Uber动态溢价分析报告PPT文档格式.pptx
《Uber动态溢价分析报告PPT文档格式.pptx》由会员分享,可在线阅读,更多相关《Uber动态溢价分析报告PPT文档格式.pptx(43页珍藏版)》请在冰豆网上搜索。
SelectapickuplocationChoosetypeofcarUberX:
cheapsedanUberSUV:
cheapSUVorvanUberBlack:
fancysedanUberXL:
fancySUVUberPool:
carpoolwithsomerandosEtcRequestandrideawaySimpleandConvenient,ExceptMarketplacesTransparentSupplierssettheirownpricesCustomersmayobserveallproductsandtheirpricesOpaqueSupplyanddemandarehiddenfromcustomersSuppliersdonotchoosetheirownpricesPricesaresetbyanalgorithmGoalsDeterminehowthesurgepricingalgorithmworksDoesitworkthewayUberclaims?
Isitresponsivetochangesinsupplyanddemand?
Cansurgesbepredictedand/oravoided?
ImpactofsurgesondriversandpassengersDataCollectionMeasuringSurgesAvoidingSurgesImpactofSurgesConclusionsDataCollectionUbersofficialsurgepricingpatentsaysthecalculationisbasedonsupply,demand,andotherfactorsHowcanwecollectthisdata?
1.UberAPIPros:
easytouse,includessurgemultipliersandEstimatedWaitTimes(EWT)Cons:
nocars,demand,orsupplyinformation2.UberRiderAppUberAppPingsUbersserversevery5seconds8nearestcarsEstimatedWaitTime(EWT)SurgemultiplierLimitationsMeasuringsupplyisstraightforwardSupply=observedcarsontheroadMeasuringdemandistrickyCarsmaygetbookedOrjustgoofflineOrdriveoutoftheareaWecanonlyestimatedemandFulfilleddemand=numberofcarsthatgoofflineUpperboundontruefulfilleddemandLimitedVisibilityRadiusr3pmonSundayRecall:
theUberapponlyseesthe8closestcarsLimitedVisibilityrr5pmonMondayHowfarapartshouldweplaceourmeasurementspoints?
RadiusMeasurementsCarsobservedbyallapps:
RadiusrRadiusMeasurementsValidationValidatedmethodologyusingground-truthdatafromNYCtaxisBuiltan“Ubersimulator”andusedourmethodstomeasurethetaxisObserved97%ofsupplyand95%ofdemandFinalDataCollectionCollectedfourmonthsofdatafromMidtownManhattanandSanFrancisco2monthsfromeachcity,43measurementpoints2ndand3rdlargestUbermarketsVerydifferentpublictransportoptionsRadiusexperiments247metersinMidtownManhattan387metersindowntownSanFranciscoExampleMeasurementGrid43measurementpointsCollected2monthsofdatainMidtownEthicsWedidnotcollectanypersonalinformationaboutUberdriversorpassengersWeneverbookedanyridesWedidnotinduceanysurgesWeplaced40“users”inrandomlocationswithnosurge,inthemiddleofthenightDidnotobservesurgesforonehourRepeated100timesatdifferentlocationsandhoursDataCollectionMeasuringSurgesAvoidingSurgesImpactofSurgesConclusionsResearchQuestionsHowmuchandhowoftendoesitsurge?
Howlongdosurgeslast?
Howdosurgepricesvarybylocation?
WhatfeaturesdoesUberusetocalculatesurgeprices?
CollectedDataHowmuchdoesitsurge?
14%ofthetimeitissurginginManhattan57%ofthetimeinSFSurgemultiplierstendtobehigherinSFHowlongdosurgeslast?
Noisiness:
70%ofsurgeslast=10minutesStaircase:
surgeslastmultiplesof5minutesHowdoessurgevarybylocation?
MatchNoMatchMatchHowissurgecalculated?
ManypossiblevariablesSupply,demand,EWT,etc.Usecross-correlationtoperformtime-seriesanalysisTime12pm12:
0512:
1012:
1512:
2012:
2512:
30SurgeSupplyHighcorrelationat-10minutesCross-correlation(SupplyDemand)vs.SurgeEWTvs.SurgeModeratelystrongcorrelationswhentimedifferenceiszeroSuggestsUberusesdatafromthelast5minutewindowwhencalculatingsurgeZerocorrelationinothertimewindowsSurgepricingalgorithmisresponsivebutnoisyDataCollectionMeasuringSurgesAvoidingSurgesImpactofSurgesConclusionsCanwepredictsurges?
UsefulvariablesSupply,demand,EWTPrevioussurgemultiplier(s)PredictivemodelsLinearandnon-linearregressionsPerformanceR2rangesfrom0.370.57Theseresultsareterrible,i.e.wecannotreliablypredictsurgesMissingsomekeyvariable(s)Unfulfilleddemand:
howmanypeopletriedtobookaridebutcouldnt?
AvoidingSurgesNoSurge!
EWT=6minNoSurge!
EWT=3min5min7minAvoidingSurges10-15%chanceyoullsavemoneySavingsupto50%byavoidingsurgesDataCollectionMeasuringSurgesAvoidingSurgesImpactofSurgesConclusionsImpactonSupplyandDemandWhydidUberimplementsurgepricing?
ToequalizesupplyanddemandReducedemandbyraisingpricesIncreasesupplybyincentivizingdriversIneconomicterms,surgepricingisaboutincentivesAretheincentivesprovidedbythesurgepricingsystemworking?
StateTransitionsIfoneareaissurging,weexpectthefollowingfivethingstohappen1.(Supply)New:
carsshouldprefertocomeonlineinthesurgingarea2.(Supply)Move-in:
carsshoulddriveintothesurgingarea3.(Supply)Move-out:
fewcarsshoulddriveoutofthesurgingarea4.(Demand)Booked:
fewercarsshouldgetbookedinthesurgingarea5.(Demand)Old:
morecarsthatbeg