Agent-based modeling of electric vehicle diffusion under the phase-out of charging infrastructure subsidies in China

Agen -based modeling o elec ic ehicle di usion unde he phase-ou o
cha ging in as uc u e subsidies in China
Lijing Zhu
a
, Runze Li
a
, Jingzhou Wang
b
, Haibo Chen
c
, Ond ej Ha an
c
,
Wen-Long Shang
d,c,*
a
School o Economics and Managemen , China Uni e si y o Pe oleum, Beijing, China
b
Depa men o Ag icul u al Economics, Sociology, and Educa ion, Pennsyl ania S a e Uni e si y, Uni e si y Pa k, PA, USA
c
Ins i u e o T anspo S udies, Uni e si y o Leeds, 34-40 Uni e si y Road, Leeds, LS2 9JT, UK
d
Cen e o T anspo S udies, Impe ial College London, SW7 2AZ, London, UK
ARTICLE INFO
Keywo ds:
Elec ic ehicle
Cha ging in as uc u e
Subsidy phase-ou s
Agen -based modeling
ABSTRACT
Go e nmen subsidies o elec ic ehicle cha ging in as uc u e (EVCI) in China ha e accele a ed he deploy-
men o cha ging s a ions and p omo ed he di usion o elec ic ehicles (EVs). Howe e , hese subsidies ha e
also imposed a subs an ial iscal bu den on public inances. While much o he exis ing li e a u e compa es
di e en ypes o EVCI subsidies, ew s udies explo e he implica ions o phasing ou EVCI- ela ed subsidies o
go e nmen spending and EV di usion. This pape de elops an agen -based model (ABM) inco po a ing EVCI
ope a o , he e ogeneous EV consume s, and he go e nmen o analyze how EVCI subsidies in luence EV
di usion and p oposes ailo ed phase-ou policy combina ions. A key inno a ion o his s udy is he in eg a ion o
p i a e cha ging pile- ela ed ac o s in o he consume decision-making p ocess h ough a disc e e choice
expe imen . Addi ionally, egional dispa i ies in EV di usion be ween u ban and subu ban a eas unde EVCI
subsidies a e explo ed, and we ind ha by 2030, he EV pene a ion a e could each 79.78 %, wi h subu ban EV
owne ship su passing ha o u ban a eas. While EVCI subsidies signi ican ly in luence ea ly and mid-s age EV
adop ion, hei e ec i eness diminishes in he la e s ages. Implemen ing phase-ou subsidies unde cu en
s anda ds can educe cumula i e go e nmen spending by app oxima ely 91 % compa ed o a no-phase-ou
scena io, wi h only a ma ginal decline o 0.05 % in EV owne ship. A compa a i e analysis o 50 subsidy
phase-ou policy combina ions e eals ha hose ea u ing high ini ial ope a ing subsidies wi h low ini ial
cons uc ion subsidies unde a apid phase-ou mode a e he mos cos -e ec i e. The policy ecommenda ions
p oposed alle ia e iscal bu dens and p omo e mo e balanced EV de elopmen be ween u ban and subu ban
a eas.
1. In oduc ion
Unde he dual objec i es o ca bon peaking and ca bon neu ali y,
accele a ing he elec i ica ion o he anspo a ion sec o has become a
key pa hway o China o ul ill i s emission educ ion commi men s
(Bao e al., 2023; Zhong e al., 2024). Wi hin his elec i ica ion agenda,
he elec ic ehicle (EV) indus y has expe ienced apid g ow h, la gely
d i en by suppo i e go e nmen subsidies. By he end o 2024, he
numbe o EVs in China had eached 31.4 million (Kong, 2025). How-
e e , he de elopmen o elec ic ehicle cha ging in as uc u e (EVCI)
o suppo EV adop ion emains hinde ed by signi ican imbalances. The
i s challenge is a quan i y imbalance. By he end o 2024, China had
ins alled app oxima ely 3.579 million public cha ging piles (PCPs) and
9.239 million p i a e cha ging piles (P CPs).
1
Wi h an EV lee o 31.4
million, he esul ing ehicle- o-cha ging pile a io is oughly 2.45:1,
which alls below he a ge ed 1:1 a io (Na ional De elopmen and
Re o m Commission [NDRC], 2015). The second challenge is spa ial
dis ibu ion imbalance (Y. Shang e al., 2025). EVCI deploymen in
China shows a clea egional dispa i y, wi h highe densi ies gene ally
obse ed in he sou h han he no h and b oade co e age in he eas
* Co esponding au ho . Cen e o T anspo S udies, Impe ial College London, SW7 2AZ, London, UK.
E-mail add ess: [email p o ec ed] (W.-L. Shang).
1
The numbe o public and p i a e cha ging piles in 2024 is a ailable a : h ps://www.e cipa.o g.cn/newsin o/8137834.h ml.
Con en s lis s a ailable a ScienceDi ec
T anspo Policy
jou nal homepage: www.else ie .com/loca e/ anpol
h ps://doi.o g/10.1016/j. anpol.2025.103876
Recei ed 21 July 2025; Recei ed in e ised o m 25 Oc obe 2025; Accep ed 28 Oc obe 2025
T anspo Policy 175 (2026) 103876
A ailable online 30 Oc obe 2025
0967-070X/© 2025 The Au ho (s). Published by Else ie L d. This is an open access a icle unde he CC BY-NC-ND license (
h p://c ea i ecommons.o g/licenses/by-
nc-nd/4.0/ ).
ela i e o he wes .
2
In majo ci ies, he a e age EVCI co e age a e in
cen al u ban a eas eaches 80.8 %, compa ed wi h less han 5.0 % in
u al a eas.
3
EVCI plays a c ucial ole in alle ia ing ange anxie y among
EV use s and is widely ecognized as a c i ical in luen ial ac o o EV
adop ion (X. Li and Liu, 2023; Z. Wang, 2020). The e ec o EVCI
deploymen on EV adop ion can be explained h ough he lens o indi-
ec ne wo k e ec (INE), which e e s o he phenomenon in which a
use ’s u ili y om adop ing a pa icula p oduc inc eases wi h he
adop ion o a complemen a y p oduc (Ka z and Shapi o, 1985). In he
EV ma ke , EVCI se es as a complemen a y good o EVs, and i s
quan i y has an indi ec ye subs an ial impac on he use expe ience o
EV owne s (Sun e al., 2018). Limi ed access o EVCI emains a majo
obs acle o EV adop ion (Y. Liu e al., 2015).
Since 2015, he Chinese go e nmen has implemen ed a se ies o
subsidy policies o EVCI ope a o s o accele a e he de elopmen o
cha ging in as uc u e (W. Shang e al., 2024). In he p elimina y
s ages, hese e o s p ima ily a ge ed cha ging s a ion cons uc ion
subsidies (CSs), aimed a alle ia ing he ini ial in es men bu den on
ope a o s and he eby s imula ing he expansion o cha ging ne wo ks.
Beginning in 2016, he go e nmen implemen ed ope a ional subsidies
(OSs) o incen i ize g ea e e iciency among cha ging s a ion ope a o s,
and unde his policy, ope a o s ecei e subsidies based on he ac ual
olume o elec ici y dispensed. Howe e , he ex ensi e EVCI subsidies
ha e imposed a g owing iscal bu den on he go e nmen s. Acco ding o
epo s om he Minis y o Finance o he P.R.C.,
4
cumula i e subsidies
o EVCI amoun ed o 7.556 billion RMB be ween 2016 and 2020. In
esponse, se e al egions (p o inces) in China ha e begun adjus ing
hei subsidy policies. Fo example, Hunan and Hainan P o inces ha e
adop ed phased educ ion policies o EVCI subsidies, g adually
lowe ing he subsidy le els on an annual basis.
5
To da e, limi ed
a en ion has been gi en o he educ ion o EVCI subsidies and hei
po en ial consequences. Mo eo e , mos exis ing esea ch has p edom-
inan ly ocused on PCPs, wi h ew s udies in es iga ing he ole o P CPs
in EV di usion. In eali y, PCPs accoun o only 27.92 % o China’s o al
EVCI s ock,
6
and he abili y o EV use s o ins all P CPs is ano he c i ical
ac o in luencing pu chase decisions (Qian e al., 2019). The e o e,
inco po a ing he ole o P CPs in o s udies o EV di usion is essen ial o
mo e accu a ely e lec eal-wo ld EV adop ion pa e ns.
In addi ion o incen i e policies and EVCI, he di usion o EVs is
shaped by o he ac o s such as ehicle a ibu es, uel p ices, and in-
di idual income le els (Su and Diao, 2025). Acco dingly, he EV di u-
sion p ocess can be concep ualized as a complex sys em in which agen s,
such as consume s, ope a o s, and policymake s, in e ac wi h each
o he o e ime wi hin a dynamic en i onmen . Agen -based modeling
(ABM) app oach p o ides a obus amewo k o analyzing such sys-
ems. By cons uc ing mic o-le el agen s wi h he e ogeneous a ibu es
and decision-making ules, ABM allows o au onomous
decision-making and in e ac ions wi hin a de ined en i onmen , he eby
enabling he explo a ion o di usion pa hways o e ime (Mehdizadeh
e al., 2022). Building on his, his s udy de elops an ABM amewo k
inco po a ing he e ogeneous EV consume s, EVCI ope a o , and he
go e nmen o examine he impac s o EVCI incen i e policies on he
di usion o EVs. The main con ibu ions o his s udy a e as ollows:
Fi s , i examines he impac o EVCI subsidy phase-ou on bo h EV
di usion and go e nmen iscal expendi u e, while also conduc ing a
compa a i e analysis o al e na i e phase-ou s a egies. To he bes o
ou knowledge, i ep esen s he i s quan i a i e e alua ion o di e en
EVCI subsidy wi hd awal scena ios, which he eby o e s aluable pol-
icy insigh s o he design o he phased subsidy educ ion. Second, his
s udy examines he spa ial dis ibu ion o EVCI by compa ing he
di usion pa e ns o EVs and cha ging piles in u ban e sus subu ban
a eas and iden i ies e ec i e policy s a egies o add ess pe sis en im-
balances in EVCI deploymen ac oss di e en geog aphic egions. Thi d,
his s udy examines he ole o P CPs in consume EV pu chase decisions
by in eg a ing P CPs ins alla ion in o he decision-making p ocess
wi hin an ABM amewo k and u he simula es he di usion o P CPs,
which add esses a no able gap in he exis ing li e a u e on p i a e
cha ging in as uc u e.
The emainde o his pape is o ganized as ollows: Sec ion 2 e-
iews he ele an li e a u e on EVCI subsidies and EV di usion
modeling. Sec ion 3ou lines he ABM amewo k. Sec ion 4de ails he
pa ame e se ings and model alida ion. Sec ion 5p esen s simula ion
esul s and conduc s sensi i i y analysis. Finally, Sec ion 6concludes
wi h he main indings.
2. Li e a u e e iew
2.1. EV and EVCI policies
The apid expansion o EVs in he Chinese ma ke has been la gely
d i en by s ong policy suppo . To encou age EV adop ion, he go -
e nmen has p ima ily elied on a se ies o policies such as license pla e
es ic ions, EV pu chase ax exemp ions, and EV pu chase subsidies.
The e ec i eness o hese policy ins umen s has been ex ensi ely
documen ed in he exis ing li e a u e (Shen e al., 2021; Sun e al., 2018;
T. Zhang e al., 2024). While license pla e con ol policies can s ongly
in luence EV adop ion, hei impac is geog aphically es ic ed as hey
a e implemen ed in only a limi ed numbe o ci ies. By con as , pu -
chase ax exemp ions and pu chase subsidies a e applied mo e b oadly
and play a mo e impo an ole on EV di usion (Zhu e al., 2022).
Howe e , wi h he apid expansion o he EV ma ke and g owing iscal
p essu es, di ec pu chase subsidies we e g adually educed and ul i-
ma ely phased ou by he end o 2022.
Gi en he phase-ou o EV pu chase subsidies, schola s ha e
inc easingly shi ed hei a en ion o EVCI subsidy policies. Li e al.
(2017) iden i ied a eedback loop be ween EV adop ion and cha ging
in as uc u e deploymen . Thei indings indica e ha , ce e is pa ibus,
CSs o EVCI ha e wice he e ec on EV di usion as EV pu chase sub-
sidies. Zhu e al. (2019) a gued ha , in he con ex o EV pu chase
subsidy wi hd awal, he go e nmen should inc ease CSs o EVCI o
sus ain EV ma ke g ow h. Zhang e al. (2024) examined he e ec s o
di e en subsidy combina ions, including EV pu chase subsidies and
EVCI CSs, on EV sales. Thei indings sugges ha , as he EV ma ke
ma u es, pu chase subsidies should be g adually edi ec ed owa d
suppo ing EVCI cons uc ion, wi h CSs p og essi ely phased ou in he
la e s ages o ma ke de elopmen . Following he in oduc ion o OSs
o cha ging s a ions, some schola s ha e examined he combined e ec s
o CSs and OSs on EVCI de elopmen and di usion. Luo e al. (2023)
assessed he e ec i eness o di e en subsidy ypes in EV adop ion and
ound ha he cos o acqui ing one addi ional EV use h ough EV
pu chase subsidies is much highe han h ough ei he CSs o OSs. Ling
e al. (2024) analyzed how di e en combina ions o CSs and OSs a ec
he decision-making o EVCI ope a o s and concluded ha CSs a e mo e
sui able o ope a o s wi h low ope a ing cos coe icien s (which mea-
su e he ope a ional capabili y, wi h lowe alues indica ing lowe cos s
2
Addi ional de ails on he spa ial pa e n o EVCI alloca ion ac oss China can
also be ound a : h ps://www.e cipa.o g.cn/newsin o/8137834.h ml.
3
Co e age a es o EVCI in u ban and u al a eas a e p o ided a : h ps://
www.cnene gynews.cn/ ocus/2025/03/13/de ail_20250313204160.h ml and
h ps://mp.weixin.qq.com/s/DAaxqE7dO0YMVBILaYj7Hw.
4
Repo s by he Minis y o Finance o he P.R.C. can be accessed a : h p://jj
s.mo .go .cn/zxzyz /jnjpbzzj/202204/ 20220420_3804302.h m and h p://jjs.
mo .go .cn/zxzyz /jnjpbzzj/202405/ 20240507_3934118.h m.
5
De ails on he phasing ou EVCI subsidies in Hunan and Hainan p o inces
a e a ailable h ough: h ps://plan.hainan.go .cn/s gw/0400/201907/a
69db13 d15463b870 e8806d d2577.sh ml and h ps://gx .hunan.go .cn/gx
/xxgk_71033/zc g/g xwj/202212/ 20221230_29171166.h ml.
6
In o ma ion on he numbe o public cha ging piles and he EV s ock in
China is sou ced om he China Elec ic Vehicle Cha ging In as uc u e P o-
mo ion Alliance(EVCIPA)’s epo : h ps://www.e cipa.o g.cn/newsin o/813
7834.h ml.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
2
o achie e he same le el o se ice), whe eas OSs a e mo e e ec i e o
hose wi h ela i ely high cos coe icien s. Chen e al. (2023) ound ha
he ma ginal e ec s o in es men subsidies, CSs, and OSs on EVCI
deploymen di e , wi h in es men subsidies exhibi ing he mos p o-
nounced diminishing ma ginal e ec .
In summa y, wi h he wi hd awal o EV pu chase subsidies and he
implemen a ion o cha ging s a ion subsidies in China, schola s ha e
inc easingly ocused on he impac o hese policy shi s. Howe e , ew
s udies ha e analyzed CSs and OSs wi hin an in eg a ed analy ical
amewo k, and e en ewe ha e quan i a i ely assessed he conse-
quences o EVCI subsidy phase-ou s. As wi h EV pu chase subsidies, he
subsidies o cha ging s a ions a e also expec ed o be g adually phased
ou as he EVCI indus y ma u es. I is he e o e essen ial o design
op imized phase-ou s a egies ha educe iscal expendi u e while
minimizing po en ial ad e se e ec s on EV de elopmen .
2.2. Indi ec ne wo k e ec s
The in luence o cha ging in as uc u e on EV di usion can be un-
de s ood h ough he lens o INE. Fi s in oduced by Ka z and Shapi o
(1985), INE desc ibes si ua ions in which he u ili y a consume de i es
om a p oduc is co ela ed wi h he a ailabili y o a complemen a y
p oduc . The exis ence and signi icance o INE ha e been con i med
ac oss a ious indus ies, such as ideo games (Clemen s and Ohashi,
2005), DVDs (Inceoglu and Pa k, 2011), and ele ision se s (X. Zhang,
2007). In ecen yea s, nume ous s udies ha e examined he INE be-
ween EVs and EVCI, which can gene ally be classi ied in o mac o-le el
and mic o-le el analyses. Mac o-le el s udies p ima ily assess he
impac o EVCI quan i y on EV adop ion using agg ega e da a. D awing
on da a om China’s EV ma ke , Li and Liu (2023) demons a ed he
posi i e impac s o EVCI on EV di usion in e ms o bo h EVCI s ocks
and densi y. Based on da a om he U.S. ma ke , Li e al. (2017) u he
con i med he signi ican in luence o INE on EV di usion h ough an
empi ical s udy. Simila ly, Koch e al. (2022) analyzed he INE in he
No wegian EV ma ke by cons uc ing an econome ic model and ound
ha i s in ensi y is lowe han in he Uni ed S a es.
Mic o-le el s udies, by con as , ocus on how he quan i y o EVCI in
a use ’s icini y a ec s pe cei ed u ili y, ypically elying on ques ion-
nai es o cap u e use p e e ences. Sun e al. (2018a,b), o example,
conduc ed a disc e e choice expe imen wi h indi idual-le el da a and
ound ha INE posi i ely in luences EV choice, wi h hei he e ogenei y
analysis u he e ealing subs an ial a ia ion in sensi i i y o INE
ac oss di e en use g oups. Likewise, Liu e al. (2015) con i med he
p esence o INE in he EV ma ke using ques ionnai e da a and no ed
ha EVCI- ela ed INE issues hinde EV pu chases. O e all, he exis ence
and impo ance o INE a e well es ablished in p e ious esea ch.
Building on his ounda ion, ou s udy inco po a es INE in o an ABM
amewo k o simula e he dynamic ela ionships be ween he di usion
o EVs and EVCI.
2.3. EV di usion esea ch me hodologies
A subs an ial body o esea ch has explo ed EV di usion h ough
me hodologies such as he Bass model, sys em dynamics (SD), and ABM.
The Bass model, o iginally p oposed by Bass (1969), is widely used o
quan i y inno a ion di usion p ocesses and has been applied by se e al
schola s o analyze EV adop ion. Fo ins ance, Shi e al. (2022) de el-
oped a Bass model inco po a ing ehicle sc appage and in e nal com-
bus ion engine ehicle (ICEV) compe i ion mechanisms o p edic EV
di usion in Shanghai unde policy in luences. Yang e al. (2025) in e-
g a ed unce ain y heo y in o he Bass model o simula e EV sales
g ow h unde dis up i e e en s, while Fan e al. (2025) p oposed a Bass
model wi h a g een p emium ac o o o ecas EV sales in he Chinese
ma ke . Ne e heless, al hough he Bass model is ela i ely simple and
da a-e icien , i s inabili y o accoun o ex e nal in luen ial ac o s has
limi ed i s applicabili y o he ea ly s ages o EV di usion esea ch.
Unlike he Bass model, SD allows o he inco po a ion o a wide
ange o in luencing ac o s om a mac o-le el pe spec i e. Li e al.
(2023) de eloped an SD-based EV di usion model o analyze he im-
pac s o ac o s such as ehicle pe o mance and cha ging con enience.
Simila ly, Zhu e al. (2024) cons uc ed an EV di usion model inco -
po a ing ehicle pe o mance, EVCI, and policy ac o s, which enabled a
compa a i e analysis o di e en policy measu es in p omo ing EV
di usion. Using an SD amewo k ha in ol es go e nmen , i ms, and
consume s, Kong e al. (2020) assessed he e ec s o phasing ou EV
pu chase subsidies on EV ma ke sha e. Shen e al. (2021) analyzed he
impac o subsidy phase-ou policies in Shanghai on local EV adop ion,
and simila ly, Kim e al. (2021) p ojec ed EV di usion ends in Sou h
Ko ean ci ies h ough an SD app oach and e alua ed he e ec i eness o
subsidy policies on EV pene a ion.
Howe e , nei he he Bass model no SD is capable o explici ly
cap u ing he impac o indi idual-le el beha io on he o e all sys em.
In con as , he ABM app oach enables he cons uc ion o mul iple
he e ogeneous agen s wi hin a s udied en i onmen . These agen s
in e ac wi h each o he and wi h hei en i onmen , he eby gene a ing
complex phenomena and pa e ns a he mac o le el (Mehdizadeh e al.,
2022). Consequen ly, a g owing numbe o schola s ha e s udied EV
adop ion wi h he ABM app oach. Fo example, Epps ein e al. (2011)
de eloped a mic o-le el ABM o EV di usion ha inco po a es agen
beha io s shaped by media in o ma ion dissemina ion and pee e ec s.
Zhang e al. (2011) combined he mul inomial logi model wi h an ABM
amewo k o ep esen he e ogeneous consume s based on empi ical
da a. O he schola s ha e u he le e aged ABM’s capabili ies o
analyze how social ela ionships among use s in luence o e all EV
de elopmen (Sha iei e al., 2012; Xu and Bi, 2024). Table 1 below
summa izes he li e a u e employing he a o emen ioned me hods o
s udy EV adop ion.
Besides, se e al schola s ha e also le e aged he spa ial modeling
capabili ies o ABM o explo e he spa ial di usion o EVs and EVCI and
o o ecas hei sp ead ac oss geog aphical egions. Wang e al. (2023)
modeled he spa ial de elopmen o he EV ma ke om 2015 o 2040
and ound ha EV use s g adually o m clus e s o e ime. Sil ia and
K ause (2016) di ided an u ban a ea in o comme cial and esiden ial
zones, alloca ing PCPs in a ying quan i ies ac oss hese zones o e lec
a ealis ic cha ging ne wo k s uc u e. Simila ly, Hu´
e ink e al. (2010)
adop ed a spa ial segmen a ion app oach and di ided he s udy egion
in o u ban and u al a eas o analyze how di e en hyd ogen e ueling
in as uc u e deploymen s a egies in luence he di usion o hyd ogen
ehicles. Mo eo e , by in eg a ing ABM wi h geog aphic in o ma ion
sys ems (GISs), some esea che s ha e de eloped spa ial en i onmen s
ha e lec u ban cha ac e is ics, he eby enabling mo e accu a e ana-
lyses o EV and EVCI di usion wi hin speci ic ci y con ex s (Luo e al.,
2023; Zhuge e al., 2021). Al hough hese s udies le e age ABM’s spa ial
modeling capabili ies o segmen he simula ion space, he demog aphic
cha ac e is ics assigned o di e en egions o en ely on subjec i e as-
sump ions a he han empi ical e idence. Table 2 summa izes he
ele an li e a u e on EV di usion wi h an ABM app oach.
This pape seeks o simula e he di usion o EVs and cha ging s a-
ions unde eal-wo ld condi ions by inco po a ing indi idualized ac-
o s such as consume loca ion, he numbe o nea by PCSs, and he
easibili y o ins alling P CPs. The ABM app oach, cha ac e ized by i s
bo om-up amewo k o modeling indi idual beha io o de i e
sys em-le el ou comes, is pa icula ly sui ed o simula ing EV di usion
unde he e ogeneous consume p e e ences. Mo eo e , by u ilizing
spa ial modeling capabili ies, his s udy analyzes he spa ial dynamics o
EV di usion and p oposes di e en ia ed s a egies o p omo ing EV
up ake in u ban and subu ban a eas based on he simula ion esul s.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
3
3. Me hodology
3.1. Agen -based modeling desc ip ion
This pape cons uc s a mul i-agen sys em wi hin an ABM ame-
wo k, inco po a ing EV consume s, EVCI ope a o , and he go e nmen ,
o e alua e he impac s o policy in e en ions on he EV ma ke . As
ehicle pe o mance and cha ging in as uc u e con inue o e ol e,
po en ial consume s may shi hei p e e ences om ICEVs o EVs,
he eby eshaping o e all cha ging demand. In esponse o his changing
demand and go e nmen subsidy policies, EVCI ope a o adjus s i s PCS
deploymen s a egies, which in u n u he in luence he de elopmen
o he cha ging en i onmen . Fu he mo e, conside ing he in luence o
P CPs on consume pu chase decisions and he di e ences in ins alla ion
condi ions be ween u ban and subu ban a eas in China, his pape
adop s an u ban-subu ban classi ica ion o cap u e egional he e oge-
nei y. Sub-models o each agen a e desc ibed in de ail in he ollowing
sec ions.
3.2. Consume sub-model
The ABM amewo k cen e s on he decision-making beha io o
consume s, assuming ha hey ollow a u ili y maximiza ion app oach
ha accoun s o ehicle a ibu es, cha ging s a ion cha ac e is ics, and
indi idual socioeconomic p o iles. Selec ion and pa ame e iza ion o
a iables in he decision model a e in o med by s a is ical analysis o
da a ob ained om a disc e e choice expe imen .
3.2.1. Disc e e choice model
This s udy assumes ha consume s beha e a ionally and conside
ac o s such as ehicle a ibu es, INE, and he accessibili y o cha ging
o gas s a ions when making pu chase decisions. To assess he in luence
o hese ac o s on pu chase in en ions, his pape employs a s a ed
p e e ence app oach wi hin a disc e e choice expe imen amewo k,
which analyzes consume p e e ences by obse ing choices ac oss hy-
po he ical pu chase scena ios wi h a ying a ibu e combina ions
(Higgins e al., 2017; Qian e al., 2019; Rudolph, 2016). Da a was
collec ed h ough a ques ionnai e comp ising wo main componen s:
demog aphic in o ma ion and a s a ed p e e ences su ey. Following
Pa e al. (2019), he demog aphic sec ion includes ques ions on e-
sponden s’ income, accep able ehicle pu chase p ice, place o esi-
dence, and a ailabili y o p i a e cha ging in as uc u e. The s a ed
p e e ences sec ion ocuses on ehicle a ibu es, INE, and P CP ins al-
la ion s a us. Speci ically, he ehicle a ibu es conside ed a e pu chase
p ice, d i ing ange, e ueling o echa ging ime, cos pe 100 km o
d i ing, and main enance and insu ance expenses (Rudolph, 2016). INE
is measu ed by he numbe o cha ging o gas s a ions wi hin a p e-
de ined adius o he esponden ’s esidence (Sun e al., 2018). A
disc e e choice model is de eloped unde he assump ion ha use s
choose be ween wo ehicle ypes: ICEVs and EVs. U ili y Uij ha use i
de i es om selec ing ehicle ype j consis s o h ee componen s, as
speci ied in Equa ion (1). Xj ep esen s he explana o y a iables,
including ehicle a ibu es (pu chase p ice, d i ing ange, e ueling/-
echa ging ime, cos pe 100 km o d i ing, and insu ance and main-
enance cos s), INE (measu ed by he numbe o cha ging s a ions wi hin
a speci ied adius), and P CP ins alla ion s a us ( ep esen ed by a bina y
a iable and equal o 1 i a P CP is ins alled and 0 o he wise). Υi deno es
indi idual-le el demog aphic cha ac e is ics, including gende , annual
income, and age (Qian e al., 2019).
ε
ij is he e o e m.
Uij =βʹXj+
α
ʹΥi+
ε
ij (1)
To ensu e he alidi y o he ques ionnai e design, we conduc ed a
one-week pilo su ey in Decembe 2024, combining in-pe son isi s o
Table 1
O e iew o li e a u e on EV adop ion.
Me hodology Re e ence Main con en Summa y
Bass model Shi e al.
(2022)
A Bass model inco po a ing ehicle sc appage and echnological
compe i ion o analyze EV di usion in Shanghai, China.
Al hough he Bass model o e s ad an ages such as ela i e simplici y
and low da a equi emen s, i s applicabili y is cons ained by limi ed
lexibili y in cap u ing he in luence o ex e nal en i onmen al ac o s.X. Yang e al.
(2025)
In eg a ion o unce ain y heo y wi h he Bass model o analyze
EV di usion unde he in luence o unce ain e en s.
Fan e al.
(2025)
A Bass model inco po a ing he g een p emium o EVs.
Sys em dynamics
(SD)
Y. Li e al.
(2023)
A mul i-agen in e ac ion SD model o analyze he impac s o
di e en subsidy schemes on EV di usion and ca bon emissions.
SD can cap u e in e ac ions among he e ogeneous agen s and allows o
a simula ion o eedback in complex sys ems. Howe e , i aces
limi a ions in explici ly ep esen ing sys em-le el phenomena ha a ise
om indi idual esponses o ex e nal ac o s.
Shen e al.
(2021)
An SD model conside ing dynamic echnological ma u i y
e olu ion and incen i e phase-ou .
Zhu e al.
(2024)
An SD model inco po a ing mul i-agen in e ac ions based on INE
o compa e he e ec i eness o di e en policies in p omo ing EV
di usion.
Kong e al.
(2020)
An SD model o EV di usion inco po a ing in e ac ions among he
go e nmen , i ms, and consume s.
Kim e al.
(2021)
An SD model o EV di usion inco po a ing policy incen i es and
en i onmen al bene i s.
Agen -based
modeling (ABM)
Zhang e al.
(2011)
An ABM explo ing he impac o exogenous ac o s on EV adop ion
in he con ex o mul i-agen in e ac ions be ween manu ac u e s,
consume s, and he go e nmen .
ABM can cap u e complex in e ac ions among he e ogeneous agen s
and be ween agen s and hei en i onmen a a mac o le el.
Epps ein
e al. (2011)
An ABM in es iga ing he nonlinea impac o social in luence and
media exposu e on EV ma ke pene a ion.
Sha iei e al.
(2012)
An ABM o ecas ing EV adop ion h ough pe cei ed u ili y and
social in luence wi hin a choice-based di usion amewo k.
Xu and Bi
(2024)
An ABM conside ing he impac o wo d-o -mou h e ec s and
social ne wo ks on consume s’ decisions o pu chase EVs.
Sun e al.
(2018)
An INE-based ABM e alua ing he in luence o spa ial EVCI
deploymen policies on EV di usion.
Y. Wang e al.
(2023)
An INE-based ABM e alua ing he impac o he go e nmen and
ope a o EVCI deploymen on EV di usion.
Luo e al.
(2023)
An ABM modeling he ole o policy and ex e nal ac o s in
o e coming he EV di usion bo leneck.

Zhuge e al.
(2021)
A GIS-based ABM e alua ing he impac s o EV p ice, ange, and
EVCI on EV di usion ac oss di e en egions in Beijing, China.

L. Zhu e al.
T anspo Policy 175 (2026) 103876
4
4S s o es ac oss mul iple dis ic s o Beijing wi h online da a collec ion.
Feedback om esponden s ega ding he cla i y and app op ia eness o
he ques ionnai e was collec ed du ing he su ey and subsequen ly
used o e ine i s con en . Du ing he o mal su ey, ques ionnai es we e
collec ed om Ap il 2 h ough Ap il 30, 2025. The egional dis ibu ion
o he esponses is epo ed in Appendix A, Table A3, while desc ip i e
s a is ics o he su ey da a a e p esen ed in Tables A1 and A2. In o al,
549 esponses we e collec ed h ough he Wenjuanxing pla o m,
7
o
which 533 we e deemed alid a e da a imming. The cleaned da a was
analyzed using S a a 16.0, wi h a mixed logi model o es ima e con-
sume p e e ences o he ele an ac o s. Es ima ed coe icien s we e
ob ained h ough he maximum likelihood me hod, and he esul s a e
summa ized in Table 3.
As shown in Table 3, se e al ehicle a ibu es, including ehicle
p ice, d i ing ange, e ueling/ echa ging ime, cos pe 100 km o
d i ing, and main enance and insu ance cos , a e ound o be s a is i-
cally signi ican . Pa icula ly, he es ima ed coe icien s o ehicle
p ice, e ueling/ echa ging ime, cos pe 100 km o d i ing, and
main enance & insu ance cos a e nega i e, sugges ing ha highe
alues o hese a ibu es educe consume u ili y, which is consis en
wi h heo e ical expec a ions and aligns wi h p io esea ch (Qian e al.,
2019). Wi h espec o he cha ging en i onmen , bo h INE and P CP
ins alla ion s a us exhibi s a is ically signi ican coe icien s, each
con ibu ing posi i ely o consume u ili y. No ably, he impac o P CP
ins alla ion is much g ea e han ha o ha ing one PCS wi hin a 5 km
adius. This inding suppo s he conclusions o Qian e al. (2019) and
Hel es on e al. (2015), who a gued ha he a ailabili y o p i a e
cha ging in as uc u e has a s onge in luence on EV adop ion
compa ed o he con enience o public cha ging in as uc u e.
Rega ding demog aphic cha ac e is ics, only he income ca ego ies o
‘150,000–200,000 RMB’ and ‘≥300,000 RMB’ exhibi s a is ically sig-
ni ican e ec s. The magni ude and di ec ion o hese es ima ed co-
e icien s indica e ha highe income le els a e associa ed wi h a g ea e
likelihood o EV pu chase, which is consis en wi h he indings o Sun
e al. (2018a,b). By con as , he coe icien s o gende and age u n ou
o be no s a is ically signi ican , indica ing ha hese cha ac e is ics
ha e li le in luence on EV pu chase decisions.
3.2.2. U ili y unc ion
In he consume sub-model, a mixed logi eg ession me hod based
on su ey da a is used o cons uc a u ili y e alua ion amewo k ha
inco po a es ehicle a ibu es and en i onmen al ac o s. As shown in
Fig. 1, he amewo k consis s o ou componen s: (1) changes in ehicle
li espan o e ime (wi hin he blue box), (2) compa ison o ehicle a -
ibu es (wi hin he ed box), (3) e alua ion o he EV cha ging en i-
onmen (wi hin he yellow box), and (4) calcula ion o EV pu chase
p obabili y (wi hin he g een box). The blue box ep esen s consume s
who al eady own a ehicle (ei he EV o ICEV). Fo hese consume s,
ehicle li espan dec eases wi h each pe iod, and once i eaches ze o,
hey a e eclassi ied as non-owne s. The ed box deno es ehicle a i-
bu e compa ison, in which EV d i ing ange and p ice a e i s e alu-
a ed. Consume s p oceed o conside he cha ging en i onmen only i
bo h a ibu es mee expec a ions; i ei he alls sho , hey shi o
pu chasing an ICEV. Gi en ha ICEVs a e no subjec o ange anxie y, i
is assumed ha consume s will pu chase an ICEV i i s p ice alls below a
Table 3
Maximum likelihood es ima ion esul s o he mixed logi eg ession.
Va iables Coe . S d.
e o
P-
alues
Vehicle p ice (10,000 RMB) −0.060*** 0.011 0.000
D i ing ange (100 km) 0.004*** 0.0004 0.000
Re ueling/Recha ging ime (minu es) −0.009*** 0.001 0.000
Cos pe 100 km (RMB) −0.026*** 0.005 0.000
Main enance and insu ance cos (10,000
RMB)
−0.540*** 0.063 0.000
Numbe o gas o cha ging s a ions (wi hin
5 km)
0.018*** 0.005 0.001
P CP ins alla ion (=1 i P CP ins alled) 0.504*** 0.065 0.000
Income (100,000–150,000 RMB) 0.193 0.158 0.221
Income (150,000–200,000 RMB) 0.305*0.167 0.066
Income (200,000–300,000 RMB) 0.182 0.170 0.285
Income (≥300,000 RMB) 0.489** 0.210 0.020
Gende (Male) 0.031 0.103 0.767
Age (25–34) 0.060 0.147 0.682
Age (35–44) 0.054 0.161 0.739
Age (45–54) 0.309 0.266 0.245
Age (≥55) 0.219 0.471 0.642
Cons an 0.357 0.249 0.153
Log likelihood −
2780.2399
 
No es: *p <0.1; **p <0.05; ***p <0.01. The numbe o obse a ions is 533.
The omi ed e e ence ca ego ies a e ‘Income (50,000–100,000 RMB)’ o in-
come le el, ‘Female’ o gende , and ‘Age (18–24)’ o age, due o
mul icollinea i y.
Table 2
ABM-based s udies on EV di usion.
Re e ence Consume
he e ogenei y
PCP P CP PCS
subsidy
PCS
subsidy
phase-
ou
PCS
spa ial
alloca ion
Zhang
e al.
(2011)
✓    
Epps ein
e al.
(2011)
✓    ✓
Sha iei
e al.
(2012)
✓    
Xu and Bi
(2024)
✓    ✓
Sun e al.
(2018)
✓ ✓    ✓
Y. Wang
e al.
(2023)
✓ ✓ ✓   ✓
Luo e al.
(2023)
✓ ✓ ✓ 
Zhuge
e al.
(2021)
✓ ✓ ✓   ✓
Zhuge
e al.
(2021)
✓ ✓    ✓
Wolbe us
e al.
(2021)
✓ ✓    ✓
Huang
e al.
(2021)
✓ ✓ ✓ 
Pagani
e al.
(2019)
✓ ✓ ✓   ✓
Sil ia and
K ause
(2016)
✓ ✓    ✓
This s udy ✓ ✓ ✓ ✓ ✓ ✓
7
This pla o m specializes in p o essional ques ionnai e dis ibu ion and
o e s access o an ex ensi e sample pool o app oxima ely 6.2 million in-
di iduals, encompassing a wide spec um o income le els, age g oups, occu-
pa ions, and o he demog aphic cha ac e is ics.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
5

p ese accep able h eshold; o he wise, hey exi he ma ke wi hou
making a pu chase. The yellow box co esponds o he assessmen o he
cha ging en i onmen . When an EV’s d i ing ange and sale p ice mee
expec a ions, consume s i s check whe he hey al eady possess a P CP
o ha e pe mission o ins all one. I ei he condi ion is sa is ied, he
model calcula es he p obabili y o pu chasing an EV wi h access o a
P CP.
8
I no , i is hen assessed whe he a PCS exis s wi hin 5 km
9
; i
a ailable, he p obabili y o pu chasing an EV wi hou P CP (bu wi h
access o PCS) is calcula ed. I no PCS is a ailable, consume s ins ead
conside an ICEV and e alua e i s p ice. Finally, he g een box e e s o
he calcula ion o EV pu chase p obabili y. A e his p obabili y is
de e mined, a andom numbe is d awn om a uni o m dis ibu ion
[0,1] o simula e he consume ’s pu chasing decision mechanism (Zhu
and Ma, 2025): I he calcula ed p obabili y is g ea e han o equal o
he andom numbe , he consume pu chases an EV; o he wise, hey
conside an ICEV and e alua e whe he i s p ice mee s expec a ions.
To accoun o EV consume he e ogenei y, we build on Wang e al.
(2023) by analyzing he co ela ions be ween annual income and key
pa ame e s such as annual d i ing mileage and he p opo ion o ime
spen cha ging a PCSs. Based on he di ec ion and s eng h o hese
co ela ions, we di e en ia e pa ame e alues ac oss income g oups.
Fu he mo e, d awing on he su ey da a, we es ablish income dis i-
bu ion p o iles o u ban and subu ban popula ions. De ails o he in-
come dis ibu ion and co ela ion analysis a e p o ided in Appendix A
(Table A2 and Table A4, espec i ely). Based on he p e iously in o-
duced disc e e choice model, he u ili y consume i de i es om
choosing ehicle ype j a ime , which is Ui,j, , can be speci ied in
Equa ion (2):
Ui,j, =β1*(Pj, −Spu , )+β2*Cd i,i,j+β3*Co he ,j+β4*Rj, +β5*Tj+β6*
Ni,j, +β7*Bi+
ε
i,j,
(2)
Whe e Pj, is he p ice o ehicle ype j a pe iod , and Spu , deno es he
EV pu chase subsidy a pe iod . Cd i,i,j ep esen s he cos pe 100 km o
d i ing o consume i using ehicle ype j, while Co he ,j is he main e-
nance and insu ance cos o ehicle ype j. Rj, deno es he d i ing ange
o ehicle ype j a pe iod , and Tj is he ime equi ed o ully e uel o
echa ge ehicle ype j. Ni,j, e e s o he numbe o cha ging o gas
s a ions wi hin a 5-km adius o consume i wi h ehicle ype j du ing
pe iod . Bi is a bina y indica o equal o 1 i consume i has access o a
P CP, and 0 o he wise. β is a ec o o es ima ed coe icien s, and
ε
i,j, is
he e o e m. I is wo h no ing ha he in luence o demog aphic a -
ibu es on consume pu chase decisions is omi ed om he model
speci ica ion in his pape . While his simpli ica ion could be seen as
p oblema ic, as a p- alue abo e 0.1 does no necessa ily imply a lack o
e ec (Sco ano and Danielis, 2025). We p oceed wi h his assump ion
o he sake o model pa simony. Acco dingly, he p obabili y use i
chooses an EV du ing pe iod , deno ed as P i,e, , is o mula ed as shown
in Equa ion (3):
P i,e, =e(Ui,e, −
ε
i,e, )
e(Ui,e, −
ε
i,e, )+e(Ui, , −
ε
i, , )(3)
3.2.2.1. Cos o 100-km d i ing. This pape assumes ha EV use s all
in o wo ca ego ies: hose who ely exclusi ely on PCS and hose who
ha e access o P CPs (al hough he la e may also use public cha ging
acili ies). The cha ging capaci ies o public and p i a e s a ions a e
deno ed as Kpub and Kp i, espec i ely. Fo use s who ely solely on PCSs,
he pe -100-km d i ing cos , Cd i,i,e,pub, is de e mined by h ee ac o s: he
composi e elec ici y p ice a PCS Pelc, he se ice ee pe uni (kWh) o
elec ici y Pse , and he pa king ee incu ed pe uni (kWh) o elec ici y
cha ged while cha ging Ppa k.
10
Fo consume s wi h P CPs, he o e all
cha ging cos , Cd i,i,e,p i, is calcula ed by combining he cos s o PCS and
P CP usage. This o al cos depends on he sha e o cha ging ime ha
indi idual i spends using PCP, ep esen ed by yi. The speci ic cos is
calcula ed using Equa ion (4) o use s who ely only on PCSs, and
Equa ion (5) o hose wi h access o P CPs.
Cd i,i,e,pub =he*(Pelc +Pse +Ppa k)(4)
Cd i,i,e,p i =he*Kpub*(Pelc +Pse +Ppa k)*yi+Kp i*Pp i*(1−yi)
Kpub*yi+Kp i*(1−yi)(5)
Whe e yi deno es he pe cen age o o al cha ging ime ha consume s
wi h P CPs spend using PCSs,
11
and he ep esen s he elec ici y con-
sump ion pe 100 km o d i ing o an EV. Pp i is he cha ging p ice o
P CPs and is assumed o be ime-in a ian .
12
The composi e elec ici y
p ice a PCS, Pelc, is calcula ed by mul iplying he sha e o cha ging
olume in each ime-o -use pe iod by he co esponding hou ly elec-
ici y a e
13
and hen summing ac oss all ime pe iods, as shown in
Equa ion (6).
Pelc =∑
23
s=0
σ
s*Pelc,s(6)
Whe e
σ
s ep esen s he sha e o cha ging olume in hou s, and Pelc,s is
he co esponding elec ici y p icing a e. Fo ICEV consume s, he cos
pe 100 km o d i ing, Cd i,i, , is calcula ed as he p oduc o he uel p ice
Pg and he uel consump ion pe 100 km o d i ing h , as shown in
Equa ion (7). In his case, Bi is se o 0, indica ing ha access o P CPs
will no a ec he u ili y om using ICEVs.
Cd i,i, =h *Pg(7)
3.2.2.2. Main enance and insu ance cos s. The o al main enance and
8
In he Chinese ma ke , mos au omobile manu ac u e s p o ide EV con-
sume s ee P CPs and ins alla ion se ices, making i ela i ely easy o he
consume s o access P CPs. Bu in ac , he ins alla ion o a P CP equi es bo h a
dedica ed pa king space and app o al om he p ope y managemen com-
pany. Bu acco ding o a epo by he China Consume Associa ion, he main
easons EV use s do no ins all P CP a e he absence o a dedica ed pa king
space o p ope y managemen company’s e usal, which is consis en wi h he
indings o EVCIPA (2020). Also, al hough some consume s do no ins all P CPs
e en he condi ions abo e a e me , such case a e no he main eason. The e-
o e, his pape makes an assump ion and simpli ica ion: once ins alla ion
condi ions a e sa is ied, consume s a e assumed o ins all P CPs. Sou ce: h
ps://www.cca.o g.cn/De ail?ca alogId=475800366178373&con en Type=a
icle&con en Id=521575306829893.
9
Acco ding o he in es iga ion by he China Consume s Associa ion, 96 % o
EV use s li e wi hin 5 km o PCSs. Based on his, his s udy assumes ha EV
consume s would conside he p esence o a PCS wi hin 5 km when making
pu chase decisions. Sou ce: h ps://www.cca.o g.cn/De ail?ca alogId=4758
00366178373&con en Type=a icle&con en Id=521575306829893.
10
Fo compu a ionally simplici y, we assume a cons an pa king ee o
app oxima ely 4 RMB pe hou . Gi en ha he capaci y o a public cha ging pile
is se a 30 kW, he co esponding pa king ee incu ed pe uni (kWh) o
elec ici y cha ged is calcula ed as 4/30 =0.13 RMB pe kWh.
11
In he main ex , yi e e s o he p opo ion o o al cha ging ime consume
i spen using PCS (i.e., Time o cha ging a PCS/(Time o cha ging a PCS +
Time o cha ging a P CP)). Howe e , in he ques ionnai e, yi was de ined in
opposi e way o ease o esponden unde s anding, namely, as he sha e o
cha ging ime spen using P CP (i.e., Time o cha ging a P CP/( ime o
cha ging a PCS +Time o cha ging a P CP)). Acco dingly, he ‘sha e o
cha ging ime a P CP’ epo ed in Table A1 does no appea in he main ex .
12
The go e nmen has manda ed egula ions on elec ici y p icing o P CPs
(NDRC, 2014). Unde his policy, a uni ied esiden ial elec ici y a e is applied,
which means ha Pp i emains cons an o e ime. Sou ce: h ps://z xxgk.nd c.
go .cn/web/i emin o.jsp?id=19564.
13
The ime-o -use cha ging olume sha es a e om: h ps://www.e cipa.o g.
cn/newsin o/8137317.h ml.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
6
insu ance cos o ehicle ype j is calcula ed by mul iplying he es i-
ma ed main enance and insu ance cos s pe pe iod by he expec ed
numbe o pe iods ha he ehicle will be in use, as shown in Equa ion
(8). Speci ically, Cins,j deno es he insu ance cos , Cmai,j ep esen s he
main enance cos , and
η
j e e s o he es ima ed se ice li e (in pe iods)
o ehicle ype j.
Co he ,j=(Cins,j+Cmai,j)*nj(8)
3.2.2.3. EV p icing. Wi h espec o EV p icing, his pape ollows he
app oach o Hu´
e ink e al. (2010) o hyd ogen ehicle p icing,
assuming ha EV p ices g adually dec ease as ma ke s ock inc eases, as
shown in Equa ion (9).
Pe, =Pe,0*(Qe,0
Qe, −1)
ω
(9)
Whe e Pe,0 is he ini ial p ice o EVs, and Qe,0 is he ini ial s ock o EVs.
Qe, −1 ep esen s he s ock o EVs in pe iod -1, and
ω
deno es he
lea ning a e, wi h highe alues o
ω
indica ing a as e decline in EV
p ices.
3.2.2.4. EV d i ing ange. To model he e olu ion o EV d i ing ange
o e ime, his pape adop s an S-shaped echnology li ecycle cu e o
ep esen echnological ma u i y, which in u n a ec s d i ing ange.
Since EV d i ing ange is closely ied o ad ances in ba e y and ela ed
echnologies (Z. Liu e al., 2023), and gi en ha he numbe o pa en s
can se e as a p oxy o echnological p og ess, we ollow he app oach
o Shen e al. (2021) by linking echnological ma u i y o d i ing ange.
In hei wo k, echnology di usion heo y is applied o i an S-cu e o
he g ow h o EV- ela ed pa en s, wi h he maximum numbe o pa en s
es ima ed. Technological ma u i y in each pe iod is hen calcula ed as
he a io o he cumula i e numbe o pa en s o his es ima ed
maximum. While he same me hod is adop ed in his s udy, adjus men s
a e made o accoun o di e ences in he simula ion pe iod: in Shen
e al. (2021), each simula ion s ep ep esen s one mon h, whe eas in his
s udy, each s ep co esponds o h ee mon hs. EV d i ing ange and
echnological ma u i y a e calcula ed using Equa ions (10) and (11),
espec i ely.
Re, =Re,0
(1−Tech +Tech0)2(10)
Tech =1
1+e−
τ
*[(3* −2)+667−θ](11)
Whe e Tech deno es he echnological ma u i y o EVs a ime , and
Tech0 is he ini ial le el o echnological ma u i y.
τ
ep esen s he
g ow h a e o he S-cu e, while θ indica es i s in lec ion poin — he
Fig. 1. The decision-making p ocess o consume s.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
7
ime a which he g ow h a e begins o slow. To a oid un ealis ic o
uncon olled g ow h in EV d i ing ange o e ime, an uppe bound is
also se o he maximum d i ing ange.
3.2.2.5. D i e popula ion. China’s au omo i e ma ke has expe ienced
s eady g ow h in bo h ehicle owne ship and he numbe o licensed
d i e s(Kong, 2025). Al hough he annual g ow h a e o new d i e s
has g adually declined, he ma ke is expec ed o emain in an expansion
phase un il app oxima ely 2028. F om a long- e m pe spec i e, he
Chinese au omo i e ma ke is p ojec ed o sus ain g ow h h ough 2050
(Hao e al., 2011). In his s udy, he simula ion pe iod co e s he
eal-wo ld ime ame om 2019 o 2031. To simpli y he model, and
based on he obse ed annual end, we assume ha he g ow h a e o
licensed d i e quan i y each pe iod is ime-in a ian and ep esen ed
by γd. The popula ion size o he simula ed space in pe iod , deno ed as
M , is calcula ed using Equa ion (12), while he numbe o newly added
po en ial consume s in each pe iod, ΔM , is de e mined using Equa ion
(13).
M =M −1*(1+γd)(12)
ΔM =M −M −1(13)
3.2.2.6. P CP ins alla ion eligibili y. The ins alla ion o P CP equi es
se e al condi ions, mos no ably su icien load capaci y and access o
dedica ed pa king spaces. To p omo e EV adop ion, he Chinese go -
e nmen issued a policy in 2015 manda ing ha newly buil neighbo -
hoods include p o isions o cha ging in as uc u e (Gene al O ice o
he S a e Council o he P.R.C, 2015).
14
Howe e , many communi ies
buil be o e 2015 we e no equipped wi h su icien elec ical in a-
s uc u e o suppo P CPs, and he majo i y o pa king spaces in hese
neighbo hoods do no mee he echnical equi emen s o P CP ins al-
la ion, pa icula ly in olde communi ies buil be o e 2000. These aging
communi ies accoun o nea ly 40 % o esiden ial a eas in 20 majo
ci ies ac oss China.
15
Based on he ongoing eno a ions o olde com-
muni ies, his s udy assumes ha he p opo ion o he popula ion
mee ing P CP ins alla ion condi ions will g adually inc ease, e lec ing
con inuous imp o emen s in esiden ial in as uc u e. Ne e heless,
due o he pe sis en sho age o dedica ed pa king spaces in China,
16
ou model cons ains his p opo ion om e en each 100 %. To simpli y
he modeling p ocess, i is u he assumed ha , in each pe iod, a
po ion o he popula ion wi hou P CP ins alla ion condi ions, deno ed
as M ,np, acqui es he necessa y condi ions a a ixed a e, γn. This
speci ica ion is o mula ed in Equa ion (14):
M +1,np =M ,np*(1−γn)(14)
Whe e γn ep esen s he con e sion a e o P CP ins alla ion eligibili y in
each pe iod.
3.3. EVCI ope a o sub-model
In his s udy, he EVCI ope a o e alua es bo h ope a ional pe o -
mance and ma ke cha ging demand o de e mine whe he o deploy
new PCSs. The ope a o ’s e enue p ima ily comes om cha ging ees
and go e nmen subsidies, while cos s consis o cons uc ion, land en ,
main enance, and elec ici y p ocu emen (Q. Zhang e al., 2018). ROI
se es as a comp ehensi e me ic o e alua ing he balance be ween
ope a ing p o i s and cos s (Sch oede and T abe , 2012). Al hough
cha ging demand is di ec ly linked o he numbe o EV use s, i is also
shaped by a complex in e play o cha ging p ices, elec ici y p ocu e-
men cos s, ope a ing expenses, and cons uc ion cos s (Q. Zhang e al.,
2018). The e o e, analyzing cha ging demand equi es a dual ocus on
bo h he scale o EV use s and he cha ac e is ics o he exis ing public
cha ging in as uc u e. The u iliza ion a e se es as an e ec i e indi-
ca o o he ela ionship be ween he numbe o consume s and he
numbe o PCSs ac oss di e en egions. By analyzing u iliza ion a es,
he ope a o can be e iden i y egional cha ging demand pa e ns and
adjus cha ging s a ion deploymen s a egies acco dingly. As he p i-
ma y objec i e o his pape is no o examine compe i i e dynamics
among ope a o s, he model is simpli ied by assuming a single ope a o
in he ma ke . The decision-making p ocess o his ope a o is illus a ed
in Fig. 2. I he ope a ing p o i in a gi en pe iod is posi i e, he ope a o
unde akes cos imp o emen measu es o PCSs. Based on he upda ed
cos s and cu en p o i , ROI is ecalcula ed. When ROI exceeds a p e-
de ined h eshold ROImin, he ope a o deploys addi ional s a ions.
Deploymen decisions a e guided by a compa ison o egional u iliza ion
a es wi h he o e all a e age, wi h new PCSs alloca ed in a eas showing
abo e-a e age demand.
3.3.1. Ope a ing p o i
The ope a ing p o i o PCSs mainly depends on se ice ees, pa king
ees, ope a ing cos s, OSs, and ad e ising e enue. Re enue om
pa king and se ice is di ec ly ied o consume cha ging demand—
highe demand leads o highe ope a ing p o i . A he same ime, as he
numbe o PCSs inc eases, ope a ing cos s ise co espondingly, while
ad e ising e enue also g ows due o he expansion o in as uc u e.
17
The ope a o ’s ope a ing p o i in pe iod ,
π
, is gi en by Equa ion (15).
π
=∑
n
i=1
Ec
i*(Pse +Ppa k)−∑
n
g=1
Cg
op +Sop, +Aads*∑
m
h=1
Nh,e, (15)
Whe e Ec
i ep esen s he elec ici y demand o use i du ing each pe iod
and is calcula ed in Equa ion (16). Aads deno es he ad e ising e enue
gene a ed by each PCS pe pe iod, and Cg
op is he ope a ing cos o PCS g
pe pe iod. Nh,e, indica es he numbe o he s a ions in egion h du ing
pe iod , while Sop, ep esen s he go e nmen OSs p o ided in pe iod .
Ec
i=Di
100*he(16)
Cg
op =γc*Cg
con (17)
As shown in Equa ion (16), he elec ici y demand o use i pe
pe iod, deno ed as Ec
i, is de e mined by he use ’s d i ing dis ance in ha
pe iod Di and he EV’s ene gy consump ion pe 100 km he. The ope a ing
cos o PCS g in each pe iod, Cg
op, is calcula ed based on i s cons uc ion
cos Cg
con and he a io o ope a ing o cons uc ion cos s γc, as exp essed
in Equa ion (17). Especially, Ec
i is ime-in a ian , and he cons uc ion
cos o PCS g, Cg
con, equals he uni cons uc ion cos in pe iod , Ccon, ,
when he s a ion was buil .
3.3.2. Resea ch and de elopmen cos
The g ow h in he numbe o EVs inc eases cha ging demand, which
equi es he ope a o o expand he numbe o PCSs o accommoda e
mo e use s. Howe e , he cons uc ion o new s a ions di ec ly a ec s
he ope a o ’s ROI. To add ess his, he ope a o alloca es a po ion o i s
unds o esea ch and de elopmen (R&D), wi h a pa icula ocus on
educing cons uc ion cos s o PCSs. Following he desc ip ion o ope -
a o R&D in es men in Sun e al. (2016), his s udy cons uc s a
14
This epo is accessed a : h ps://www.go .cn/zhengce/con en /2015-10
/09/con en _10214.h m.
15
The in o ma ion is om he Beike Resea ch Ins i u e epo : h ps:// ese
a ch.ke.com/121/A icleDe ail?id=274.
16
Rele an in o ma ion is accessed om: h ps://capi al.people.com.
cn/n1/2020/0708/c405954-31775313.h ml.
17
The ins alla ion o PCSs does no di ec ly a ec he ope a o ’s p o i . This is
because, based on ou ield in es iga ion, mos PCSs a e p o ided by au omo-
bile manu ac u e s a he ime o ehicle pu chase, wi h bo h equipmen and
ins alla ion se ices o e ed o use s ee o cha ge.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
8
simpli ied cos -imp o emen R&D model. In his model, he ope a o
alloca es a po ion o i s ope a ing p o i in pe iod o R&D in es men ,
deno ed as RD , as de ined in Equa ion (18).
RD =λ *
π
(18)
Whe e λ ep esen he p opo ion o p o i alloca ed o R&D in es men
in pe iod . The R&D in es men in pe iod , deno ed as RD , de e mines
he cons uc ion cos Ccon, o ha pe iod. Howe e , cons uc ion cos s
canno dec ease inde ini ely and a e subjec o a lowe bound, deno ed
as Cmin
con . The educ ion in cons uc ion cos o each pe iod, ΔCcon, , is
calcula ed using Equa ion (19).
ΔCcon, =
μ
*RD * *(Ccon, −1−Cmin
con )(19)
Whe e
μ
ep esen s he p opo ion o R&D in es men alloca ed spe-
ci ically o cos educ ion, and e lec s he e ec i eness o R&D in
lowe ing cons uc ion cos s. A e calcula ing he cos educ ion wi h
Equa ion (19), he ope a o upda es he cons uc ion cos acco ding o
Equa ion (20).
Ccon, =Ccon, −1−ΔCcon, (20)
3.3.3. Public cha ging s a ions deploymen
The deploymen o PCSs by EVCI ope a o s is p ima ily d i en by wo
ac o s: RIO and cha ging demand. The o me is e alua ed agains a
minimum accep able h eshold, ROImin, which mus be exceeded o new
in es men s o p oceed. The la e is assessed h ough egional u iliza-
ion a es.
18
Following Luo e al. (2023), ROI pe pe iod, ROI , is
calcula ed as he a io o ope a ing p o i o ini ial cons uc ion cos , as
exp essed in Equa ion (21).
ROI =
π
Ccon, − (z*K1*scon)(21)
Whe e z deno es he numbe o PCPs, K1 is he cha ging capaci y o a
single s a ion, scon ep esen s he CS o each s a ion, and
π
is he
a e age ope a ing p o i pe s a ion, which is calcula ed based on
Equa ion (22). Fo compu a ional simplici y, i is assumed ha all
cha ging s a ions ha e iden ical cha ging capaci y and an equal numbe
o cha ging piles, wi h a ia ions ac oss s a ions limi ed o cons uc ion
and ope a ion cos s.
π
=
π
∑
m
h=1
Nh,e,
(22)
When he ROI exceeds he h eshold ROImin, he ope a o p oceeds wi h
he cons uc ion o a new PCS. The o e all a e age u iliza ion a e o all
PCSs in he simula ed space, deno ed as , is calcula ed using Equa ion
(23).
=∑m
h=1∑n
i=1Ec
i,h
∑
m
h=1(Nh,e, *Emax)
(23)
Whe e Emax deno es he maximum cha ging olume ha a single s a ion
can o e pe pe iod. Once he o e all a e age u iliza ion a e is ob-
ained, he egional u iliza ion a e o egion h in pe iod , h , is
de e mined h ough Equa ion (24).
h, =∑n
i=1Ec
i,h
Nh,e, *Emax
(24)
The a e age u iliza ion a e o PCSs e ec i ely e lec s EV cha ging
demand wi hin a egion. A highe u iliza ion a e indica es s onge
demand and signals an unde supply o cha ging in as uc u e, he eby
indica ing he need o addi ional PCS deploymen . The ope a o iden-
i ies egions wi h u iliza ion a es highe han he a e age by
compa ing he egional a e age u iliza ion a e h, wi h he o e all
a e age , and deploys new PCSs in hese egions.
19
Howe e , he
expansion o PCSs is inhe en ly cons ained, as addi ional cons uc ion
and ope a ion lead o highe cos s. Gi en his ac , i is assumed ha he
ope a o deploys addi ional PCSs only in egions wi h abo e-a e age
demand, wi h he numbe o newly deployed s a ions calib a ed o
ensu e ha egional u iliza ion does no exceed he o e all a e age a e,
while no new PCSs a e deployed in egions wi hou excess demand. To
ensu e ha egional u iliza ion a es do no su pass he o e all a e age,
ope a o s apply Equa ion (25) o calcula e he numbe o addi ional
s a ions, ΔNh,e, , needed in each egion. The esul s a e ounded up o he
nea es in ege o a oid ac ional alues in ΔNh,e, . Since he p ima y
objec i e o his s udy is o examine he impac o public policies on EV
di usion a he han he ma ke e ec s o speci ic PCS loca ions, he
model assumes ha newly added PCSs a e andomly dis ibu ed wi hin
a ge ed egions. Fu he mo e, cha ging demand is assumed o be
e enly dis ibu ed ac oss all PCSs in each egion.
ΔNh,e, =⌈∑n
i=1Ec
i,h
*Emax
−Nh,e, ⌉(25)
3.4. Go e nmen sub-model
The go e nmen p o ides subsidies o bo h EV consume s and EVCI
ope a o , which aligns wi h eal-wo ld policy p ac ices. Consume s
ecei e EV pu chase subsidies, while ope a o ob ains wo o ms o
inancial suppo : OSs and CSs o cha ging in as uc u e.
3.4.1. EV pu chase subsidy
The simula ion begins in 2019, coinciding wi h he pe iod when EV
pu chase subsidies signi ican ly boos ed EV adop ion be o e being
Fig. 2. The decision-making p ocess o he EVCI ope a o .
18
Rele an in o ma ion is a ailable h ough: h ps:// esea ch.gszq.com/ ese
a ch/ epo ? id=8ae505846943c3450169537578e22c1a.
19
This s udy assumes ha egional u iliza ion a e dec eases as addi ional
cha ging s a ions a e buil in he egion.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
9
sho un. This phenomenon also sheds ligh on why many cha ging
in as uc u e p o ide s in China expe ienced inancial di icul ies du -
ing he ea ly s ages o EV ma ke expansion.
33
Faced wi h pe sis en
ope a ional losses, se e al cha ging in as uc u e p o ide s suspend
se ices, esul ing in he p oli e a ion o so-called “zombie cha ge s”,
which e e o he inac i e o non- unc ional PCSs ha hinde EV
di usion. In oducing OSs du ing his s age can help alle ia e he issue
by incen i izing con inued ope a ion. The e o e, a policy shi om CSs
o OSs eme ges as a necessa y measu e o p omo e he sus ainable
de elopmen o he cha ging in as uc u e indus y.
5.3. Impac analysis o subsidy phase-ou policies
5.3.1. Se ing o subsidy phase-ou modes
Gi en he subs an ial iscal bu den associa ed wi h cha ging in a-
s uc u e subsidies, hei g adual phase-ou has become an ine i able
policy end. Indeed, many Chinese ci ies ha e al eady announced
speci ic wi hd awal plans. Fo example, Shanghai has ou lined a wo-
s age educ ion o OSs beginning in 2025, wi h a ull e mina ion
scheduled by 2028. Hainan P o ince has e ained CSs exclusi ely o
Fig. 7. Tempo al e olu ion o EV pene a ion a e by egion unde baseline scena io.
Fig. 8. Tempo al e olu ion o EV pu chase p ice and d i ing ange.
33
Mo e de ails on his can be assessed a : h ps:// epo .i esea ch.cn/ epo
/202006/4456.sh ml.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
16

u al a eas, while lowe ing he subsidy a e om 10 %–15 % o 5 %–10
%. Simila ly, Chongqing has announced ha , beginning in 2026, i s
cu en OS o 0.1 RMB/kWh will be educed by an addi ional 20 %,
placing addi ional p essu e on EVCI ope a o s o enhance ope a ional
e iciency.
In he sensi i i y analysis o EVCI subsidies, we ind ha hei impac
on he di usion o EVs and EVCI is mos p onounced in he ea ly and
middle s ages o he simula ion (be o e pe iod 16). Building on his
obse a ion, and in line wi h eal-wo ld policy ends owa d subsidy
phase-ou s, we design wo al e na i e phase-ou scena ios ha begin in
he middle s age (pe iod 17), as summa ized in Table 6. Mode 1 (g adual
phase-ou ) assumes ha subsidies a e educed by 25 % om hei ini ial
le el s a ing in pe iod 17, ollowed by addi ional 25 % educ ions a
egula in e als un il a ull wi hd awal is achie ed by pe iod 37. Mode
2 ( apid phase-ou ) assumes ha subsidies a e immedia ely hal ed in
pe iod 17 and ully elimina ed by pe iod 33.
To u he in es iga e he e ec s o di e en ini ial subsidy le els on
EV adop ion, EVCI deploymen , and go e nmen expendi u e, his pape
adop s a policy-mix app oach. Speci ically, we combine mul iple ini ial
le els o CSs and OSs wi h he wo p e iously de ined phase-ou s a-
egies. By pai ing he i e CS le els and i e OS le els lis ed in Table 5
wi h he wo phase-ou modes, a o al o 50 dis inc policy combina ions
is yielded. Fo ins ance, Policy Combina ion 1 (0.05, 150) ep esen s a
scena io wi h an ini ial OS le el o 0.05 RMB/kWh and a CS le el o 150
RMB/kW, ollowed by a g adual phase-ou . Simila ly, Policy Combina ion
2 (0.25, 250) ep esen s ano he scena io wi h an ini ial OS le el o 0.25
Fig. 9. Tempo al e olu ion o ehicle ma ke s ock by ype and egion.
Fig. 10. Tempo al e olu ion o EVCI scale and densi y by ype and egion.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
17
RMB/kWh and a CS le el o 250 RMB/kW, unde a g adual phase-ou
s a egy as well. The ollowing sec ion p o ides a compa a i e anal-
ysis o ou comes ac oss all 50 policy combina ions.
5.3.2. Compa ison o policy combina ions
Fig. 16 p esen s he cumula i e subsidy expendi u e and EV s ock
le els a he end o he simula ion pe iod o each o he 50 policy
combina ions. O e all, he esul s indica e ha , ega dless o he phase-
ou mode, highe ini ial subsidy le els always lead o g ea e EV
owne ship. Howe e , when compa ing he wo phase-ou modes unde
iden ical ini ial subsidy con igu a ions, he g adual phase-ou mode
esul s in signi ican ly highe cumula i e go e nmen spending, on
a e age 22.30 % mo e han he apid phase-ou mode. By con as , he
co esponding inc ease in EV s ocks is negligible, wi h an a e age di -
e ence o only 0.028 %.
To e alua e he e iciency o di e en subsidy phase-ou s a egies,
which is de ined as achie ing highe EV adop ion while minimizing
Fig. 11. Tempo al e olu ion o EVCI ope a o p o i abili y and OS- o-P o i a io by egion.
Fig. 12. Tempo al e olu ion o cumula i e subsidies by ype and egion.
Table 5
Speci ica ion o OS and CS le els.
Subsidy
ype
Le el Uni
Ve y
low
Low Baseline High Ve y
high
OS 0.05 0.10 0.15 0.20 0.25 RMB/
kWh
CS 50 100 150 200 250 RMB/kW
L. Zhu e al.
T anspo Policy 175 (2026) 103876
18
go e nmen expendi u e, we conduc ed an addi ional analysis o he 50
policy combina ions. The baseline subsidy le el is de ined as an OS o
0.15 RMB/kWh and a CS o 150 RMB/kW. Ini ial subsidy le els below
his baseline a e ca ego ized as low subsidies, which include OS alues
o 0.05 RMB/kWh and 0.10 RMB/kWh, and CS alues o 50 RMB/kW
and 100 RMB/kW. Con e sely, OS alues o 0.20 RMB/kWh and 0.25
Fig. 13. Tempo al e olu ion o PCP quan i y and p opo ional di e ence by OS le el.
Fig. 14. Tempo al e olu ion o EV owne ship and p opo ional di e ences by OS le el.
Fig. 15. Tempo al e olu ion o PCP quan i y and ope a o p o i abili y pe cen age di e ences by CS le el.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
19
RMB/kWh, along wi h CS alues o 200 RMB/kW and 250 RMB/kW, a e
classi ied as high subsidies. Unde bo h phase-ou modes, we ocus on
wo dis inc g oups o policy combina ions: (1) low OS wi h high CS, and
(2) high OS wi h low CS. Fo ease o compa ison, hese combina ions a e
plo ed as connec ed poin s in Fig. 16. As shown in he igu e, policy
combina ions wi h high ini ial OS and low ini ial CS clus e in he uppe -
le egion o he g aph, whe eas hose wi h low ini ial OS and high
ini ial CS appea in he lowe - igh egion. The compa a i e analysis
e eals ha , on a e age, combina ions wi h high ini ial OC and low
ini ial CS educe cumula i e go e nmen expendi u e by 32.15 %, while
simul aneously achie ing 0.54 % highe EV owne ship compa ed o he
combina ions wi h low ini ial OS and high ini ial CS. Taken oge he ,
hese esul s sugges ha policy combina ions in ol ing a apid phase-
ou s a egy wi h high ini ial OS and low ini ial CS a e he mos cos -
e ec i e.
5.3.3. Compa ison o phase-ou modes
Fig. 17 p esen s he EV s ocks and EVCI quan i ies co esponding o
each o he 50 subsidy phase-ou policy combina ions. Fo ease o
compa ison, wo e e ence lines a e added o indica e he le els o EV
owne ship and EVCI deploymen unde he baseline non-phase-ou
scena io (whe e baseline subsidy le els a e main ained h oughou he
simula ion). These e e ence lines di ide he igu e in o ou sub-
egions. Policy combina ions alling in he uppe - igh egion esul in
inc eases in bo h EV adop ion and EVCI deploymen compa ed o he
baseline. This indica es ha , p o ided ini ial subsidy le els a e ela i ely
high, implemen ing a phase-ou policy does no hinde he di usion o
EVs o cha ging in as uc u e. Howe e , i is also no ewo hy ha he
gap in cumula i e go e nmen spending be ween phase-ou and non-
phase-ou policies is subs an ial. Compa ed wi h he baseline non-
phase-ou scena io, subsidy phase-ou combina ions educe o al go -
e nmen subsidy expendi u e by 83.94 %–97.19 %. In o he wo ds, he
uppe - igh egion o Fig. 17 highligh s a se o policy op ions ha no
only sus ain o enhance EV adop ion and EVCI deploymen , bu also
d ama ically alle ia e iscal bu dens.
Speci ically, we compa e EV owne ship and cumula i e subsidy
expendi u e unde he policy combina ion (0.15, 150) ac oss bo h
g adual and apid phase-ou scena ios, using he baseline non-phase-ou
scena io as a benchma k. Compa ed o he baseline, cumula i e subsidy
expendi u es dec ease by 90.07 % unde he g adual phase-ou and by
91.88 % unde he apid phase-ou . Meanwhile, changes in EV owne -
ship emain wi hin 0.05 %, a di e ence ha is p ac ically negligible.
These indings a e consis en wi h ou esul s in Sec ion 5.1, which
indica e ha he impac o subsidies on EV di usion is concen a ed in
he ea ly and middle s ages, wi h only limi ed e ec s obse ed in he
la e s ages. I cu en subsidy le els we e o be main ained h oughou
he en i e simula ion pe iod, he apid inc ease in EVCI deploymen
du ing he mid- o-la e s ages would esul in subs an ially highe
expendi u es on bo h OSs and CSs. By con as , he phase-ou policies
p oposed in his s udy subs an ially educe iscal bu dens wi hou
comp omising he long- e m di usion o EVs.
5.4. Impac analysis o ex e nal ac o s
5.4.1. Gasoline and elec ici y p ices
Fig. 18 illus a es he sensi i i y o EV owne ship o changes in
gasoline and elec ici y p ices. The esul s indica e ha bo h ac o s
signi ican ly in luence EV adop ion decisions. As shown in Fig. 18(a), EV
owne ship is posi i ely co ela ed wi h gasoline p ices, while Fig. 18(b)
illus a es a nega i e co ela ion be ween elec ici y p ices and EV
adop ion. These indings a e consis en wi h p e ious esea ch, such as
Luo e al. (2023) and Sha iei e al. (2012).
5.4.2. Elec ici y consump ion pe hund ed kilome e s
Fig. 19(a) shows he sensi i i y analysis o EV owne ship wi h
espec o a ia ions in elec ici y consump ion pe 100 km. Simila o
elec ici y p ices, EV owne ship is nega i ely co ela ed wi h elec ici y
consump ion pe 100 km, as bo h ac o s in luence consume pu chase
decisions h ough hei e ec on he d i ing cos o EVs. Meanwhile, as
shown in Fig. 19(b), lowe elec ici y consump ion pe 100 km is asso-
cia ed wi h a educ ion in he numbe o PCSs. This occu s because
imp o ed ene gy e iciency educes o e all cha ging demand, he eby
discou aging he u he deploymen o PCSs. In ligh o ongoing im-
p o emen s in ene gy e iciency, he cha ging in as uc u e indus y
should s a egically shi i s ocus om simply expanding he numbe o
acili ies owa d enhancing se ice quali y. Such a ansi ion would
signal he eme gence o a alue-d i en de elopmen phase, cha ac e -
ized by ewe bu mo e e icien ly ope a ed cha ging acili ies, a he
han con inued scale-o ien ed expansion.
6. Conclusions and implica ions
6.1. Conclusions
This pape examines he e ec s o EVCI subsidies and hei phase-ou
in China wi hin an ABM amewo k and yields h ee main conclusions:
(1) The Chinese EV ma ke is p ojec ed o sus ain apid g ow h, wi h
he pene a ion a e expec ed o each 79.78 % by 2030. A
no able end is he accele a ed di usion o EVs in subu ban
a eas, whe e EV owne ship is an icipa ed o e en ually su pass
ha o u ban a eas. In e ms o cha ging in as uc u e, P CPs a e
expec ed o emain dominan , while go e nmen subsidies will
con inue o play a c ucial ole in suppo ing he expansion o
public cha ging acili ies. Al hough subu ban a eas ini ially lag
behind u ban a eas in PCP deploymen , hey a e expec ed o
o e ake u ban le els as EV adop ion g ows, albei a a lowe
spa ial densi y. F om a p o i abili y pe spec i e, subu ban
cha ging piles may incu inancial losses in he ea ly s ages bu
a e likely o yield highe long- e m p o i s o ope a o s han hei
u ban coun e pa s.
(2) EVCI subsidies a e ound o be mos e ec i e in p omo ing EV
di usion du ing he ea ly and middle s ages, wi h hei in luence
diminishing in he mid- o-la e s ages. Speci ically, ou esul s
indica e ha aising he OC le el om he cu en 0.15 RMB/
kWh o 0.25 RMB/kWh can enhance cha ging pile deploymen by
up o 20 % in he ea ly and middle s ages bu yields only abou a
2 % inc ease in he la e s ages. While CSs can acili a e ea ly
in as uc u e expansion, hey isk causing o e in es men and
ope a ional ine iciencies, which may unde mine he long- e m
sus ainabili y o he indus y. I cu en subsidy le els we e
main ained wi hou phase-ou , annual go e nmen spending in
2030 would be app oxima ely 4.14 imes highe han in 2024.
These indings unde sco e he impo ance o shi ing om CSs o
Table 6
EVCI subsidy phase-ou modes.
Pe iod G adual subsidy phase-ou (Mode
1)
Rapid subsidy phase-ou (Mode 2)
OS CS Max OS OS CS Max OS
[1,16] Ini ial
le el
400 kW h/kW/
pe iod
Ini ial
le el
400 kW h/kW/
pe iod
[17,28] Ini ial
le el*75
%
300 kW h/kW/
pe iod
Ini ial
le el*50
%
200 kW h/kW/
pe iod
[29,32] Ini ial
le el*50
%
200 kW h/kW/
pe iod
Ini ial
le el*25
%
100 kW h/kW/
pe iod
[33,36] Ini ial
le el*25
%
100 kW h/kW/
pe iod
0 0 0
[37,52] 0 0 0 0 0 0
L. Zhu e al.
T anspo Policy 175 (2026) 103876
20
OSs and adop ing a phased educ ion in subsidy policies in he
la e s ages o ma ke de elopmen .
(3) To assess he e ec i eness o subsidy phase-ou s a egies, his
s udy cons uc s 50 policy combina ions by a ying ini ial CS and
OS le els ac oss wo phase-ou modes. The simula ion esul s
indica e ha main aining cu en subsidy s anda ds while
adop ing a phase-ou mode can educe cumula i e go e nmen
spending by 91 % compa ed o a no-phase-ou scena io, while
esul ing in only a ma ginal 0.05 % decline in EV owne ship. A
compa ison o he 50 policy combina ions u he e eals ha ,
unde iden ical ini ial CS and OS le els, he apid phase-ou mode
educes o al expendi u e by 22.30 % compa ed o he g adual
mode. Mo eo e , a policy combina ion wi h high ini ial OS and
low ini ial CS lowe s cumula i e expendi u e by 32.15 %
compa ed o he opposi e se up (a combina ion wi h low ini ial
OS and high ini ial CS). In summa y, hese indings sugges ha
he mos cos -e ec i e policy mix consis s o a apid phase-ou
mode combined wi h high ini ial OS and low ini ial CS, as i
signi ican ly educes iscal expendi u e wi hou impeding EV
di usion.
6.2. Implica ions
Ou indings p o ide insigh s o policymake s and gene a e p ac-
ical guidance o EVCI ope a o s. Based on he simula ion esul s, he
ollowing speci ic ecommenda ions a e p oposed:
(1) When designing subsidy policies, go e nmen s should accoun
o he g owing iscal bu den a ising om apid EV di usion. In
he ea ly s ages o ma ke de elopmen , ci ies wi h adequa e
iscal budge s and an u gen need o accele a e EV adop ion may
implemen high le els o bo h CSs and OSs. Such a s a egy
Fig. 16. Cumula i e subsidy expendi u e and EV owne ship ac oss policy combina ions.
Fig. 17. Compa ison o phase-ou policy combina ions and he baseline non-phase-ou scena io.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
21

suppo s he apid deploymen o public cha ging in as uc u e
and helps ope a o s o e come p o i abili y challenges in he ea ly
s age. Howe e , i is also essen ial o ha e s ic supe ision o
CSs o he sake o e iciency in using public esou ces. In o-
ducing capaci y u iliza ion assessmen s o ope a o s can help
p e en o e in es men in egions wi h limi ed cha ging de-
mand. Fo ci ies acing igh e iscal cons ain s, a s a egy wi h
ela i ely highe OS and lowe CS, wi h a ocus on enhancing
ope a ional pe o mance, would be mo e app op ia e. I is also
ecommended o link subsidy amoun s o key pe o mance in-
dica o s such as u iliza ion a es and aul a es, which can
acili a e he iden i ica ion and elimina ion o unde pe o ming
cha ging s a ions. As he ma ke de elops, go e nmen s should
adop a dynamic subsidy phase-ou mechanism ha balances he
educ ion o subsidy expendi u e and he p omo ion o EV
di usion.
(2) To p omo e balanced EVCI de elopmen ac oss u ban and sub-
u ban a eas, policymake s should accoun o egional he e o-
genei y and adop a ge ed measu es. In e ms o subsidy policy
design, g ea e inancial suppo should be alloca ed o subu ban
EVCI o accele a e in as uc u e deploymen and enhance
ope a o p o i abili y du ing he ea ly s ages o ma ke de elop-
men . On he echnical and planning aspec s, he go e nmen can
le e age he ela i e ease o P CP ins alla ion in subu ban a eas
by expanding cha ging pile capaci y and s eamlining ins alla ion
equi emen s. Such e o s would help add ess issues such as
limi ed g id capaci y and cumbe some app o al p ocesses ha
cu en ly cons ain P CP expansion. Th ough hese measu es,
policymake s can os e a subu ban EVCI de elopmen mode
cen e ed on P CPs, wi h PCSs playing a complemen a y ole,
he eby ully le e aging he po en ial o P CPs while o e coming
key ba ie s o EV and EVCI adop ion in subu ban a eas.
(3) EVCI ope a o s should ake ad an age o he cu en subsidy
window by p io i izing cha ging pile deploymen in high-
po en ial subu ban a eas, le e aging high OS le els o o se
inancial losses in he ea ly s ages. A he same ime, imple-
men ing dynamic p icing s a egies can help enhance u iliza ion
a es o piles and mi iga e he impac o u u e subsidy phase-
ou s. Beyond p icing, ope a o s should explo e inno a i e busi-
ness models o educe ope a ional and main enance cos s in
subu ban a eas. Fo example, gi en he widesp ead p esence o
P CPs in subu ban a eas, ope a o s could collabo a e wi h
homeowne s o pilo sha ed cha ging schemes. Fu he mo e,
conside ing he unique cha ac e is ics o subu ban land owne -
ship, a c owd- unded ins alla ion model, whe e homeowne s
p o ide space and EVCI ope a o s con ibu e equipmen and
echnical expe ise, could be an e ec i e solu ion.
Fig. 18. EV owne ship unde a ying gasoline and elec ici y p ices.
Fig. 19. EV owne ship and PCP quan i y unde a ying elec ici y consump ion le els.
L. Zhu e al.
T anspo Policy 175 (2026) 103876
22
6.3. Limi a ions and u u e wo k
This s udy is subjec o se e al limi a ions. Fi s , i emains chal-
lenging o accu a ely model a ia ions in EVCI capaci y due o he
complexi y o in luencing ac o s and he he e ogeneous composi ion o
EVCI wi h di e en capaci y le els. As a esul , EVCI capaci y in his
s udy is se o be s a ic (exogenous) a he han dynamically e ol ing.
Fu u e esea ch could apply al e na i e o ecas ing app oaches o be e
cap u e he dynamic e olu ion o cha ging pile capaci y.
Second, he elec ici y consump ion pe 100 km o EV d i ing is also
ea ed as an exogenous a iable in his s udy because o he lack o
eliable o ecas ing me hods, and hus a sensi i i y analysis was con-
duc ed in his s udy o pa ially add ess his limi a ion. In addi ion, his
model ea s EV cha ging ime and ICEV a ibu es (such as ene gy
consump ion pe 100 km o d i ing, d i ing ange, and ehicle p ice) as
s a ic. This simpli ica ion may lead o ei he o e es ima ion o unde -
es ima ion o he long- e m a ac i eness o EVs. Fu u e esea ch could
add ess his limi a ion by de eloping dynamic models o bo h EVs and
ICEVs o imp o e accu acy and mo e closely e lec eal-wo ld echno-
logical p og ess and usage condi ions.
Thi d, au omobile manu ac u e s a e no inco po a ed in o he
model due o i s complexi y, as he p ima y ocus o his s udy is on
EVCI. Fo he sake o simpli ica ion, his pape also ollows p e ious
s udies by conside ing only a single ope a o wi hou accoun ing o
compe i ion among mul iple ope a o s. Fu u e esea ch could add ess
hese limi a ions by inco po a ing au omobile manu ac u e s and
in oducing compe i i e mechanisms among mul iple ope a o s o mo e
accu a ely e lec he e olu ion o he EV ma ke in he eal wo ld.
CRediT au ho ship con ibu ion s a emen
Lijing Zhu: Concep ualiza ion, Me hodology, W i ing – o iginal
d a . Runze Li: Da a cu a ion, Me hodology. Jingzhou Wang: So -
wa e, Valida ion. Haibo Chen: Da a cu a ion. Ond ej Ha an: W i ing
– e iew & edi ing. Wen-Long Shang: Supe ision, W i ing – o iginal
d a .
Decla a ion o compe ing in e es
The au ho s decla e ha hey ha e no known compe ing inancial
in e es s o pe sonal ela ionships ha could ha e appea ed o in luence
he wo k epo ed in his pape .
Acknowledgemen s
This wo k was suppo ed by Minis y o Educa ion in China (MOE)
P ojec o Humani ies and Social Sciences (No. 24YJA790105) and
Beijing Na u al Science Founda ion, China (No. 9232003). Besides, his
esea ch was also pa ially suppo ed by he ZEV-UP and ePowe Mo e
p ojec s co- unded by he Eu opean Union unde G an ag eemen ID:
101138721 and 101192753.
Appendix A
Table A1
Sample cha ac e is ics o su ey esponden s
Va iable Ca ego y Sample size Pe cen age
Gende Female 305 57.2 %
Male 228 42.8 %
Age 18–24 78 14.6 %
25–34 244 45.8 %
35–44 162 30.4 %
45–54 37 6.9 %
≥55 12 2.3 %
Income <100,000 RMB 73 13.7 %
100,000–150,000 RMB 154 28.9 %
150,000–200,000 RMB 131 24.6 %
200,000–300,000 RMB 109 20.4 %
≥300,000 RMB 66 12.4 %
Place o esidence Cen al U ban a ea 261 49.0 %
Inne subu ban a ea 195 36.6 %
Ou e subu ban a ea 30 5.6 %
Ru al a ea 47 8.8 %
EV owne ship and P CP ins alla ion s a us Yes 173 32.5 %
EV only 111 20.8 %
No 249 46.7 %
P CP cha ging ime
(P CP +PCS)cha ging ime
No applicable 360 67.5 %
≤20 % 14 2.6 %
21–40 % 54 10.1 %
41–60 % 51 9.6 %
61–80 % 35 6.6 %
>80 19 3.6 %
Vehicle pu chase in en ion in he nex yea Yes 385 72.2 %
No 148 27.8 %
Annual d i ing mileage (uni : 10,000 km) <1 64 12.0 %
1–1.5 116 21.8 %
1.5–2 250 46.9 %
2–3 89 16.7 %
≥3 14 2.6 %
L. Zhu e al.
T anspo Policy 175 (2026) 103876
23
Table A2
Income cha ac e is ics o esponden s om di e en egions
Va iable Ca ego y Sample size Pe cen age
Income (U ban a ea) <100,000 RMB 18 6.9 %
100,000–150,000 RMB 57 21.9 %
150,000–200,000 RMB 70 26.9 %
200,000–300,000 RMB 68 26.2 %
≥300,000 RMB 47 18.1 %
Income (Subu ban and u al a eas) <100,000 RMB 52 19.0 %
100,000–150,000 RMB 98 35.9 %
150,000–200,000 RMB 63 23.1 %
200,000–300,000 RMB 41 15.0 %
≥300,000 RMB 19 7.0 %
Table A3
Geog aphic Dis ibu ion o he Responses
P o ince Sample size P o ince Sample size
Beijing 29 Hubei 32
Tianjin 10 Hunan 15
Hebei 32 Guangdong 57
Shanxi 18 Guangxi 6
Inne Mongolia 7 Hainan 2
Liaoning 23 Chongqing 11
Jilin 7 Sichuan 19
Heilongjiang 8 Guizhou 16
Shanghai 23 Yunnan 9
Jiangsu 40 Shaanxi 13
Zhejiang 30 Gansu 8
Anhui 15 Qinghai 1
Fujian 25 Ningxia 2
Jiangxi 10 Xinjiang 2
Shandong 40 Henan 23
Table A4
Co ela ion ma ix o demog aphic a iables
Income Annual d i ing
mileage
Minimum accep able EV
d i ing ange
Maximum accep able ehicle
pu chase p ice
Sha e o cha ging ime a
P CP
Income 1   
Annual d i ing mileage 0.259*** 1   
Minimum accep able EV d i ing
ange
0.275*** 0.106*** 1  
Maximum accep able ehicle
pu chase p ice
0.614*** 0.242*** 0.329*** 1 
Sha e o cha ging ime a P CP 0.254*** 0.176*** 0.112*** 0.306*** 1
No es: *p <0.1; **p <0.05; ***p <0.01.
Appendix B
Table B1
Desc ip ion o pa ame e s and a iables
Ca ego y Symbol Desc ip ion Symbol Desc ip ion
Pa ame e s P ICEV pu chase p ice Pj,0Ini ial pu chase p ice o ehicle ype j
Rj,0Ini ial d i ing ange o ehicle ype j TjRe ueling/ echa ging ime o ehicle ype j
heElec ici y consump ion pe 100 km o d i ing o an EV h Fuel consump ion pe 100 km o an ICEV
Pelc Composi e elec ici y p ice a a public cha ging s a ion Ppa k Pa king ee while cha ging
Pse Se ice ee pe uni (kWh) o elec ici y cha ged Pp i Cha ging p ice o p i a e cha ging piles
Pgas Gasoline p ice Kpub Capaci y o public cha ging s a ion
Kp i Capaci y o p i a e cha ging s a ion
σ
sSha e o cha ging olume in hou s
njEs ima ed se ice li e o ehicle ype j Cins.jInsu ance cos o ehicle ype j pe pe iod
Cmai,jMain enance cos o ehicle ype j pe pe iod θIn lec ion poin o he echnological ma u i y cu e
τ
G ow h a e o he echnological ma u i y cu e γdG ow h a e o licensed d i e quali y pe pe iod
γnCon e sion a e o p i a e cha ging pile ins alla ion eligibili y pe pe iod Aads Ad e ising income o a public cha ging s a ion pe pe iod
γcOpe a ing- o-cons uc ion cos a io pe pe iod Cmin
con Minimum cons uc ion cos o a cha ging s a ion
(con inued on nex page)
L. Zhu e al.
T anspo Policy 175 (2026) 103876
24
Table B1 (con inued)
Ca ego y Symbol Desc ip ion Symbol Desc ip ion
zNumbe o cha ging piles in a public cha ging s a ion sop Ope a ional subsidy a e pe uni o elec ici y cha ged
scon Cons uc ion subsidy o each public cha ging s a ion sop,max Maximum subsidized cha ging olume pe uni o cha ging pile
capaci y (kW)
Emax Maximum cha ging olume o a s a ion pe pe iod  
Va iables Ui,j, U ili y o use i om ehicle ype j in pe iod Pe, Pu chase p ice o an EV in pe iod
Qj, Ma ke s ock o ehicle ype j in pe iod Re, D i ing ange o an EV in pe iod
spu , EV pu chase subsidy in pe iod Cd i,i, Cos pe 100 km o d i ing o consume i when using ICEV
Cd i,i,e,pub Cos pe 100 km o d i ing o consume i wi hou p i a e cha ging piles
when using EV
Cd i,i,e,p i Cos pe 100 km o d i ing o consume i wi h p i a e cha ging
piles when using EV
Ni,j, Numbe o cha ging o gas s a ions wi hin a 5-km adius o consume i wi h
ehicle ype j in pe iod
BiP i a e cha ging piles access indica o (1 i yes, and
0 o he wise)
yiTime o cha ging a PCS/(Time o cha ging a PCS +Time o cha ging a
P CP)
P i,j, Pu chase p obabili y o ehicle ype j in pe iod o consume i
Tech Technological ma u i y o EVs in pe iod ΔM Newly added licensed d i e s in pe iod
M Numbe o licensed d i e s in pe iod M ,np Numbe o consume s wi hou p i a e cha ging pile ins alla ion
eligibili y in pe iod
π
Ope a ing p o i o EVCI ope a o in pe iod Ec
iElec ici y demand o use i pe pe iod
Cg
op Ope a ing cos o public cha ging s a ion g pe pe iod Nh,e, Numbe o public cha ging s a ions in egion h in pe iod
DiD i ing dis ance o consume i pe pe iod Cg
con Cons uc ion cos o public cha ging s a ion g
RD R&D in es men in pe iod λ R&D in es men sha e o p o i in pe iod
μ
Sha e o R&D o cos educ ion
ν
E ec i eness o R&D in lowe ing cons uc ion cos s
Ccon, New public cha ging s a ion cons uc ion cos in pe iod ΔCcon, Public cha ging s a ion cons uc ion cos educ ion in pe iod
ROI Ra e o e u n o EVCI ope a o in pe iod
π
A e age ope a ing p o i o a public cha ging s a ion in pe iod
Sop,h, Amoun o ope a ional subsidy in egion h in pe iod Sop, To al amoun o ope a ional subsidy in pe iod
Scon, To al amoun o cons uc ion subsidy in pe iod ΔNh,e, Numbe o newly buil cha ging s a ions in egion h in pe iod
O e all a e age u iliza ion a e o public cha ging s a ions ac oss he
simula ed egion in pe iod
h, Regional a e age u iliza ion a e o public cha ging s a ions in
egion h in pe iod
Appendix C
Table C1
Consume a ibu e se ings
Pa ame e Desc ip ion Value Sou ce
IAnnual income (uni : 10,000 RMB) U ban a ea Uni o m [5,10),
p ob =7 %
The su ey s udy in Appendix A
Uni o m [10,15),
p ob =22 %
Uni o m [15,20),
p ob =27 %
Uni o m [20,30),
p ob =26 %
Uni o m [30,100],
p ob =18 %
Subu ban
a ea
Uni o m [5,10),
p ob =19 %
Uni o m [10,15),
p ob =36 %
Uni o m [15,20),
p ob =23 %
Uni o m [20,30),
p ob =15 %
Uni o m [30,100],
p ob =7 %
DiD i ing dis ance o use i pe pe iod
(uni : 1000 km)
Uni o m [1.25,3.75] i I =[5,10) The su ey s udy in Appendix A
Uni o m [2.50,5.00] i I =[10,15)
Uni o m [3.75,6.25] i I =[15,20)
Uni o m [4.375,6.875] i I =[20,30)
Uni o m [5.00,7.50] i I =[30,100]
yiP opo ion o use ’s o al cha ging
ime a PCP
Uni o m [0.6,0.8] i I =[5,10) The su ey s udy in Appendix A
Uni o m [0.5,0.7] i I =[10,15)
Uni o m [0.4,0.6] i I =[15,20)
Uni o m [0.3,0.5] i I =[20,30)
Uni o m [0.2,0.4] i I =[30,100]
Pke Maximum accep able EV pu chase
p ice (uni : 10,000 RMB)
1.8 *I a. The su ey s udy in Appendix A
b. (Sun e al., 2018)
Pk Maximum accep able ICEV pu chase
p ice (uni : 10,000RMB)
1.4 *I a. The su ey s udy in Appendix A
b. (Sun e al., 2018)
RkMinimum accep able EV d i ing
ange (uni : km)
200 i I =[5,10) The su ey s udy in Appendix A
350 i I =[10,15)
(con inued on nex page)
L. Zhu e al.
T anspo Policy 175 (2026) 103876
25