TI

MESERI
ESANALYSI
S

TimeSe r
ies
Times erie
si sda t
atha thasbeenc l
assi
fiedchronologi
callyornume ri
c a
ldatathatisdescri
bed
overauni forms etoft i
me .
Times er
iesa nalysisi stheevaluat
iona nde xtra
c t
ionoft hec ompone nt
sofamode lthat
‘breakdown”apa rticularser
iesintounde rs
tandableande xplaina
blepor t
ions.Times erie
s
mode lsatt
e mpttopr edictthefut
ureusinghi s
tori
caldata.Thesemode l
sma kethea s
sumption
thatwha tha ppensi nt hef ut
ureisaf uncti
onofwha tha ppenedi nthepa st.Times erie
s
analysiscons is
tsof( a)t i
mes eri
esde composit
ionorbr eaki
ngt hes eri
esdowni nt
oi t
s
character
isti
cva ri
ations.(b)proj
ecti
nge achcharacter
ist
icintothefuture.(c)addi
ngt oget
he r
allt
hei ndi
vi dualprojecti
onstoarri
vea toneforecastfi
guref orthecompl e
tetimes e
ries.

Compo nent
sofaTi meSe rie
s
(
a)Thet r
end( T).Thisisthege ner
alwa yinwhichthefigure
sa removingorthegenera
l
dir
ecti
oninwhi chgr aphofTi mes eri
esappear
stobegoi ngove ralonginter
valofti
me.
Trendisalsotheunde rl
ying,long-t
ermt e
ndenc
yoft heda t
a(toriseorfal
l).Thetr
endis
usual
lytheresultoflong-
termf act
orssuchaschangesi
nt hepopulati
onsiz
e,demographi
c
chara
cter
ist
ic
soft hepopulat
ion,tec
hnologyandcons
ume rpref
erences
.

Ba
sict
rend ba
sict
rendpl
usc
ycl
ic
alva
ria
ti
ons t
rend+c
ycl
ic
al+s
eas
ona
lva
ria
ti
ons

Theob j
e ctoff i
ndi ngt hetimes eriest rendi st oe na bletheunde rlyingt ende ncyoft heda t
at o
behi ghlight ed.Re asonsf ors tudyi ngs ec ulart rends–as tudyoft het rendsa llowsust o
describeahi storic alpa tt
ern.Thet re ndc a nbeus edt oe va l
ua tethes uc cessofpr eviouspol icy-
Itpe rmi tsust opr oje ctpa stpa tt
e r
nsort re ndsi nt othef uture.-studyi ngt het rendofat ime
seri
e sa llowsust oe l
imi natet het rendc ompone ntf romt hes eri
e s
,t husma kingi te asierforus
tostud yt heot he rthre ecompone ntsoft het imes erie s
.
(b)Se as ona lVa riations( S) .The sea res hor t
- t
er m pe riodicf luc tuati
onsi nva luesduet o
diff
e rentc ircums tanc es.The yc ana l
sobede sc rib
e da ss easona lfluctua t
ionsora lmos t
identica lpa tternswhi chat imes erie sf ollowsdur ingc orrespondi ngi nterva l
sofs uccessive
periods .Se asonsha sawi de rme a ningunde rt hist opi c.Re asonsf ors tudyi ngs ea
s onal
variations-s owec ane s
tablishpa t
te rnofpa stcha nge s-topr ojectpa stpatt
e rnsint othef uture
–on c et hes e
a sona lpa tternt hate xistsha sbe e ne stablishedi tse f
fec tsca nbee l
imi na t
edf rom
thetimes eries.
(c)Cyc licalinf lue nce sorVa r
iati
ons ,(C) .The sea reme di umt ermc hangesc aus edbyf actors
thata pplyf orawhi le,thengoa wa y,a ndt he nc omeb a cka ga i
ni nar epetit
ivec yc le.Theyc an
alsobevi ewe da sl ongt erm os cillati
onsors wi ngsa boutt het rendl ineorc urve.The
economi cc ycle( boom,r eces si
on,bo om)i sa ne xa mp le.Gove rnme ntpol i
cynor ma llyca uses
thesebus ine s
sc yc l
e sofbo om,r ec ession,de pr essiona ndr e covery.
(d)Ra ndom/ Re s idua lva r
ia t
ions( R) .The ya rea ls oc alledI rre
gul arva r
iat
ions( I).The sea re
non- recur ri
ngr a ndom va ri
ationors por adicmot ionsoft imes e riesduet oc ha ncee ve nts
.
The sei nc ludef loods ,s t
rikes ,elections ,wa r,f ir
e ,pol it
ica lunrest,t ransportbr e akdowna nd
we athe rc onditions .The sea rege nera llyuns eenunl essoneha souts i
deknowl edge .

1
TimeSe riesMode l
s
Thet woma inmode l
sus eda r
e
(1)Addi t
iveMode l:Ti mes eriesvalue ,Y=T+S+C+Rwhe reS, C,andRa reexpresseda s
a bsoluteva lues.TheAddi tiveismor es uit
ableifthecompone ntfactorsa r
eindepe ndent
e .gseasona lvariat
ioni snota f
fectedbyt hevalueofthetrend.
(2)Mul t
iplicati
ve:Ti mes eri
esva l
ue ,Y=T S C Rwhe r
eS, C, Raree xpres
seda s%or
pr oportionswhi letheTr end( T)i sinthes a
meuni tsast
het i
mes e r
iesvalue.
Themul tiplicat
ivemode lismos tcommonl yusedinpr act
icea ndi smor ea ppropri
atei fthe
charac t
erist
icsi nt
eracte .
gwhe reah ighertre
ndva lueincr
e a
sest hes easonalva r
iat
ion.Thi s
mos tge nerallya c
ce pt
eda ndus edmode lspecif
iesthatval
ue soft het i
mes eri
es(Y)a ret he
produc tofa l
lthec ompone ntsi.e
.Y=T S C R.Foras erie
sc ompos edonlyofa nnual
data,therei snos easonalc ompone nt,Y=T C R
TheAddi tiveismor esuitableifthec ompone ntfact
orsareindepende nte.gse as
onalva r
ia t
ion
isnota ffectedbyt heva lueoft hetrend.

Me t
hodsofCa lculatingt het rend
Thet r
endi sthec or ec ompone ntoft het i
mes er
ie smode laboutwhi cht heot he rc ompone nt
s
fl
uctuat
e.Thet r e
ndc a nbef oundbyi de nt ifyings epa r
a t
eTr endva l ues,e ac hc orrespondi ngt o
ati
mepoi nt .
Me t
hodstha tc anbeus e dt oe xt r
a ctthet renda re:
1.Me thodofmovi nga ve rage s
2.Me thodofs emi -ave rage s
3.Me thodofl eas tsqua res.
(1)Movi ngAve rage -t hisi sthemos tc ommonl yu sedme thodus edfori de ntifyingat rend
andi nvol ve sc alcu lati
onofas e tofa ve rages.Thet rendwhe nobt a ine da ndc ha rt
ed
cons i
stsofs t
ra ight -
lines egme nt s .Ea cha verager epres ents/ corres pondst oat rend( t)
valuef orat imes e ries.Amov inga ve r
agec anbeus e da saf orec astbutwhe ngr aphi ng
movi nga ve rage si ti simpor tantt or e a
lizet hatbe i
nga ve rage sthe ymus tbepl otte da t
themi dpoi ntoft hepe riodst owhi cht heyr ef
e r
.Ma the ma ticallyt hemovi nga ve rage
(whi chs ervesa sa ne stima t
eoft hene xtpe r
iodsde ma nd,s ale,pr oduc t
ion)c anbe
foundus i
ngt hef o rmul a :Movi nga verage=Sum ofva lue si npr evi ousnpe riods/
numbe rofpe r iods .Theme th odi sus edi nor dert oe l
imi na t
et hedi stortionoft he
seasona lf igure sbyt het rend.Thei de ai st ofindoutwha tt hef igur ef ore a chs ea son
woul dbei fthe rewe renos easona lva r
iationsa ndt henr elatet hea ctua lf i
gur etot his
asape rc entage .Th ea ve ragesa ret het rend( t
)va l
ue sre quire d.
Thisi nvol ve sc a l
cul atingas etofa ve rages,e achonec or r
e spondi ngt oat rend( t)va l
ue
forat imepoi ntoft hes eries.Thea ve r
age sa r
ec alledmovi nga ve rage s .NB.t otalsa nd
average sa rewr it
te ndowni nl inewi t
ht hemi ddleva lueoft hes etbe ingwor ke don.
The sea ver a
ge sa r et het r
e nd( t)va lue sre quired.Theme thodofmovi nga ve rage si s
mos tl
ypr e ferredf ort i
mes eriest ha tha veas ignificants eas ona le ffect.I ts houl da ls
o
benot edt hatt her ei snouni ques e toft rendva luesf orat imes e ries .Ea c hme thod
yieldsadi fferentt re nd.Amo vi nga veragei sc onstruc tedbyr epla cinge a
c hva luei n
thes eriesbyt heme anofi tselfa nds omeoft h eva l
ue sdi rec tl
ypr ece di nga nddi re ctl
y
foll
owi ngi t.Ift hea vera gingi sdoneove ra neve nnumbe rofpe riodss a y4ye arsor12
mont hs ,themovi nga ve ragewi lli nitiall
yf a l
lbe t
we ens ucc essiveye a rsormont hs .In
suchc as e
s ,theva lue sa rec ust oma rilybr oughtba cki nl i
ne( orc entr ed)byt aki nga
subs e
que nt2ye ar( or2mon th)movi nga verage .Or dina r
i lythepur pos eoff ittinga
movi nga ve r
a gei st oe li
mi na tei ns of a raspos sibles omes or tofunwa ntedor
dist
r a
c tingf luc tuationsi ntheda ta.

2
 Disa
dvant
agesofMovingAverage
s:-
Dat
aatthebe gi
nni
ngande ndoft hese
rie
sa r
elost
;henc et
heydonotc overthe
comple
teper
iod-themovingaveragesmaygener
atecyc
lesorothermovement
sthat
werenotpr
esentint
heori
ginaldat
a–t heaver
agesar
estrongl
ya f
fec
tedbyext
reme
val
ues.

Procedur
eintheca lcula
tionofmovinga verage:
(a)Lis
tthese
riesve rti
cally
(b)Computemovi ngt otal
sa ndpl
acethesea tmidpoi
ntsofr
ele
vantper
iods
.
(c)Computethemo vinga ver
ages
.
(d)Ifther
eisane vennumbe rofseasons,a ve
ragetheadj
acentmovi
nga ve
rage
sto
givecent
redavera
g esc entre
done ac
hs eason.
(
e)Co
mput
e( )
.Thi
sgi
vest
hei
ndi
vidua
lse
asona
lva
ria
ti
ons
.

(
f)Findthemeanoftheindi
vidualse
asonalvari
at
ionsforea
c hs e
ason.
(
g)Ad justt
hes
eme a
nss othatthesum ofallseas
onalvari
ationsis100xnumberof
s
easons.Thefi
gur
esarri
vedata f
tert
hisadj
ustmentarethef
ina ls
easona
lva
ria
ti
ons.

Thesa l
esforMATCOa regivena sbel
ow:
Table40.MATCOSa les1994- 1997
Quart
er
Yea r 1 2 3 4
1994 72 112 115 90
1995 97 150 200 100
1996 105 140 160 130
1997 95 100 112 98
Required(i)computethetr
endva l
uesusi
ngthemovingave
rageme t
hod.
(i
i)Compu teseasonalvar
iat
ioni ndi
ces
(i
ii
)De seasonal
is
et he1997figuresonly(adj
ustfors
eas
onalvari
at
ion)
.

3
Ta
ble41.Fi
ndi
ngt
het
renda
ndSe
asona
lInde
x
Year Quar
ter Sal
es Movi
ngAnnual
Tot
al Addin Tr
end/
cent
redaver
age SeasonalVar
iat
ion
Pai
rs
1 72

2 112
1994 389
3 115 803 100.
37 114.
58
414
4 90 866 108.
25 83.
14
452
1 97 989 123.
62 78.
46
537
2 150 1084 135.
5 110.
7
1995 547
3 200 1102 137.
75 145.
19
555
4 100 1100 137.
5 72.
73
545
1 105 1050 131.
25 80
505
2 140 1040 130 107.
69
1996 535
3 160 1060 132.
5 120.
75
525
4 130 1010 126.
25 102.
97
485
1 95 922 115.
25 82.
43
437
2 100 842 105.
25 95.
01
1997 405
3 112

4 98

Calculationofthes eas
ona lindices
Havin gfoundt hes e
asonalva riat
ion(=Ac tua
lVal
ue/
Ce nt
redValuex100)
Ther esultanti
s
Table42Se a
sonalIndic
esc a lc
ulate
d
Q1 Q2 Q3 Q4
1994 - - 114.58 83.
14
1995 78.46 110. 70 145.19 72.
73
1996 80.00 107. 69 120.75 102.97
1997 82.43 95.01 - -
Totals 240. 89 313. 40 380.52 258.84
Una djustedaverage 80. 30 104. 47 126.84 86.
28
Adjus t
e d a verage. 81 105 127 87
(s
eas ona li
ndex)
Theme ansoft hes e
asona lva ri
ati
onsa ddupt o397.89.Ther
ei sther
efor
ene
edt
o
adj
us ttheva ri
ati
onsus i
ngt hef ormula

Adj
ust
ment
= .Forqua
rte
rlys
tat
is
ti
cst
hent
henumbe
r

ofs
eas
onsi
s4butf
ormont
hlys
tat
is
ti
cst
henumbe
rofs
eas
onsi
s12.Ther
equi
red

4
 tot
ali
s4x100f or4s easonseri
esand12x100f or12s easons eri
es.Thea djust
menti
s
the
na ddedtotheu na
djustedaverageyieldi
ngt hes ea
sona lindex(seasona li
ndi
ces
)
Wh er
er equi
redt ot
al<ac t
ualtota
lt he
nt hea dj
ustmentist obes ubt
ractedf r
om t
he
unadj
ustedaver
a ge(mean).Inoure xampletheadjustment=( 400-397.89)/4=0.53.
Dependingonyourpr ef
e r
ences,ana l
ter
na t
iveme thodofa djus
ti
ngt heva l
uesi
sto
multi
plye ach oft hes easonalindice
sbe for
et hey ha vebe ena djusted by t
he
a
djus
tme
ntf
act
orof (
unr
ounde
d).

Fro
mt hetabl
eabovetheindi
ces81and105inQ1andQ2indic
atet
hatsal
esare81%
ofthetre
nd(i.e
.1 9% bel
owt r
end)and105% oft
het
rend(or5% abovethetr
end)
bec
a us
eofthese
asonalinf
lue
nces.

Desea sonal
isi
ngDa ta
-Thi si stheproc essofs e
parati
ngoutt het r
enda ndseasonal
varia
tion.I fseasonalf l
uc tuati
onsa r
er emove dfr
om theor i
ginalda t
a,theresultis
deseas onali
sedda taors e as
o nal
lya djust
e dda t
a.De ali
ngwi thf i
gureswhi cha re
subjectt oseasonalvariationi tisdif
ficulttoknowwhe therforexamplear e
lat
ively
highf igureinabus yse asoni sduewhol l
yt oseas
ona lf
actor
sorot herfac
torsarealso
involve d.Oncet hesea s
ona lfa c
torshavebe enre
move di ti
sthenpos si
bletos e
et he
infl
ue nc eofotherfactorse xcludingseasona lvar
iat
ions.Suchfiguresarethenc a
lled
deseas onali
sedfigures
.
Thef or mulaforde s
easona lis
ingfiguresisgivenby
De
sea
sona
li
sedY= .Thes
eas
ona
lva
ria
ti
onus
edi
nthi
sfor
mul
a

ist
hes eas
onalindex(t hea dj
ust
e dave
rage)–thef i
nalseas
onalvar
iat
ion.Inothe r
wordstheseasonalinf
luencesma ybere movedf r
om at i
mes er
iesbydi vi
dingthe
act
ualY va l
ue f or each peri
od by i tscor r
espondi
ng seas
onali ndex.The
des
easonal
ise
dYva lue
s,whic haremeasuredi
nt hesamephysic
alunit
sa stheact
ua l
yvalues
,ref
lectt
hec ol
lect
iveinf
luenc
eoft het
rend,cycli
cala
ndirr
egul
arf or
cesonly.

Ta
ble50.Adj
ust
mentf
orseasona
lva
ria
tion
1997 Ac t
ual Seas
ona
lVa r
iat
ion De
sea
sonal
is
ed
1 95 81 117
2 100 105 95
3 112 127 88
4 98 87 113

(
2)Theme thodofs emia ver
age
s-Thistechni
quei nvol
vesthec alc
ulat
ionoft wo
ave
rages
,whi c
hwhe nplot
tedonagraphastwos epar
atepoi
ntsform astr
a i
ghtl
ine
whenjoi
ne d.I
fthe
reisa noddnumberofye a
rsorpe r
iods
,ignorethemi ddl
eone.
Cal
cula
tet hemeanvaluefortheea
rlie
rha l
fa ndtheme anforthelat
e rhalfFor
exa
mp l
e

Ta
ble43.Sale
sofXYZLt
d.
Ye ar
s Sa
les($000)
2004 90 Findmeanhe
rewhi
ch
2005 100 i
s95
2006 115 ignor
e
2007 125 Findmea
n her
ewhi
ch
2008 135 i
s130

5
 Calcula
tetheme anpointforeachgroupandplotthe
sepoint
s.Fortheperi
od2004
and2005,c hoosethemi ddl
eposit
ionbe t
ween2004a nd2005andplotagains
t95
(sa
les)the
nf ort
heye a
rs2007a nd2008,choos
ethemiddleposi
ti
onandplotagai
nst
130.Jointhetwopoint
sbyme ansofas t
rai
ghtl
inet
ogiveyouthetr
endli
ne.

160
150
140 Tr
endl
ine
130
120
Sa
les110
100
90
80

0
2003 2004 2005 2006 2007 2008
Ye
ars

(
3)Theme thodofl e as
ts quares.Themos twi delyus e
dme thodoff i
ttingtrendlinesto
times er
iesi stheme t
hodofl easts quares
.Ta ket hephys i
ca lti
mepoi ntsasva l
ues
cod e
da s1,2,3e t
co fthei ndepende ntva r
iable(x).Ta ketheda tavaluesthemse l
vesas
valuesoft hede pendentva r
iable(y) .Ca l
culatethele as
ts qua r
esregr es
sionlineofy
onx ,t ha sy=a+bx (Se
ti er egress
ione quati
ons )
.Tr anslat
et heregressi
online
ast=a+bxwhe reanygi venva lueoft imepoi ntxwi llyi
e ldac orr
e spondingvalueof
thetre ndt.
Ta bl
e44Codi ngme t
hod
Ye ar Sa les(y) Code dx
1999 14500 1
2000 14657 2
2001 17685 3
2002 23711 4
Fort he2003f ore ca
stthec orrespondi ngxva l
uet obes ubs t
itutedintothey=a+bx
equ a
tioni s5.Pl easenot etha tthes amef orecas
tc anbeobt ai
nedi fyous t
artyour
cod e
dxva luesat0.Somepe oplee ve ngot othee xtentofus ingtheye arsa st
heya re
.
Thea nswe robtainedisc orrectbuta tagr e
a tcostintermsoft ime !

Ŷ=a+bx.Thex’ spract
ica l
lyalwaysref
ertosuccess
ivepe ri
ods( usual
lyye ar
s)a nd
whe r
ethisistheca s
ewec ans i
mpl i
fyt
hewo r
koff it
ti
ngal ea
sts quar
estrendlineby
perf
orming ac hangeofs cale,orc odi
ngthex’s,sotha ti
nthene ws c
alethesumof
thex’siszero.Iftheseri
e shasa noddnumbe rofperi
ods ,wec ountf r
omt hemi ddle
ofthei nunitsof1pe rioda s
signi
ngx=0t othemi ddleperioda nd1, 2,3tot he
fol
lowingpe ri
odsa nd–1, -2,
-3t otheprece
dingpe ri
ods .Ifthes eri
esha sa ne ven
numbe rofpe ri
ods,though,t herearetwomi ddl
epe riods,andt hemi dpointoft he
seri
esfall
sbe t
we enthem.As si
gningx=0t othispointintime a ndcountingfrom

6
herei
nunit
sofs i
xmonths(orhal
fyear
s)t
hex’sare1,3,
5… fort
hef
oll
owingpe
riods
and–1,-
3,-
5..
fortheprec
edingper
iods
.Ineit
hercase(anoddnumberandaneve n
numberofper
iodsintheser
ies
)x=0a ndsubs
ti
tuti
ngthisi
ntot
henor
malequa
tions
foraandb

a
nds
olvi
ngf
or a
nd weg
et

a
nd

val
idonl
yifx=0
Theadva
ntageoft
hiski
ndofc
odi
ngi
sevi
dent
.

Exa
mple
Tabl
e45Le
a s
tsquar
eswor
kings
Year Sal
es(Y) x xy x2
1990 10 -
9 -90 81
1991 12 -
7 -84 49
1992 14 -
5 -70 25
1993 11 -
3 -33 9
1994 15 -
1 -15 1
1995 14 1 14 1
1996 12 3 36 9
1997 14 5 70 25
1998 16 7 112 49
1999 14 9 126 81
Tot
al 132 0 66 330

Thel
easts
quarest
rendequa
ti
onis
YT=13.2+0.2x (Tr e
pres
ent
str
end)or
Ŷ=13.2+0.2x

7
Fore
cast
ing
Usethef
oll
owingdat
atoca
lcul
at
etrenda
ndpr
epa
ref
ore
cas
ts
Tabl
e46Sa l
esofa
nitembyquart
er

Ste
psina na l
ysingtheda taandprepari
ngaf orecast
.
1.Calculatethet r
endi nt hedat
aus i
ngl eas
ts quaresme thod.[t a
kephysica
ltimepointsas
val
uesc odeda s1,2,3etc.oftheindependentva r
iable(x).Taket hedatava
luesthe
ms el
vesas
val
uesoft hede pendentva ri
a ey y=a+bx
bl
2.Est
ima tethes al
esfore achquart
erusingregressi
onf ormula.
3.Cal
cul at
et he%va r
iationofeachquarter
’sa ct
ualsalesfromt heesti
mate
sin2a bove.
4.Averaget hepercentageva r
iat
ionsfr
om3.Thi sestimatesthea ver
agesea
sonalvari
at
ions.
5.Pre
pa reforecastbasedo ntre
ndxpe rcent
ages easonalva r
iat
ions.
Useequa tony=a+bx.(onec
i antr
anslat
et heregressi
onl i
nea st=a+bxwhe reanygiven
val
ueoft imepoi ntxwi llyiel
dac orr
espondingva l
ueoft hetrend.

Ta
ble47Ca
lcul
at
ionofEs
ti
mat
es

Tr
endli
ne=y=28.74+1.84x
Est
imate
dsal
eforQ1inyear1=28.
74+1.84(1)
Est
imate
dsal
eforQ4inye a
r4=28.74+1.84(16)
Avera
geper
cent
ageva
ria
ti
onstofi
ndsea
sonalvari
ati
ons

Ta
ble48Adj
ust
edI
ndi
ces

8
Oc cas
ionall
ya djustment
sne e
dtobema deiftot
ali
snot400%
Seasonall
ya djustedfore
c a
st=trendxs e
asonalva
ria
ti
on%.Fore
xampl
eforQ1f
ore
cas
t=
30.58x0. 56=17. 12.
Finalfore
c as
tsarede t
ail
edbelow
Table49.Se asonallyAdjust
edFor e
cas
ts

Oncethea bovehavebeencalc
ulat
edthe
yc a
nbeusedtofor
eca
st(
ext
rapol
at
e)f
utur
esa
les
.
ForexampleforQuart
er17Trend=28.74+1.84(
17)=60.
02
Theseasonaladj
ust
mentinthefi
gurei
s56%a ndt
her
efor
e
theadj
us t
edforec
ast=60.02x56%=33. 61
Q18=55.67
Q19=108.29
Q20=55.05

TrendAna lysi
s
Trendmo veme ntsa reattri
butedt otwobr oa dc l
assesofc aus alf a
ctors,popul at
iona nd
technology.Popul a t
ionf act
orsinc l
udes ucht hingsa st otalpopul ati
ongr owth,gr owt hin
particul
ara r
eas ,a nd numbe rsin va ri
ousa gegr oups .Te chnologicalf actorsmus tbe
cons i
deredbo thi nternaltoande xternaltothec ompa ny/industr
ybe ingstudied.Pur pos esf
or
fit
tingtrendline sareba si
ca l
ly(1)asa naidtof oreca
sting.(2)forhi stor
icalde sc
r i
ptiona nd(3)
toa idinstudyingc ycles.
Whe nstudyingt renda sanaidtof orecas
tingorf orhistoricaldescripti
on,t heempha sisison
thet re
ndwhi chi sus uall
yr epre
se nt
edbyt hema t
h ema ti
cale quationofal i
ne .Ifwea re
int
e res
tedint r
e nda sanaidins t
udyingc ycles,weus ea nyfor moft rendre prese
nta ti
onwhi ch
all
owsf orane a sys eparat
ionoft r
enda ndc yc l
e.Thisma ybeama thema ticall
ine ,ori tmay
beamovi nga ve rage .Thege ner
alobj e
cti
vei stof i
ndat r
endr epresentat
ionwhi c hbi s
e ctst
he

9
c
ycl
esthrought
ime,a
ndwhi
leamathe
mati
call
inemaydothi
sifthecyc
lesa
ref
air
lyr
egul
ar,
t
hef
lex
i bi
lit
yoft
hemovi
ngaver
agemaybeneededf
orot
herproble
ms.

Use softr
e ndlines
The ycorrespondt ot hepurposesf orwhi cht heyarefi
t.Tous eatrendlinea sana i
dt o
forecas
ti
ng,oneme re
lysub s
ti
tut
e stheappropriat
etimevalueint
hetre
ndequa t
ionandsolves
forthetrende sti
ma te
.Con st
antsoft heequa t
ionssuchasba ndrat
esofgrowthma ybeus ed
toshowt hemor einter
esti
ngfeaturesofthet r
endgrowth.Ifthet
rendl
inehasbe encomputed
toa i
dins tudyingc ycle
s,theoriginalvaluesofthed a
taaredivi
dedbythea ppropr
iat
etrend
valuestore movethet re
ndf r
omt heor i
ginalda t
a.

Ratesofc hange
Avera gerat
esofc hangera reaus e
f ulme as
ureforsumma rizi
ngt hegr owthpa
tte
rnsoftr
end
l
ines.The ygivether ateatwhi c
ht hetrendwoul dha vegr owni fi tha dgrownatac ons
tant
r
a t
e(i.e.i
fithada ppea r
eda sastr
aightli
neons emi-logarithmicpa per)Theme asure
sgivea n
aver
a ger at
eofgr owt hwhi c
hwoul dbec ompar
ablet ot hecons tantr at
eofgrowthofa
l
ogarithmics t
rai
ghtlinetrend.Formul a
sf ordet
erminingt hea ve
ra ger e l
at
ivec
hange ,a nd
t
hea veragerateofcha nge ona nannua lbasi
sforanyt ypeoft r
endl inea r
e

a
nd

wheeni
r sthenumberofpos
sibl
ea nnualt
rendva
lue
s,Tni
sthet
rendva
lueoft
hel
astye
ar
andT1i
sthet
rendva
lueoft
hefir
styear.

Sea s
onalAna lysi
s
De te
rmine same asureofs easonalvari
at
i onstart
ingwi t
ht heda t
a.Itdi f
fersfromt r
e ndlines
(comput eddi rect
lyf r
om t heda t
a)whe r
e ass eas
onalsisc omputedbye l
imina t
ingtheot her
compone ntsfromt heda t
as ot ha
tonlyse asonalremains.Se ve
ralme thodsa r
ei nus ebutt he
variat
ionsoft herati
ot omovi nga ver
ageme thodismos tl
yus ed.Theme thode mployst hree
st
e psint h ecomput at
iono ft
hei ndex(1)c omput i
nga na nnualmovi nga verageoft hes er
ies
beings tud ie
d( 2)di vidingt heva luesoft heor igi
nalda tathroughbyt hec orre
s ponding
movi nga verageva l
ue sa nd( 3)a ver
aginga nda djust
ingt her a
ti
osc ompute di nstep2t o
obtainthes e
asonalinde x.

Usesofseasonalindexe
s
Seas
onalindexes( SI
)areusedtoadj
ustati
mes eri
esforsea
sonalvar
iat
ion(i
.e.r
emovethe
se
asonaleff
e ct
sont heser
ies
)andtoa i
dinshort-
ter
mf or
ecas
ti
ng.Accordi
ngtoourmode l
,
wema yremovet heeff
ect
sofseas
onalvar
iat
ionsfr
omat imeser
iesbydividi
ngtheor
igi
nal
dat
afortheseriesthr
oughbytheseas
onali
ndex.Insymbol
sweha ve:

Theref
oreifwewi s
ht ostudyt hemovementofat imes e
rie
sf r
eef
r om sea
sona le f
fect
s,we
divi
de each month orqua rt
eroft he ser
iesthrough by the seas
onali nde xf ort he
corr
espondi
ngmonthorqua r
ter.St
udyofthemove me nt
sofat i
mes er
iesfr
e efroms easonal
vari
at
ionisdesi
rabl
ebecaus e
,intheshor
trun,se
a s
onaleff
ectsmayoutwe i
gha ndr uncounter
tosomelongert
erme f
fect
s ,suchascycl
eandtrend,whosemove mentwea retryingtostudy.
Suchwouldbet hecase,fore xampl
e,nearthebottom ofac yc
lewhe nonewa se val
uating

10
whethera nincreaseins al
esore mployme nti nagi venmont hwa sas ignofrecoveryor
me r
e l
yt heexpect
eds e
asonalincr
eas
ef ort ha tperi
od.Knowle dgeofs e
a s
onalpa
tt
ernsandt he
i
nde xest he
ms el
vesi sus efulinshort-termf or
ecasti
nga ndpl anning.Ifac ompa nyi s
expecti
ngas a l
espe a
kinJ uly(hi
ghs easona li nde
xf orJul
ys ales
),itcanplanit
sproduc t
ion
i
na dva ncetome etthispeakbyma nufac turinge xcesspr
oduc tearl
ierintheyearduringa
sl
acks aless
ea s
ona ndkeepingiti
ninve ntoryf orJuly.

CycleAnalysi
s
Thea nal
ysisofc ycl
icala ndi r
regul
ari nfl
uencesonda taisuse
fulfordesc
ribi
ngpast
var
iati
onsbutbecaus
eoft heirunpredi
ct
a blenat
ure,t
hei
rvaluei
nfore
cast
ingi
sveryli
mit
ed.
Ifwei gnoretheCyc lea ndI rr
egula
rc omponents
,sinc
ebyde fi
nit
ionthe
yc annotbe
pre
dict
ed,thefor
ecas
tingmode lwillbecome .

Mor eatt
enti
onha spr oba blybeende votedt othes t
udyoft hec yc li
calmove me ntsinatime
seri
esthantoa nyoft heot hercompone nts.The reisnoc ompl e teagree
me ntont hewayto
bestme a
surec yc
lesoront hec aus
esofc ycle
s.Themos tgene rallyaccepteda ndus e
dtime
seri
esmode lspeci
fi
e st hatvaluesofthet i
mes eries(Y)arethepr oductofallthec omponent
s
i.
e.Y=TxCxSxIa nds incethesepa r
a t
ionofc ycleandirre gularmove me ntsisdif
fi
cul
t
andtheeffectofanirre gularfact
oronthes e
riesissimil
artotha tofac yclethemode lmaybe
rewri
tt
ena s
Y=T S CIwhe r
et hec ycleandirregulararec ombinedasas i
nglecompone nt.Thecycl
e
ir
regul
arc omponentc anbec a
lcul
atedbydi vidi
ngt heorigina lda t
athroughbyme ansof
tr
endands easonal
,the rebye li
minat
ingt hema ndl ea
vingonlyt hec ycleir
regular.Insymbols

Re s
idualMe thod
Fora nnualda t
a,thec l
assi
calt
imeseri
esmodeli
sY=T CI .Toobt
ainame as
ureofthe
cycleint hea nnualda t
awef irs
tcomputeameasur
eoft
rend.Theorigina
lda
taarethen
divi
de dthroughbyt hismeasureoftr
end,c
ance
li
ngoutt
renda ndl
eavingonl
ythecycle
ir
regularcompone nt.Insymbolswehave:

Themode lformonthl
yorqua r
ter
lydat
aisY=TxSxCI .Toobt
a i
nthecyc
lica
lresi
dualin
thi
sc a
sewemus tfirs
tcomputebothameasur
eoft r
endandase
a s
onali
ndex.Divi
dingthe
ori
ginalda
tabybothagai
nleavesonl
ythecycl
eir
regul
arc
omponent
.

General
ly,noattempti sma detos eparat
ethec ycl
ea ndi r
regul
arcomponentsinannualdat
a,
buttheirre
gularinmont hlyda t
ai ss ometi
me ssmo othedoutbyus inga3- mont hmovi
ng
aver
age giving doubl e we ightt ot he centrali t
em.The r es
iduala pproach,while
st
rai
ghtf
orwa r
di nit
sl ogic,issubjecttoseveraldif
ficulti
es.Tobesucces
sfulitrequi
rest
hat
oneobtainatrendlinet hatfit
swe ll,bise
cti
nga llthec ycl
es.Thi
sisnota l
wa yseasytodo.
Alsoifmonthlyda t
aa rebe i
ngus ed,ana dequateseasonalindexmustbedetermined.Tothe
ext
entthatours easona lindexa ndt rendlinea reincompl et
e,wes ha
llnotobt ainatrue

11
me as
ureofc ycle.Ina ddit
iontherei sthemor eba si
cpr oble
mofwhe the
rweha vethecorr
ect
mode lata l
l.De s
pitethesedi f
fic
ul t
ies,theme thoddoe sof t
e nyieldaus ef
ulme asureofthe
apparentcyclic
ale l
e mentinma nys eri
e s.
Toi l
lustr
atethec omput a
ti
onofame a
s ureofcyc l
icalvaria
tionweus ethedatausedintre
nd
proje
c t
ion.Inaddi t
ionthetrendva luesoft heleastsquarestrendlinehavebeenc omputedfor
eachye ar.Thef i
na lcolumnoft het ablec ont
ainso urme asureofc yclic
alvariat
ion.These
valuesa r
et e
rme d“ cycl
icalrel
atives”a nda reobtainedbydi vidi
ngthes al
esfigur
ef oreach
yearbyt hec or
respondingt r
endf iguref ortheyear.Ther esultsareconventi
onall
ye xpr
essed
aspercentages.

Ta
ble51Cyc
leRe
lat
ive
Ye
ar Sale
s Le
asts
qua
rest
rendva
lue
s
Cycli
c alRelat
ive
(
100Y/ YT)
1990 10 11.
4 87.
7
1991 12 11.
8 101.7
1992 14 12.
2 114.8
1993 11 12.
6 87.
3
1994 15 13.
0 115.4
1995 14 13.
4 104.5
1996 12 13.
8 87.
0
1997 14 14.
2 98.
6
1998 16 14.
6 109.6
1999 14 15.
0 93.
3
Toill
ustr
atethei
nte
rpret
at
ionofas
peci
fi
cye a
r’sva
luel
ookat1992’
sr elat
iveof114.8%.
Thisi
ndicat
esthats
alesi
n1992were14.
8% abovetr
endbe
caus
eofc yc l
ica
la ndirr
egul
ar
fa
ctor
sinthedata
.

Use sofRe sultsofCyc leAna lys i

s
Cyc lea nal
ys i
si spe r
forme dc hiefl
yt oga inknowl edgewhi chwi llaidi nf orecast
ing.The
for
mt hisa i
dt akesma ybequi teva r
ied,howe ver.Tot hebus ines
sma nager,thepr oblem of
cyclefor ec
a st
ingdi f
fer
sf romt hatoftrenda ndseasona linthat,whil
et hela t
terfactorstendto
bes pecifi
ct othec ompa nyori ndustr
y,as pecifi
ccyc l
ef orecastbasedonc ompa nyda taa l
one
woul dbeofdoubt f
ulva lue.Mos tcompa niesandi ndus tr
iesa r
ea ffect
edbyt hec yclesin
gene r
albus inessconditi
ons ,andt hei
rownf orecas
tss houldber e
lat
e dtof orecastsofge nera
l
businessc onditi
ons .Ac ompany’ sanalysi
sofi tsownpa stcyc l
eswi llhelpt oe sta
bl i
shthe
natureofi tsrelat
ionshipt ogene r
albusinesscycles.Cor rectantic
ipati
onoft het urningpoi nt
s
isve ryi mportantt osuc cess
fulbus i
nes sde c
isi
onsa ndf ore
c ast
sofc ycli
calc onditionsa r
e
ma dea ndre vi
seda lmostc onst
a ntl
y.

Expone nti
alSmoothing
Althoughmovi nga veragesha vebeenwi delyusedt osmoothc ert
ainunwante dfl
uctuati
ons
outofat imes e
ries,e xponent
ials moothing i
sa l
soa vail
ablefort hes amepur pose
.
Expone nti
alsmoothingi sapopularforecast
ingtechniquethatcomputesthesmoothe dva l
ue
oft hetimes er
iesinonep eri
odtof orec
asttheva l
uei nthene xtperi
od.Int hi
srespectthe
exponentialsmoothingc anbevi ewe dasas pe
cialkindofamovi nga ver
age ,al
thoughi t
s
natureisquitedif
ferentfromt heconventi
onalmovi nga ve
rageprocessdesc
ribedpreviousl
y.
Thes implestexponentia
ls moothi
ngmode l(cal
leds i
ngleparameterexponentia
lsmoot hi
ng)
postula
testhat:

12
Ne
wf or
ecast=Ol
dfore
cast+Err
orfra
cti
on.
WheeFt+1=t
r hef
ore
cas
tedval
ueforper
iodt
+1

=t hes moot hingc ons tant(0<<1)

Yt=a ctualva luef orpe riodt
Ft=Thef orec astedva luef orperiodt
Thust hene xtpe riod’ sfore castedva lueFt+1 iscalculate
df rom thea c t
ualvalueYta ndthe
forecas t
edva luef ort hepe ri
odFt,bywa yofthes moot hingc onstant.Al pha i salsoa
singlewe i
ght i
ngf a ctor.Ft+1 thusdoe si nfactdepe ndona llpa stvaluesint heser
ies.The
we ighta ss
igne dtoe achobs ervedva luei nthes er
iesde creasesexpone nt
iall
ya ndthi
siswhy
wer ef
e rt
ot hepr oce ssase xpone nti
alsmoot hing.Th ef ormulaa bovec anbeputa s:
Ne wFor ecast-=( ac t
ua lde mand)+( 1-)(oldfore c
ast).
Itise asie
rtoke eptra ckoft hec alc
ulationsinvolvedifweus ethisformofe quati
on.
Smoot hedfor ecastfort hismont h’
ss ales=( saleslastmont h)+( 1-)(previ
ousforecas
tof
lastmo nt
h’ss ales).Fore xampl eifactua lsal
eslastmont hwe re$15000a ndweha dforecas
t
$16000s alesforl astmont hifwewe reus ing=0. 4,forecastforthismont his
Fore cast=0. 4(15000)+( 0.6)(16000)=$15600

Expone nt
ialsmoot hingisnormallyus edinpr oductiona ndi nventor
yc ont
rol.Thet
echni
que
isideallys ui
tedtos hortrunforec as
ti
ngf ori nventoryc ontrol;Itwa sorigi
nal
lyusedin
businessma i
nlytopr ovideane f
fici
enta nde conomi c
alwa yt of or
ecastit
em –byitem
dema ndf ori t
emsc overedbya utomatedi nventoryc ontrols yste
ms .Thisisparti
cul
arl
y
importantinl argec ompa ni
eswhe r eofte
nl ite
r a
llyt housandsa ndt hous
andsofi t
emsa r
e
st
ocke dan ditisnece ss
arytoma ker out
ineforecastsofs aywe e
kt owe ekormonthtomonth
dema ndfore achit
e m.

Choi ceoft hes moothi

ngc ons ta
nt.
Thes moothingc onsta
ntc anbec ha ngedt ogi vemor ewe ighttor ece
ntda ta(whe nitishi gh)
ormor ewe i
ghtt opa stdata( whe ni tislow) .Fore xa mpl ewhe n=0. 5itc a nbes hown
ma thema t
icall
yt hatthene wf orec asti sba seda lmos te ntire l
yonde ma ndint helastt hree
periods.Whe n =0. 1t hef orecas tpl acesl it
tl
ewe ightonr ecentdema nda ndt akesma ny
periods( about19)ofhi st
oricva lue sintoa ccount.Ma nybooksa dvi seust ous eva lues
betwe en0.1a nd0.3.So,t hes moot hingc ons t
a ntcont rol
st henumbe rofpa stda tafrom t he
seri
e swhic ha regoingtoinf l
uenc eourf ore cast
.As ma llc ons ta
ntgiveswe ightst oolderda t
a
andr e s
ult
si nas ma l
lr e
spons eofourf orecaststoa nyc ha nges.A la rgercons t
antfol l
ows
mor ec loselya nyc hanges,butc oul dbeove rs e
nsit
ivet os omei r
regul arcompone ntst hat
mightoc c
ura ta nydatapoi nt.Int hisc asei twoul do ve r-
r eac tandforec astas udde ngr owth,
alt
hought he reisnor easonf ors uchr eacti
on.Expone nt ia
ls moot hi
ngi sava l
ua bleme thodi f
wea redealingwi thas ta
ti
ona rys erie s
,a ndt hisiswh yweha vea c
cepte dthes ugge st
iont hat
thec onsta
nts houldbebe twe en0. 1a ndo. 3.I npi c
kingava luef orthesmoot hi
ngc onstant,the
over al
lobje ct
iveistoobt a
int hemos taccura t
ef orec
a st.Theove ral
lacc uracyofaf oreca s
ting
mode lcanbede t
erminedbyc ompa ringt hef orecastedva lue swi t
ht hea ct
ua lorobs erved
values.

Tabl
e52.UseofDif
fer
entsmoot
hingcons
tant
s
Year Obser
vedValueorac
tualva
lue Ft(=0.
2) Ft(
=0.
5)

13
 1995 9.
9 9.
9 9.9
1996 22.2 12.4 16.0
1997 11.4 12.2 13.7
1998 14.8 12.7 14.2
1999 19.7 14.1 17.0
2000 14.9 14.3 16.0
2001 15.9 14.6 16.0
2002 13.4 14.4 14.7
2003 12.0 13.9 13.4
Theseri
eswith=0.2isagoodde a
lmor est
ablethant
heseri
eswit
h=0.5a ndt
hisiswhat
wewa ntwhe nourobject
ivei stoaver
ageoutt heyeart
oye a
rchange
si nthedata
(ifwe
as
sumet hemtober andom f l
uct
uat
ions
)r a
therthantores
pondtothemthewa ytheser
ies
wit
h=0. 5does
.

Short
comingsofexponentia
lsmoothi
ng
It
sa c
cura
c yisgenera
ll
ypoorhe ncethene e
dt ouseothert
imese
riesmethods
.Itfa
il
sto
re
spondtotrendsorseasonalandcycl
icalvar
iat
ions.Adva
nceds
moothingmodelst
ohandl
e
t
rendsandseasona
lfact
or sar
eusuall
yc ompute
rize
d.

14