u
ohe
m:aktL, lhen dhe

placed
dime
agons he
inder val H
2
3 518
515
490
5005 495. 15
CoYers ie 1: ktI 4 467
506.15 503-625
d Pven, ie m:aK,1hen {he 5 502 516-5
ird moving avg K placed tlun 054D 5425 524-6
tuo middle valuer tho tme 557 5635 558-O
iterval it cover ie : Ok andtor l. 8 571 58| 512:5
5
However, 1he movirg avg dops nol 9 586
cojncide wth the oriqidal ime pericd I0 Cia

hence we tetre the moving a vg by

raubacks)
aking a mo ving 0vg
Hhe movi ng ayg
extenl
pting the find
| 4heue value aqainr {ktl.
) Hdoer nl
for all terms
provide rend valuer

The qraph obtaned on plathng th e a)H connel be uced 4o focus of

moving avg- againttime gives Prid fure drend (Moin ob
Hhe trend. trend analysu)

aple: ReMark Meving avq melhod g've

|Acudy demand (d) 10r the paud a rerrec pdure he long
|12 yearsi 4,310,has indiroled em rend
Ahe qolloui ng ) The rend
4pprox
c

ir

lineqr
a erier H)
lingar or
 Ccllaery mouement
Ahe dala are feth
regulr in perid
and ampltude
i) lhe ren d is nol linear,
correspondr to middle lerm
Ahen the moving
a bias in he trenqvq intart
value
The 4ormula (inds trend values

moving av xlen m:a11

when a linear polnemial is fied
ehenoticalleomem lhencving Frem (i),
Ayg aelld
A weigted procecr will m
A
onist
L0,1 1-)/
FHing a polynomial degree P<m
m

Aodhe fiel m -erms he cerie

i)Uang the polynem ialto artimale
he yalue inthe middle o 1he
TOrge
u) m
aing
Re peartrg

A
operations with
rom the
avg
o ',3" e
can be
erm ExAnple
talculote he irond
|1ollpuomg 4ime uerier by a A
val uer he
eharaderized by the ertend m and where a linear polynomial is
pthe deoqree he Hted to ccHeauve e-
We wre m.a s ie s, 1.
3
5ppint

5
X143819.2C1&.4317 26C-S45 12
Let a. at . Teol.7lr7c
9 1
2 10Jn
1
and
: 2K+I tr ms be
n5 k+

Xx
e
,, x-a,
4= -K2,-,0, ,2-K.
X-,X0, , ..X 5
Putding -o in (i)

To ectimate X. (he m ddle term), 45

ue have to erhmate he value
from a) by using he principle
lead uarer 10

fer t/1H ua)

ur
norgal
22oea24
di wr
0 ao
are,
and a ,lhe
I8. +3

4-K 4:
for -4 ard m.a)

I813

(i) 15 49

From o0, u qiven by

 Case a m,] 3 akat-)-A
-ao +a.t+ qa
14.
Kt) (2K-)(ak13)
an d
-k, -t-1), ,X, . The ueg C4 c ,

The (oK)(ak-)(3k13
1 ind moving avg Coinside a le
lo ermate do, ) Cu= C-4 The weighs are
0grPguire he ymmeric
19llouoin g
24-Aota,
normal equalions: R) Sum o weiqhtr Ik unty = 1
t a 2°
Exasple
2t0 2 Suppse we have ls,alfindihe

2
i.e odd pouer 1 |weg
Hence ) m5 =aktl
K=a
a, °.8
Weghsi
21°- ao 24.+a 4, (íu)
C- a, C-1 Co, Ci, Ca
Ca 3a (4) t3(a)- - s-a)2)
l(41) (4-) (4+6)
3

2241 Rxt) 2ia. i q2 Z2 5x3 x1

2211- Z1*Z{* (axt) Zt Ta.- 31316-1 -5

|. [okr224- (2+)- z2,

Co = 3
1214)
513 X1
(akt)* -(24)

|
21' 1+3+...1 n
:nlnt)(2nt)
Ca-C- 35
- nln)anti) 3n°gn-1)
30
Ls,al ,,,
35 35 35 35 35
but,

Ihe ueightc are symmetric o

he midale value

ilenlbaahe 30 tend alue tned

ft a
5 poin
6.axlka)lakailaethae-l
X,
Jueightod

Ls,al
X1-
avg
,,.,
4he
dh uil
30 implia hol the trend val vet
Bt(knlakal zi"
Ihis

1he d Ftem K.
35
30
Consider dhe follo wing reries
36
 1
e+ 12r c|elhed meauring-Eaonal

343 52129 1)Mellod Smple avelageu

8 ) Rorie {o irend mhed.

)
3) Rotio to moving
COM put e Ahe,irend volue ot overggt
t:4 relative methd.
5,]: 1-3 12
4) Link

35
|DMelhad impleauETaget
Trend yalue a {9 tepr n volved;

as
3
|(-gx) 162x8)1 (i1ra1) t
L (12xe+)-
(ax12s)
t
Arranqe -he dala Ly Parr
Ahe year
t16 06r7c8-315|) CoMpue dhe averagerfer hece
35
porl he 4eor ie
and i:1,a-+
N,i:2.
for mehly
Jor quterly dola:
ti) Coupute e combined me an
m2Kt| p=3 Momihly
Concider the polynomial

¿isQuarterly
and
i) 9btain cearonal indicec for

)
Normalequalions arp, dfferent partr the yeQr by
2 (art)., a248, a)
epreusing {he ovg. 1he part
214 -a, 2+ ' 0, 1he a ya age le
combin ed meao

0
S

Mo y
Since Quore .
2'0 for inodd 14 s4he cimple allThe
|Uolving for ae usng o) ond ui) whic melheds.
are he cau e ens used in case 2 Houe ver, based on {he
Cm,a] = [m,8] acsumplion that dhe dala doar
(-a .2,11, 12,-3): ls,31 nd conlain any drendtyclic

|
eect and rreqular comonete
aro removed by averag ng
MEAS REMFTS OF SEASOH AL |here acsumplions are nel in
FLVCTVATIONS qeneral Arue since masl econ omid
4 ir impo rto nt to study Geasonal time ceries have trend.
poHernc co as |An index shous increare or
. decrease
i
leterminp dhe efece úoaco |in aalven quantity
On the rize 1ho variobler.

) To dtudu ol
Nolue
should be he
he
variable
was no seaDnal variation
To wolch dh e value
For ihe Gudy
there
Exauple
The dola
|9uerly

eaconal varial icn 1 |1020_a

4he dodo mue beqlven for parls 1G7 403
1468 50
the yeor ip ueekly, mol ly, J9G0 49. 501 521 55
below qiver he avg
Price

53
| |
40r
55 595
cemedit Jo

4o_
& + +
a

quoery or cemi-annually.
1910 56- 59D G G G53 +

Caleulate 4he seson al variahen

and iderprae
- Adur{. ahoe ndi r oa
40dt5pL4731 554 1200 Te HonlLly dale
dclol -|

4825 cr 400 (or qualery dala.

t48 t 53- t50 -em ba conrlal
59 muHiplying
ohere
K 120o
:61-15

4
-53 95
-4S3159.5ts52tC5:3_ Etaple
Calculoe lhe seascngl varialion
ralo to rend melhad.

3
19 34
:52.9 1912 40 St
1913 5t
S = 48-QS x I00 97 8o 9a
52-79 |beerm,ne the trend values 4i

a-5.7 for he meled Oy

59-19
: 9S.
10o..

o3 7.
lead 4ocord

1970
t49 4X,
3a
S

Ioa.7
8
59-19
. 1RO

:a+bd
Ratic do dcend meho ŽX4 : na t b 24
Thic method
improves melhod he
Gimple averq 9er and i baued
on he asumpti Dn that ceasonal a 80 -5a Q5Ç
Yariodion tr
any qiven monh,
|9ulpr or par thE(eor o a i b- |a
|cond and 1ador thetrend.
|Grp involved , X4 56 +124
0
)
Oblain he drend values by
leat a9 mhod
Elimin ade the trend by Prprecing
Ahe oriqunal a4a
valu e.
Assuming
4hase ls
he irend
the
will
. aa
1910

multiplicalive m odel, 11
Conlain
b- 12early increamerl

Sead onal, 912 5I 5

3A55
3 - Qucterly

5 |3D5

54 5 551.5 Go 5
30
4t
increampnt
s),e3)
33536
48 5
45.5
5

Cyclit and irregular coponerdt973 c3.5GG 5 68 512 5

)Eminale cyclic ond irrequlor 197+ 55 8.5 8 8I5 s4 S|
coMpon enta by avernging the 74
fo dy moohs laualre to getTrend climinate values,
indite 4erspason al varialions Expres Ahe eriqinal as l the
 calculated Irend values
00 I09.1

ear
1970
I09 131 5 1015 93 I where
S€.I I22.4 l09.9
|1a1 71.1 I06 34 0-3 n-i
14 73 85 |43 97 8 85-5
105-5
Ava. |2 78 |16.0 I09 12 81-3D. Snce T measureu tbe corr Cospficeal
| Tol arq: 0.20
|lun Guccecsive ohservothonu
Llq9 1, t ic called teriad
Core lolion coeficiem or
Adjurted ag,S k. aulo correlal ton coefIcIend llag1
K400
Totol
avg here

Adjucled
401

S:q
29

18 xo.49&s
|Thug ,T 12(-R) n-I
(X -)
92. 43

S3 ioR.92xo.9968 n
-I0259 Yn a-x)
87 3R XOq62.
67-04 Var (X4
S,- 1O0--152, 11 s8,259, -12 96 n(- )(Xit -)
PROBABILTY MODELS oE IIME 2(x-x)
SERLES Lim
Jhese are used to analyse time n-i

series corlaining random components

Auiocenelaion Fundion (acr) ,

|Similarly lag 3,
The correlatipn COficient
|Hunordinay
a vaiabler and 9ven n X3 («,X)
paing of obtervations is givgn by

A
) (ta )n.s
Cimlor idpa con be applhed to see
whether succerIVe oerer vot ions are
,Xn) xn-1) pairs
|lage
Th
)(h-x)
bulne

correlatpd.at
|lo T
wn.
be the
o brErvation
1 distance apart
correlalion
a 1drance
coeficie
part
|Aulocorrelolicn
lpn ortlpn
coeficeml
a;
logb

|(Lag 1).
Var(X)
 Cov( X4r) ECr, Xt)- E(x)E
ueesclatienacily
(ir)Ame' eier u
clnd talicnay
doe y1he joint
Cov (xx):E(x)- tGa)EK) dutriGrlio
E (x)-E )]" Same 1in dlbutin o
- Var(x) ,a., X4n th,Y1 l.
and h

whcre h: birlanc Hon

Aulerorre lotion cbcervothenc
Ceeftiert at log o
-4 le A Hime cerier ic ctricly
wlolicnary y the probabiliic
-pl) l clruclure dos nol change
wlh ime
Corellagam. wealy clahnay
Thu is a
graph Ph) oted agnntr
A
s
procesc

covariance
mean
is
conctant and
no
dt

on nciondors
bul bnly de pend
|deprnd
on he dtance on lhe var abler

|Inpractice, rten u®u io

|deine ctati onarity in the weak
The graph
auo
hy provi
ic useful in ideprating sense
rorrelalion coeficierls
an objective créterion ematk:
the nrture 1he. Mact 1he proboblrty lheoy
lernoal cuuro {le corier +imn eori ic roncorned a
th Hime serie
Eomele ctotie,on
and 4or thr mpgeriet
atme Terier contain Seaspnpl analyei ylen requires oneto
fluctuatitas,then the correlogram transor m a nonatalionary
oll alo evhibit ascillaton al the ime Ceriar mto dalionary
Lame fre quen cy and oill have the one so asto uce this lheory
tollowmg
Fzgmple
Slorle {he nadure the follpuonq
Aime seriec
)X - Q+br
here e, ~ H (o,c)
E l4) E (ablte)
0+
1 no tal1onary since
mean i cenelar
Stotnany
|A Hime Terier
Time
is
eras
brogdly qid to be SL
he
X e, n no a
tea Ces
there ir noy4temortic for Eo be a ime vories
stationary
charg
i
mean (Nore nd, i here where e o id o,8)
is nd emoic chonge n variance E(4) E le,sin kt t e, Carlt

and i ,ridly. peridic vori otions

have bepn rpmoved
This mplhec hat a ime eriAS N

totionor 4 t doe ne4 have Le. {IA) ea

Hrend ieeauonal and cyclic
vO riatipn
 ELe.e, nn Ki Sn A (tth) e
te,eafn q,e
Cos (4h) e,e. Cos kt cos A(4+h)
e"g,ca
) eta,et
te.ea cos cin k {th) r a,
we ausume independen te

CEle.e): o a,lcos -in + a3 (cark

E(e. e)

e
E Ce,) E(e.) =o
var (e) : E (e.)? - [E (e)]
E(e)'- 6 2tca-n)

E (x.X)- 04n dtin h)+

B:Co (A-B) 3 sin A Sin Bt Cos A Car B
eA (A)

Polar form coordinale,

Cov (A,)- E(x.) re
:6° (o [) ]
- e: tan
:6ce ( + Ah -)

:lcos hndependert ey | Frequency,

hen
A and pha e pe>o,

in 1he weak enSe

Slatienary
Let q(A) - |A.ktiAa Anpltue
05E0F FlLTERS INTMEEPE :4an'Aak
ALYL A
|A flder i cimply a 1rancormalion Hence,
Y,ea
another cequpnce
yaceriRc otionary, we Can
|anal4se a finte cection 4he cerie Frequency:
and hen qeneralise he con cluri on P hase angle l)
to the en-tirg ceries |In conclusCn, the antplitude change
Thie implikr converting the from I to g),the freauency
1ime tories imoa emaint in varip nt the phare changd
non-ctation ary
ctotinary Aime ceries bejore analys from D to
|A4iHercan be uled to elimnate |The -unction q) s colled the
Arend and ceasenal efecl \na |gain funticn he flHer and e
9iven time eries. the phae fundion

Linpar Filter.
biffe rencin q FHer
14 can be ihpd do convert one time|H 1Sa cpecIol dype Her
|cerie ino an
|1ine peEator
ar
her b a ewhich ie ucefulfer renmo
Hhe trend Snpl by deenl
a qven ceries until

,3
uhere is the oriqinal 1i me
Example
Serie.
afbt +e,where e ore
:e-q:
Suppare
cor ti sin
|

id va lo,)
1}: di^t-)
E4):atbt
 :
|xa+ ,€, [o.bi-) ,ea]
- Such a çunion ic he outocorrelHion

E
Var
(1,) -b
(Y) t8
ne dependemd 4
unrtion (Arr)
p (h):

Coy
()
5

(,)
cov

Cov (, ,
(, Ke)

26ndependent 1 Nor
|Aulecrrre lolicn coficienls rel

difforent diclan ce and

apart
Gme the
|Given
E(x,)
Erample
a+b
a,b
,
.
,e,
provider on inporHart
Hhe
Thee
propertier
Qu de lo
a ime
cocgicienl4 previde insigkt
erie:

:arb1'+e -
tc-o
(.
latba-),c
he probnbildy model hat
|gencraled he dola
Given he obrer valien.r ,xa
1b 1, e-R -b41 -e', o-C- on a dicero{e -ime oric. U
Can fofm (n-) par cbcervatien

E0:b-2C-C we lel he fird variable to

bependeml on lf
t
Frcl dufer ence do es nol make the |be{he fin cbeervolcn () and
Hime cerier ta-icnar and {hprerprethe p Con d obser vation i
up hove do do aecond deront

E(Z,)
b 20-c*e -P -L2c(t-)-(+

eC-te-at2C
2C Contant
he

where
and
(
corre lelion ce ficient bton

Yar (,):0+ 40+c

66 Constann

Zi ie
(ctationary

)
general, if the trend is a polyn oma r me asurer -lhe corre lation blon.
dpgree n,lhen we dterncE the |ucrpcuive obrervation. and -
Series n imnpc in ordor Ho romcy e I Called Ahe comple atocorrelaton
Hhe Irend lag 1or firc seal cnrrelollon

|Seasonal ed rar oro be climnate &, -R- x an d

|bydiyferenting
AuIOCavaRiACE AuIocCRPELAILOt
z(.)" (xt-
FUHCTIDNS. r. (i)(-)

log
,
lera a ionary Procpoc , we have
neled tha eor (xth)dependr on |l a imilar manner, autprerrelcn
cocHieied lag.h

Cov (x, ) - 6 (h) rn - PCh) - (x-(xh-)

6)1 he theoralical aulotovariancd

funcdion
6(0)
Td cempore the baric properlier
Hhe time series, iH is oHen weful
to have a untion hal ic not
inluen tEd by unit 4 reaurpmerdl
 TuE sERIES MOD ELS
1hle noise
iu a purely
lhic
rande m procer
A discrpdp ctochadic Process (Time = u?. u? 6

Cerie) is sad 4o bo a uhite noie 4 -6°>h:e Condart

(onr a EOuence variable l

erond crde 1i
alionary.

et) which are mudually independentExomple

Shore pricu on cuccRSsivE days
and identi carlly dtribuled oith
meon Co) and variance E")
euhere e~ HI (o,)
E («):L(e) Conclart
k)- ccv (x h)
= E (x4 X¢h)
=gle en)
Lo h0
|lh)ic ndependend t
pure ly ra ndom procAr is CPcond
A
prder otionacy
P(h) (h)
h= 0

Random ualki
urpece a random
sequence
ducrete random
r lh e a
|variabler
Proc mean and Yariance c'

|{hen a process randcm, ual k

|i x. e wlhere oc
e,~NI (,d)
X,= Ko + e. I
mally

Indegender

4,e tea teat. +€e

E(x)- 2Ele,) "{u H tonctant

i non-crtiona
-,A-X41

e
E(1]- E(e) - > Concat