Download full Intermittent demand forecasting: context, methods and applications 1st edition aris a.

Page 1


Visit to download the full and correct content document: https://ebookmass.com/product/intermittent-demand-forecasting-context-methods-an d-applications-1st-edition-aris-a-syntetos/

More products digital (pdf, epub, mobi) instant download maybe you interests ...

Intermittent Demand Forecasting: Context, Methods and Applications Syntetos

https://ebookmass.com/product/intermittent-demand-forecastingcontext-methods-and-applications-syntetos/

Applied economic forecasting using time series methods Ghysels

https://ebookmass.com/product/applied-economic-forecasting-usingtime-series-methods-ghysels/

Forecasting with Artificial Intelligence: Theory and Applications Mohsen Hamoudia

https://ebookmass.com/product/forecasting-with-artificialintelligence-theory-and-applications-mohsen-hamoudia/

Flood forecasting : a global perspective 1st Edition Adams

https://ebookmass.com/product/flood-forecasting-a-globalperspective-1st-edition-adams/

Bus Transportation: Demand, Economics, Contracting, and Policy 1st Edition David A. Hensher

https://ebookmass.com/product/bus-transportation-demandeconomics-contracting-and-policy-1st-edition-david-a-hensher/

Bioassays: Advanced Methods and Applications 1st Edition Donat Hader (Editor)

https://ebookmass.com/product/bioassays-advanced-methods-andapplications-1st-edition-donat-hader-editor/

Business Forecasting Michael Gilliland

https://ebookmass.com/product/business-forecasting-michaelgilliland/

Bioinformatics: Methods and Applications Dev Bukhsh

Singh

https://ebookmass.com/product/bioinformatics-methods-andapplications-dev-bukhsh-singh/

Small

Molecule Drug Discovery: Methods, Molecules and Applications 1st Edition Andrea Trabocchi (Editor)

https://ebookmass.com/product/small-molecule-drug-discoverymethods-molecules-and-applications-1st-edition-andrea-trabocchieditor/

IntermittentDemandForecasting

IntermittentDemandForecasting

Context,MethodsandApplications

JohnE.Boylan

LancasterUniversity Lancaster,UK

ArisA.Syntetos

CardiffUniversity Cardiff,UK

Thiseditionfirstpublished2021 ©2021JohnWiley&SonsLtd

Allrightsreserved.Nopartofthispublicationmaybereproduced,storedinaretrievalsystem,or transmitted,inanyformorbyanymeans,electronic,mechanical,photocopying,recordingorotherwise, exceptaspermittedbylaw.Adviceonhowtoobtainpermissiontoreusematerialfromthistitleisavailable athttp://www.wiley.com/go/permissions.

TherightofJohnE.BoyanandArisA.Syntetostobeidentifiedastheauthorsofthisworkhasbeen assertedinaccordancewithlaw.

RegisteredOffices

JohnWiley&Sons,Inc.,111RiverStreet,Hoboken,NJ07030,USA

JohnWiley&SonsLtd,TheAtrium,SouthernGate,Chichester,WestSussex,PO198SQ,UK

EditorialOffice 9600GarsingtonRoad,Oxford,OX42DQ,UK

Fordetailsofourglobaleditorialoffices,customerservices,andmoreinformationaboutWileyproducts visitusatwww.wiley.com.

Wileyalsopublishesitsbooksinavarietyofelectronicformatsandbyprint-on-demand.Somecontentthat appearsinstandardprintversionsofthisbookmaynotbeavailableinotherformats.

LimitofLiability/DisclaimerofWarranty

Thecontentsofthisworkareintendedtofurthergeneralscientificresearch,understanding,anddiscussion onlyandarenotintendedandshouldnotberelieduponasrecommendingorpromotingscientificmethod, diagnosis,ortreatmentbyphysiciansforanyparticularpatient.Inviewofongoingresearch,equipment modifications,changesingovernmentalregulations,andtheconstantflowofinformationrelatingtothe useofmedicines,equipment,anddevices,thereaderisurgedtoreviewandevaluatetheinformation providedinthepackageinsertorinstructionsforeachmedicine,equipment,ordevicefor,amongother things,anychangesintheinstructionsorindicationofusageandforaddedwarningsandprecautions. Whilethepublisherandauthorshaveusedtheirbesteffortsinpreparingthiswork,theymakeno representationsorwarrantieswithrespecttotheaccuracyorcompletenessofthecontentsofthisworkand specificallydisclaimallwarranties,includingwithoutlimitationanyimpliedwarrantiesofmerchantability orfitnessforaparticularpurpose.Nowarrantymaybecreatedorextendedbysalesrepresentatives,written salesmaterialsorpromotionalstatementsforthiswork.Thefactthatanorganization,website,orproduct isreferredtointhisworkasacitationand/orpotentialsourceoffurtherinformationdoesnotmeanthat thepublisherandauthorsendorsetheinformationorservicestheorganization,website,orproductmay provideorrecommendationsitmaymake.Thisworkissoldwiththeunderstandingthatthepublisheris notengagedinrenderingprofessionalservices.Theadviceandstrategiescontainedhereinmaynotbe suitableforyoursituation.Youshouldconsultwithaspecialistwhereappropriate.Further,readersshould beawarethatwebsiteslistedinthisworkmayhavechangedordisappearedbetweenwhenthisworkwas writtenandwhenitisread.Neitherthepublishernorauthorsshallbeliableforanylossofprofitorany othercommercialdamages,includingbutnotlimitedtospecial,incidental,consequential,orother damages.

LibraryofCongressCataloging-in-PublicationDataappliedfor

ISBN978-1-119-97608-0(hardback);LCCN-2021011006

CoverDesign:Wiley

CoverImage:MataAtlantica-AtlanticForestinBrazil©FGTrade/GettyImages,Turbine©Brasil2/Getty Images

Setin9.5/12.5ptSTIXTwoTextbySPiGlobal,Chennai,India 10987654321

ForJanandRachel

Contents

Preface xix

Glossary xxi

AbouttheCompanionWebsite xxiii

1EconomicandEnvironmentalContext 1

1.1Introduction 1

1.2EconomicandEnvironmentalBenefits 3

1.2.1After-salesIndustry 3

1.2.2DefenceSector 4

1.2.3EconomicBenefits 5

1.2.4EnvironmentalBenefits 5

1.2.5Summary 6

1.3IntermittentDemandForecastingSoftware 6

1.3.1EarlyForecastingSoftware 6

1.3.2DevelopmentsinForecastingSoftware 6

1.3.3OpenSourceSoftware 7

1.3.4Summary 7

1.4AboutthisBook 7

1.4.1OptimalityandRobustness 7

1.4.2BusinessContext 8

1.4.3StructureoftheBook 9

1.4.4CurrentandFutureApplications 10

1.4.5Summary 10

1.5ChapterSummary 11 TechnicalNote 11

2InventoryManagementandForecasting 13

2.1Introduction 13

2.2SchedulingandForecasting 13

2.2.1MaterialRequirementsPlanning(MRP) 13

2.2.2DependentandIndependentDemandItems 14

2.2.3MaketoStock 15

2.2.4Summary 15

2.3ShouldanItemBeStockedatAll? 15

2.3.1Stock/Non-StockDecisionRules 16

2.3.2HistoricalorForecastedDemand? 18

2.3.3Summary 18

2.4InventoryControlRequirements 19

2.4.1HowShouldStockRecordsbeMaintained? 19

2.4.2WhenareForecastsRequiredforStockingDecisions? 22

2.4.3Summary 24

2.5OverviewofStockRules 25

2.5.1ContinuousReviewSystems 25

2.5.2PeriodicReviewSystems 26

2.5.3PeriodicReviewPolicies 28

2.5.4Variationsofthe (R, S) PeriodicPolicy 29

2.5.5Summary 30

2.6ChapterSummary 30 TechnicalNotes 31

3ServiceLevelMeasures 33

3.1Introduction 33

3.2JudgementalOrdering 34

3.2.1RulesofThumbfortheOrder-Up-ToLevel 34

3.2.2JudgementalAdjustmentofOrders 34

3.2.3Summary 35

3.3AggregateFinancialandServiceTargets 35

3.3.1AggregateFinancialTargets 36

3.3.2ServiceLevelMeasures 36

3.3.3RelationshipsBetweenServiceLevelMeasures 38

3.3.4Summary 39

3.4ServiceMeasuresatSKULevel 39

3.4.1CostFactors 39

3.4.2UnderstandingofServiceLevelMeasures 40

3.4.3PotentialServiceLevelMeasures 40

3.4.4ChoiceofServiceLevelMeasure 41

3.4.5Summary 42

3.5CalculatingCycleServiceLevels 42

3.5.1DistributionofDemandOverOneTimePeriod 43

3.5.2CycleServiceLevelsBasedonAllCycles 44

3.5.3CycleServiceLevelsBasedonCycleswithDemand 45

3.5.4Summary 47

3.6CalculatingFillRates 48

3.6.1UnitFillRates 48

3.6.2FillRates:StandardFormula 49

3.6.3FillRates:Sobel’sFormula 51

3.6.4Summary 53

3.7SettingServiceLevelTargets 53

3.7.1ResponsibilityforTargetSetting 53

3.7.2Trade-offBetweenServiceandCost 54

3.7.3SettingSKULevelServiceTargets 55

3.7.4Summary 56

3.8ChapterSummary 56

TechnicalNote 57

4DemandDistributions 59

4.1Introduction 59

4.2EstimationofDemandDistributions 60

4.2.1EmpiricalDemandDistributions 60

4.2.2FittedDemandDistributions 62

4.2.3Summary 64

4.3CriteriaforDemandDistributions 64

4.3.1EmpiricalEvidenceforGoodnessofFit 64

4.3.2FurtherCriteria 64

4.3.3Summary 65

4.4PoissonDistribution 65

4.4.1ShapeofthePoissonDistribution 66

4.4.2Summary 67

4.5PoissonDemandDistribution 67

4.5.1Poisson:APrioriGrounds 67

4.5.2Poisson:EaseofCalculation 67

4.5.3Poisson:Flexibility 68

4.5.4Poisson:GoodnessofFit 69

4.5.5TestingforGoodnessofFit 70

4.5.6Summary 72

4.6IncidenceandOccurrence 72

4.6.1DemandIncidence 72

4.6.2DemandOccurrence 73

4.6.3Summary 74

4.7PoissonDemandIncidenceDistribution 75

4.7.1APrioriGrounds 75

4.7.2EaseofCalculation 75

4.7.3Flexibility 76

4.7.4GoodnessofFit 76

4.7.5Summary 79

4.8BernoulliDemandOccurrenceDistribution 79

4.8.1BernoulliDistribution:APrioriGrounds 79

4.8.2BernoulliDistribution:EaseofCalculation 80

4.8.3BernoulliDistribution:Flexibility 81

4.8.4BernoulliDistribution:GoodnessofFit 81

4.8.5Summary 82

4.9ChapterSummary 82

TechnicalNotes 83

x Contents

5CompoundDemandDistributions 87

5.1Introduction 87

5.2CompoundPoissonDistributions 88

5.2.1CompoundPoisson:APrioriGrounds 89

5.2.2CompoundPoisson:Flexibility 89

5.2.3Summary 89

5.3StutteringPoissonDistribution 90

5.3.1StutteringPoisson:APrioriGrounds 91

5.3.2StutteringPoisson:EaseofCalculation 91

5.3.3StutteringPoisson:Flexibility 93

5.3.4StutteringPoisson:GoodnessofFitforDemandSizes 93

5.3.5Summary 95

5.4NegativeBinomialDistribution 96

5.4.1NegativeBinomial:APrioriGrounds 96

5.4.2NegativeBinomial:EaseofCalculation 96

5.4.3NegativeBinomial:Flexibility 97

5.4.4NegativeBinomial:GoodnessofFit 98

5.4.5Summary 99

5.5CompoundBernoulliDistributions 100

5.5.1CompoundBernoulli:APrioriGrounds 100

5.5.2CompoundBernoulli:EaseofCalculation 100

5.5.3CompoundBernoulli:Flexibility 100

5.5.4CompoundBernoulli:GoodnessofFit 101

5.5.5Summary 101

5.6CompoundErlangDistributions 101

5.6.1CompoundErlangDistributions:APrioriGrounds 103

5.6.2CompoundErlangDistributions:EaseofCalculation 104

5.6.3CompoundErlang-2:Flexibility 104

5.6.4CompoundErlang-2:GoodnessofFit 104

5.6.5Summary 105

5.7DifferingTimeUnits 105

5.7.1PoissonDistribution 106

5.7.2CompoundPoissonDistribution 106

5.7.3CompoundBernoulliandCompoundErlangDistributions 107

5.7.4NormalDistribution 108

5.7.5Summary 110

5.8ChapterSummary 110 TechnicalNotes 111

6ForecastingMeanDemand 117

6.1Introduction 117

6.2DemandAssumptions 118

6.2.1ElementsofIntermittentDemand 119

6.2.2DemandModels 119

6.2.3AnIntermittentDemandModel 120

6.2.4Summary 121

6.3SingleExponentialSmoothing(SES) 121

6.3.1SESasanError-correctionMechanism 122

6.3.2SESasaWeightedAverageofPreviousObservations 122

6.3.3PracticalConsiderations 125

6.3.4Summary 126

6.4Croston’sCritiqueofSES 126

6.4.1BiasAfterDemandOccurringPeriods 126

6.4.2MagnitudeofBiasAfterDemandOccurringPeriods 128

6.4.3BiasAfterReviewIntervalswithDemands 128

6.4.4Summary 129

6.5Croston’sMethod 129

6.5.1MethodSpecification 129

6.5.2MethodApplication 130

6.5.3Summary 131

6.6CritiqueofCroston’sMethod 132

6.6.1BiasofSize-intervalApproaches 132

6.6.2InversionBias 132

6.6.3QuantificationofBias 133

6.6.4Summary 134

6.7Syntetos–BoylanApproximation 134

6.7.1PracticalApplication 134

6.7.2FrameworkforCorrectionFactors 135

6.7.3InitialisationandOptimisation 135

6.7.4Summary 138

6.8AggregationforIntermittentDemand 138

6.8.1TemporalAggregation 138

6.8.2Cross-sectionalAggregation 141

6.8.3Summary 142

6.9EmpiricalStudies 143

6.9.1SingleSeries,SinglePeriodApproaches 143

6.9.2SingleSeries,MultiplePeriodApproaches 144

6.9.3Summary 145

6.10ChapterSummary 145 TechnicalNotes 146

7ForecastingtheVarianceofDemandandForecastError 151

7.1Introduction 151

7.2MeanKnown,VarianceUnknown 151

7.2.1MeanDemandUnchangingThroughTime 152

7.2.2RelatingVarianceOverOnePeriodtoVarianceOvertheProtection Interval 152

7.2.3Summary 153

7.3MeanUnknown,VarianceUnknown 153

7.3.1MeanandVarianceUnchangingThroughTime 154

7.3.2MeanorVarianceChangingThroughTime 155

7.3.3RelatingVarianceOverOnePeriodtoVarianceOvertheProtection Interval 156

7.3.4DirectApproachtoEstimatingVarianceofForecastErrorOverthe ProtectionInterval 158

7.3.5ImplementingtheDirectApproachtoEstimatingVarianceOverthe ProtectionInterval 160

7.3.6Summary 160

7.4LeadTimeVariability 161

7.4.1ConsequencesofRecognisingLeadTimeVariance 161

7.4.2VarianceofDemandOveraVariableLeadTime(KnownMean Demand) 162

7.4.3VarianceofDemandOveraVariableLeadTime(UnknownMean Demand) 163

7.4.4DistributionofDemandOveraVariableLeadTime 164

7.4.5Summary 165

7.5ChapterSummary 165 TechnicalNotes 166

8InventorySettings 169

8.1Introduction 169

8.2NormalDemand 170

8.2.1Order-up-toLevelsforFourScenarios 170

8.2.2Scenario1:MeanandStandardDeviationKnown 170

8.2.3Scenario2:MeanDemandUnknownStandardDeviationKnown 172

8.2.4Scenario3:MeanDemandKnownStandardDeviationUnknown 175

8.2.5Scenario4:MeanandStandardDeviationUnknown 176

8.2.6Summary 177

8.3PoissonDemand 177

8.3.1CycleServiceLevelSystemwhentheMeanDemandisKnown 177

8.3.2FillRateSystemwhentheMeanDemandisKnown 178

8.3.3PoissonOUTLevelwhentheMeanDemandisUnknown 179

8.3.4Summary 181

8.4CompoundPoissonDemand 181

8.4.1StutteringPoissonOUTLevelwhentheParametersareKnown 181

8.4.2NegativeBinomialOUTLevelswhentheParametersareKnown 183

8.4.3StutteringPoissonandNegativeBinomialOUTLevelswhentheParameters areUnknown 183

8.4.4Summary 184

8.5VariableLeadTimes 184

8.5.1EmpiricalLeadTimeDistributions 184

8.5.2Summary 185

8.6ChapterSummary 185 TechnicalNotes 186

9AccuracyandItsImplications 193

9.1Introduction 193

9.2ForecastEvaluation 194

9.2.1OnlyOneStepAhead? 194

9.2.2AllPointsinTime? 194

9.2.3Summary 195

9.3ErrorMeasuresinCommonUsage 195

9.3.1PopularForecastErrorMeasures 195

9.3.2CalculationofForecastErrors 197

9.3.3MeanError 197

9.3.4MeanSquareError 198

9.3.5MeanAbsoluteError 198

9.3.6MeanAbsolutePercentageError(MAPE) 198

9.3.7100%MinusMAPE 199

9.3.8ForecastValueAdded 199

9.3.9Summary 200

9.4CriteriaforErrorMeasures 200

9.4.1GeneralCriteria 200

9.4.2AdditionalCriteriaforIntermittence 201

9.4.3Summary 201

9.5MeanAbsolutePercentageErroranditsVariants 201

9.5.1ProblemswiththeMeanAbsolutePercentageError 202

9.5.2MeanAbsolutePercentageErrorfromForecast 202

9.5.3SymmetricMeanAbsolutePercentageError 203

9.5.4MAPEFFandsMAPEforIntermittentDemand 204

9.5.5Summary 205

9.6MeasuresBasedontheMeanAbsoluteError 205

9.6.1MAE:MeanRatio 205

9.6.2MeanAbsoluteScaledError 206

9.6.3MeasuresBasedonAbsoluteErrors 207

9.6.4Summary 208

9.7MeasuresBasedontheMeanError 208

9.7.1DesirabilityofUnbiasedForecasts 209

9.7.2MeanError 209

9.7.3MeanPercentageError 210

9.7.4ScaledBiasMeasures 210

9.7.5Summary 211

9.8MeasuresBasedontheMeanSquareError 211

9.8.1ScaledMeanSquareError 212

9.8.2RelativeRootMeanSquareError 212

9.8.3PercentageBest 213

9.8.4Summary 213

9.9AccuracyofPredictiveDistributions 214

9.9.1MeasuringPredictiveDistributionAccuracy 214

9.9.2ProbabilityIntegralTransformforContinuousData 215

9.9.3ProbabilityIntegralTransformforDiscreteData 215

9.9.4Summary 217

9.10AccuracyImplicationMeasures 218

9.10.1SimulationOutline 218

9.10.2ForecastingDetails 218

9.10.3SimulationDetails 219

9.10.4ComparisonofSimulationResults 220

9.10.5Summary 221

9.11ChapterSummary 221 TechnicalNotes 221

10Judgement,Bias,andMeanSquareError 225

10.1Introduction 225

10.2JudgementalForecasting 225

10.2.1EvidenceonPrevalenceofJudgementalForecasting 226

10.2.2JudgementalBiases 226

10.2.3EffectivenessofJudgementalForecasts:EvidenceforNon-intermittent Items 229

10.2.4EffectivenessofJudgementalForecasts:EvidenceforIntermittent Items 230

10.2.5Summary 231

10.3ForecastBias 232

10.3.1MonitoringandDetectionofBias 232

10.3.2BiasasanExpectationofaRandomVariable 234

10.3.3ResponsetoDifferentCausesofBias 235

10.3.4Summary 236

10.4TheComponentsofMeanSquareError 236

10.4.1CalculationofMeanSquareError 236

10.4.2DecompositionofExpectedSquaredErrors 236

10.4.3DecompositionofExpectedSquaredErrorsforIndependentDemand 238

10.4.4Summary 239

10.5ChapterSummary 240 TechnicalNotes 240

11ClassificationMethods 243

11.1Introduction 243

11.2ClassificationSchemes 244

11.2.1ThePurposeofClassification 244

11.2.2ClassificationCriteria 245

11.2.3Summary 245

11.3ABCClassification 246

11.3.1ParetoPrinciple 246

11.3.2ServiceCriticality 246

11.3.3ABCClassificationandForecasting 247

11.3.4Summary 248

11.4ExtensionstotheABCClassification 248

11.4.1CompositeCriterionApproach 249

11.4.2Multi-criteriaApproaches 250

11.4.3ClassificationforSpareParts 250

11.4.4Summary 251

11.5ConceptualClarifications 251

11.5.1DefinitionofNon-normalDemandPatterns 251

11.5.2ConceptualFramework 252

11.5.3Summary 253

11.6ClassificationBasedonDemandSources 254

11.6.1DemandGeneration 254

11.6.2AQualitativeClassificationApproach 254

11.6.3Summary 255

11.7Forecasting-basedClassifications 255

11.7.1ForecastingandGeneralisation 256

11.7.2ClassificationSolutions 257

11.7.3Summary 258

11.8ChapterSummary 259 TechnicalNotes 260

12MaintenanceandObsolescence 263

12.1Introduction 263

12.2MaintenanceContexts 264

12.2.1Summary 265

12.3CausalForecasting 265

12.3.1CausalForecastingforMaintenanceManagement 266

12.3.2Summary 268

12.4TimeSeriesMethods 268

12.4.1ForecastinginthePresenceofObsolescence 269

12.4.2ForecastingwithGranularMaintenanceInformation 272

12.4.3Summary 273

12.5ForecastinginContext 273

12.6ChapterSummary 275 TechnicalNotes 276

13Non-parametricMethods 279

13.1Introduction 279

13.2EmpiricalDistributionFunctions 280

13.2.1Assumptions 281

13.2.2LengthofHistory 281

13.2.3Summary 282

13.3Non-overlappingandOverlappingBlocks 282

13.3.1DifferencesBetweentheTwoMethods 282

13.3.2MethodsandAssumptions 284

13.3.3PracticalConsiderations 284

13.3.4PerformanceofNon-overlappingBlocksMethod 285

13.3.5PerformanceofOverlappingBlocksMethod 285

13.3.6Summary 286

13.4ComparisonofApproaches 286

13.4.1TimeSeriesCharacteristicsFavouringOverlappingBlocks 286

13.4.2EmpiricalEvidenceonOverlappingBlocks 287

13.4.3Summary 289

13.5ResamplingMethods 289

13.5.1SimpleBootstrapping 289

13.5.2BootstrappingDemandSizesandIntervals 290

13.5.3VZBootstrapandtheSyntetos–BoylanApproximation 292

13.5.4ExtensionofMethodstoVariableLeadTimes 293

13.5.5ResamplingImmediatelyAfterDemandOccurrence 293

13.5.6Summary 294

13.6LimitationsofSimpleBootstrapping 294

13.6.1AutocorrelatedDemand 294

13.6.2PreviouslyUnobservedDemandValues 295

13.6.3Summary 296

13.7ExtensionstoSimpleBootstrapping 296

13.7.1Discrete-timeMarkovChains 296

13.7.2ExtensiontoSimpleBootstrappingUsingMarkovChains 297

13.7.3Jittering 299

13.7.4LimitationsofJittering 300

13.7.5FurtherDevelopments 300

13.7.6EmpiricalEvidenceonBootstrappingMethods 300

13.7.7Summary 302

13.8ChapterSummary 302 TechnicalNotes 303

14Model-basedMethods 305

14.1Introduction 305

14.2ModelsandMethods 305

14.2.1ASimpleModelforSingleExponentialSmoothing 306

14.2.2CritiqueofWeightedLeastSquares 307

14.2.3ARIMAModels 307

14.2.4TheARIMA(0,1,1)ModelandSES 308

14.2.5Summary 309

14.3IntegerAutoregressiveMovingAverage(INARMA)Models 309

14.3.1IntegerAutoregressiveModelofOrderOne,INAR(1) 310

14.3.2IntegerMovingAverageModelofOrderOne,INMA(1) 312

14.3.3MixedIntegerAutoregressiveMovingAverageModels 312 14.3.4Summary 313

14.4INARMAParameterEstimation 313

14.4.1ParameterEstimationforINAR(1)Models 313

14.4.2ParameterEstimationforINMA(1)Models 314

14.4.3ParameterEstimationforINARMA(1,1)Models 314

14.4.4Summary 315

14.5IdentificationofINARMAModels 315

14.5.1IdentificationUsingAkaike’sInformationCriterion 315

14.5.2GeneralModelsandModelIdentification 316

14.5.3Summary 317

14.6ForecastingUsingINARMAModels 317

14.6.1ForecastingINAR(1)MeanDemand 318

14.6.2ForecastingINMA(1)MeanDemand 318

14.6.3ForecastingINARMA(1,1)MeanDemand 319

14.6.4ForecastingUsingTemporalAggregation 319

14.6.5Summary 319

14.7PredictingtheWholeDemandDistribution 319

14.7.1ProtectionIntervalofOnePeriod 320

14.7.2ProtectionIntervalofMoreThanOnePeriod 320

14.7.3Summary 322

14.8StateSpaceModelsforIntermittence 322

14.8.1Croston’sDemandModel 323

14.8.2ProposedStateSpaceModels 324

14.8.3Summary 325

14.9ChapterSummary 325 TechnicalNotes 325

15SoftwareforIntermittentDemand 329

15.1Introduction 329

15.2TaxonomyofSoftware 330

15.2.1ProprietarySoftware 330

15.2.2OpenSourceSoftware 332

15.2.3HybridSolutions 333

15.2.4Summary 333

15.3FrameworkforSoftwareEvaluation 333

15.3.1KeyAspectsofSoftwareEvaluation 334

15.3.2AdditionalCriteria 335

15.3.3Summary 336

15.4SoftwareFeaturesandTheirAvailability 336

15.4.1SoftwareFeaturesforIntermittentDemand 336

15.4.2AvailabilityofSoftwareFeatures 337

15.4.3Summary 338

15.5Training 339

15.5.1Summary 340

15.6ForecastSupportSystems 340

15.6.1Summary 341

15.7AlternativePerspectives 341

15.7.1BayesianMethods 342

15.7.2NeuralNetworks 342

15.7.3Summary 343

15.8WayForward 343

15.9ChapterSummary 345

TechnicalNote 345

References 347

AuthorIndex 365

SubjectIndex 367

Preface

Theimagesonthefrontofthisbookhighlightacrucialtensionforalladvancedeconomies. Thereisadesiretotravelmoreandconsumemore,butalsoagrowingawarenessofthe detrimentaleffectsthatthisishavingontheenvironment.Thereisabelatedrealisation thatthoseofuslivingincountrieswithdevelopedeconomiesneedtoconsumelessand wasteless.

Wastecanoccuratallstagesofthesupplychain.Consumersmaybuyfoodtheynevereat orclothestheyneverwear.Retailersandwholesalersmayordergoodsfrommanufacturers thatneversell.Thesewastagescanbesignificantlyreducedbybetterdemandforecasting andinventorymanagement.Someitemsconformtoregulardemandpatternsandarerelativelyeasytoforecast.Otheritems,withirregularandintermittentdemandpatterns,are muchharder.

Wastagecanbeaddressedbychangesinproduction,movingawayfrombuilt-inobsolescenceandtowardsproductsthatcanbemaintainedandrepairedeconomically.Forthisto beanattractiveproposition,sparepartsneedtobereadilyavailable.Unfortunately,these itemsareoftenthemostdifficulttoforecastbecausemanyofthemaresubjecttothesporadicnatureofintermittentdemand.Althoughtherehavebeensignificantadvancesin intermittentdemandforecastingoverrecentdecades,thesearenotallavailableincommercialsoftware.Inthefinalchapterofthisbook,wehighlighttheprogressthathasbeen made,includingmethodsthatarefreelyavailableinopensourcesoftware.

Thereasonsfortheslowadoptionofnewforecastingmethodsandapproachesincommercialsoftwarearevaried.Webelievethatoneofthereasonsisalackofappreciationof thebenefitsthatmayaccrue.Becauseintermittentdemanditemsaresodifficulttoforecast, itmaybethoughtthat highlyaccurate forecastingmethodscanneverbefound.Thismaybe true.However,itispossibletofind moreaccurate methods,whichcancontributetowards significantimprovementsininventorymanagement.

Thereisalsoaneedforgreaterawarenessofthemethodsthathavebeendevelopedin recentyears.Informationonthemisscatteredamongstavarietyofacademicjournals,and someofthearticlesarehighlytechnical.Therefore,wehavesetourselvesthechallenge ofsynthesizingthisbodyofknowledge.Wehaveendeavouredtobringtogetherthemain strandsofresearchintoacoherentwhole,andassumingnopriorknowledgeofthesubject.

Therearevariousperspectivesfromwhichdemandforecastingcanbeaddressed.One optionwouldbetotakeanoperationsmanagementview,withafocusonforecastingand planningprocesses.Anotherwouldbetotakeamorestatisticalperspective,startingwith

mathematicalmodelsandworkingthroughtheirproperties.Whilesomeofourmaterial hasbeeninfluencedbytheseorientations,thedominantperspectiveofthisbookisthatof operationalresearch(OR).ThestartpointofORshouldalwaysbethereal-lifesituation thatisencountered.Thismeansthatitisessentialtogainanin-depthunderstandingof inventorysystemsandhowforecastsinformthesedecisions.Suchanappreciationenablesa sharperfocusonforecastingrequirementsandtheappropriatecriteriafora‘goodforecast’.

Inthisbook,thefirstthreechaptersfocusontheinventorymanagementcontextinwhich forecastingoccurs,includingtheinventorypoliciesandtheservicelevelmeasuresthatare appropriateforintermittentdemand.Recognisingtheinterconnectionbetweeninventory policies,demanddistributions,andforecastingmethods,thenexttwochaptersfocuson demanddistributions,includingevidencefromstudiesofreal-worlddata.Thefollowing twochaptersconcentrateonforecastingmethods,withdiscussionofpracticalissuesthat mustbeaddressedintheirimplementation.Wethenturntothelinkagebetweenforecasts andinventoryavailability,andreviewhowforecastaccuracyshouldbemeasuredandhow itsimplicationsforinventoriesshouldbeassessed.Wealsolookathowstockkeepingunits shouldbeclassifiedforforecastingpurposes,andexaminemethodsdesignedspecifically toaddressmaintenanceandobsolescence.Thenexttwochaptersdealwithmethodsthat cantacklemorechallengingdemandpatterns.Weconcludewithareviewofforecasting softwarerequirementsandourviewsonthewayforward.

Wearegratefultothosepioneerswhoinspiredustostudythissubject,andwhohave givenusvaluableadviceovertheyears,especiallyJohnCroston,RoyJohnston,andTom Willemain.Wewouldliketoexpressourthankstothosewhocommentedondraftchapters ofthisbook:ZiedBabai,StephenDisney,RobertFildes,ThanosGoltsos,MatteoKalchschmidt,StephanKolassa,NikosKourentzes,MonaMohammadipour,EricaPastore,Fotios Petropoulos,DennisPrak,Anna-LenaSachs,andIvanSvetunkov;andtoNicoleAyiomamitouandAntonisSiakalliswhohelpedwiththefigures.

LancasterandCardiff January2021

JohnE.Boylan ArisA.Syntetos

Glossary

ADIDAaggregate–disaggregateintermittentdemandapproach

AICAkaikeinformationcriterion

ARautoregressive

ARIMAautoregressiveintegratedmovingaverage

ARMAautoregressivemovingaverage

APEabsolutepercentageerror

BObackorder

BoMbillofmaterials

BSBrierscore

CDFcumulativedistributionfunction

CFEcumulativeforecasterror

CSLcycleservicelevel(allreplenishmentcycles)

CSL+ cycleservicelevel(replenishmentcycleswithsomedemand)

CVcoefficientofvariation

EDFempiricaldistributionfunction

ERPenterpriseresourceplanning

FMECAfailuremode,effects,andcriticalityanalysis

FRfillrate

FSSforecastsupportsystem

FVAforecastvalueadded

HEShyperbolicexponentialsmoothing

INARintegerautoregressive

INARMAintegerautoregressivemovingaverage

INMAintegermovingaverage

IPinventoryposition

KSKolmogorov–Smirnov(test)

LTDlead-timedemand

MAmovingaverage

MADmeanabsolutedeviation

MAEmeanabsoluteerror

MAPEmeanabsolutepercentageerror

MAPEFFmeanabsolutepercentageerrorfromforecast

MASEmeanabsolutescalederror

MEmeanerror

MMSEminimummeansquareerror

MPEmeanpercentageerror

MPSmasterproductionschedule

MROmaintenance,repair,andoperations

MRPmaterialrequirementsplanning

MSEmeansquareerror

MSOEmultiplesourceoferror

MTOmaketoorder

MTSmaketostock

NBDnegativebinomialdistribution

NNneuralnetwork

NOBnon-overlappingblocks

OBoverlappingblocks

OUTorderupto

PISperiodsinstock

PITprobabilityintegraltransform

RMSErootmeansquareerror

rPITrandomisedprobabilityintegraltransform

S&OPsalesandoperationsplanning

SBASyntetos–BoylanApproximation(method)

SBCSyntetos–Boylan–Croston(classification)

SCMsupplychainmanagement

SESsingle(orsimple)exponentialsmoothing

SKUstockkeepingunit

SLAservicelevelagreement

SMAsimplemovingaverage

sMAPEsymmetricmeanabsolutepercentageerror

sMSEscaledmeansquareerror

SOHstockonhand

SOOstockonorder

SSOEsinglesourceoferror

TSBTeunter–Syntetos–Babai(method)

VZViswanathan–Zhou(method)

WMHWrightModifiedHolt(method)

WSSWillemain–Smart–Schwarz(method)

AbouttheCompanionWebsite

Thisbookisaccompaniedbyacompanionwebsite.

www.wiley.com/go/boylansyntetos/intermittentdemandforecasting

Thiswebsiteincludes:

● Datasets(withaccompanyinginformation)

● LinkstoRpackages

EconomicandEnvironmentalContext

1.1Introduction

Demandforecastingisthebasisformostplanningandcontrolactivitiesinanyorganisation. Unlessaforecastoffuturedemandisavailable,organisationscannotcommittostaffinglevels,productionschedules,inventoryreplenishmentorders,ortransportationarrangements. Itisdemandforecastingthatsetstheentiresupplychaininmotion.

Demandwilltypicallybeaccumulatedinsomepre-defined‘timebuckets’(periods),such asaday,aweek,oramonth.Thedeterminationofthelengthofthetimeperiodthatconstitutesatimebucketisaveryimportantdecision.Itisachoicethatshouldrelatetothenature oftheindustryandthevolumeofthedemanditselfbutitmayalsobedictatedbytheIT infrastructureorsoftwaresolutionsinplace.Regardlessofthelengthofthetimebuckets, demandrecordseventuallyformatimeseries,whichisasequenceofsuccessivedemand observationsovertimeperiodsofequallength.

Onmanyoccasions,demandmaybeobservedineverytimeperiod,resultinginwhat issometimesreferredtoas‘non-intermittentdemand’.Alternatively,demandmayappear sporadically,withnodemandatallinsomeperiods,leadingtoanintermittentappearance ofdemandoccurrences.Shouldthatbethecase,contributiontorevenuesisnaturallylower thanthatoffaster-movingdemanditems.Intermittentdemanditemsdonotattractmuch marketingattention,astheywillrarelybethefocusofapromotion,forexample.However, theyhavesignificantcostimplicationsforasimplereason:thereareoftenmanyofthem! Serviceorsparepartsareveryfrequentlycharacterisedbyintermittentdemandpatterns. Theseitemsareessentiallycomponentsor(sub-)assembliescontributingtothebuild-up ofafinalproduct.However,theyface‘independentdemand’,whichisdemandgenerated directlyfromcustomers,ratherthanproductionrequirementsforaparticularnumberof unitsofthefinalproduct.Intheafter-salesenvironment(or‘aftermarket’),wedealexclusivelywith‘independentdemand’items.Servicepartsfacingintermittentdemandmay representalargeproportionofanorganisation’sinventoryinvestment.Insomeindustries, thisproportionmaybeashighas60%or70%(Syntetos2011).Themanagementofthese itemsisaveryimportanttaskwhich,whensupportedbyintelligentinventorycontrolmechanisms,mayyielddramaticcostreductions. IntermittentDemandForecasting:Context,MethodsandApplications, FirstEdition. JohnE.BoylanandArisA.Syntetos. ©2021JohnWiley&SonsLtd.Published2021byJohnWiley&SonsLtd. CompanionWebsite:www.wiley.com/go/boylansyntetos/intermittentdemandforecasting

Industriesthatrelyheavilyonafter-salessupport,includingtheautomotive,IT,and electronicssectors,aredominatedbyintermittentdemanditems.Thecontributionsof theafter-salesservicestothetotalrevenuesoforganisationsintheseindustrieshavebeen reportedtobeashighas60%(Johnstonetal.2003).Thissignifiesanopportunitynot onlytoreducecostsbutalsotoincreaserevenuesthroughacarefulbalancingofkeeping enoughinstocktosatisfycustomersbutnotsomuchastounnecessarilyincreaseinventory investments.Therearetremendouseconomicbenefitsthatmayberealisedthroughthe reappraisalofmanagingintermittentdemanditems.

Therearealsosignificantenvironmentalbenefitstoberealisedbysuchareappraisal. Becauseoftheirinherentslowmovement,intermittentdemanditemsareatthegreatest riskofobsolescence.Theproblemisexacerbatedbythegreatlyreducedproductlifecycles inmodernindustry.Thisaffectstheplanningprocessforallintermittentdemanditems (bothfinalproductsandsparepartsusedtosustaintheoperationoffinalproducts).Better forecastingandinventorydecisionsmayreduceoverallscrapandwaste.Furthermore,the sustainedprovisionofsparepartsmayalsoreduceprematurereplacementoftheoriginal equipment.

Theareaofintermittentdemandforecastinghasbeenneglectedbyresearchersandpractitionersfortoolong.Fromabusinessperspective,thismaybeexplainedintermsofthelack offocusonintermittentdemanditemsbythemarketingfunctionoforganisations.However,thetougheconomicconditionsexperiencedfromaround2010onwardshaveresulted inaswitchofemphasisfromrevenuemaximisationtocostminimisation.Thisswitchrepositionsintermittentdemanditemsasthefocusofattentioninmanycompanies,aspartof thedrivetodramaticallycutdowncostsandremaincompetitive.Inaddition,themore recentemergenceoftheafter-salesbusinessasamajordeterminantofcompanies’success hasalsoledtotherecognitionofintermittentdemandforecastingasanareaofexceptional importance.

FollowingaseminalcontributioninthisareabyJohnCrostonin1972,intermittent demandforecastingreceivedverylittleattentionbyresearchersoverthenext20years.This wasincontrasttotheextensiveresearchconductedonforecastingfaster-movingdemand items.Researchactivitygrewrapidlyfromthemid-1990sonwards,andwehavenow reachedastagewhereacomprehensivebodyofknowledge,boththeoreticalandempirical, hasbeendevelopedinthisarea.Thisbookaimstoprovidepractitioners,students,and academicresearcherswithasinglepointofreferenceonintermittentdemandforecasting. Althoughthereareconsiderableopeningsforfurtheradvancements,thecurrentstateof knowledgeoffersorganisationssignificantopportunitiestoimprovetheirintermittent demandforecasting.Numerousreports,tobediscussedinmoredetaillaterinthischapter, indicatethatintermittentdemandforecastingisoneofthemajorproblemsfacingmodern organisations.Specialisedsoftwarepackagesoffersomeforecastingsupporttocompanies buttheyoftenlagbehindnewdevelopments.Therearegreatbenefitsthathavenotyet beenachievedinthisarea,andwehopethatthisbookwillmakeacontributiontowards theirrealisation.

Therearethreemainaudiencesforthisbook:

1.Supplychainmanagement(SCM)practitioners,broadlydefined,whowishtorealisethe fullbenefitsofmanagingintermittentdemanditems.

2.Softwaredesignerswantingtoincorporatenewdevelopmentsinforecastingintotheir software.

3.Studentsandacademicswishingtolearnandincorporateintotheircurricula,respectively,thestateoftheartinintermittentdemandforecasting.

Insummary,businesspressurestoreducecostsandenvironmentalpressurestoreduce scrap(oftenintroducedintheformofprescribedpoliciesimposedbynationalgovernments ortheEUforexample)renderintermittentdemanditems,andforecastingtheirrequirements,oneofthemostimportantareasinmodernorganisations.

Therearegreatbenefitsassociatedwithforecastingintermittentdemandmoreaccurately,andthosebenefitsarefarfrombeingrealised.Thismaybeexplainedbythewell reportedinnovation–adoptiongap,whicharisesfromthedivergencebetweeninnovationsandreal-worldpractices.Organisationalpracticestypicallylagbehindsoftware developments,andsoftwaredevelopmentstypicallylagbehindthestateoftheartinthe academicliterature.Itistheaimofthisbooktobridgethesegapsandshowhowintelligent intermittentdemandforecastingmayresultinsignificanteconomicandenvironmental benefits.

Intheremainderofthischapter,wefirstdiscussinmoredetailthepotentialbenefitsthat mayberealisedthroughimprovedintermittentdemandforecasting.Wethenprovidean overviewofthecurrentstateofsupplychainsoftwarepackagesandenterpriseresource planning(ERP)solutionswithregardtointermittentdemandforecasting.Thisisfollowed byasectionwhereweelaborateonboththestructureofthisbookandtheperspectivethat wetakeregardingthematerialpresentedhere.Weclosewithasummaryofthechapter.

1.2EconomicandEnvironmentalBenefits

Intermittentdemandforproductsappearssporadically,withsometimeperiodsshowing nodemandatall.Moreover,whendemandoccurs,thedemandsizemaybeconstantor variable,perhapshighlyso,leadingtowhatisoftentermed‘lumpydemand’.Laterinthis chapter,wediscusswhyforecastingsporadicandlumpydemandpatternsisaverydifficult task.Specificcharacterisationsofintermittentdemandseriesareconsideredindetailin Chapters4and5.

1.2.1After-salesIndustry

Intermittentdemanditemsdominateserviceandrepairpartsinventoriesinmanyindustries(BoylanandSyntetos2010).AsurveybyDeloitteResearch(2006)benchmarkedthe servicebusinessesofmanyoftheworld’slargestmanufacturingcompanieswithcombined revenuesreachingmorethan$1.5trillion;serviceoperationsaccountedforanaverageof 25%ofrevenues.Inadditiontotheircontributiontorevenues,theseitemspresentadistinct opportunityforcostreductions.Maintenance,repair,andoperations(MRO)inventories typicallyaccountforasmuchas40%oftheannualprocurementbudget(Donnelly2013). Increasedrevenuesandreducedcostsnaturallyleadtoincreasedprofits.Manyorganisationshaverepeatedlytestifiedtotheimportanceofafter-salesservicesfortheirbusinesses

andtheprofitstheygenerate.CompaniessuchasBeretta,Canon,DAFTrucks,Electrolux, EPTA,GEOil&Gas,andLavapiuhavereportedcontributionsoftheafter-salesservices totheirtotalprofitofupto50%(Syntetos2011).Comparablenumbershavebeenreported byGaiardellietal.(2007),Kimetal.(2007),andGluecketal.(2007),whileafter-salesservicehasbeenidentifiedasakeyprofitleverinthemanufacturingsector(Manufacturing Management2018).

Intermittentdemanditemsareatthegreatestriskofobsolescence.Manycasestudies (e.g.Molenaersetal.2012)havedocumentedlargeproportionsof‘dead’(obsolete)stockin avarietyofindustries,withseriousenvironmentalimplications.However,under-stocking situationsmaybeasharmful,giventhepotentiallyhighcriticalityoftheitemsinvolved. Incivilaviation,forexample,lackofsparepartsisoneofthemajorcausesof‘aircrafton ground’events(problemsseriousenoughtopreventaircraftfromflying).Badkook(2016) foundthataquarteroftheaircraftinan(un-named)airline’sBoeing777fleetwereaffected bysuchaircraftongroundeventsoverayear.

1.2.2DefenceSector

Defenceinventories,whicharehighlyreliantonspareparts,havebeenrepeatedlyidentified asahighriskareawithadirectimpactonanation’seconomy.IntheUnitedStatesfor example,theDepartmentofDefense(DoD)managesaroundfivemillionsecondaryitems. Theseincluderepairablecomponents,subsystems,assemblies,consumablerepairparts, andbulkitems.Theyreportedthat,asofSeptember2017,thevalueoftheinventorywas$93 billion(GAO2019).Althoughamatterofconcern,therehadbeennosubstantialreductions ininventoryvaluesoverthepreviousdecade(being,forexample,$95billionin2013and 2010;GAO2012,2015).

Amajordeterminantoftheperformanceofaninventorysystemistheforecasting method(s)beingusedtopredictdemand.Inaccurateforecastsleadtoeitherexcess inventoryorshortfalls,dependingonthedirectionoftheforecasterror.Over-forecasting canleadtoholdingstocksthataresimplynotneeded.AccordingtotheUSGovernment AccountabilityOffice(GAO2011,p.11),‘Ourrecentworkidentifieddemandforecasting astheleadingreasonwhytheservicesandDLA[DefenseLogisticsAgency]accumulate excessinventory’.

Unfortunately,progressinimprovingforecastingandinventorymanagementhasbeen slowinmanyindustries,withthedefenceindustrybeingacaseinpoint.TheGAOofthe UnitedStatesreported,‘Since1990,wehaveidentifiedDoD[DepartmentofDefense]supplychainmanagementasahigh-riskareadueinparttoineffectiveandinefficientinventory managementpracticesandprocedures,weaknessesinaccuratelyforecastingthedemand forspareparts,andothersupplychainchallenges.Ourworkhasshownthatthesefactors havecontributedtotheaccumulationofbillionsofdollarsinsparepartsthatareexcess tocurrentneeds’(GAO2015,p.2).Progressininventorymanagementhasbeenmade sincethen,especiallywithregardtothevisibilityofphysicalinventories,receiptprocessing,andcargotracking(GAO2019).Theseimprovementsininformationsystemshaveled toinventorymanagementbeingremovedfromthelistofhigh-riskareas.However,itis notablethatnoclaimshaveyetbeenmadeforcorrespondingimprovementsindemand forecasting.

Movingbeyondtheafter-salesindustry,andthedefencesector,wenowexaminethepotentialbenefitsthatmayresultfromintelligentintermittentdemandforecastingforthewider economy.Purchasedgoodsinventoriesandtheirmanagementaresignificantconcernsfor firmswishingtoremaincompetitiveandsurviveinthemarketplace.Accordingtothe26th AnnualStateofLogisticsreport(CSCMP2015,statisticsreferringto2014),theUnitedStates alonehasbeensittingonapproximately$2trillionworthofgoodsheldforsale.Accordingto thesamereport,theinventorycarryingcosts(taxes,obsolescence,depreciation,andinsurance)areestimatedtobearound$0.5trillion(i.e.about25%ofthevalueofthegoods). Thetotalvalueofinventorywasequivalenttoapproximately14%oftheUSgrossdomestic product(GDP)in2014.Althoughsimilarstatisticshavenotbeengiveninsubsequentpublications,the30thAnnualStateofLogisticsReport(CSCMP2019)revealedthatinventory carryingcostsintheUnitedStatesincreasedby14.8%between2014and2018.

Thesefiguresshowthatahugeamountofcapitalistiedupinwarehouses.Theyalsoindicatethatsmallimprovementsinmanaginginventoriesmaybetranslatedintoconsiderable costbenefits.Weshould,therefore,notbesurprisedtolearnthatfirms,frommanufacturing towholesaletoretail,arecurrentlyintensifyingtheirsearchformoreefficientandeffective inventorymanagementapproaches.Theiraimistominimisenotonlytheirdirectinvestmentsinpurchasedgoodsinventorybutalsotheindirectcostincurredinmanagingthis inventory.Inamaketostock(MTS)environment(discussedinSection1.4.2),ifthereisno decouplingintermsoftheownershipandlocationoftheinventories,thentheseindirect costsbecomemoresignificantthelongerthestockremainsunsold.Thehighvolumesof stocksofintermittentdemanditems,andtheirhighriskofobsolescence,shouldputthem veryhighupthelistofprioritiesformodernbusinesses.

1.2.4EnvironmentalBenefits

Obsolescenceisaveryimportanttopicforsupplychainmanagement.Thecomplexityof supplychains,inconjunctionwithincreasinglyreducedproductlifecycles,isresultingin highlevelsofobsolescence.Molenaersetal.(2012)discussedacasestudywhere54%of thepartsstockedatalargepetrochemicalcompanyhadseennodemandforthelastfive years.Syntetosetal.(2009b)evaluatedtheinventorypracticesemployedintheEuropean sparepartslogisticsnetworkofaJapanesemanufacturer.Theyfoundonecase,reportedin Sweden,wheresomepartsinstockhadnot‘moved’atalloverthepreceding10years.The valueoftheon-handexcess(spareparts)inventoryoftheUSAirForce,Navy,andArmyhas beenestimatedtobeof$1.7billion,$1.4billion,and$2.5billion,respectively(GAO2015). Muchofthisexcessstockisatriskofobsolescence.

Whenobsolescent(or‘dead’)stockiscreated,thereisconsiderableenvironmentalwaste. Firstly,thereisanenvironmentalcostassociatedwithproducinggoodsthatareneverused. Secondly,thereareenvironmentalcostsoftransportingthesegoodstonational,regional, orlocalstockingpoints.Finally,thereareenvironmentalcostsofdisposingofthesestocks. Thepreventionoftheaccumulationofdeadstockreliesonaccuratedemandforecasts.Consequently,moreaccurateandrobustforecastingmethodsmaybetranslatedtosignificant reductionsinwastageorscrap,withconsiderableenvironmentalbenefits.

1.2.5Summary

Moreaccurateforecastingofintermittentdemandpresentsorganisationswithadistinct opportunitytoreducecostsandaddressmajorissuesontheirenvironmentalagenda.In theafter-salescontext,intelligentintermittentdemandforecastingisofparamountimportance,asmanyitemshavedemandpatternsthatareintermittentinnature.Otherinventory settingsthataredominatedbyspareparts(e.g.themilitary,publicutilities,andaerospace) wouldalsobenefitdirectlyfrommoreaccurateintermittentdemandforecastingmethods.

1.3IntermittentDemandForecastingSoftware

Giventherelevanceofintelligentforecastingmethodsinmodernorganisations,itisvital thattheyareincludedinsoftwaresolutions.Thecontinuousupdateofsoftwaretoreflect researchdevelopmentsintheareaofintermittentdemandforecastingisofgreatfinancial andenvironmentalimportance.Forecastingsoftwaresolutionsarebrieflyreviewedinthis sectionandrevisitedingreaterdetailinChapter15.

1.3.1EarlyForecastingSoftware

Earlyforecastingsoftwaresolutionsinthe1950sand1960swerebasedonsingleexponential smoothing(SES)(amethodthatisdiscussedindetailinChapter6),meaningthatintermittentdemanditemswerenottreatedanydifferentlyfromfast-movingitems.SESisa methoddevisedforfast-movingitemsthatexhibitnotrendorseasonality.Itisaverypracticalforecastingmethodfortheseitems,andisincludedinthevastmajorityof(inventory) forecastingsoftwareapplications.Itisstillusedforintermittentdemand,althoughweshall seeinChapter6thatitisnotanaturalmethodfortheseitemsanditdoessufferfromsome majorweaknesses.

1.3.2DevelopmentsinForecastingSoftware

Softwarepackageshavesincemovedon,withmost,butnotall,packagesofferingmethodsthataredesignedforintermittentdemand.Croston’s(1972)method,forexample,was developedspecificallyforintermittentdemanditems,andisincorporatedinstatisticalforecastingsoftwarepackages(e.g.ForecastPro),anddemandplanningmodulesofcomponent basedenterpriseandmanufacturingsolutions(e.g.IndustrialandFinancialSystems,IFS AB).Itisalsoincludedinintegratedreal-timesalesandoperationsplanningprocesses(e.g. SAPAdvancedPlanningandOptimisation[SAPAPO]andSAPDigitalManufacturing).

Similarly,morerecentdevelopmentsindemandcategorisation(rulesthatdistinguish betweenvarioustypesofdemandpatternsandsignifywhenapatternshouldbetreatedas intermittent)havealsobeenadoptedinsomecommercialsoftware(e.g.BlueYonder,SyncronInternational),allowingtheirclientsthecapabilitytoachievesomedramaticinventory costreductions(ResearchExcellenceFramework2014).However,theadoptionofrecent developmentshasnotbeenwidespread,andtherearemanysoftwarepackagesthathave limitedfunctionality.Overall,therehavebeenratherminorimprovementsincommercial softwaresincearound2000despitesomemajorimprovementsinempiricallytestedtheory sincethattime.

1.3.3OpenSourceSoftware

Anotherimportantdevelopment,tobediscussedindetailinChapter15,istheavailabilityofopensourcesoftwareofrecentlyproposedintermittentdemandforecastingmethods. Thisenablescompaniestoincorporatethemintheirownin-housedevelopedsolutions,or forcommercialsoftwarecompaniestoextendtheirrepertoireofmethodsmorereadily.Furthermore,sophisticateddatabasesystemsareenablingcompaniesto‘sliceanddice’their datamoreeasily.Thismeansthatdatamaybeexaminedmorereadilybysegments,such asgeographicalregionsorproductgroupings,inforecastingandplanningsoftware(e.g. ForecastPro,Smoothie).Thisprovidesthegroundworkforimplementingdevelopmentsin forecastingatdifferentlevelsofaggregation(tobediscussedindetailinChapter6).However,whilstsoftwaresolutionsaremovingaheadbyembracingslicinganddicing,theydo notdosointermsofnewforecastingmethods(includingthosethattakeadvantageofslicinganddicing).Therearesignificantopportunitiesofferedbyopensourcesoftwareand moderndataanalyticstoimprovetheforecastingfunctionalityofcommercialsoftware.

1.3.4Summary

Therehavebeensomeverypromisingadvancesintheareaofintermittentdemandforecasting,someofwhichhavefoundtheirwayintosoftwareapplications.However,muchstill remainstobedoneintermsofsoftwarecompanieskeepingupwithimportantmethods thathaverecentlybeendevelopedandparticularlythosethathavebeenempiricallytested andshowntoyieldconsiderablebenefits.

1.4AboutthisBook

Inthissectionwebrieflyreviewthestancetaken,thescopeofdiscussion,andthestructure ofthebook.

1.4.1OptimalityandRobustness

Intermittentdemandpatternsareverydifficulttomodelandforecast.Itisthegenuinelack ofsufficientinformationassociatedwiththeseitems(duetothepresenceofzerodemands) thatmayprecludetheidentificationofseries’componentssuchastrendandseasonality. Demandhistoriesarealsoveryoftenlimited,whichmakesthingsevenworse.Demand arrivessporadicallyand,whenitdoesso,itmaybeofaquantitythatisdifficulttopredict. Theactualdemandsizes(positivedemands)maysometimesbealmostconstantorconsistentlysmallinmagnitude.Alternatively,theymaybehighlyvariable,leadingto‘erratic’ demand.Intermittencecoupledwitherraticnessleadstowhatisknownas‘lumpy’demand. ThegraphinFigure1.1showsexamplesofintermittentandlumpydemandpatterns,based onannualdemandhistoryfortwoservicepartsusedintheaerospaceindustry. FromFigure1.1,twothingsbecomeapparent:(i)theannualdemandhistorycontains onlyfivepositivedemandobservationsand(ii)variabilityreferstoboththedemand arrivals(howoftendemandarrives)andthesizeofthedemand,whendemandoccurs.The

Figure1.1 Intermittentandlumpydemand.Source:BoylanandSyntetos(2008).©2008,Springer Nature.

lackofinformationassociatedwithintermittentdemandpatternscoupledwiththisdual sourceofvariabilitycallsforsimplifyingassumptionswhenmodellingthesepatterns.A commonsimplifyingassumptionisthatthedemandisnon-seasonal.Suchsimplifications mayimpedethedevelopmentofsolutionsthatareoptimalinastatisticalsense,but doallowforthedevelopmentofmethodsthatpotentiallyareveryrobustandeasyto implement.Robustnessisdefinedhereasa‘sufficientlygood’performanceacrossawide rangeofpossibleconditions.Optimalityisdefined,forparticularconditions,asthe‘best’ performance.

Weshallreturntorobustnessandstatisticaloptimalityinlaterchaptersbut,forthetime being,itissufficienttosaythatrobustnessisessentialinpracticalapplications.Whileoptimalityisdesirable,itshouldnotbeattheexpenseofrobustness.Manyofthemethods tobediscussedinthisbookhavebeenfoundtoberobustbysuchsoftwarecompaniesas BlueYonder,LLamasoft,Slimstock,andSyncronInternational,helpingtheircustomersto dramaticallyreduceinventorycosts.

1.4.2BusinessContext

Withrobustnessinmind,thisbookpresentsarangeofapproachestointermittentdemand forecastingthatareapplicableinanyindustrialmaketostock(MTS)setting.Inadditionto anMTSsetting,unlessotherwisespecified,wefocusonsinglestockkeepingunit(SKU), singlestockinglocationenvironments,asexplainedbelow.

Maketostock.InanMTSenvironment,customersarewillingtowaitnomorethanthe timeittakestodelivertheparticularitemtothemandsotheitemneedstobeavailablein stock,readytobedispatched,or,inthecaseofretailing,itneedstobeavailableontheshelf. Inthiscase,demandisnotknownandneedstobepredicted.Thealternativeenvironment isknownasmaketoorder(MTO),wheretheproductsarenotassumedtobeinstock,and thecustomermustwaituntilthemanufacturerassemblestheproductforthem.Inthiscase,

customerdemandisknownanddoesnotneedtobepredicted.Thissituationiscommonfor someproducts(e.g.furniture)butnotforothers(e.g.automotiveoraerospacespareparts). Thereisalsoamoveto3Dprintingofproductsinsomeindustries,whichisaformofMTO butwithshorterdelays(TechnicalNote1.1).

Singlestockkeepingunit(SKU)approaches.Wearelookingatforecastingtherequirements(andmanagingtheinventories)ofsingleSKUs.Althoughsomeofthemethodsto bediscussedinthisbookrelyuponcollectiveconsiderations(acrossagroupofSKUs),the restofthematerialconsiderssingleSKUproblems.Thisisbecausehigherlevelsofaggregationare,typically,notassociatedwithintermittentdemand.Consider,forexample,10 intermittentdemanditems,allofwhicharereplenishedfromthesamesupplier.Itmakes sensetoconsidertheaggregatedemandofthoseitemstofacilitateefficienttransportation arrangements.However,althoughdemandattheindividualSKUlevelmaybeintermittent, aggregatedemand(acrossall10SKUs),mostprobably,willnotbeintermittent.

Singlestockinglocationapproaches.Wefocusondetermininginventoryreplenishment requirementsateachsinglelocation,withouttakingintoaccountinteractionsbetween locations.Assuch,wedonotconsiderthepossibilityofsatisfyingdemandbylateraltransshipmentsofstocksbetweenstores.Thisisbecausethesedecisionsrelateexplicitlytojoint inventory-transportationoptimisation,whichisbeyondthescopeofthisbook.Further,and asdiscussedabove,aggregatedemand(acrossdifferentlocationsinthiscase)istypicallynot associatedwithintermittence.

Weshouldalsomentionthat,althoughtheterm‘demand’isbeingusedinthisbookwhen referringtoforecasting,demandwillnotalwaysbeknownand,inthiscase,actualsales mustbeusedasaproxy.Theterms‘demand’and‘sales’areusedinterchangeablyinthis bookalthough,strictlyspeaking,thelatterisoftenusedasanapproximationfortheformer.

1.4.3StructureoftheBook

Thisbookstartsbycontextualisingintermittentdemandforecastinginthewiderscholarshipandpracticeofinventorymanagement.WebegininChapter2withadiscussionof inventorymanagementandsomeofitsimplicationsforforecasting.Then,inChapter3,we examinetheservicedriversofinventoryperformance.ThefocusshiftsinChapters4and 5tothecharacterisationofintermittentdemandpatternsbydemanddistributions.This formsanaturalfoundationforthenexttwochapters,whichfocusonforecastingmethods. Chapter8takesusbacktoinventoryreplenishmentandthelinkagebetweenforecasting andinventorycontrol.Inthenextchapter,wemoveontothemeasurementofforecasting accuracyandinventoryperformance.Forecastingaccuracyassessmentisanotoriouslydifficultproblemforintermittentseries,andthechapterhighlightsthetrapsfortheunwary andgivessomepointerstogoodpractice.

Althoughthemainemphasisofthisbookisonforecasting,classificationmethodsare alsoimportantinpracticalapplications.InChapter10,welaysomeofthegroundworkfor classificationmethods,discussedinChapter11,whichhavebeendesignedspecificallyto addressintermittence.Inthenextchapter,weturnourattentiontoobsolescenceandforecastingmethodsthatareparticularlysuitedtothisstageofthelifecycle.Chapter13presents analternativeperspectiveondemandforecasting,concentratingonmethodsthatdonot assumeanyparticularformofdemanddistribution.Bycontrast,Chapter14delvesmore

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.