Withacknowledgementofthesource,reproductionofallorpartofthepublicationisauthorized,exceptforcommercial
purposes.
Legaldeposit‐D/2010/7433/4
Responsiblepublisher‐HenriBogaert
Federal Planning Bureau
Kunstlaan/Avenue des Arts 47-49, 1000 Brussels
WORKING PAPER 2-10
The PLANET model
Methodological Report:
The Car Stock Module
February2010
IngeMayeres,MaudNautet,AlexVanSteenbergen,
Abstract‐The vehicle stock module calculates the size and composition of the car stock. Its
outputisafulldescriptionofthecarstockineveryyear,by vehicle type,ageand(emission)
technology of the vehicle. The vehicle stock is
represented in the detail needed to compute
transportemissions.Theintegrationofthecarstockmodulein
PLANETwillallowtobettercap‐
turetheimpactofchangesinfixedandvariabletaxesleviedoncars.Amongtheseimpacts,the
effectontheenvironmentisofparticularinterest.
JelClassification‐R41,R48
Keywords‐Passengerroadtransport,vehiclestockmodelling
Lestravauxprésentésdanscedocumentontétéréalisés
danslecadred’unecollaborationavecleSPFMo‐
bilitéetTransports.
Hetwerkinditrapportmaaktdeeluitvaneensamenwerkingmetde
FODMobiliteitenVervoer.
WORKING PAPER 2-10
Contents
Introduction 1
1. Modelling approach 2
2. The total desired stock 3
3. Vehicle scrappage 4
3.1. Methodology 4
3.2. Observed scrappage rates 4
3.3. Estimation results 6
4. The composition of car sales 8
4.1. The nested logit model for car sales 8
4.1.1. Level 3 9
4.1.2. Level 2 10
4.1.3. Level 1 11
4.1.4. Scale parameters 11
4.2. The calibration of the nested logit model for car sales 12
4.2.1. Data 12
4.2.2. Methodology 13
4.2.3. Calibrated elasticities 14
5. Output of the car stock module 16
6. Links of the car stock module with the other modules 17
7. References 18
WORKING PAPER 2-10
List of tables
Table 1: Estimated parameters of the loglogistic hazard function (t-statistic between brackets) 6
Table 2: The reference equilibrium 12
Table 3: Target elasticity values of conditional annual mileage with respect to monetary income
and variable costs 13
Table 4: Calibrated elasticity values for average annual mileage of newly purchased cars 15
Table 5: Calibrated elasticity values for car sale probabilities 15
Table 6: The impacts of doubling the fixed or variable costs of different car sizes 15
Table 7: The impacts of doubling the fixed or variable costs of gasoline and diesel cars 15
Table 8: Input in the car stock module of year t from the other PLANET modules 17
Table 9: Output of the car stock module of year t to the other PLANET modules 17
List of figures
Figure 1: Average scrappage rates at age 0 to 30 during the period 2000 to 2005 for diesel
and gasoline cars 5
Figure 2: Observed and estimated scrappage rates for diesel cars between 0 and 20 years old 6
Figure 3: Observed and estimated scrappage rates for gasoline cars between 0 and 20 years old 7
Figure 4: Decision structure for car purchases 9
WORKING PAPER 2-10
1
Introduction
Thecarstock module calculatesthe sizeandcomposition of thecarstock.Itsoutputisafullde‐
scriptionofthecarstockineveryyear,byvehicletype(fuel),ageand(emission)technologyof
the vehicle. The vehicle stock is represented in the detail needed to compute the
transport
emissions.
Forbuses, coaches, roadfreightvehicles, inlandnavigation and railthe carstock is not mod‐
elledindetail.Inthesecasesthemodelusesinformationaboutthevkmandtkmratherthanthe
vehiclestocktodetermineresourcecosts,environmentalcosts,etc.
Thepastversionofthe
PLANETmodelusedanexogenousevolutionofthecarstocktakenfrom
otherresearchprojects.Fromnowonthevehiclestockmoduleisintegratedintherestofthe
PLANETmodel.
Theassumptionsthataremadearedescribedinadetailedwayinthereportonthebusiness‐as‐
usualscenario
1
.Inthispaperwedescribetheworkthathasbeendonetoendogenisetheevolu‐
tionofthevehiclestock.
Thisdocumentdescribesthefirstversionofthecarstockmodule.Themethodologypresented
heremightundergosomechangesinthefuture
2
.
1
Desmet,R.,B.Hertveldt,I.Mayeres,P.MistiaenandS.Sissoko(2008),ThePLANETModel:MethodologicalReport,
PLANET1.0,Studyfinancedbythefr ameworkconvention “Activitiestosupportthefederalpolicyonmobilityand
transport,2004‐2007”betweenthe
FPSMobilityandTransportandtheFederalPlanningBureau,WorkingPaper10‐
08,FederalPlanningBureau,Brussels.
2
Forexample,intheactualversionofthemodel,thedefinitionofcarsiz eislinkedtocylindersize.Inthefuture,we
willlookatthepossibilitytodefinecarsizelinkedtopower.
WORKING PAPER 2-10
2
1. Modelling approach
Severalapproachesexisttomodelthemagnitudeandcompositionofthecarstock.DeJonget
al.(2002)giveareviewoftherecent(since1995)internationalliteratureoncarownershipmod‐
elling.In
PLANETwewilluseanaggregateapproach.Otherexamplesofthisapproach canbe
foundin
TREMOVE(DeCeusteretal.,2007)andASTRA(Rothengatteretal.,2000).
Wefirstdescribethegeneralprinciples,andthendiscussthedifferentstepsinmoredetail.The
generalapproachissimilarasin
ASTRAandTREMOVE.Foreachcartypethevehiclestockisde‐
scribedby vintage and vehicletype. IfStock
i(t,T)represents the vehicle stock oftype i (diesel
andgasolinecar)inyeartandofageT,thetwobasicequationsare:
Stock
i(t,0)=Salesi(t)
Stock
i(t,T)=Stocki(t‐1,T‐1)–Scrapi(t,T)forT>0
Sales
i(t)standsforthesalesofnewcarsoftypeiinyeartandScrapi(t,T)isthescrappageof
vehiclesoftypeiandageTinyeart.
Ineachyeartthestockofvehiclessurvivingfromyeart‐1iscomparedwiththedesiredstockof
vehiclesneededby the transport users.Ifthedesiredstockis largerthan
thesurvivingstock,
newvehiclesarebought.Thisapproachrequiresthedeterminationineachyearofthetotalde‐
siredvehiclestock(Section2),thenumberofvehiclesofeachtypethatisscrapped(Section3)
andthecompositionofthevehiclesales(Section4).
Themodel includesvehicles from age
0 until theagetheyare scrappedor leavethe country.
Anychangesinownershipinbetweenarenotmodelled.Noseparatecategoriesareconsidered
fornewandsecondhandvehicles.
Inafirststagenodistinctionismadebetweencarsownedbyprivatebusiness,governmentand
utilitiesonthe
onehandandpersonalcarsontheotherhand.Thisdistinctioncouldbeuseful
becausethepolicyinstrumentscanbedifferentinbothcasesandbecausechangesinthecom‐
positionofthefleetstockeventuallyfilterdowntothepersonalcarstock.Includingaseparate
categoryof fleetcars would
require modellingthe transitionof thesecars tothe personalcar
stock. Account shouldalso be taken of exports and imports.The National Energy Modelling
System(
NEMS)ofthe US DepartmentofEnergy (US DoE, 2001)isan example ofa model that
incorporatesthedistinctionbetweenfleetandpersonalcars.
WORKING PAPER 2-10
3
2. The total desired stock
Inordertoderivethetotaldesiredstockwecanconsiderthefollowingtwoapproaches:
– toderivethedesiredstockfromthevkm,ascalculatedinthe
MODALandTIMECHOICEmod‐
ule,andtheevolutionoftheannualmileagepervehicle.Thisistheapproachthatistakenin
the
TREMOVEmodel.
– torelatethedesiredcarstocktoeconomicdevelopment,transportcostsandpopulation.The
functionrelatingthedesiredstocktoitsexplanatoryvariablesmayeitherbecalibrated(cf.
the
ASTRA model; Rothengatter et al., 2000) or estimated (cf. for example, Medlock and
Soligo,2002).Fortheothervehiclesthesameapproachasin
TREMOVEcontinuestobeused.
Thefirstapproachhasthedrawbackthatassumptionsneedtobemadeabouttheaveragean ‐
nualvehiclemileage.Thesecondapproachallowstoderiveforcarsanaverageannualmileage
byconfrontingthecarstockwiththe cartransportdemand thatisderivedinthe
MODALand
TIMECHOICEmodule.
In the firstversionof
PLANET thefirstapproach wasused.In thenewversion ofPLANET, the
secondapproachisused.Withthefirstapproachwestartfromthetotalvkmpercarthatisde‐
rivedinthe
MODALandTIMECHOICEmodule.Thenumberofvkmisthendividedbytheaver‐
ageannualmileagetogetthedesirednumberofcarsforagivenyear.Thedeterminationofthe
averageannualmileageforcarswillbediscussedinSection5.
WORKING PAPER 2-10
4
3. Vehicle scrappage
Inordertoknowthesurvivingcarstockinyeartascrappagefunctionneedstobedetermined.
Inthisversionofthemodelscrappageisassumedtobeexogenous.Inalaterstageanendoge‐
nousscrappagefunctionwillbeconsidered
3
.
3.1. Methodology
Thescrappagefunctionisestimatedforthefollowingcartypes:dieselcarsandgasolinecars.
Thescrappagerateofthesevehiclesisestimatedaccordingtotheageofthevehicle(T),witha
scrappagefunctiondeterminedby a loglogisticdistribution.Thefollowingequationgivesthe
hazard function of the loglogistic
distribution which describes the rate at which cars are
scrappedatageTgiventhattheystayinthevehiclestockuntilthisage.
()
ρ
λ
ρ
λλρ
)(1
1
)(
T
T
consTh
+
−
+=
whereλandρareshapeandscaleparametersandconsisaconstantterm.Ifthevalueofthe
shapeparameters(λ)liesbetween0and1,theshapeofthehazardfunctionfirstincreasesand
thendecreases withage.Theloglogistic hazardfunctionis alsoconcave at first,
andthenbe‐
comesconvex.Theshapeofthishazardfunctionisclosetotheshapeofthescrappageratesfor
allvehicletypesobservedduringtheyears2000to2005
4
.Theparametersλandρandthecon‐
stanttermareestimatedonthebasisofdataobtainedfromthe
DIV.Thesearedescribedinthe
followingparagraph.
3.2. Observed scrappage rates
The DIV hasprovided uswith timeseries ofthe age distribution ofthe carfleetaccording to
fuel.Thetimeseriesrefertotheyears1997to2005(except1999).Thesedataareusedtocalcu‐
latescrappageratesaccordingtofuelandageforallreportedyears.Theobservednumber
of
scrappedvehiclesofageTisdefinedasthedifferencebetweenthenumberofvehiclesofageT
inyeartandthenumberofvehiclesofageT+1inyeart+1.Thescrappagerateisthenobtained
bydividingthenumberofscrappedvehiclesperageinyear
tbythetotalnumberofvehicleof
thisageinthefleetduringthesameyear.
3
Ingeneral,scrappingdependsonthetechnicallifetimeofavehicle,theprobabilityofbreakdownbeforetheendof
the planned technical life and policies that directly or indirectly affect vehicle costs such as purchase taxes and
scrapping incentives. The following studies could prove to be useful for modelling endogenous
scrappage rates:
HamiltonandMacauley(1998),DeJongetal.(2001),Loggheetal.(2006).
4
AWeibulldistributionisoftenusedtomodeldurationdata,buttheshapeofitshazardfunction‐“s‐shape”‐does
notcorrespondwelltotheshapeoftheobservedscrappagerates.
WORKING PAPER 2-10
5
Thenextfigurepresents the average scrappageratesderivedfromthedataofthe DIV for the
differenttypesofcarsfrom1to30yearsold.Theaveragesarecalculatedovertheperiod2000‐
2005.
Figure 1: Average scrappage rates at age 0 to 30 during the period 2000 to 2005 for diesel and
gasoline cars
-5%
0%
5%
10%
15%
20%
25%
30%
0 5 10 15 20 25 30
Diesel cars Gasoline cars
Source:
FPB based on DIV.
Thedataforgasolineanddieselcarsreferto“ordinarypassengercars”and“mixedcars”.Based
onthedataofthe
DIV,wenotesomefindings:
– Thecardatapresentsomeirregularitiesduringthefirstyearofregistration
5
.
– Thedatashowthatthescrappageratesarerelativelyhighduringthe4firstyearsofregistra‐
tion,inparticularfordieselcars.Thiscanbeexplainedbyleasedandcompanycarsleaving
thestockbeforebeing4yearsold.
– Weobservethatthescrappageratesarehigher
fordieselthanforgasolinecarsas,atagiven
age,themileageofdieselcarsishigher.
– Carsof25yearsandolderhavenegativescrappageratesbecause“old‐timers”arereentering
thestock(astaxesandinsurancecostsbecomecheaper).Manyofthosearegasolinecars.
–
Duringtheperiod1997‐2005,themarketshareofgasolinecarshas fallenfrom60%to50%.
Furthermore,thedieselstockisyoungerthanthegasolinestock.So,thereisaphenomenon
of“dieselisation”ofthecarstock.
– For the period 1997‐2005, 97% of the car stock was between
0 and 30years old, 96% was
youngerthan20years.
5
Somecardealers realize“fictive registrations” inordertoincreasetheir salesfigures. Vehiclesareregistered and
retiredofthestockafterlessthanamonth.So,registrationsfornewcarsareoverestimated.
WORKING PAPER 2-10
6
3.3. Estimation results
Basedontheobservedscrappageratespresentedabove,theconstantandtheparametersλand
ρoftheloglogistichazardfunctionwereestimatedbymeansofanonlinearleastsquaresesti‐
matorin
TSP.Theestimationonlytakesintoaccountvehiclesof20yearsandyounger.Thisis
donebecausethestockafterthisagebecomeslessrepresentativeasthenumberofoldvehicles
becomessmallerandsmaller
6
.Table1presentstheestimatedvaluesoftheparametersλ,ρand
consandthecorrespondingt‐statistic.ItalsogivestheR‐squaredoftheestimatedmodels.
Table 1: Estimated parameters of the loglogistic hazard function (t-statistic between brackets)
Diesel cars Gasoline cars
λ 0,075 0,076
(68,28) (53,58)
ρ 4,816 4,734
(56,23) (44,97)
Cons 0,051 0,020
(14,48) (4,63)
R
2
0,990 0,983
Figure2and3presenttheobservedandestimatedscrappageratesforthe2vehicletypes.
Figure 2: Observed and estimated scrappage rates for diesel cars between 0 and 20 years old
0%
5%
10%
15%
20%
25%
30%
0 5 10 15 20
Diesel cars : observed rates Diesel cars : estimated rates
Source:
FPB.
6
Intheperiod2000to2005,96%ofthecarstockwasbetween0to20yearsold.
WORKING PAPER 2-10
7
Figure 3: Observed and estimated scrappage rates for gasoline cars between 0 and 20 years old
0%
5%
10%
15%
20%
25%
30%
0 5 10 15 20
Gasoline cars : observed rates Gasoline cars : estimated rates
age of vehicle
Source:
FPB.
Thecomparisonoftheobservedandestimatedscrappageratesshowsthattheestimatedscrap‐
pageratesareabletoreflectratherwellthespecificitiesofthecarfleetevolution.Nevertheless,
forthe4firstyearsofregistration,theestimatedscrappageratecannotreproducethefluctua‐
tionsoftheobservedscrappagerate.
WORKING PAPER 2-10
8
4. The composition of car sales
Inthissectionwedescribethewayinwhichthetechnologychoicefornewvehiclesismodelled.
Wemodelthechoicebetweenthreecarsizes(small,mediumandbig)
7
andbetweendifferent
technologies(diesel,gasoline, hybrid diesel, hybrid gasoline,
LPGandCNG). TheEUROtypeof
thecarsisassumedtobedeterminedbytheyearinwhichitisbought.Thecarchoiceismod‐
elledbymeansofanestedlogitmodel
8
.
4.1. The nested logit model for car sales
The decision structure for determiningthe share of thedifferent car typesincar salesis pre‐
sentedinFigure4.Simultaneouslywiththechoiceofthecartype, the model alsodetermines
theannualmileageofthenewcars.InFigure4Level1describesthechoicebetweensmalland
mediumcarsontheonehandandbigcarsontheotherhand.Conditionalonthischoice,the
categoryofsmallandmediumcarsissplitintosmallcarsandmediumcars(Level2).Finally,
giventhedecisiononthecarsize,thechoicebetweendieselandgasolinecarsis
determinedat
Level3.Finally,thenumberofhybridandconventionaldieselcarsisdeterminedbyapplying
exogenous shares of these two subtypes in total diesel car sales. Similarly, total gasoline car
salesare split intoconventionalandhybridgasoline cars,
CNG cars andLPGcarsby applying
exogenoussharesforthesefoursubtypes.
Inmoreformaltermswearedealingwithamultidimensionalchoiceset:
C
1 size(small,medium,big) indexedbys
C
2 fueltype(gasoline,diesel) indexedbyf
andwetakeintoaccountthatelementsofthechoicesetC
1shareunobservedattributes.There‐
fore,wemodelanadditionallevel,wherethechoiceismadeamongdifferentcompositesizes
(small+medium,big),indexedbycs.
7
Thecarsizesaredefinedasfollows:0‐1400cc=small,1401‐2000cc=medium,>2000ccbig.
8
Formoreinformationaboutnestedlogitmodelsseee.g.,KoppelmanandWen(1998),Heiss(2002)andHensherand
Greene(2002).
WORKING PAPER 2-10
9
Figure 4: Decision structure for car purchases
New cars
Small
Small + Medium
Medium
DieselGasoline
Big
Big
DieselGasoline DieselGasoline
Level 1
Level 2
Level 3
Hybrid
Conventional
CNG LPG
Hybrid
Conventional
Exogenous
shares
4.1.1. Level 3
Level3describes,conditionalonthepurchaseofacarofagivensizesandcompositesizecs,
thechoiceofthefueltype.Consistentwiththediscretechoiceliterature,indirectutilityv(f|cs,s)
ofselectingalternativefgivensizesandcompositesizecsiswritten
as
()()()
scsfscsfVscsfv ,,,
η
+=
whereV(f|cs,s)isthedeterministic‘universal’indirectutilityfunction,assumedtobethesame
for everyone, and η(f|cs,s) is an individual‐specific component that reflects idiosyncratic taste
differences.BasedondeJong(1990),thefollowingfunctionalformisusedforthedeterministic
componentofindirect
utility,conditionalonthepurchaseofacaroftypef(givensandcs):
()
()
()()()()
()
()()()
()
()
scsf
ref
scsf
scsfFscsfKY
scsf
scsfMVCscsfscsf
scsf
scsfV ,
,1
,,
,1
1
,,,exp
,
1
,
ξ
α
α
βδ
β
⎥
⎥
⎥
⎥
⎦
⎤
⎢
⎢
⎢
⎢
⎣
⎡
−
−−
−
+
−
=
where MVC(f|cs,s) is themonetary variable costof travel forhouseholds buying a car of fuel
typef(givensandcs)andYrepresentsmonetaryincome.K(f|cs,s)andF(f|cs,s)aretheannual
fixedresourcecostandtheannualfixedtaxfor
acaroftypef(givensandcs).Finally,
α
(f|cs,s),
β
(f|cs,s)and
δ
(f|cs,s)
areparameters.Notethatweexpressindirectutilityinmonetarytermsby
dividingbyξ
ref(f|cs,s),themarginalutilityofincomeinthereferenceequilibrium.
WORKING PAPER 2-10
10
Theconditionalannualmileagetravelledbyanewlyboughtcarcanbeobtainedbyapplying
Roy’sidentitytotheconditionalindirectutilities:
() ()()()
[]
()()()
(
)
scsf
scsfFscsfKYscsfMVCscsfscsfscsfX
,
,,,,,exp,
α
βδ
−−−=
Themodelthereforeallowsnotonlytodeterminethetypeofvehiclethatisbought,butalsothe
annualmileagedrivenbynewlypurchasedvehicles.Fromthepreviousequationitcaneasily
bederivedthattheelasticityoftheconditionalannualmileagew.r.t.themonetaryvariablecost
equals‐β(f|
cs,s)GVC(f|cs,s).Inaddition,α(f|cs,s)equalstheelasticityofconditionalannualmile‐
agew.r.t.monetarydisposableincome.
Assuming thatpeople selectthe fuel type that yieldshighest utility, and that the individual‐
specificcomponentsη(f|cs,s) aredistributed Gumbel i.i.d., theprobabilityofchoosing
acar of
typef(conditionalonsizesandcompositesizecs)isthengivenbythelogitexpression:
()
(
)
(
)
[
]
()()
[]
(
)
(
)
[
]
()
[]
cssIV
scsfVcss
cssf
scsfVcss
scsfVcss
scsfP
exp
,exp
,'
,'exp
,exp
,
μ
μ
μ
==
∑
μ(s|cs)isascaleparameterandisinverselyrelatedtothestandarddeviation.Ahigherμindi‐
cateslessindependenceamongtheunobservedportionsofutilityforalternativesinthesubnest
(s|cs).IV(s|cs)istheinclusivevalue:
() ()()
[]
⎟
⎟
⎟
⎠
⎞
⎜
⎜
⎜
⎝
⎛
=
∑
f
scsfVcsscssIV ,expln
μ
Itlinksthelowerandmiddlelevelofthemodelbybringinginformationfromthelowermodel
tothemiddlemodel.Itismoreorlesstheexpectedextrautilityfromsbybeingabletochoose
thebestalternativeins|cs.
4.1.2. Level 2
Level2describesthechoicebetweensmallandmediumcars,conditionaluponthechoiceofa
carbelongingtothecompositesize“small+medium”.Theconditionalprobabilityofchoosings,
givencs,isgivenby:
()
(
)
()
()
()
()
()
(
)
()
()
()
[]
csIV
cssIV
cssµ
cs
css
cssIV
cssµ
cs
cssIV
cssµ
cs
cssP
exp
exp
'
'
exp
exp
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
=
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
=
∑
λ
λ
λ
WORKING PAPER 2-10
11
λ
(cs)isascaleparameter.IV(cs)istheinclusivevalueofthesubsetofalternativesincs:
()
()
()
()
⎟
⎟
⎟
⎠
⎞
⎜
⎜
⎜
⎝
⎛
⎥
⎥
⎦
⎤
⎢
⎢
⎣
⎡
=
∑
css
cssIV
cssµ
cs
csIV
λ
expln
For big cars, this decisionlevel is irrelevant, since the composite size ‘big’is associated with
onlyonesize(big).ThereforeP(s=big|cs=big)=1.
4.1.3. Level 1
Finally,atlevel1,themarginalprobabilityofchoosingcompositesizecsisgivenby:
()
()
()
()
()
IV
csIV
cs
cs
csIV
cs
csIV
cs
csP
exp
)(
exp
'
'
)'(
exp
)(
exp
⎥
⎦
⎤
⎢
⎣
⎡
=
⎥
⎦
⎤
⎢
⎣
⎡
⎥
⎦
⎤
⎢
⎣
⎡
=
∑
λ
ρ
λ
ρ
λ
ρ
Theoverallinclusivevalue(IV)canbewrittenas:
()
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
⎥
⎦
⎤
⎢
⎣
⎡
=
∑
cs
csIV
cs
IV
)(
expln
λ
ρ
Veryoftenρisnormalisedtoequalunity.
Theprobabilitythatonebuysacarofcompositesizecs,sizesandfueltypefisgivenby:
()
()()
()
csPcssPscsfPfscsP ,,, =
4.1.4. Scale parameters
Forthemodeltobeconsistentwithutilitymaximisationthefollowingconditionsmustbesatis‐
fiedforthescaleparameters:
()
()
()
(
)
csscscsscscsscs ,0,,and ∀>≤≤
μλρμλρ
Thismeansthatthevarianceoftherandomutilitiesatthelowestlevelsho uldnotexceedthe
varianceatthemiddlelevel,whichshouldnotexceedthevarianceatthetoplevel.Thecondi‐
tionsimplythat:
()
()
()
cs
cs
andcsscs
css
cs
∀≤<∀≤< 10,10
λ
ρ
μ
λ
With
ρ
=
λ
(cs)=
μ
(s|cs)thenestedlogitmodelreducestoajointlogitmodel( MNLmodelforthe
jointchoiceofvehiclesizeandfueltype).
WORKING PAPER 2-10
12
4.2. The calibration of the nested logit model for car sales
4.2.1. Data
Inordertoconstructareferenceequilibriumonwhichtocalibratethemodel,wecollecteddata
oncarsales,annualmileageofnewcars,variableandfixedcostsandmonetaryincomeforthe
year 2005. Table 2 gives an overview of the different cost components, monetary income
(
GDP/capita),annualmileageandsharesofthedifferentvehicletypesintotalcarsales.
Table 2: The reference equilibrium
Reference equilibrium
Small Medium Big
Fixed taxes (€/car/year)
(1)
Gasoline 448 645 1364
Diesel 447 769 1381
Variable taxes (€/100vkm)
(2)
Gasoline 6.5 8.4 10.7
Diesel 3.4 4.0 5.2
Fixed monetary costs excl. taxes Gasoline 1163 1924 4109
(€/car/year)
(3)
Diesel 1323 1955 3476
Variable monetary costs excl. taxes
(€/100vkm)
(4)
Gasoline 8.3 11.7 17.7
Diesel 6.9 7.4 9.7
GDP/capita (€/person/year) 26085 26085 26085
A
nnual mileage (km/year) Gasoline 12393 13747 17432
Diesel 14808 22731 30588
Sale probabilities
(5)
Gasoline 19.97% 8.51% 0.78%
Diesel 27.33% 39.85% 3.56%
(1)
Includes registration tax, traffic tax, radio tax and indirect taxation on purchase, insurance and control.
(2)
Includes indirect taxation on maintenance and fuel (plus the fuel excise).
(3)
Includes purchase, control and insurance costs net of taxes.
(4)
Includes fuel and maintenance costs net of taxes.
(5)
Observed shares of the different vehicle types in total car sales in base year.
Sources:
BFP, CBFA, DIV, IEA, INS, FPS Economics, SPF Mobility and Transport, VITO.
Thedataforthereferenceequilibriumshowmonetarycostsrisingwithsize. Thevariablecosts
ofdieselcarsarelowerthanthoseofgasolinecars.Thefixedcostsofdieselcarsarehigherthan
forgasolinecars,exceptforthebiggestcars.Inthecaseofbigcarsthisisbecause
theaverage
sizeofbiggasolinecarsislargerthanthatofbigdieselcars.Monetarycostscannotfullyexplain
theobservedbehaviour.Aswewillsee,somecharacteristicsorhiddentastedifferencescannot
beaccountedforbyusingcostdataalone.Aconstanttermisthereforeintroducedinthe
calibra‐
tiontotheequationforindirectutility(moredetailinthenextsection).
WORKING PAPER 2-10
13
4.2.2. Methodology
Calibrationofthemodelrequiresfurtherinformationonthevalueoftheincomeelasticityand
theelasticityw.r.t.variablecostsofannualmileage.Ourincomeandcostelasticitiesarebased
onDeJong(1990).TheincomeandcostelasticitiesofDeJong(1990)areadjustedtoaccountfor
differentiationby car
size. This permits toobtainreasonableelasticitiesand greatlyimproves
theresults.
Table 3: Target elasticity values of conditional annual mileage with respect to monetary income
and variable costs
Elasticity w.r.t. monetary income
Small car 0.22
Medium car 0.23
Big car 0.39
Elasticity w.r.t. monetary variable costs
Small car -0.14
Medium car -0.22
Big car -0.45
Given these target elasticities and information for the base year, the parameters α(f|cs,s),
β(f|cs,s)andδ(f|cs,s)canbeeasilyobtained:
α(f|cs,s)=Targetmonetaryincomeelasticityofannualmileage
β(f|cs,s)=Targetelasticity
ofannualmileagew.r.t.monetarycost/MVCref(f|cs,s)
δ(f|cs,s)=ln[X
ref(f|cs,s)]+β(f|cs,s)*MVCref(f|cs,s)‐α(f|cs,s)*ln[Yref–Kref(f|cs,s)–Fref(f|cs,s)]
ξ(f|cs,s)=α(f|cs,s)*[Y
ref–Kref(f|cs,s)–Fref(f|cs,s)]
Giventhesevalues,indirectutilitiesV(f|cs,s)canbecalculated.Wenotethatinthefinalcalibra‐
tion,weaddedaconstant(chosentoobtainreasonableelasticityvalues)totheequationforin‐
directutility.Thisstepwasnecessary,sincewefoundthatsome
featuresofthedatainthebase
yearcouldnotbeproperlyexplainedwithinthediscretechoiceframeworkbythecostandin‐
comedataalone.Tocapturetheexistenceofunobservablecharacteristics,weaddedaconstant
totheindirectutilityfunction.Thevaluesoftheseconstantshavebeentakenas
smallaspossi‐
ble.
Giventheseparameters,thescalingparameterscanbedetermined.Forexample,givenindirect
utilities,thelowernestscalingparametersμ(s|cs)canbecalculatedasfollows:
1/μ(s|cs)=[V(“dies”|cs,s)–V(“gas”|cs,s)]/ln[P(“dies”|cs,s
)/P(“gas”|cs,s)]
Thescalingparametersfortheotherlevelscanbeobtainedinasimilarway.
WORKING PAPER 2-10
14
4.2.3. Calibrated elasticities
Table4andTable5givethecalibratedelasticitiesofconditionalannualmileageofnewlypur‐
chasedcarsandoftheprobabilitiesofbuyingthedifferentcartypes,inbothcasesw.r.t.mone‐
taryvariablecosts,fixedcostsandmoneyincome.
Theelasticityvaluesofannualmileagew.r.t.tomonetarycosts
andincomeinTable4largely
reflect the elasticities chosen in the calibration procedure. Elasticities of mileage w.r.t. fixed
costs aresignificantly smaller, ascan be expected sincethese changes onlyinfluence mileage
indirectlythroughchangesinmoneyincome.
TheelasticityvaluesinTable5reflecttoalargeextent
thestructureofthenestedutilityfunc‐
tion.Bychoosingthisfunctionalform,oneimposescertainrestrictionsonthebehaviouralreac‐
tions thatare allowed bythe model.Forexample, theclose–to–zero incomeelasticities ofex‐
penditure shares are a direct result of the homogeneity of the utility function. Similarly, the
nestingstructurewillensurethatreactionsofdifferentcategoriesinonenestoftheutilityfunc‐
tiontochangesinanothernestwillbeequal.Forexample,thecross‐priceelasticitiesofmedium
gasolineanddieselcarsw.r.t.tochangesinthepriceofsmallcarsareequal.Reactionsof
me‐
diumandsmallcarsalestochangesinthepriceofbigcarsareequal,owingtotheseparationof
bigcarsontheonehandandsmallandmediumcarsontheotherhandintheuppernestofthe
utilityfunction.
Theownpriceelasticityislesspronounced
fordieselcarsthanforgasolinecars,whilecross–
priceelasticitiesaremarkedlylarger.Smallcarsareingenerallesssensitivetochangesinthe
ownpricethanbigcars.
Togetabetterfeelforthebehaviouralimpactofachangeincarcosts,wehaveperformedsome
additionalsimulations
withthe model. Table 6summarisesthe impactsofdoublingthefixed
andgeneralisedvariablecostsofcarsofdifferentsizes.For example,doublingthefixedcostsof
smallcarswouldreducetheirsharefrom47%to36%ofcarsales.Ascanbeexpected,sucha
pricechangewould
primarilybenefitthesaleofmediumcars.
Table 7 presents the impacts of doubling the fixed andgeneralised variablecosts of gasoline
and diesel cars. Doubling the generalised variable costs of gasoline cars would decrease the
shareofgasolinecarsinsalesfrom29%toabout0%.For dieselcarsthe
doublingofthesecosts
wouldbringtheirsharefrom70%to1%.Thedoublingofthefixedcostsofdieselwouldresult
inadieselshareof0%ratherthan70%.Thismeansthatinsidethesamesizecategorydieseland
gasolinecarsarenearperfectsubstitutes.
WORKING PAPER 2-10
15
Table 4: Calibrated elasticity values for average annual mileage of newly purchased cars
Monetary variable costs Fixed costs Mone-
tary
income
Small Medium Big Small Medium Big
Gaso-
line
Diesel Gaso-
line
Diesel Gaso-
line
Diesel Gaso-
line
Diesel Gaso-
line
Diesel Gaso-
line
Diesel
Small cars
Gasoline
Diesel
-0.14
-
-
-0.14
-
-
-
-
-
-
-
-
-0.01
-
-
-0.02
-
-
-
-
-
-
-
-
0.23
0.24
Medium cars
Gasoline
Diesel
-
-
-
-
-0.22
-
-
-0.22
-
-
-
-
-
-
-
-
-0.03
-
-
-0.03
-
-
0.25
0.26
Big cars
Gasoline
Diesel
-
-
-
-
-
-
-
-
-0.45
-
-
-0.45
-
-
-
-
-
-
-
-
-0.10
-
-
-0.09
0.49
0.48
Table 5: Calibrated elasticity values for car sale probabilities
Monetary variable costs Fixed costs Mone-
tary
income
Small Medium Big Small Medium Big
Gaso-
line
Diesel Gaso-
line
Diesel Gaso-
line
Diesel Gaso-
line
Diesel Gaso-
line
Diesel Gaso-
line
Diesel
Small cars
Gasoline -3.00 2.28 0.06 0.28 0.01 0.03 -2.63 2.66 0.06 0.30 0.01 0.03 0.00
Diesel 2.00 -1.88 0.06 0.28 0.01 0.03 1.76 -2.19 0.06 0.30 0.01 0.03 0.00
Medium cars
Gasoline 0.10 0.11 -17.58 18.59 0.01 0.03 0.09 0.13 -16.37 19.55 0.01 0.03 0.00
Diesel 0.10 0.11 3.68 -4.33 0.01 0.03 0.09 0.13 3.43 -4.55 0.01 0.03 0.00
Big cars
Gasoline 0.07 0.08 0.04 0.20 -16.52 16.38 0.06 0.09 0.04 0.21 -18.18 17.55 -0.01
Diesel 0.07 0.08 0.04 0.20 3.44 -4.44 0.06 0.09 0.04 0.21 3.79 -4.75 0.00
Table 6: The impacts of doubling the fixed or variable costs of different car sizes
Share in
reference
equilibrium
Generalised variable costs x 2 Fixed costs x 2
Small Medium Big Small Medium Big
Small cars 47.29% 37.27% 61.72% 48.38% 36.04% 63.76% 48.64%
Medium cars 48.37% 57.81% 33.05% 49.48% 58.97% 30.89% 49.74%
Big cars 4.34% 4.92% 5.24% 2.14% 4.99% 5.36% 1.62%
Table 7: The impacts of doubling the fixed or variable costs of gasoline and diesel cars
Share in
reference
equilibrium
Generalised variable cost x 2 Fixed cost x 2
Gasoline Diesel Gasoline Diesel
Gasoline cars
29.26% 0.30% 98.70% 0.39% 99.52%
Diesel cars
70.74% 99.70% 1.30% 99.61% 0.48%
WORKING PAPER 2-10
16
5. Output of the car stock module
Foreachyearofthesimulation,thevehiclestockmoduleprovidesthecompositionofnewve‐
hiclesalesandcalculatesaveragecostdata.
Asdescribedabove,newvehiclesalesarecalculatedeachyearbycomparingthetotaldesired
vehiclestock (defined astotalvehicle km dividedbyaverageannualmileageof
the previous
year)totheremainingvehiclestockofthepreviousyearafterscrappage.
Salesofnewcarsarethendividedamonggasolineanddieselcarsofdifferentsizesaccordingto
theabovedemandsystem.Afinalstepcalculatestheshareof
LPG,CNGandhybridgasolineand
dieselcarsusingexogenouslydefinedshares.
Forallroadvehicletypesthevehiclestockmoduleprovidesoutputsonthreeclassesofmone‐
tarycosts whichserveasaninputof theModal andTime Choice module ofthenextyear. It
concernsweightedaverages,wherethe
weightsarethesharesofeachfuel,sizeandEurocate‐
goryintotalmileagedriven.
Thecostcategoriesare:
– Taxespaidpervkm(includingalltaxes:indirecttaxes,excisesandfixedtaxes)
– Fuelcostspervkm(Fuelexpenditureincludingexcisesandtaxes)
– Totalmonetarycostspervkm
(Allmonetarycosts–fixedandvariable–includingtaxes)
Inaddition,thevehiclestockmoduledeterminestheannualmileageofthenewlyboughtcars.
Thisiscombinedwiththeannualmileageoftheoldercars,todeterminetheaverageannualcar
mileage.Thisisusedinthenextperiod
todeterminethetotaldesiredcarstock(bydividingthe
numberofcarvkmbytheaverageannualcarmileage).
WORKING PAPER 2-10
17
6. Links of the car stock module with the other modules
Table8andTable9summarisethelinksbetweenthecarstockmoduleandtheotherPLANET
modules.
Table 8: Input in the car stock module of year t from the other PLANET modules
Input from Yea
r
Total vehicle km of cars, LDV and HDV Modal and time choice
t
Generalised income per capita Macro
t
Taxes on the various car types Policy
t
A
verage annual mileage of cars Vehicle stock
t-1
Table 9: Output of the car stock module of year t to the other PLANET modules
Output to Yea
r
A
verage emission factors per road transport mode Welfare
t+1
A
verage monetary costs, fuel costs and taxes per road mode Modal and time choice
t+1
WORKING PAPER 2-10
18
7. References
DeCeuster,G.,B.vanHerbruggen,O.Ivanova,K.Carlier,A.MartinoandD.Fiorello(2007),
TREMOVE,ImprovementoftheDataSetandModelStructure,FinalReport,EuropeanCom‐
mission,
DGEnvironment,Brussels.
deJong,G.(1990),AnIndirectUtilityModeloftheDemandforCars,EuropeanEconomicRe‐
view34,pp.971‐985.
deJong,G.,C.VellayandJ.Fox(2001),VehicleScrappage:LiteratureandaNewStatedPrefer‐
enceTheory,PaperforEuropeanTransportConference2001,
PTRC,Cambridge.
de Jong, G.,J. Fox, M.Pieters,L. Vonkand A.Daly(2002), AuditofCar OwnershipModels,
RANDEurope.
Desmet, R., B. Hertveldt, I. Mayeres, P. Mistiaen and S. Sissoko (2008), The
PLANET Model:
MethodologicalReport,
PLANET1.0,Studyfinancedbytheframeworkconvention“Activities
tosupportthefederalpolicyonmobilityandtransport,2004‐2007”betweenthe
FPSMobility
andTransportandtheFederalPlanningBureau,WorkingPaper10‐08,FederalPlanningBu‐
reau,Brussels.
Hamilton,B.,W.andM.K.Macauley(1998),CompetitionandCarlongevity,Discussionpaper
98‐20,ResourcesfortheFuture.
Heiss,F.(2002),SpecificationsofNestedLogitModels,DiscussionPaper16‐2002,
MEA,Univer‐
sityofMannheim.
Hensher,D.A.andW.H.Greene(2002),SpecificationandEstimationoftheNestedLogitModel,
TransportationResearchB,vol.36,pp.1‐17.
Koppelman, F.S. and C‐H. Wen (1998), Alternative nested logit models: structure, Properties
andEstimation,TransportationResearchB,vol.32,no.5,pp.289
‐298.
Logghe,S.,B.VanHerbruggenandB.VanZeebroeck(2006),EmissionsofRoadTrafficinBel‐
gium,Reportby
T.M.LeuvenonbehalfofFEBIACandFPSMobilityandTransport.
MedlockIII,K.B.andR.Soligo(2002),CarOwnershipandEconomicDevelopmentwithFore‐
caststotheYear2015, JournalofTransportEconomicsandPolicy, vol.36,part 2, pp. 163‐
188.
Rothengatter,W.,W.Schade,A.Martino,M.Roda,A.Davies,L.Devereux,I.Williams,P.
Bry‐
ant,D.McWilliams(2000),
ASTRAMethodology,ASTRADeliverable4,projectfundedbythe
EuropeanCommission,4thresearchprogramme.
USDepartmentofEnergy(2001),TheTransportationSectorModeloftheNationalEnergyMod‐
elingSystem,ModelDocumentationReport,February2001.