Sei sulla pagina 1di 17

9/1/2015

Forecasting
Tuesday,September01,201510:55:34PM

Management&DevelopmentCenter

AboutUs|OurServices|Training|Consulting|

M&DCPurchasing&SupplyChain:MaterialManagement

Forecasting

IntroductiontoMaterialManagement

MasterScheduling

Contents

MaterialRequirementsPlanning
CapacityManagement

1.

Introduction

ProductionActivityControl

2.

DemandManagement

Purchasing

3.

DemandForecasting

4.

CharacteristicsOfDemand

5.

PrinciplesOfForecasting

6.

CollectionAndPreparationOfData

7.

ForecastingTechniques

Forecasting
InventoryFundamentals
OrderQuantities

IndependentDemandOrderingSystems8.

SomeImportantIntrinsicTechniques

9.
Seasonality
PhysicalInventoryandWarehouseManagement
PhysicalDistribution

10.

TrackingTheForecast

1.Introduction
Forecastingisapreludetoplanning.Beforemakingplans,anestimatemustbemadeofwhat
conditionswillexistoversomefutureperiod.Howestimatesaremade,andwithwhataccuracy,is
anothermatter,butlittlecanbedonewithoutsomeformofestimation.
Whyforecast?Therearemanycircumstancesandreasons,butforecastingisinevitableindeveloping
planstosatisfyfuturedemand.Mostfirmscannotwaituntilordersareactuallyreceivedbeforethey
starttoplanwhattoproduce.Customersusuallydemanddeliveryinreasonabletime,and
manufacturersmustanticipatefuturedemandforproductsorservicesandplantoprovidethecapacity
andresourcestomeetthatdemand.Firmsthatmakestandardproductsneedtohavesaleablegoods
immediatelyavailableoratleasttohavematerialsandsubassembliesavailabletoshortenthe
deliverytime.Firmsthatmaketoordercannotbeginmakingaproductbeforeacustomerplacesan
orderbutmusthavetheresourcesoflaborandequipmentavailabletomeetdemand.
Manyfactorsinfluencethedemandforafirmsproductsandservices.Althoughitisnotpossibleto
identifyallofthem,ortheireffectondemand,itishelpfultoconsidersomemajorfactors:
Generalbusinessandeconomicconditions.
Competitivefactors.
Markettrendssuchaschangingdemand.
Thefirmsownplansforadvertising,promotion,pricing,andproductchanges.

2.DemandManagement
Theprimepurposeofanorganizationistoservethecustomer.Marketingfocusesonmeeting
customerneeds,butoperations,throughmaterialsmanagement,mustprovidetheresources.The
coordinationofplansbythesetwopartiesisdemandmanagement.

a.Demandmanagement
Demandmanagementisthefunctionofrecognizingandmanagingalldemandsfor
products.Itoccursintheshort,medium,andlongterm.inthelongterm,demand
projectionsareneededforstrategicbusinessplanningofsuchthingsasfacilities.In
themediumterm,thepurposeofdemandmanagementistoprojectaggregatedemand
forproductionplanning.Intheshortrun,demandmanagementisneededforitemsand
isassociatedwithmasterproductionscheduling.Wearemostconcernedwiththe
latter.
Ifmaterialandcapacityresourcesaretobeplannedeffectively,allsourcesofdemand
mustbeidentified.Theseincludedomesticandforeigncustomers,otherplantsinthe
samecorporation,branchwarehouses,servicepartsandrequirements,promotions,
distributioninventory,andconsignedinventoryincustomerslocations.
Demandmanagementincludesfourmajoractivities:
Forecasting.
Orderprocessing.

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

1/17

9/1/2015

Forecasting
Makingdeliverypromises.Theconceptofavailabletopromisewasdiscussed
inChapter3.
Interfacingbetweenmanufacturingplanningandcontrolandthemarketplace.
Figure8.1showsthisrelationshipgraphically.

b.Orderprocessing.
Orderprocessingoccurswhenacustomersorderisreceived.Theproductmaybe
deliveredfromfinishedgoodsinventoryoritmaybemadeorassembledtoorder.if
goodsaresoldfrominventory,asalesorderisproducedauthorizingthegoodstobe
shippedfrominventory.Iftheproductismadeorassembledto

order,thesalesdepartmentmustwriteupasalesorderspecifyingtheproduct.This
mayberelativelysimpleiftheproductisassembledfromstandardcomponentsbut
canbealengthy,complexprocessiftheproductrequiresextensiveengineering.A
copyofthesalesorderstatingthetermsandconditionsofacceptanceoftheorderis
senttothecustomer.Anothercopy,senttothemasterplanner,isauthorizationtogo
aheadandplanformanufacture.Themasterplannermustknowwhattoproduce,
howmuch,andwhentodeliver.Thesalesordermustbewritteninlanguagethat
makesthisinformationclear.

3.DemandForecasting
Forecastsdependuponwhatistobedone.Theymustbemadeforthestrategicbusinessplan,the
productionplan,andthemasterproductionschedule.AsdiscussedinChapter2,thepurpose,
planninghorizons,andlevelofdetailvaryforeach.
Thestrategicbusinessplanisconcernedwithoverallmarketsandthedirectionoftheeconomyover
thenexttwototenyearsormore.Itspurposeistoprovidetimetoplanforthosethingsthattake
longtochange.Forproduction,thestrategicbusinessplanshouldprovidesufficienttimefor
resourceplanning:plantexpansion,capitalequipmentpurchase,andanythingrequiringalonglead
timetopurchase.Thelevelofdetailisnothigh,andusuallyforecastsareinsalesunits,sales
dollars,orcapacity.Forecastsandplanningwillprobablybereviewedquarterlyoryearly.
Productionplanningisconcernedwithmanufacturingactivityforthenextonetothreeyears.For
manufacturing,itmeansforecastingthoseitemsneededforproductionplanning,suchasbudgets,
laborplanning,longleadtime,procurementitems,andoverallinventorylevels.Forecastsaremade
forgroupsorfamiliesofproductsratherthanspecificenditems.Forecastsandplanswillprobablybe
reviewedmonthly.
Masterproductionschedulingisconcernedwithproductionactivityfromthepresenttoafewmonths
ahead.Forecastsaremadeforindividualitems,asfoundonamasterproductionschedule,individual
iteminventorylevels,rawmaterialsandcomponentparts,laborplanning,andsoforth.Forecasts
andplanswillprobablybereviewedweekly.

4.CharacteristicsOfDemand
Inthischapter,thetermdemandisusedratherthansales.Thedifferenceisthatsalesimplieswhat
isactuallysoldwhereasdemandshowstheneedfortheitem.Sometimesdemandcannotbe
satisfied,andsaleswillbelessthandemand.
Beforediscussingforecastingprinciplesandtechniques,itisbesttolookatsomecharacteristicsof
demandthatinfluencetheforecastandtheparticulartechniquesused.

a.DemandPatterns
Ifhistoricaldatafordemandareplottedagainstatimescale,theywillshowany
shapesorconsistentpatternsthatexist.Apatternisthegeneralshapeofatime
series.

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

2/17

9/1/2015

Forecasting

Althoughsomeindividualdatapointswillnotfallexactlyonthepattern,theytendto
clusteraroundit.
Figure8.2showsahypotheticalhistoricaldemandpattern.Thepatternshowsthat
actualdemandvariesfromperiodtoperiod.Therearefourreasonsforthis:
trend,seasonality,randomvariation,andcycle.

Trend.
Figure8.2showsthatdemandisincreasinginasteadypatternof
demandfromyeartoyear.Thisgraphillustratesalineartrend,but
therearedifferentshapes,suchasgeometricorexponential.The
trendcanbelevel,havingnochangefromperiodtoperiod,oritcan
riseorfall.

Seasonality.
ThedemandpatterninFigure8.2showseachyearsdemand
fluctuatingdependingonthetimeofyear.Thisfluctuationmaybe
theresultoftheweather,holidayseasons,orparticulareventsthat
takeplaceonaseasonalbasis.Seasonalityisusuallythoughtofas
occurringonayearlybasis,butitcanalsooccuronaweeklyor
evendailybasis.Arestaurantsdemandvarieswiththehourofthe
day,andsupermarketsalesvarywiththedayoftheweek.

Randomvariation.
Randomvariationoccurswheremanyfactorsaffectdemandduring
specificperiodsandoccuronarandombasis.Thevariationmaybe
small,withactualdemandfallingclosetothepattern,oritmaybe
large,withthepointswidelyscattered.Thepatternofvariationcan
usuallybemeasured,andthiswillbediscussedinthesectionon
trackingtheforecast.

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

3/17

9/1/2015

Forecasting

Cycle.
Overaspanofseveralyearsandevendecades,wavelikeincreases
anddecreasesintheeconomyinfluencedemand.However,
forecastingofcyclesisajobforeconomistsandisbeyondthe
scopeofthistext.

b.StableversusDynamic
Theshapesofthedemandpatternsforsomeproductsorserviceschangeovertime
whereasothersdonot.Thosethatretainthesamegeneralshapearecalledstable
andthosethatdonotarecalleddynamic.Dynamicchangescanaffectthetrend,
seasonality,orrandomnessoftheactualdemand.Themorestablethedemand,the
easieritistoforecast.Figure8.3showsagraphicalrepresentationofstableand
dynamicdemand.Noticetheaveragedemandisthesameforbothstableand
dynamicpatterns.Itisusuallytheaveragedemandthatisforecast.

c.DependentversusIndependentDemand
Chapter4discusseddependentandindependentdemand.Itwassaidthatdemand
foraproductorserviceisindependentwhenitisnotrelatedtothedemandforany
otherproductorservice.Dependentdemandforaproductorserviceoccurswhere
thedemandfortheitemisderivedfromthatofaseconditem.Requirementsfor
dependentdemanditemsneednotbeforecastbutarecalculatedfromthatofthe
independentdemanditem.
Onlyindependentdemanditemsneedbeforecast.Theseareusuallyenditemsor
finishedgoodsbutshouldalsoincludeservicepartsanditemssuppliedtoother
plantsinthesamecompany(intercompanytransfers).

5.PrinciplesOfForecasting
Forecastshavefourmajorcharacteristicsorprinciples.Anunderstandingofthesewillallowusto
makemoreeffectiveuseofforecasts.Theyaresimpleand,tosomeextent,commonsense.

a.

Forecastsareusuallywrong.Forecastsattempttolookintotheunknownfutureand,
exceptbysheerluck,willbewrongtosomedegree.Errorsareinevitableandmustbe
expected.

b.

Everyforecastshouldincludeanestimateoferror.Sinceforecastsareexpectedtobe
wrong,therealquestionis,Byhowmuch?Everyforecastshouldincludeanestimate
oferroroftenexpressedasapercentage(plusandminus)oftheforecastorasarange
betweenmaximumandminimumvalues.Estimatesofthiserrorcanbemade
statisticallybystudyingthevariabilityofdemandabouttheaveragedemand.

c.

Forecastsaremoreaccurateforfamiliesorgroups.Thebehaviorofindividualitemsina
groupisrandomevenwhenthegrouphasverystablecharacteristics.Forexample,the

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

4/17

9/1/2015

Forecasting
marksforindividualstudentsinaclassaremoredifficulttoforecastaccuratelythanthe
classaverage.Highmarksaverageoutwithlowmarks.Thismeansthatforecastsare
moreaccurateforlargegroupsofitemsthanforindividualitemsinagroup.
Forproductionplanning,familiesorgroupsarebasedonthesimilarityofprocessand
equipmentused.Forexample,afirmforecastingthedemandforknitsocksasaproduct
groupmightforecastmenssocksasonegroupandwomensasanothersincethe
marketsaredifferent.However,productionofmensandwomensanklesockswillbe
doneonthesamemachinesandkneesocksonanother.Forproductionplanning,the
forecastshouldbefor(a)mensandwomensanklesocksand(b)mensandwomens
kneesocks.

d.

Forecastsaremoreaccuratefornearertimeperiods.Thenearfutureholdsless
uncertaintythanthefarfuture.Mostpeoplearemoreconfidentinforecastingwhatthey
willbedoingoverthenextweekthanayearfromnow.Assomeoneoncesaid,tomorrow
isexpectedtobeprettymuchliketoday.
Inthesameway,demandfortheneartermiseasierforacompanytoforecastthanfor
atimeinthedistantfuture.Thisisextremelyimportantforlongleadtimeitemsand
especiallysoiftheirdemandisdynamic.Anythingthatcanbedonetoreduceleadtime
willimproveforecastaccuracy.

6.CollectionAndPreparationOfData
Forecastsareusuallybasedonhistoricaldatamanipulatedinsomewayusingeitherjudgmentora
statisticaltechnique.Thus,theforecastisonlyasgoodasthedataonwhichitisbased.Togetgood
data,threeprinciplesofdatacollectionareimportant.

I.

Recorddatainthesametermsasneededfortheforecast.Thisisaproblemindeterminingthe
purposeoftheforecastandwhatistobeforecast.Therearethreedimensionstothis:

a.

Ifthepurposeistoforecastdemandonproduction,databasedondemand,not
shipments,areneeded.Shipmentsshowwhengoodswereshippedandnotnecessarily
whenthecustomerwantedthem.Thusshipmentsdonotnecessarilygiveatrue
indicationofdemand.

b.

Theforecastperiod,inweeks,months,orquarters,shouldbethesameastheschedule
period.Ifschedulesareweekly,theforecastshouldbeforthesametimeinterval.

c.

Theitemsforecastshouldbethesameasthosecontrolledbymanufacturing.For
example,ifthereareavarietyofoptionsthatcanbesuppliedwithaparticularproduct,
thedemandfortheproductandforeachoptionshouldbeforecast.

Supposeafirmmakesabicyclethatcomesinthreeframesizes,threepossiblewheelsizes,a3,5.
or10speedgearchanger,andwithorwithoutdeluxetrim.Inall,thereare54(3x3x3x2)individual
enditemssold.Ifeachwereforecast,therewouldbe54forecaststomake.Abetterapproachisto
forecast(a)totaldemandand(b)thepercentageofthetotalthatrequireseachframesize,wheelsize,
andsoon.Thatwaythereneedbeonly12forecasts(threeframes,threewheels,fivegears,andthe
bikeitself).
Inthisexample,theleadtimetomakethecomponentswouldberelativelylongincomparisontothe
leadtimetoassembleabike.Manufacturingcanmakethecomponentsaccordingtocomponent
forecastandcanthenassemblebikesaccordingtocustomerorders.Thiswouldbeidealforsituations
wherefinalassemblyschedulesareused.

II.

Recordthecircumstancesrelatingtothedata.Demandisinfluencedbyparticularevents,and
theseshouldberecordedalongwiththedemanddata.Forinstance,artificialbumpsindemand
canbecausedbysalespromotions,pricechanges,changesintheweather,orastrikeata
competitorsfactory.Itisvitalthatthesefactorsberelatedtothedemandhistorysotheymay
beincludedorremovedforfutureconditions.

III.

Recordthedemandseparatelyfordifferentcustomergroups.Manyfirmsdistributetheirgoods
throughdifferentchannelsofdistribution,eachhavingitsowndemandcharacteristics.For
example,afirmmayselltoanumberofwholesalersthatorderrelativelysmallquantities
regularlyandalsoselltoamajorretailerthatbuysalargelottwiceayear.Forecastsof
averagedemandwouldbemeaningless,andeachsetofdemandsshouldbeforecast
separately.

7.ForecastingTechniques
Therearemanyforecastingmethods,buttheycanusuallybeclassifiedintothreecategories:
qualitative,extrinsic,andintrinsic.

a.QualitativeTechniques
Qualitativetechniquesareprojectionsbasedonjudgment,intuition,andinformed
opinions.Bytheirnature,theyaresubjective.Suchtechniquesareusedtoforecast
genera!businesstrendsandthepotentialdemandforlargefamiliesofproductsover
anextendedperiodoftime.Assuch,theyareusedmainlybyseniormanagement.
Productionandinventoryforecastingisusuallyconcernedwiththedemandfor
particularenditems,andqualitativetechniquesareseldomappropriate.
Whenattemptingtoforecastthedemandforanewproduct,thereisnohistoryon
whichtobaseaforecast.Inthesecases,thetechniquesofmarketresearchand
historicalanalogymightbeused.Marketresearchisasystematic,formal,and
consciousprocedurefortestingtodeterminecustomeropinionorintention.Historical
analogyisbasedonacomparativeanalysisoftheintroductionandgrowthofsimilar
productsinthehopethatthenewproductbehavesinasimilarfashion.Another
methodistotestmarketaproduct.

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

5/17

9/1/2015

Forecasting
Thereareseveralothermethodsofqualitativeforecasting.One,calledtheDelphi
method,usesapanelofexpertstogivetheiropiniononwhatislikelytohappen.

b.ExtrinsicTechniques
Extrinsicforecastingtechniquesareprojectionsbasedonexternal(extrinsic)
indicatorswhichrelatetothedemandforacompanysproducts.Examplesofsuch
datawouldbehousingstarts,birthrates,anddisposableincome.Thetheoryisthat
thedemandforaproductgroupisdirectlyproportional,orcorrelates,toactivityin
anotherfield.Examplesofcorrelationare:
Salesofbricksareproportionaltohousingstarts.
Salesofautomobiletiresareproportionaltogasolineconsumption.
Housingstartsandgasolineconsumptionarecalledeconomicindicators.They
describeeconomicconditionsprevailingduringagiventimeperiod.Somecommonly
usedeconomicindicatorsareconstructioncontractawards,automobileproduction,
farmincome,steelproduction,andgrossnationalincome.Dataofthiskindare
compiledandpublishedbyvariousgovernmentdepartments,financialpapersand
magazines,tradeassociations,andbanks.
Theproblemistofindanindicatorthatcorrelateswithdemandandonethatpreferably
leadsdemand,thatis,onethatoccursbeforethedemanddoes.Forexample,the
numberofconstructioncontractsawardedinoneperiodmaydeterminethebuilding
materialsoldinthenextperiod.Whenitisnotpossibletofindaleadingindicator,it
maybepossibletouseanonleadingindicatorforwhichthegovernmentoran
organizationforecasts.Inasense,itisbasingaforecastonaforecast.
Extrinsicforecastingismostusefulinforecastingthetotaldemandforafirms
productsorthedemandforfamiliesofproducts.Assuch,itisusedmostoftenin
businessandproductionplanningratherthantheforecastingofindividualenditems.

c.IntrinsicTechniques
Intrinsicforecastingtechniquesusehistoricaldatatoforecast.Thesedataare
usuallyrecordedinthecompanyandarereadilyavailable.Intrinsicforecasting
techniquesarebasedontheassumptionthatwhathappenedinthepastwillhappen
inthefuture.Thisassumptionhasbeenlikenedtodrivingacarbylookingoutthe
rearviewmirror.Whilethereissomeobvioustruthtothis,itisalsotruethatlacking
anyothercrystalball,thebestguidetothefutureiswhathashappenedinthepast.
Sinceintrinsictechniquesaresoimportant,thenextsectionwilldiscusssomeofthe
moreimportanttechniques.Theyareoftenusedasinputtomasterproduction
schedulingwhereenditemforecastsareneededfortheplanninghorizonoftheplan.

8.SomeImportantIntrinsicTechniques
AssumethatthemonthlydemandforaparticularitemoverthepastyearisasshowninFigure8.4.
SupposeitistheendofDecember,andwewanttoforecastdemandforJanuaryofthecomingyear.
Severalrulescanbeused:
Demandthismonthwillbethesameaslastmonth.Januarydemandwouldbeforecastat84,
thesameasDecember.Thismayappeartoosimple,butifthereislittlechangeindemand
monthtomonth,itprobablywillbequiteusable.
Demandthismonthwillbethesameasdemandthesamemonthlastyear.Forecastdemand
wouldbe92,thesameasJanuarylastyear.Thisruleisadequateifdemandisseasonaland
thereislittleupordowntrend.
Rulessuchasthese,basedonasinglemonthorpastperiod,areoflimitedusewhenthereismuch
randomfluctuationindemand.Usuallymethodsthataverageouthistoryarebetterbecausethey
dampenoutsomeeffectsofrandomvariation.
Asanexample,theaverageoflastyearsdemandcanbeusedasanestimateforJanuarydemand.
Suchasimpleaveragewouldnotberesponsivetotrendsorchangesinlevelofdemand.Abetter
methodwouldbetouseamovingaverage.
Averagedemand.Thisraisesthequestionofwhattoforecast.Asdiscussedearlier,demandcan
fluctuatebecauseofrandomvariation.Itisbesttoforecasttheaveragedemandratherthansecond
guesswhattheeffectofrandomfluctuationwillbe.Thesecondprincipleofforecastingdiscussed
earliersaidthataforecastshouldincludeanestimateoferror.Aswewillseelater,thisrangecanbe
estimated.Thus,aforecastofaveragedemandshouldbemade,andtheestimateoferrorappliedto
it.

a.MovingAverages
Onesimplewaytoforecastistotaketheaveragedemandfor,say,thelastthreeor
sixperiodsandusethatfigureastheforecastforthenextperiod.Attheendofthe
next
January92
July84

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

6/17

9/1/2015

Forecasting
February83
August81
March66
September75
April74
October63
May75
November91
June84
December84

period,thefirstperioddemandisdroppedandthelatestperioddemandaddedto
determineanewaveragetobeusedasaforecast.Thisforecastwouldalwaysbe
basedontheaverageoftheactualdemandoverthespecifiedperiod.
Forexample,supposeitwasdecidedtouseathreemonthmovingaverageonthe
datashowninFigure8.4.OurforecastforJanuary,basedonthedemandinOctober,
November,andDecember,wouldbe:

63+91+84
.=79
3
NowsupposethatJanuarydemandturnedouttobe90insteadof79.Theforecastfor
Februarywouldbecalculatedas:

91+84+90
=88
3

b.ExampleProblem
Demandoverthepastthreemonthshasbeen120,135,and114units.Usingathree
monthmovingaverage,calculatetheforecastforthefourthmonth.

Answer
120+135+114369
Forecastformonth4===
123
33
Actualdemandforthefourthmonthturnedouttobe129.Calculatetheforecastfor
thefifthmonth.

135+114+129
Forecastformonth5==
126
3
Inthepreviousdiscussion,theforecastforJanuarywas79,andtheforecastforFebruarywas88.The
forecasthasrisen,reflectingthehigherJanuaryvalueandthedroppingofthelowOctobervalue.Ifa
longerperiod,suchassixmonths,isused,theforecastdoesnotreactasquickly.Thefewermonths
includedinthemovingaverage,themoreweightisgiventothelatestinformation,andthefasterthe
forecastreactstotrends.However,theforecastwillalwayslagbehindatrend.Forexample,consider
thefollowingdemandhistoryforthepastfiveperiods:
Period

Demand
1

1000

2000

3000

4000

5000

Thereisarisingtrendtodemand.Ifafiveperiodmovingaverageisused,theforecastforperiod6is
(1000+2000+3000+4000+5000)5=3000.Itdoesnotlook
veryaccuratesincetheforecastislaggingactualdemandbyalargeamount.However,
ifathreemonthmovingaverageisused,theforecastis(3000+4000+5000)3
4000.Notperfect,butsomewhatbetter.Thepointisthatamovingaveragealwayslags
atrend,andthemoreperiodsincludedintheaverage,thegreaterthelagwillbe.
Ontheotherhand,ifthereisnotrendbutactualdemandfluctuatesconsiderablyduetorandom
variation,amovingaveragebasedonafewperiodsreactstothefluctuationratherthanforecaststhe
average.Considerthefollowingdemandhistory:
Period

Demand
1

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

2000

7/17

9/1/2015

Forecasting
2

5000

3000

1000

4000

Thedemandhasnotrendandisrandom.Ifafivemonthmovingaverageisused,theforecastforthe
nextmonthis3000.Thisreflectsallthevalues.Ifatwomonthaverageistaken,theforecastsforthe
third,fourth,fifth,andsixthmonthsare:
Forecastforthirdmonth=(2000+5000)

2=3500

Forecastforfourthmonth=(5000+3000)2=4000
Forecastforfifthmonth=(3000+1000)2=2000
Forecastforsixthmonth=(1000+4000)22500

Withatwomonthmovingaveragetheforecastreactsveryquicklytothelatestdemandandthusis
notstable.
Movingaveragesarebestusedforforecastingproductswithstabledemandwherethereislittletrend
orseasonality.Movingaveragesarealsousefultofilteroutrandomfluctuations.Thishassome
commonsensesinceperiodsofhighdemandareoftenfollowedbyperiodsoflowdemand.
Onedrawbacktousingmovingaveragesistheneedtoretainseveralperiodsofhistoryforeachitem
tobeforecast.Thiswillrequireagreatdealofcomputerstorageorclericaleffort.Also,the
calculationsarecumbersome.Acommonforecastingtechnique,calledexponentialsmoothing,
givesthesameresultsasamovingaveragebutwithouttheneedtoretainasmuchdataandwith
easiercalculations.

c.ExponentialSmoothing
Itisnotnecessarytokeepmonthsofhistorytogetamovingaveragebecausethe
previouslycalculatedforecasthasalreadyallowedforthishistory.Therefore,the
forecastcanbebasedontheoldcalculatedforecastandthenewdata.
UsingthedatainFigure8.4,supposeanaverageofthedemandofthelastsix
months(80units)isusedtoforecastJanuarydemand.IfattheendofJanuary,
actualdemandis90units,wemustdropJulysdemandandpickupJanuarys
demandtodeterminethenewforecast.However,ifanaverageoftheoldforecast
(80)andtheactualdemandforJanuary(90)istaken,thenewforecastforFebruaryis
85units.Thisformulaputsasmuchweightonthemostrecentmonthasontheold
forecast(allpreviousmonths).Ifthisdoesnotseemsuitable,lessweightcouldbe
putonthelatestactualdemandandmoreweightontheoldforecast.Perhapsputting
only10%oftheweightonthelatestmonthsdemandand90%oftheweightonthe
oldforecastwouldbebetter.Inthatcase,
Februaryforecast=0.1(90)+0.9(80)=81
Noticethatthisforecastdidnotriseasmuchasourpreviouscalculationinwhichthe
oldforecastandthelatestactualdemandweregiventhesameweight.One
advantagetoexponentialsmoothingisthatthenewdatacanbegivenanyweight
wanted.
Theweightgiventolatestactualdemandiscalledasmoothingconstantandis
representedbytheGreekletteralpha(a).Itisalwaysexpressedasadecimalfrom0
to1.0.
Ingeneral,theformulaforcalculatingthenewforecastis:
Newforecast=(a)latestdemand)+(1a)previousforecast)

d.ExampleProblem
TheoldforecastforMaywas220,andtheactualdemandforMaywas190.Ifalpha
(a)is0.15,calculatetheforecastforJune.IfJunedemandturnsouttobe218,
calculatetheforecastforJuly.

Answer
Juneforecast=(0.15)(190)+(10.15)(220)=215.5
Julyforecast=(0.15)(218)+(0.85)(215.5)=215.9

Exponentialsmoothingprovidesaroutinemethodforregularlyupdatingitemforecasts.Itworksquite
wellwhendealingwithstableitems.Generally,ithasbeenfoundsatisfactoryforshortrange
forecasting.Itisnotsatisfactorywherethedemandisloworintermittent.

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

8/17

9/1/2015

Forecasting

Exponentialsmoothingwilldetecttrends,althoughtheforecastwilllagactualdemandifadefinite
trendexists.Figure8.5showsagraphoftheexponentiallysmoothedforecastlaggingtheactual
demandwhereapositivetrendexists.Noticetheforecastwiththelargerafollowsactualdemand
moreclosely.
Ifatrendexists,itispossibletouseaslightlymorecomplexformulacalleddoubleexponential
smoothing.Thistechniqueusesthesameprinciplesbutnoteswhethereachsuccessivevalueofthe
forecastismovingupordownonatrendline.Doubleexponentialsmoothingisbeyondthescopeof
thistext.
Aproblemexistsinselectingthebestalphafactor.Ifalowfactorsuchas0.1isused,theold
forecastwillbeheavilyweighted,andchangingtrendswillnotbepickedupasquicklyasmightbe
desired.Ifalargerfactorsuchas0.4isused,theforecastwillreactsharplytochangesindemand
andwillbeerraticifthereisasizablerandomfluctuation.Agoodwaytogetthebestalphafactoristo
usecomputersimulation.Usingpastactualdemand,forecastsaremadewithdifferentalphafactorsto
seewhichonebestsuitsthehistoricaldemandpatternforparticularproducts.

9.Seasonality
Manyproductshaveaseasonalorperiodicdemandpattern:skis,lawnmowers,bathingsuits,and
Christmastreelightsareexamples.Lessobviousareproductswhosedemandvariesbythetimeof
day,week,ormonth.Examplesofthesemightbeelectricpowerusageduringthedayorgrocery
shoppingduringtheweek.Powerusagepeaksbetween4and7p.m.,andsupermarketsaremost
busytowardtheendoftheweekorbeforecertainholidays.

a.SeasonalIndex
Ausefulindicationofthedegreeofseasonalvariationforaproductistheseasonal
index.Thisindexisanestimateofhowmuchthedemandduringtheseasonwillbe
aboveorbelowtheaveragedemandfortheproduct.Forexample,swimsuitdemand
mightaverage100permonth,butinJulytheaverageis175andinSeptember,35.
TheindexforJulydemandwouldbe1.75andforSeptember,0.35.
Theformulafortheseasonalindexis:

Period
averagedemand
Seasonalindex=
.
Averagedemand
forallperiods
Theperiodcanbedaily,weekly,monthly,orquarterlydependingonthebasisfor
theseasonalityofdemand.
Theaveragedemandforallperiodsisavaluethataveragesoutseasonality.Thisis
calledthedeseasonalizeddemand.Thepreviousequationcanberewrittenas:

periodaverage
demand
Seasonalindex=
.
deseasonalized
demand

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

9/17

9/1/2015

Forecasting

b.ExampleProblem
Aproductthatisseasonallybasedonquarterlydemandandthedemandforthe
pastthreeyearsisshowninFigure8.6.Thereisnotrend,butthereisdefinite
seasonality.Averagequarterlydemandis100units.Figure8.6alsoshowsagraph
ofactualseasonaldemandandaveragequarterlydemand.Theaveragedemand
shownisthehistoricalaveragedemandforallperiods.Rememberweforecast
averagedemand,notseasonaldemand.

Answer
Theseasonalindicescannowbecalculatedasfollows:

128
Seasonalindex==1.28
(quarter1)
100
102
==1.2
(quarter2)
75
75
==0.75
(quarter3)
100
95
==0.95
(quarter4)
100

Totalofseasonalindices=4.00
Notethatthetotalofalltheseasonalindicesequalsthenumberofperiods.Thisis
agoodwaytocheckwhetherthecalculationsarecorrect.

Year

Quarter
1

Total

122

108

81

90

401

130

100

73

96

399

132

98

71

99

400

Average

128

102

75

95

400

c.SeasonalForecasts
Theequationfordevelopingseasonalindicesisalsousedtoforecastseasonal
demand.Ifacompanyforecastsaveragedemandforallperiods,theseasonal
indicescanbeusedtocalculatetheseasonalforecasts.Changingtheequation
aroundweget:
Seasonaldemand=(seasonalindex)(deseasonalizeddemand)

d.ExampleProblem
Thecompanyinthepreviousproblemforecastsanannualdemandnextyearof420
units.Calculatetheforecastforquarterlysales.

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

10/17

9/1/2015

Forecasting

Answer
420
Forecastaveragequarterlydemand==105
units
4
Expectedquarterdemand=(seasonalindex)
(forecastquarterlydemand)
Expectedfirstquarterdemand=1.28x105=134.4
units
Expectedsecondquarterdemand=1.02x105=
107.1units
Expectedthirdquarterdemand=0.75x105=78.75
units
Expectedfourthquarterdemand=0.95x105=99.75
units

Totalforecastdemand=420
units

e.DeseasonalizedDemand
Forecastsdonotconsiderrandomvariation.Theyaremadeforaveragedemand,
andseasonaldemandiscalculatedfromtheaverageusingseasonalindices.Figure
8.7showsbothactualdemandandforecastaveragedemand.Theforecastaverage
demandisalsothedeseasonalizeddemand.Historicaldataareofactualseasonal
demand,andtheymustbedeseasonalizedbeforetheycanbeusedtodevelopa
forecastofaveragedemand.
Also,ifcomparisonsaremadebetweensalesindifferentperiods,theyare
meaninglessunlessdeseasonalizeddataareused.Forexample,acompanyselling
tennisracketsfindsdemandisusuallylargestinthesummer.However,some
peopleplayindoortennis,sothereisdemandinthewintermonthsaswell.If
demandinJanuarywas5200unitsandinJunewas24,000units,howcould
JanuarydemandbecomparedtoJunedemandtoseewhichwasthebetterdemand
month?Ifthereisseasonality,comparisonofactualdemandwouldbemeaningless.
Deseasonalizeddataareneededtomakeacomparison.
Theequationtocalculatedeseasonalizeddemandisderivedfromtheprevious
seasonalequationandisasfollows:

actualseasonal
demand
Deseasonalizeddemand=
.
seasonalindex

f.ExampleProblem
AcompanysellingtennisracketshasaJanuarydemandof5200unitsandaJuly
demandof24,000units.IftheseasonalindicesforJanuarywere0.5andforJune
were2.5,calculatethedeseasonalizedJanuaryandJulydemand.Howdothetwo
monthscompare?

Answer
DeseasonalizedJanuarydemand=52000.510,400units
DeseasonalizedJunedemand=24,0002.5=9600units
JuneandJanuarydemandcannowbecompared.Onadeseasonalizedbasis,
JanuarydemandisgreaterthanJunedemand.

Deseasonalizeddatamustbeusedforforecasting.Forecastsaremadeforaveragedemand,andthe
forecastforseasonaldemandiscalculatedfromtheaveragedemandusingtheappropriateseason
index.
Therulesforforecastingwithseasonalityare:
Onlyusedeseasonalizeddatatoforecast.
Forecastdeseasonalizeddemand,notseasonaldemand.
Calculatetheseasonalforecastbyapplyingtheseasonalindextothebaseforecast.

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

11/17

9/1/2015

Forecasting

g.ExampleProblem
Acompanyusesexponentialsmoothingtoforecastdemandforitsproducts.For
April,thedeseasonalizedforecastwas1000,andtheactualseasonaldemandwas
1250units.TheseasonalindexforAprilis1.2andforMayis0.7.Ifcxis0.1,
calculate:

a.

ThedeseasonalizedactualdemandforApril.

b.
c.

ThedeseasonalizedMayforecast.
TheseasonalforecastforMay.

Answer

1250
a.DeseasonalizedactualdemandforApril=
=1042

1.2
b.DeseasonalizedMayforecast=cx(latestactual)+
1cx)
(previous
forecast)
=0.1(1042)+
0.9(1000)=1004
c.SeasonalizedMayforecast=(seasonalindex)
(deseasonalized
forecast)
=0.7(1004)=703

10.TrackingTheForecast
Asnotedinthediscussionontheprinciplesofforecasting,forecastsareusuallywrong.Thereare
severalreasonsforthis,someofwhicharerelatedtohumaninvolvementandotherstothebehavior
oftheeconomy.Iftherewereamethodofdetermininghowgoodaforecastis,forecastingmethods
couldbeimprovedandbetterestimatescouldbemadeaccountingfortheerror.Thereisnopointin
continuingwithaplanbasedonpoorforecastdata.Weneedtotracktheforecast.Trackingthe
forecastistheprocessofcomparingactualdemandwiththeforecast.

a.ForecastError
Forecasterroristhedifferencebetweenactualdemandandforecastdemand.Error
canoccurintwoways:biasandrandomvariation.
Bias.Cumulativeactualdemandmaynotbethesameasforecast.Considerthedata
inFigure8.8.Actualdemandvariesfromforecast,andoverthesixmonthperiod,
cumulativedemandis120unitsgreaterthanexpected.
Biasexistswhenthecumulativeactualdemandvariesfromthecumulativeforecast.
Thismeanstheforecastaveragedemandhasbeenwrong.IntheexampleinFigure
8.8,theforecastaveragedemandwas100,buttheactualaveragedemandwas720

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

12/17

9/1/2015

Forecasting
6=120units.Figure8.9showsagraphofcumulativeforecastandactualdemand.
Biasisasystematicerrorinwhichtheactualdemandisconsistentlyaboveorbelow
theforecastdemand.Whenbiasexists,theforecastshouldbechangedtoimprove
itsaccuracy.
Month

ForecastActual
Monthly

Cumulative

Monthly

Cumulative

100

100

110

110

100

200

125

235

100

300

120

355

100

400

125

480

100

500

130

610

100

600

110

720

Total

600

600

720

720

Thepurposeoftrackingtheforecastistobeabletoreacttoforecasterrorby
planningarounditorbyreducingit.Whenanunacceptablylargeerrororbiasis
observed,itshouldbeinvestigatedtodetermineitscause.
Oftenthereareexceptionalonetimereasonsforerror.Examplesaremachine
breakdown,customershutdown,largeonetimeorders,andsalespromotions.These
reasonsrelatetothediscussiononcollectionandpreparationofdataandtheneedto
recordthecircumstancesrelatingtothedata.Ontheseoccasions,thedemand
historymustbeadjustedtoconsidertheexceptionalcircumstances.
Errorscanalsooccurbecauseoftiming.Forexample,anearlyorlatewinterwill
affectthetimingofdemandforsnowshovelsalthoughthecumulativedemandwillbe
thesame.
Trackingcumulativedemandwillconfirmtimingerrorsorexceptionalonetime
events.Thefollowingexampleillustratesthis.NotethatinAprilthecumulative
demandisbackinanormalrange.

Month

Cumulative Cumulative
Forecast
Actual

Forecast

Actual

February

100

110

200

205

March*

100

155

300

360

April

100

45

400

405

May

100

90

500

495

*Customerforesawapossiblestrikeandstockpiled.
Month

Forecast

Actual

Variation(error)

100

105

100

94

100

98

100

104

100

103

100

96

Total

600

600

Randomvariation.
http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

13/17

9/1/2015

Forecasting
Inagivenperiod,actualdemandwillvaryabouttheaveragedemand.
Thevariabilitywilldependuponthedemandpatternoftheproduct.
Someproductswillhaveastabledemand,andthevariationwillnot
belarge.Otherswillbeunstableandwillhavealargevariation.
ConsiderthedatainFigure8.10,showingforecastandactual
demand.Noticethereismuchrandomvariation,buttheaverageerror
iszero.Thisshowsthattheav
erageforecastwascorrectandthere
wasnobias.ThedataareplottedinFigure8.11.

b.MeanAbsoluteDeviation
Forecasterrormustbemeasuredbeforeitcanbeusedtorevisetheforecastorto
helpinplanning.Thereareseveralwaystomeasureerror,butonecommonlyusedis
meanabsolutedeviation(MAD).
ConsiderthedataonvariabilityinFigure8.10.Althoughthetotalerror(variation)is
zero,thereisstillconsiderablevariationeachmonth.Totalerrorwouldbeuselessto
measurethevariation.Onewaytomeasurethevariabilityistocalculatethetotal
errorignoringtheplusandminussignsandtaketheaverage.Thisiscalledmean
absolutedeviation:
meanimpliesanaverage,
absolutemeanswithoutreferencetoplusandminus,
deviationreferstotheerror:

sumofabsolute
deviations
MAD=
.
numberof
observations

c.ExampleProblem
GiventhedatashowninFigure8.10,calculatethemeanabsolutedeviation.

Answer
Sumofabsolutedeviations=5+6+2+4
+3+4=24

24
MAD==4
6

d.Normaldistribution
Themeanabsolutedeviationmeasuresthedifference(error)betweenactualdemand
andforecast.Usually,actualdemandisclosetotheforecastbutsometimesisnot.A

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

14/17

9/1/2015

Forecasting
graphofthenumberoftimes(frequency)actualdemandisofaparticularvalue
producesabellshapedcurve.Thisdistributioniscalledanormaldistributionandis
showninFigure8.12.Chapter11givesamoredetaileddiscussionofnormal
distributionsandtheircharacteristics.
Therearetwoimportantcharacteristicstonormalcurves:thecentraltendency,or
average,andthedispersion,orspread,ofthedistribution.InFigure8.12,thecentral
tendencyistheforecast.Thedispersion,thefatnessorthinnessofthenormalcurve,
ismeasuredbythestandarddeviation.Thegreaterthedispersion,thelarger

thestandarddeviation.Themeanabsolutedeviationisanapproximationofthe
standarddeviationandisusedbecauseitiseasytocalculateandapply.
Fromstatisticsweknowthattheerrorwillbewithin:
1MADoftheaverageabout60%ofthetime
2MADoftheaverageabout90%ofthetime
3MADoftheaverageabout98%ofthetime.

e.Usesofmeanabsolutedeviation.
Meanabsolutedeviationhasseveraluses.Someofthemostimportantfollow.
Trackingsignal.Biasexistswhencumulativeactualdemandvariesfromforecast.
Theproblemisinguessingwhetherthevarianceisduetorandomvariationorbias.If
thevariationisduetorandomvariation,theerrorwillcorrectitself,andnothingshould
bedonetoadjusttheforecast.However,iftheerrorisduetobias,theforecast
shouldbecorrected.Usingthemeanabsolutedeviation,wecanmakesome
judgmentaboutthereasonablenessoftheerror.Undernormalcircumstances,the
actualperioddemandwillbewithin3MADoftheaverage98%ofthetime.Ifactual
perioddemandvariesfromtheforecastbymorethan3MAD,wecanbeabout98%
surethattheforecastisinerror.
Atrackingsignalcanbeusedtomonitorthequalityoftheforecast.Thereareseveral
proceduresused,butoneofthesimplerisbasedonacomparisonofthecumulative
sumoftheforecasterrorstothemeanabsolutedeviation.Followingistheequation:

algebraicsumofforecast
errors
Trackingsignal=
.
MAD

f.ExampleProblem
Theforecastis100unitsaweek.Theactualdemandforthepastsixweekshas
been105,110,103,105,107,and115.IfMADis7.5,calculatethesumofthe
forecasterrorandthetrackingsignal.

Answer
Sumofforecasterror=5+10+3+5+7+15=45
Trackingsignal=457.5=6

g.ExampleProblem
Acompanyusesatriggerof4todecidewhetheraforecastshouldbereviewed.
Giventhefollowinghistory,determineinwhichperiodtheforecastshouldbe
reviewed.MADfortheitemis2.
Period Forecast

Actual

Deviation

Cumulative

Tracking

Deviation

Signal

2.5

100

96

100

98

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

15/17

9/1/2015

Forecasting
3

100

104

100

110

Actual

Deviation

Cumulative

Tracking

Deviation

Signal

2.5

Answer
Period Forecast

100

96

0.5

100

98

0.5

100

104

1.5

100

110

10

13

6.5

Theforecastshouldbereviewedinperiod4.

Contingencyplanning.Supposeaforecastismadethatdemandfordoorslammerswillbe100units
andthatcapacityformakingthemis110units.Meanabsolutedeviationofactualdemandaboutthe
forecasthistoricallyhasbeencalculatedat10units.Thismeansthereisa60%chancethatactual
demandwillbebetween90and110unitsanda40%chancethattheywillnot.Withthisinformation,
manufacturingmanagementmightheabletodeviseacontingencyplantocopewiththepossibleextra
demand.
SafelystockThedatacanbeusedasabasisforsettingsafetystock.Thiswillbediscussedin
detailinChapter11.

h.P/DRatio
Becauseoftheinherenterrorinforecasts,companieswhorelyonthemcanruninto
avarietyofproblems.Forexample,thewrongmaterialmaybeboughtandperhaps
processedintothewronggoods.Amorereliablewayofproducingwhatisreally
neededistheuseoftheP/Dratio.
P,orproductionleadtime,isthestackedleadtimeforaproduct.Itincludestime
forpurchasingofrawmaterialstoarrive,manufacturing,assembly,delivery,and
sometimesthedesignoftheproduct.Figure1.1.onpage4showsvarioustimesin
differenttypesofindustriesandisreproducedhereasFigure8.13
"D,"ordemandleadtime,isthecustomersleadtime.itisthetimefromwhena
customerplacesanorderuntilthegoodsaredelivered,itcanbeveryshort,asina
maketostockenvironment,orverylong,asanengineertoordercompany.
Thetraditionalwaytoguardagainstinherenterrorinforecastingistoincludesafety
stockininventory.Thereisanaddedexpensetotheextrainventorycarriedjustin
case.Oneotherwayistomakemoreaccuratepredictions.Therearefivewaysto
moveinthisdirection.

a.

ReducePtime.ThelongerthePtime,themorechancethereisforerror.
Ideally,PwillbelessthanD.

b.

ForceamatchbetweenPandD.Movinginthisdirectioncanbedoneintwo
ways:
a.MakethecustomersDtimeequaltoyourPtime.Thisiscommonwith
customproductswhenthemanufacturermakestheproductaccordingtothe
customersspecification.

c.

Sellwhatyouforecast.Thiswillhappenwhileyoucontrolthemarket.One
goodexampleistheautomobilemarket.Itiscommontoofferspecial
inducementstowardtheendoftheautomotiveyearinordertosellwhatthe
manufacturershavepredicted.

d.

Simplifytheproductline.Themorevarietyintheproductline,themoreroom
forerror.

e.

Standardizeproductsandprocesses.Thismeansthatcustomizationoccurs
closetofinalassembly.Thebasiccomponentsareidentical,orsimilar,forall
components.Figure8.14showsthisgraphically.

f.

Forecastmoreaccurately.Makeforecastsusingawellthoughtout,well
controlledprocess.

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

16/17

9/1/2015

Forecasting

http://www.mdcegypt.com/Pages/Purchasing/Material%20Management/Forecasting.asp

17/17

Potrebbero piacerti anche