Transport Policy 37 (2015) 44–56

Contents lists available at ScienceDirect

Transport Policy
journal homepage: www.elsevier.com/locate/tranpol

Simulating the market penetration of cars with alternative

fuelpowertrain technologies in Italy
Eva Valeri n, Romeo Danielis
Department of Economics, Business, Mathematics and Statistics (DEAMS), Univesity of Trieste, P.le Europa, n. 1, 34100 Trieste, Italy

art ic l e i nf o a b s t r a c t

Article history: This paper evaluates the market penetration of cars with alternative fuelpowertrain technologies in Italy
Received 26 February 2014 under various scenarios. Seven cars on sale in 2013 are considered: the Ford Fiesta (diesel), the VW Polo
Received in revised form (gasoline), the Fiat Punto Evo (bi-fuel – CNG), the Natural Power Alfa Romeo Mito (bi-fuel – LPG), the
16 July 2014
Toyota Yaris (hybrid – gasoline), the Peugeot iOn (BEV – owned battery), the Renault Zoe (BEV – leased
Accepted 10 October 2014
Available online 11 November 2014
battery). A Mixed Error Component Logit model is estimated based on data collected via a stated pre-
ference choice survey administered in 2013 in various Italian cities. The model's parameters are then used
Keywords: to build a Monte Carlo simulation model which allows evaluating, under different scenarios, the market
Car choice penetration of the seven cars. The main findings are that (a) the subsidies enacted by the Italian gov-
Market scenario ernment in favour of the low CO2 emitting cars appear to favour mostly the Ford Fiesta (diesel); (b) a
Stated preference
three-fold increase in the BEVs range would not change their market share significantly (about 2%); and
Discrete choice
(c) only a combination of changes such as the introduction of a subsidy equal to €5000, the decrease of
Policy simulation
the purchase price for BEVs by €5000, the increase in the battery range, and the increase in the con-
ventional fuel price would significantly increase the BEVs' market share, raising it to about 15%.
& 2014 Elsevier Ltd. All rights reserved.

1. Introduction opposed to the more traditional and largely established gasoline

and diesel cars.
The car market is extremely diversified by segment and by In our view, in relation to these developments, there are some
fuelpowertrain technology. Auto manufacturers continuously im- very relevant research questions such as: how will the consumers
prove their models and increase their differentiation in an effort to react to the new fuelpowertrain technologies? Will they gradually
satisfy customers' needs, to gain a competitive edge over their penetrate the market and with which speed? In which market
competitors and to meet the various energy and environmental segments? Do public policies interested in gaining independence
regulatory constraints. An interesting recent development is the from oil, reducing energy consumption or local and global pollu-
process of electrification that is gradually gaining momentum in tion influence consumers' choices and by how much? All these
the passenger car market. The hybrid, the plug-in and the battery questions are of great significance to the auto manufacturers, to
the policy makers and to the general public and, most likely, the
electric vehicle (HEV, PHEV, BEV) are the new entrants in the car
answers will be quite differentiated among countries and popu-
market with gradually growing market shares, with the hydrogen
lation segments. The discrete choice methodology, based on data
fuel cell vehicles (HFCV) in the process of going from the concept
collected at individual level on both stated and revealed choices, is
stage to the manufacturing one. These engine technologies, to-
a largely established methodology which could help understand
gether with the Compressed Natural Gas vehicles (bi-fuel CNGV) and predict consumers' choices.
and the Liquefied Petroleum Gas Vehicles (bi-fuel LPGV) ones, Our specific interest is to provide some answers for the Italian
make up what is known as Alternative Fuel Vehicles (AFV) as car market, a task not yet performed in the scientific literature.
Compared with other western countries, Italy appears to be lag-
Abbreviations: ; AFV, alternative fuel vehicle; BEV, battery electric vehicle; BV, ging behind in the penetration of the AFVs, though having high air
biofuel vehicle; CNGV, compressed natural gas vehicle; CV, conventional vehicle and noise pollution levels and a strong economic dependence from
(including SGV and DV); DV, diesel vehicle; HEV, hybrid electric vehicle; HFCV, oil imports. Only bi-fuel CNGVs and bi-fuel LPGVs have recently
hydrogen fuel cell vehicle; LPGV, liquefied petroleum gas vehicles (bi-fuel); SGV, gained relevant market shares in some regions of the country.
standard gasoline vehicle
n
Corresponding author.
This paper focuses on seven specific cars on sale in Italy with
E-mail addresses: [email protected] (E. Valeri), the following fuelpowertrain technologies: the Ford Fiesta (diesel),
[email protected] (R. Danielis). the VW Polo (gasoline), the Fiat Punto Evo (bi-fuel – CNG), the

http://dx.doi.org/10.1016/j.tranpol.2014.10.003
0967-070X/& 2014 Elsevier Ltd. All rights reserved.
 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56 45

Natural Power Alfa Romeo Mito (bi-fuel – LPG), the Toyota Yaris

Forecasting predict a BEVs market increase to 2.5% in 10 years (2020) and to

results from “Expected” scenario A show that BEVs obtain a market share of

estimate a new car market share of BEVs of 0.04% and 0.7% for HEVs (2010).

In Denmark a 3% BEV market share is not reached until 2021 but after this
year, a higher increase is seen each year. The highest increase in the share of
Chinese consumers are more likely to consider switching from SGVs to HEVs

5.7% after 15 years (2025). HEVs have a considerable higher share compared
In forecasting analysis only two car types are compared (CV and BEV). The

Austrian consumers prefer CVs compared with both BEVs and HEVs. They
(hybrid – gasoline), the Peugeot iOn (BEV – owned battery), the

The market share of CGV is 68.3%, Renault CV (gasoline or diesel) cars is

48%. In “Probable” scenario B this market share has increased to 72%.
Renault Zoe (BEV – leased battery). The data collected via a Stated
Preference (SP) choice survey and administered in 2013 in three
Italian cities are used to estimate a Multinomial Logit (MNL) model
and a Mixed Error Component Logit (MECL) model. The estimated
parameters are used in a Monte Carlo simulation model to eval-
uate, under different scenarios, the potential market penetration
of the seven cars in Italy.
Notwithstanding the national dimension of the market ana-
lysed, the techniques used in the paper to collect the data, to es-
timate the econometric model and to perform policy simulation
represent a potential contribution to the international literature,

Forecasting market penetration

and the results obtained might be of interest not only for the

adopters is seen in 2023 (6%).

Italian policy makers.

4.6%, Renault BEVs is 27.1%.

The paper is organised as follows: a review of the literature is

with the BEVs (11.5%).

presented in Section 2; the SP experiment is illustrated Section 3,
the descriptive, econometric and simulation results are reported
and discussed in Section 4, and the conclusions and policy im-

than to BEVs.
plications are drawn in Section 5.

2. Literature review

Purchase price, annual running cost, incentives, availability

Purchase price, cost of driving 100 km, governmental in-

performance, CO2 emission, charging options, EV battery

CV, HFCV, HEV, BV, Purchase price, annual cost, Range, refuelling frequency,

Purchase price, range, charging time, running cost, CO2

Purchase price, driving distance, driving costs, driving

Car choice and its use is certainly not a new topic. It has been

of charging facilities, vehicle range with full charging

studied extensively since the pioneering contributions by Lave and

centive, monthly cost of leasing the BEV battery

Train (1979), Manski and Sherman (1980), Berkovec and Rust
(1985). Yet, it is still a difficult topic as the paper by Daziano and
Chiew (2012) theoretically, methodologically and empirically de-
monstrates. From a theoretical point of view, the car is a durable
good to be used for mobility reasons but with obvious social and
cultural implications. The importance attributed to the car changes
over time and varies among individuals: from a symbol of in-
acceleration time, service

emission, engine power

dependence of the early years, to a status symbol of driving a
Ferrari or a SUV, to a mere tool to be used and not owned with
carsharing. Car choice is also a family matter, discussed with the
other family members, in a context where more than one car is
Attributes

available. From a methodological point of view, various ap-

lifetime
proaches has been used to study car choice and car use, including
(macro) time series analysis, (macro) diffusion models, (micro)
Overview of selected SP studies on consumer preferences and market penetration of AFVs.

economic preference models and attitudinal (hybrid) models.

CGV, Renault CV,

Empirically, with special reference to the last two types of models,

SGV, HEV, BEV

CVn, HEV, BEV

and considering the newly introduced engine technologies (hy-

Renault BEV
Car types

brid, plug-in, electric, fuel cell) the SP data have prevailed over the
CV, BEV

Notes: conventional vehicles (CV, including SGV and DV), n ¼ own vehicle.

revealed preference ones, simply because these are a not yet-ex-

BEV

isting- or sufficiently widespread technology.

The focus of this review is on the economic preference and
SP logit choice model and Re-

attitudinal (hybrid) models, based on individual data, aimed at

SP Multinomial Logit Model

Hybrid choice and diffusion

SP Mixel Logit Model and

forecasting or policy analysis more than at analysing the pre-

Qian and Soopramanien (2011) SP MNL and NL model

ference structure.
Table 1 lists some recent studies focusing on forecasting market
nault market data

penetration. Qian and Soopramanien (2011), with reference to the

Methodology

Chinese market, reach the conclusion that the Chinese consumers

simulation

are more likely to consider switching from SGVs to HEVs than to

model

BEVs. Mabit and Fosgerau (2011) estimated a ML model with four

alternatives (CV, HFCV, HEV, BV, BEV) and performed a market
penetration analysis taking into account the CV and the BEV. They
Glerum and Themans (2011)

reach the optimistic conclusion that the BEV could gain a market
Mabit and Fosgerau (2011)

share between 48% and 72%. On the contrary, Link et al. (2012)
predict that the BEVs could increase in Austria to a modest 2.5% in
Jensen et al. (2014)

2020 and to 5.7% in 2025, whereas HFCVs have a considerable

Link et al. (2012)

higher potential (11.5%). Glerum and Themans (2011) use a logit

model to analyse the market potential of the Renault cars com-
Authors

pared to the other brands. They conclude that the Renault BEVs
Table 1

have a stronger potential that the conventional Renault cars.

Jensen et al. (2014) add to the hybrid model a diffusion model in
 46
Table 2
Overview of selected SP studies with policy simulations.

Authors Methodology Car types Attributes Evaluated policy Simulations

Ewing and Sarigöllü SP Multinomial logit SGV, more fuel-effi- Purchase price, maintenance cost, refuel- 1) Subsidies for purchase, 2) Better electric batteries. The results reveal a large potential demand for
(1998) model cient gasoline or ling time, emission rate, acceleration, cleaner fuel-efficient and BEVs among suburban
AFV, BEV range, commuting cost, commuting time Montreal car commuters, if these can compete with
CVs in price and performance. AVF from 51 in the
base case scenario to max 57. BEV from 24 in the
base case scenario to 50, when range increases to
300 miles.
Horne et al. (2005) MNL parameters derived SGV, CNG, HEV, Capital cost, fuel cost, fuel availability, ex- 1) A $50/tonne carbon tax; 2) SGV disincentives The impact on CO2 emissions is simulated. Policy 3
from the literature used in BEV, HFC press lane access, power (increasing the fuel availability of methanol, ethanol, (SOV disincentives) is found to be more effective,
CIMS natural gas, propane, diesel, and hydrogen fuel cell Policy 1 (carbon tax) and 2 (gas disincentives) would
vehicles by 25%, giving express lane access to HFCVs not generate a significant reduction.
and BEVs, and increasing the power of every vehicle
except diesels and low-efficiency gas. A surcharge of
$1000, and $3000 was also applied to high, and low-
efficiency gasoline vehicles, respectively), 3) Single
occupancy vehicle disincentives.

E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56

Achtnicht et al. SP Nested logit model DV, SGV, HEV, Purchase price, variable cost, engine 1) Service station availability. In the case of a full service station network of all
(2008) LPGV/CNGV, BV, power, CO2 emissions, service station technologies, LPG/CNG and hydrogen cars would
BEV, HFC network, fuel type achieve substantial market shares (about 14%),
whereas the biofuels and electric power trains only
10%.
Daziano and Bolduc Baysian hybrid choice SGV, AFV, HEV, HFC Purchase price, monthly operating costs, 1) Increasing in fuel network density; 2) Tax on Relative to the baseline scenario (11.35% SVG, 3.73%
(2013) model fuel availability, express lane access, fossil fuel; 3) Purchase subsidies; 4) Increase price of AFV, 48.85% HEV and 36.07% HFC), the increase fuel
emissions data, power HEV and HFC by 50%; 5) Concerns and awareness for network density for the AFV and HFC to a level
the natural environment (women only). equivalent to the SGV, would increase their share to
5.16% and 50%, respectively. A 25% fuel tax increase
to 4.1% and 38%. A purchase price subsidy on green
vehicles would produce only a minor effect to 3.79%
and 35.55%. A social marketing campaign would
have a large impact, raising the market share of AFV
to 6% and of HFC to 48.6%. Women have a slightly
higher share of AFV equal to 3.9%.
Daziano and Acht- Bayes estimates of a mul- SGV, DV, HEV, Purchase price, fuel costs, engine power, 1) Increasing qualitatively charging infrastructure Relative to the baseline scenario derived from the
nicht (2013) tinomial probit model LPGV/CNGV, BV, CO2 emissions, fuel availability, and fuel density. experimental market shares of SP studies (SGV
HFC, BEV type 21.9%, LPG/CNG 11.43%, HEV 29.97%, EV 2.78%, BV
6.01%, HFC 7.28%, DV 26.55%), the density of the
charging network for EVs or HFC, would raise their
share up to 11.97% and 22.44%, respectively.
Hackbarth and MNL, Mixed error compo- CV (SGV and DV) Purchase price, fuel cost, driving range, 1) Incentives (vehicle circulation tax exemption, bus Base case: CV-30.35%, NGV-17.82%, HEV-20.08%,
Madlener (2013) nent Logit Model. CNGV, HEV, PHEV, fuel availability, CO2 emissions, refuelling lane access, and free parking) for PHEVs, BEVs, BVs, PHEV-10.85%, BEV-2.24%, BV-12.47%, FCEV-6.19%.
BEV, BV, HFC time, battery recharging time, policy and HFCs, 2) Purchase premiums of €2500 for Their simulations show that CVs can be expected to
incentives PHEVs, and €5000 for BEVs and HFCs, 3) Purchase further dominate the vehicle market, with CNGs and
price of €21,800 for all vehicles, 4) Battery leasing HEVs being the most likely AFVs, although policy
contract of €80/month for an annual mileage of initiatives can affect preferences for the various
10,000 km and purchase price reduction of €10,000 AFVs. An increase in the battery BEVs' range to
for BEVs, 5) 750 km driving range for BEVs, 6) 100% 750 km would increase their market shares from
fuel availability for all AFVs, 7) Battery recharging 2.24% to 5.45%.
time of 5 min, 8) Combination of Scenarios 1, 2,
6 and 7.
 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56 47

Table 3
Example of a SP choice experiment.

Car features

Ford Fiesta VW Polo Fiat Punto Evo (bi- Alfa Romeo Mito Toyota Yaris (hy- Peugeot iOn (BEV Renault Zoe (BEV –
(diesel) (gasoline) fuel – CNG) (bi-fuel – LPG) brid – gasoline) – own battery) leased battery)

Purchase price (€) 14,000 11,900 15,425 20,600 18,650 30,369 21,650
Range (km) 980 900 800 1200 1000 150 210
Acceleration (0–100 15 13 15 15 13 12 12
in s)
Annual operating 1894 2081 1757 1784 1920 1681 2553
cost (€)
Refuelling (km) 1 1 5 5 1 0 0
Which car would
you buy?

order to better take into account the technology diffusion process. one with an owned battery (similarly to Hackbarth and Madl-
They state that in Denmark a 3% BEVs market share would not be ener 2013);
reached until 2021. In summary, although no consensus predic- b) based on the MECL model's parameters, a Monte Carlo policy
tions are reached, many studies warn again too optimistic pre- simulation model is developed. The model is designed to pre-
diction for the market share penetration of the AFVs in general and dict the car market share in 32 market segments. The segments
the BEVs specifically. are derived from 5 socio-economic and behavioural dichot-
A set of recent studies, listed in Table 2, perform simulative omous variables: gender, income, number of cars, number of
analysis – based on economic preference or attitudinal models car trips longer than 400 km, and garage availability. Italian
estimated with SP data – in order to understand which policies National Statistics data are used to estimate the national ag-
would be more effective in altering the current car market shares gregate market share;
in favour of the AFVs. The policies analysed are the ones described c) policy simulations are then performed considering five sce-
by Daziano and Chiew (2012, Table 1), including subsidies for narios combining several instrumental variables (subsidies for
purchase, feebates, tax credits, gas taxes, rising gas prices, tighter purchase, rising gas prices, better electric batteries, and de-
energy-efficiency standards, better electric batteries, investments crease in the purchase price for electric cars as a result of a
in charging infrastructure, specific incentives (access to HOV/HOT1 decrease in the production costs or of an aggressive market
lanes, devoting parking), marketing campaigns or mechanical penetration strategy).
improvements. An unsolved issue of most of these studies is that
the baseline scenarios, derived from the experimental SP scenarios
is not close to the real world market shares. Daziano and Achtnicht
(2013) and Hackbarth and Madlener (2013) try to solve this pro- 3. Methodology and data collection
blem by estimating the simulated market shares under the average
attribute levels in the German market. However, the results do not The methodology consists of two steps. First, individual pre-
match the current market shares, in which, for instance the HFCVs ferences towards the selected 7 cars are estimated via MNL and
play no role. Hence, the simulations should be interpreted as MECL models based on the data collected with a SP survey. Then, a
variations from the experimental market shares and not as rea- Monte Carlo simulation model is developed and used for scenario
listic market shares which could be realized in the real world. The analysis.
main results are that the pricing policies and the density of the Disaggregate demand analysis based on the random utility
charging network have a strong influence on AFVs penetration. In theory is one of the most established approaches to estimate de-
the only study that considers the impact of raising the social mand (Anderson et al., 1992; Ben-Akiva and Lerman, 1985; Lou-
awareness, the one by Daziano and Bolduc (2013), it is found that viere and Hensher, 1983; McFadden, 1981; Train, 2002). The
also social campaigns could be extremely effective. However, the probability that an individual chooses the alternative with the
conventional vehicles appear to keep playing a relevant role in the highest utility, among a specific set of choice profiles, is estimated
future, unless very dramatic changes in technology and relative and the main factors that influence her/his choice are identified.
prices occur. Assuming that the parameters in utility function have a random
Compared with the above-mentioned literature, our study is nature, the MECL model allows for preference heterogeneity, cor-
characterized by the following features: relation between alternatives and correlation between the ex-
planatory coefficients of variables (McFadden and Train, 2000).
a) more than one hundred face-to-face interviews are adminis- Table 3 shows an example of a SP choice experiment, con-
tered in the first half of the year 2013. It is the first time that the taining the 7 cars: Ford Fiesta (diesel), VW Polo (gasoline), Fiat
Italian car market is researched in relation to the AFVs. The SP Punto Evo (bi-fuel-CNG), Natural Power Alfa Romeo Mito (bi-fuel-
data collected are used to estimate a MNL and MECL car choice
LPG), Toyota Yaris (HEV-gasoline), Peugeot iOn (BEV – owned
models. Contrary to the existing literature seven specific cars
battery), Renault Zoe (BEV – leased battery). These specific cars
(labelled alternatives) with different fuelpowertrain technolo-
were chosen because they are very popular in Italy and re-
gies are tested, including two BEV types, one with a leased and
presentative of their fuelpowertrain technology. They all belong to
the same car segment (B segment), apart from the Peugeot iOn
1
HOV-high occupancy vehicle; HOT-high occupancy toll. that belongs to the A segment.
48 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56

Table 4
Descriptive indicators of the socio-economic data of the sample.

Features: Levels:

Gender (%): M: 50 F: 50
Level of Education (%): Primary: 3 High school: 26 Bachelors/graduate Post-doc experi-
degree: 50 ence: 20
Current employment (%): Employee: 11 Manager: 3 Freelancer: 14 Student: 30 Student- Retired: 5 Housewife: 4 Unemployed: 26
worker: 9
Family size (%): 1 Person: 18 2 People: 13 3 People: 25 4 People: 26 5 People: 4 6 People: 14
Net yearly household in- o €30,000: 21 From €30,000 to 4€70,000: 22 Missing value: 5
come (%): €70,000: 52
Car expertise level (%): 1: 18 2: 12 3: 18 4: 14 5:14 6: 11 7: 13
(1 ¼None, 7¼ Very high)

Face-to-face interviews were administered in three cities up to 6 people. The net yearly household income is between
(Trieste, Bologna, Pesaro) with different size and availability of €30,000 and €70,000 for 52% of respondents. With regard to the
refuelling station. self-evaluated level of expertise with cars, half of the sample
During each interview, the following data were collected: deems to have medium or high knowledge of cars.
(i) socio-economic characteristics of the respondent (gender, level Table 5 contains some descriptive indicators of the family cars'
of education, current employment, family size, net yearly house- features in the sample. Seventy percent of the sample owns 2 or
hold income, self-evaluated level of expertise with cars); (ii) more cars, mainly with gasoline- (67%) or diesel- (25%) operated
characteristics of cars owned and used by the respondent's family
engine. More than half of the sampled individuals own a garage.
such as number, age and type of car engine technologies, avail-
Overall, the sample seems to over-represent the better edu-
ability of a private car garage and mobility habits of interviewees
cated, young, well-off segment of the population: this might lead
(for instance: transport mode mainly used, main travel purpose,
to an over-prediction in favour of the AFVs.
etc.); and, (iii) 12 stated choice experiments for each respondent.
Table 6 presents some descriptive indicators of the car mobility
Each interview lasted about 45 min. Due to time and budget
constraints only 121 interviews could be collected in the first se- habits of the sample. The transport mode mainly used is the car
mester of the year 2013. Although the sample size is admittedly (55%). For 49% of respondents, the numbers of weekly (return) car
small, we decided not to carry out other interviews in 2013 and to trips varies from 5 to 15, and for 28%, it varies from 15 to 30 trips.
devote more resources for the year 2014 as new AFVs enter in the With regards to the average distance per trip, 63% of them are less
Italian car market and the consumers get acquainted to the new than 10 km long, 32% are from 10 to 40 km long.
technologies. The car is mainly used for leisure purposes (50%), followed by
Drawing from the literature, 5 attributes were included in the family management, visiting relatives and parents (20%), and
SP choice experiments: purchase price (€), annual operating cost business reasons (19%). Finally, half the sample made from 5 to 10
(gasoline, insurance, tax, maintenance) (€), acceleration (seconds), journeys over 400 km per year and 37% from 2 to 4 trips.
range (kilometres), and refuelling distance (kilometres). The Status
Quo (SQ) attributes for each car were set equal to the Italian
average values reported in Table 9. They were varied as follows: (i) 4.2. The multinomial and mixed error component Logit models
purchase price:  20%, SQ, þ20%, þ40%; (ii) annual operating
cost:  20%, SQ2, þ20%; (iii) range: SQ, þ20%, þ40%; (iv) accel- This section reports the results obtained estimating (1) a MNL
eration: SQ,  10%,  20%; (v) refuelling distance: gasoline, diesel model, allowing us to evaluate in a simple manner the monetary
and hybrid (1 km, 5 km, 10 km); CNGV and LPGV (5 km, 20 km, value of the nonmonetary attributes, and (2) a MECL model, al-
50 km) and BEVs (0 km, 5 km, 10 km). lowing us to take into account the random nature of the model
An efficient experimental design strategy was used with
coefficients; to explore the role played by the socio-economic
4 waves in order to minimise the asymptotic standard error
variables in determining the model coefficients, and; to account
(Bliemer & Rose, 2010, 2011; Huber & Zwerina, 1996; Yu et al.,
for the correlation among alternatives and the panel features of
2009).
the data set.

4. Descriptive and econometric results

Table 5
4.1. Descriptive statistics of the sample Descriptive indicators of the family cars' features in the sample.

Table 4 reports some descriptive indicators of the socio-eco- Features: Levels:

nomic data of the sample. Men and women are equally re-
No. of owned 0 Cars: 10 1 Car: 21 2 Cars: 3 Cars: 12 44
presented in the sample. The prevailing level of education is the cars (%): 41 Cars:
bachelors/graduate degree (50%), followed by high school (26%) 16
and post-doc experience (20%). Thirty per cent of the sample are Type of engine SGV: 67 DV: 25 CNGV: 2 LPGV: 5 HEV: BEV: 0
technology 1
students (plus an additional 9% of student-workers) and 26% are
(%):
unemployed. The sample includes both singles and families with Availability of Yes: 65 No: 31 Missing
an owned values: 3
garage (%):
2
The SQ operating cost are calculated in Rusich and Danielis (2013)
 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56 49

Table 6
Descriptive indicators of the car mobility habits of the sample.

Features: Levels:

Frequency use of transport modes (%): Bike/walking: 10 Scooter/motorbike: 8 Car: 55 Public transport: 28
Number of weekly (return) car trips (%): o 5: 23 5–15: 49 15–30: 28
Average distance by car per trip (%): o 10 km: 63 10–40 km: 32 440 km: 5
Reason for travelling by car (%): Business: 19 Study: 11 Family management, visiting relatives, parents: Leisure: 50
20
Number of yearly return trips by car over 400 km 1 trip/year: 13 2–4 trips/year: 37 5–10 trips/year: 50
(%):

The MNL model is specified as follows: Table 7

MNL model's estimates.
UGasoline.car = β G ASCG + β PP PurchasePrice G + β R RangeG
Variables Estim. coeff. Std. err. t-Ratio P-value
+ β A AccelerationG + β AOC AnnualOperatingCost G
Alternative specific constants:
+ β RD Refuelling Distance G + εG UDiesel.car ASC DV (Ford Fiesta) (βD) 0.124 0.129 0.964 0.335
ASC CNGV (Fiat Punto Evo 0.337 0.180 1.871 0.061
= β D ASCD + β PP PurchasePrice D + β R RangeD NP) (βCNG)
ASC LPGV (Alfa Romeo 0.385 0.194 1.979 0.048
+ β A AccelerationD + β AOC AnnualOperatingCost D Mito) (βLPG)
ASC HEV (Toyota Yaris) (βH)  0.151 0.117  1.292 0.196
+ β RD RefuellingDistanceD + β LDTD LongDistanceTripsD
ASC BEV – owned battery  0.465 1.041  0.447 0.655
+ εD UCNG. (Peugeot iOn) (βE-ob)
ASC BEV – leased battery  1.526 0.927  1.646 0.100
= β CNG ASCCNG + β PP PurchasePrice CNG + β R Rangei (Renault Zoe) (βE-lb)

+ β A AccelerationCNG + β AOC AnnualOperatingCost CNG

Generic attributes:
+ β RDCNG RefuellingDistanceCNG + εCNG ULPG. Purchase price (€1000) (βpp)  0.208 0.010  20.575 0.000
Non-BEV range (1000 km) (βR) 1.554 0.241 6.448 0.000
= β LPG ASCLPG + β PP PurchasePrice LPG + β R RangeLPG Non-BEV acceleration (βA) 0.005 0.024 0.192 0.848
+ β A AccelerationLPG + β AOC AnnualOperatingCost LPG Annual operating cost  1.287 0.079  16.326 0.000
(€1000) (βAOC)
+ β RDLPG RefuellingDistanceLPG + εLPG UHybrid.car Refuelling distance (βRD) 0.013 0.010 1.317 0.188

= β H ASCH + β PP PurchasePrice H + β R RangeH

Fuel-specific attributes:
+ β A AccelerationH + β AOC AnnualOperatingCost H Refuelling distance  LPGV  0.017 0.005  3.554 0.000
(βRDLPG )
+ β RD RefuellingDistanceH + εH UElectric.car − owned.battery
Refuelling distance  CNGV  0.015 0.004  3.540 0.000
= β E−ob ASCE−ob + β PP PurchasePrice E−ob + RangeE−ob (βRDCNG )
Refuelling distance  BEVs 0.005 0.020 0.254 0.800
+ β AE AccelerationE−ob + β AOC AnnualOperatingCost E−ob (βRDE )
+ β RDE RefuellingDistanceE−ob + β CEE CarExpert E−ob BEV acceleration (β AE ) 0.018 0.054 0.326 0.745
BEV range (1000 km) (βRE ) 10.485 2.556 4.102 0.000
+ β CGE CarGarageE−ob + εE−ob UElectric.car − leased.battery
= β E−lb ASCE−lb + β PP PurchasePrice E−lb + RangeE−lb Socio-economic variables:
Long distance trips  DV  0.002 0.026  0.084 0.933
+ β AE AccelerationE−lb + β AOC AnnualOperatingCost E−lb
(βLDTD)
+ β RDE RefuellingDistanceE−lb + β CEE CarExpert E−lb Car expert  BEVs (βCEE )  0.472 0.182  2.590 0.010
Car garage  BEVs ( βCGE )  0.123 0.166  0.740 0.459
+ β CGE CarGarageE−lb + εE−lb
Adjusted Rho squared no 0.15
coefficients:
Table 7 reports the results for the MNL model. Number of observations: 1452
The SP alternatives present jointly the brand of the car and the Log likelihood function:  2152.25
fuelpowertrain technology. Consequently, it is not possible to
disentangle the brand effect from the fuelpowertrain effect and
the ASCs should be interpreted as a preference for the brand and the Renault strategy of selling the car at a lower price but with the
the fuel type jointly considered.3 Keeping this in mind, taking the need to spend monthly a fixed amount for the battery.
gasoline VW Polo as a reference point, ceteris paribus, the diesel All attributes except acceleration and refuelling distance are
Ford Fiesta is deemed equivalent, the CNG Fiat Punto Evo Natural significant with the expected sign. The purchase price and the
Power and the LPG Alfa Romeo Mito are considered superior, annual operating cost are highly significant. As expected, the
whereas the hybrid Toyota Yaris and the electric Peugeot iOn with coefficient of the annual operating cost is much larger than the
owned battery are, again, considered equivalent or slightly worse. coefficient of the purchase price. It is estimated that the former is
The electric Renault Zoe with leased battery is the least preferred six times larger than the latter. The respondents recognise that
€1000 purchase price is a one off cost whereas €1000 annual
among the 7 cars, showing that the Italian buyers do not favour
operating cost is spent for the entire lifetime of the car.
The range generic attribute is also highly significant (the WTP
3
We thank an anonymous referee for pointing this out. for a 1 km range increase is equal to €7.47). For the BEVs the range
50 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56

Table 8 for the BEVs priced below and above 20,000€, respectively, and
Results of the mixed error component logit model. values 8.32 and 16.82 km, for the non-BEVs priced below and
above 20,000€, respectively. Jensen et al. (2013) distinguish be-
Variables Coeff. Std. t-Ratio P-
err. value tween single car and multicar households, before and after testing
the BEV and find values equal to 65€ (before) and 134€ (after) per
Random parameters km for single car households and values equal to 46€ (before) and
Purchase price (€1000) (βpp ) a  0.633 0.053  11.8780.0000
84€ (after) per km for multiple car households. Dimitropoulos
Annual operating cost (€1000) (βAOC ) a  4.616 0.333  13.841 0.0000 et al. (2012) find values very close to ours, ranging for the BEVs
Non-BEV range (1000 km) (βR ) a 3.417 0.488 6.9990.0000 from €52 to €58 per km.
Non-BEV acceleration (βA ) b  0.126 0.066  1.912 0.0559 Acceleration is not significant for the choice between cars,
BEV range (1000 km) (βRE ) a 21.695 2.769 7.8360.0000 neither as a generic attribute nor when tested specifically for the
BEV acceleration (βAE ) a  0.113 0.038  2.9530.0031 BEVs. Since the acceleration performance of the BEVs is much
Non-random parameters discussed and promoted in the non-scientific literature, this is a
Refuelling distance (βRD )  0.002 0.012  0.183 0.8550 somewhat surprising result, although common to other discrete
ASC DV (Ford Fiesta)(βD ) 0.357 0.153 2.3300.0198 choice studies (e.g. Mabit & Fosgerau, 2011). This result is to be
Long distance trips  DV (βLDTD ) 0.032 0.033 0.9690.3325 further verified since it might reflect the lack of actual experience
ASC CNGV (Fiat Punto Evo NP)(βCNG ) 0.406 0.214 1.9000.0574 with the BEVs of the Italian users.
Refuelling distance  CNGV (βRDCNG )  0.018 0.005  3.551 0.0004 In contrast with the recent literature (Achtnicht et al., 2008;
ASC LPGV (Alfa Romeo Mito)(βLPG ) 0.596 0.243 2.4580.0140 Daziano and Bolduc, 2013; Daziano and Achtnicht, 2013; Hack-
Refuelling distance  LPGV (βRDLPG )  0.017 0.005  3.171 0.0015 barth, Madlener, 2013), we find that the refuelling distance attri-
ASC HEV (Toyota Yaris)(βH ) 0.138 0.145 0.951 0.3415 bute for the SGV, DV, HEV and BEVs plays no role in the car choice-
ASC BEV-owned battery (Peugeot  1.512 1.229  1.2300.2186 making process. However, it is significant when the LPGV and
iOn)(βE − ob ) CNGV are considered. This latter result is very realistic since the
Car Expert  BEVs (βCEE )  0.294 0.504  0.5830.5598 Italian refuelling stations network for these two fuel types is not
Car Garage  BEVs (βCGE )  0.815 0.382  2.137 0.0326 sufficiently dense. Furthermore, we have tested whether the re-
Refuelling distance  BEVs (βRDE )  0.001 0.028  0.0470.9622 spondents who make long distance trips (at least 4 longer than
ASC BEV-leased battery (Renault  1.462 1.119  1.3070.1912 400 km per year) would favour the more fuel-efficient diesel cars,
Zoe)(βE − lb ) finding a negative answer, and whether the car experts or the
garage owners are more likely to buy a BEV. We find that the
Heterogeneity sources
Purchase price: older age  0.013 0.038  0.3360.7366 Italian “car experts” do not like BEVs and that no evidence exists
Purchase price: female 0.132 0.031 4.319 0.0000 that garage ownership has a role in buying a BEV. The stability of
Purchase price: higher income 0.151 0.041 3.6890.0002 these results, as BEVs enter the market, should be subject to a
Annual operating cost: older age 0.404 0.285 1.418 0.1561
future test and investigated with more sophisticated discrete
Annual operating cost: female 1.221 0.192 6.3590.0000
Annual operating cost: higher income 0.702 0.296 2.374 0.0176 choice models (e.g. hybrid choice model).
Non-BEV range: older age  0.142 0.484  0.2940.7687 Next, by using a MECL model, we tested whether there is
Non-BEV range: female  1.740 0.377  4.613 0.0000 heterogeneity in the attribute parameters and what determines it,
Non-BEV range: higher income 0.948 0.491 1.9320.0533 using the socio-economic data collected during the interview.
Non-BEV acceleration: older age 0.070 0.039 1.813 0.0698
Non-BEV acceleration: female 0.072 0.038 1.8900.0588
Furthermore, this model accounts for the correlation among al-
Non-BEV acceleration: higher income 0.00040.039 0.0090.9930 ternatives and the panel features of the data set.
BEV range: old age 0.547 3.164 0.173 0.8627 Having tried several specifications, the best one is reported in
BEV range: female  11.08 2.771  4.0000.0001 Table 8.
BEV range: higher income  6.150 3.413  1.8020.0716
The purchase price, the annual operating cost, the non-BEV and
BEV acceleration: older age 0.06830.049 1.3810.1673
BEV acceleration: female 0.088 0.041 2.166 0.0303 BEV range, the non-BEV and BEV acceleration are assumed ran-
BEV acceleration: higher income 0.087 0.049 1.748 0.0804 domly distributed. Except the non-BEV acceleration, all variables
Spreads and standard deviations of parameter distributions
are assumed to have a constrained triangular distribution so that
Ts purchase price 0.633 0.053 11.8780.0000 the average is equal by construction to the spread to the dis-
Ts annual operating cost 4.616 0.333 13.8420.0000 tribution. The socio-economic variables such as age, gender and
Ts non-BEV range 3.417 0.488 6.9990.0000 household income are used to explain the parameters' hetero-
Ns non-BEV acceleration 0.090 0.016 5.7860.0000
geneity. An error component is included to capture the un-
Ts BEV range 21.695 2.769 7.8360.0000
Ts BEV acceleration 0.113 0.038 2.9530.0031 observed error correlation between the two BEVs.4
Accounting for heterogeneity, largely improves the goodness-
Standard deviations of latent random effects
SigmaE01 (E-ob/E-lb) 0.295 0.042 6.9880.0000
of-fit of the model (the adjusted Rho squared improves from 0.150
Adjusted Rho no coefficients 0.377 to 0.380). The error component term is lower than one but highly
significant.
Number of observations 1.452
Log likelihood function  2,825 Starting from the ASCs and having taken the gasoline VW Polo
as a reference point, ceteris paribus, the diesel Ford Fiesta, the CNG
Notes: ASC ¼ alternative specific constant. The ASC base is the BEV with an owned Fiat Punto Evo Natural Power and the LPG Alfa Romeo Mito are
battery.
considered superior to the gasoline VW Polo, whereas the hybrid
a
Random variables with a restricted triangular distribution.
b
Toyota Yaris is equivalent and the two BEVs (Peugeot iOn and
Random variable with a normal distribution.
Renault Zoe) are less preferred.

specific attribute is much higher (equal to 10.49) which corres-

4
ponds to a WTP of €50.4 per a 1 km range increase. These results As explained in the Nlogit version 4.0 – Reference Guide (N-14), the error
component is an individual specific random effect that is distributed across alter-
are intermediate relative to the ones obtained by Hackbarth and natives according to a tree structure. A sigma variable is estimated that account for
Madlener (2013) and by Jensen et al. (2013). Hackbarth and choice situation invariant variation that is unobserved and not accounted for by the
Madlener (2013) find WTP values equal to 16.21 and 32.76 per km, other model components.
 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56 51

Fig. 1. Framework of the simulation model.

All attributes except refuelling distance are significant with the whereas garage ownership counter-intuitively has a negative
expected sign. The purchase price and the annual operating cost preference for the BEVs.
are highly significant and, as before, the coefficient of the annual The socio-economic variables explain the random parameters'
operating cost is much larger than the coefficient of the purchase heterogeneity as follows: (a) respondents older than 30 years care
price. The range generic attribute is also highly significant. For the slightly less about the non-BEV acceleration properties; (b) wo-
BEVs the range specific attribute is again much higher. Accelera- men are less sensitive to the purchase price, the annual operating
tion is significant in this model with an expected sign both for the cost, the range and the acceleration both of the non-BEVs and
non-BEVs and the BEVs with a coefficient of a similar magnitude. BEVs. In short, they differentiate less among the cars; (c) re-
The refuelling distance attribute is again only significant for the spondents with an annual household income higher than €30,000
LPGV and CNGV. are less sensitive to the purchase price and to the annual operating
The respondents who make long-distance trips (at least cost, although they would prefer non-BEVs with a higher range,
4 longer than 400 km per year) do not revealed a preference for they are more willing to accept BEVs with a lower range, most
the diesel car. Italian “car experts” are indifferent to the BEVs probably as a second car.

Table 9
Estimate of the Italian market demand: base case scenario.
Source: Unione Nazionale Rappresentanti Autoveicoli Esteri (2013).

Type of engine Purchase Annual operat- Range Acceleration Refuelling Estimated mar- Actual market Actual market share
technology: price ing cost distance ket shares share by brand by fp technology

List: € € km 0–100 km in s km % % %
VW Polo (gasoline) 11,900 2081 900 13 1 38.8 22 31.03
Ford Fiesta (diesel) 14,000 1894 980 15 1 31.1 30 53.83
Fiat Punto Evo (bi-fuel – 15,425 1757 800 15 5 16.1 25 5.11
CNG)
Natural Power Alfa Ro- 20,600 1784 1,200 15 5 9.8 19 8.91
meo Mito (bi-fuel –
LPG)
Toyota Yaris (Hybrid – 18,650 1920 1000 13 1 4.1 4 1.07
gasoline)
Peugeot iOn (electric – 30,369 1681 150 12 0 0.17 0.21 0.06
owned battery)
Renault Zoe (electric – 21,650 2553 210 12 0 0.03 0.00 0.00
leased battery)
Total: 100 100 100
52 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56

Table 10
Estimated market share: introduction of a state subsidy.

Type of supply power: Purchase price (base) State subsidy Net purchase price Estimated market Estimated market shares (base
shares scenario)
€ € € % %

VW Polo (gasoline) 11,900 – 11,900 20.4 38.8

Ford Fiesta (diesel) 14,000 2000 12,000 51.9 31.1
Fiat Punto Evo (bi-fuel – CNG) 15,425 – 15,425 9.9 16.1
Natural Power Alfa Romeo Mito 20,600 – 20,600 7.8 9.8
(bi-fuel – LPG)
Toyota Yaris (hybrid – gasoline) 18,650 3560 15,090 9.4 4.1
Peugeot iOn (BEV – owned battery) 30,369 5000 25,369 0.36 0.17
Renault Zoe (BEV – leased battery) 21,650 4330 17,320 0.19 0.03
Total: 100 100

4.3. Simulation model 4.3.1. Base case scenario with current market values
Table 9 shows an estimate of the Italian market demand for the
On the basis of the parameters obtained in the MECL model, 7 cars under the base case scenario. The first five columns report
Monte Carlo simulations are performed using the Frontline risk the purchase price, the annual operating cost, the range, the ac-
solver5 software. The structure of this model is illustrated in Fig. 1. celeration and the refuelling distance for each car. The purchase
The utility of each car is estimated multiplying the attributes' va- price is the prevailing market price in Italy in 2013. The annual
lues Xi by the corresponding attribute parameter. Each attribute operating cost is calculated by Rusich and Danielis (2013) assum-
parameter is specified by its mean and standard deviation or ing an annual driving distance of 10,000 km. The range and ac-
spread. For each Monte Carlo simulation a random draw is taken celeration are the ones reported by the car manufacturers. The
from the distribution function of each parameter. The value of each refuelling distance is estimated based on the current Italian fuel
attribute coefficient is the sum of its mean, randomly taken from distribution network.
the distribution, and its covariates. Thirty-two (¼ 25 ) population The last three columns report the estimated and the actual
segments are identified combining the 5 socio-economic and be- market shares of the cars registered in the year. The second col-
havioural dichotomous variables (gender, income, number of cars, umn before the last shows the market shares as estimated by our
number of car trips longer than 400 km, and garage availability). model. The first column before the last shows the actual market
The utility of each of the 7 cars is then computed for each of the 32 share in 2012 by brand, that is we have considered how many VW
population segments. For every simulation and for each popula- Polo, Ford Fiesta, Fiat Punto Evo, etc. have been registered in 2012.
tion segment, the market share of each car is estimated using the The last column reports the actual market share by car fuelpo-
logit formula. Ten thousand Monte Carlo draws are performed and wertrain technology, that is we have reported the percentage of
the market share averages for each car are calculated for each gasoline, diesel, bi-fuel – CNG, etc. cars have been registered in
population segment. 2012.
The representativeness of each market segment at national It can be seen that the model is able to sufficiently approximate
level is then computed on the basis of the data available from the the actual market share by brand, but it overestimates the VW
National Statistical Institute (Table 16 in Annex 1). Multiplying Polo (gasoline) market share by brand.6 When the market share by
each car choice probability for the representativeness of the po- fuelpowertrain technology is considered, the model under-
pulation segment and summing up across all population segments, estimated the DV and over-estimates the bi-fuel CNGV, bi-fuel
we obtained the total estimated national market share of each car. LPGV and HEV. The BEVs are pretty well approximated both by car
Based on SQ data, we determine the current estimated market brand and fuelpowertrain technology.
shares for Italy for the 7 cars considered (‘base case' scenario).
Subsequently, we evaluate the impact of potential policy inter- 4.3.2. Scenario 1 – Introduction of a state subsidy for the purchase of
ventions or technological improvements as variations from the low emission cars
base case scenario. More specifically, we tested the impact of: A recently policy adopted by several European countries is the
introduction of purchase subsidies to encourage people to buy
1) the introduction of a state subsidy for the purchase of low cars. The Italian Parliament recently allocated funds to subsidise
emission cars (scenario 1); the purchase of cars with low environmental impact, even though
2) the introduction of a state subsidy for the purchase of low
the total amount of funds made available was quite limited (€40
emission cars and a threefold increase of the range for BEVs
million in 2013).
(scenario 2);
Economic incentives are provided for the purchase of cars with
3) the introduction of a state subsidy for the purchase of low
different levels of (low) CO2 emissions. The subsidy is differ-
emission cars and the increase in the price of fossil-based fuels
entiated between our 7 cars as follows:
(þ20%) (scenario 3);
4) the introduction of a state subsidy for the purchase of low  €2000 for the Ford Fiesta (diesel).
emission cars and the decrease of the purchase price of the  €3560 for the Toyota Yaris (hybrid).
BEVs cars (€5000) (scenario 4) and  €5000 for the Peugeot iOn (electric – owned battery).
5) the joint introduction of all previous policies and technological  €4330 for the Renault Zoe (electric – leased battery).
improvements (scenario 5).

6
The sum of the squared errors is equal to 447 when the estimated shares are
compared to the actual share by brand, while it is equal to 708 when the estimated
5
For further details see: http://www.solver.com/risk-solver-platform. shares are compared to the actual share by engine technology.
 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56 53

Table 11
Estimated market share: increase of range for BEVs.

Type of supply power: Range (km) New Range (km) Estimated market shares Estimated market shares (base scenario)

Ford Fiesta (diesel) 980 980 30.7 31.1

Fiat Punto Evo (bi-fuel – CNG) 800 800 15.7 16.1
Natural Power Alfa Romeo Mito (bi-fuel – LPG) 1200 1200 9.3 9.8
Toyota Yaris (hybrid – gasoline) 1000 1000 3.8 4.1
Peugeot iOn (BEV – owned battery) 150 450 1.08 0.17
Renault Zoe (BEV – leased battery) 210 630 0.93 0.03
Total: 100 100

Table 12
Estimated market share: increase in the price of fossil-based fuel.

Type of power supply: Annual operating cost New annual operating cost Estimated market shares Estimated market shares (base
( þ20%) scenario)
€ € % %

VW Polo (gasoline) 2081 2497 35.4 38.8

Ford Fiesta (diesel) 1894 2273 31.6 31.1
Fiat Punto Evo (bi-fuel – CNG) 1757 2108 18.0 16.1
Natural Power Alfa Romeo Mito (bi-fuel 1784 2141 10.0 9.8
– LPG)
Toyota Yaris (hybrid – gasoline) 1920 2304 3.8 4.1
Peugeot iOn (BEV – owned battery) 1681 1681 1.03 0.17
Renault Zoe (BEV – leased battery) 2553 2553 0.17 0.03
Total: 100 100

Table 13
Estimated market share: decrease in the purchase price of €5000 for BEVs

Type of supply power: Purchase price (base) Final purchase price Estimated market shares Estimated market shares (base
scenario)
€ € % %

VW Polo (gasoline) 11,900 11,900 38.7 38.8

Ford Fiesta (diesel) 14,000 14,000 31.0 31.1
Fiat Punto Evo (bi-fuel – CNG) 15,425 15,425 15.9 16.1
Natural Power Alfa Romeo Mito (bi-fuel – 20,600 20,600 9.7 9.8
LPG)
Toyota Yaris (hybrid – gasoline) 18,650 18,650 4.0 4.1
Peugeot iOn (BEV – owned battery) 30,369 25,369 0.38 0.17
Renault Zoe (BEV – leased battery) 21,650 16,650 0.28 0.03
Total: 100 100

Table 14
Estimated market share: Introduction of a state subsidy, the decrease in the purchase price for BEVs (€5000), increase in range and increase in fossil fuel price.

Type of engine technology: Purchase Annual operating Range Acceleration Refuelling Estimated market Estimated market shares
price cost distance shares (base scenario)

List: € € km 0–100 km in s km % %
VW Polo (gasoline) 11,900 2497 900 13 1 14.3 38.8
Ford Fiesta (diesel) 12,000 2273 980 15 1 42.5 31.1
Fiat Punto Evo (bi-fuel – CNG) 15,425 2108 800 15 5 8.2 16.1
Natural Power Alfa Romeo 20,600 2141 1,200 15 5 6.4 9.8
Mito (bi-fuel – LPG)
Toyota Yaris (hybrid – 15,090 2304 900 13 1 7.3 4.1
gasoline)
Peugeot iOn (BEV – owned 20,369 1681 450 12 0 6.47 0.17
battery)
Renault Zoe (BEV – leased 12,320 2553 630 12 0 14.89 0.03
battery)
Total: 100 100

The simulation model estimates the impact of the economic up to 51.9% at the expense of the VW Polo (gasoline). The Toyota Yaris
incentive as reported in Table 10. (hybrid – gasoline) increases by 5% at the expense of the other cars
Contrary to the general expectations, the Ford Fiesta (diesel) ap- whereas the BEVs increase their market share only marginally,
pears to be the one that benefits the most, increasing its market share although they are the recipients of the largest state subsidies.
54 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56

Table 15
Market share variations relative to the base case scenario.

Type of cars: Scenario 1: Scenario 2: threefold range Scenario 3: 20% fossil- Scenario 4: €5000 price Scenario 5: scenario 1–4
subsidy increase for electric cars based fuel price increase reduction for the BEVs considered jointly

VW Polo (gasoline)  18.4  0.3  3.4  0.1  24.5

Ford Fiesta (diesel) 20.8  0.4 0.5  0.1 11.4
Fiat Punto Evo (bi-fuel –  6.2  0.4 1.9  0.2  7.9
CNG)
Natural Power Alfa Romeo 2  0.5 0.2  0.1  3.4
Mito (bi-fuel – LPG)
Toyota Yaris (hybrid – 5.3  0.3  0.3  0.1 3.2
gasoline)
Peugeot iOn (BEV – owned 0.19 0.91 0.86 0.21 6.3
battery)
Renault Zoe (BEV – leased 0.16 0.9 0.14 0.25 14.86
battery)

4.3.3. Scenario 2 – Threefold increase of the range for BEVs 5. Discussion and conclusions
Table 11 shows the potential demand as a result of an increase
in the range of BEVs thanks to technological developments. This This paper reports the results of a simulation model calibrated
innovation would solve the well-known range anxiety problem for the Italian (segment B) car market. The simulation model is
associated with the BEVs, that is, the fear of being left with a dead based on parameters derived from a MECL model estimated with
battery. The assumption that we test is that the range of both BEVs data collected via SP interviews in three Italian cities and on 32
would triple. market segments aggregated according to their representativeness
The BEVs would increase their market shares by 1% at the ex- at national level.
Taking as reference the base case scenario (Table 9), the fol-
pense of the other cars.
lowing 5 scenarios were tested: the introduction of a state subsidy
for the purchase of low emission cars (scenario 1); a threefold
4.3.4. Scenario 3 – Increase in the price of fossil fuels ( þ20%)
range increase for the BEVs (scenario 2); a 20% price increase for
Table 12 illustrates the impact of a hypothetical increase of the
the fossil-based fuels (scenario 3); a €5000 purchase price de-
price of fossil fuel. Such increase is modelled as a 20% increase of
crease for the BEVs (scenario 4); joint changes of the previous
the annual operating cost for all cars except the BEVs.
scenarios (scenario 5).
Notwithstanding the common cost increase (except for the
Table 15 summarizes the simulation results as variations from
BEVs), the impact is differentiated, reflecting the relative cost
the base case scenario.
structure between fixed and variable costs. The VW Polo (gasoline) The choice made by the Italian government to subsidise the
is estimated to loose more than 3%, due to its high variable costs BEVs, the hybrid cars and – albeit with a lower sum – the diesel
whereas the Fiat Punto Evo (bi-fuel-CNG) and the Natural Power cars, greatly favors the Ford Fiesta (diesel) at the expense of the
Alfa Romeo Mito (bi-fuel-LPG) would increase their shares. The VW Polo (gasoline). The Toyota Yaris (hybrid – gasoline) gains
impact on the BEVs is mostly on the Peugeot iOn (electric – owned whereas the BEVs are only slightly affected. A threefold range in-
battery), which would gain almost 1% market share. crease for BEVs has the largest impact on their joint market share
but still not larger than 2%. It is also interesting to note that the
4.3.5. Scenario 4 – Decrease in the purchase price for BEVs (€5000) changes assumed in scenario 3 and 4, both in favour of the BEVs,
Table 13 illustrates the impact of a decrease in the purchase would also only slightly affect their market shares. A 20% fossil fuel
price for electric cars (€5000) as a result of a decrease in the price increase impacts negatively the VW Polo (gasoline) and po-
production costs or of an aggressive market penetration strategy sitively the Fiat Punto Evo (bi-fuel – CNG) and the Ford Fiesta
undertaken by the car manufacturers. (diesel).
It is estimated that the impact of such a price cut would be very It is noteworthy, however, that none of the single changes in
marginal since the current disadvantages of the BEVs cannot be the economic or technology factors significantly affect the BEVs'
sufficiently compensated for by such a price cut. market share. Only when all these changes take place simulta-
neously (scenario 5) a large improvement in the BEVs' market
4.3.6. Scenario 5 – Introduction of a state subsidy, the decrease of the share materialize. The loser would be the VW Polo (gasoline), the
purchase price for BEVs (€5000), increase in range and increase in Fiat Punto Evo (bi-fuel – CNG) and the Natural Power Alfa Romeo
fuel price Mito (bi-fuel – LPG), whereas the Ford Fiesta (diesel) and the
Scenario 5 assumes that all of the above changes occur si- Toyota Yaris (hybrid – gasoline) do, like with the BEVs, increase
their market shares.
multaneously (Table 14).
Compared with the results achieved in previous research (Ta-
It is estimated that the impact would be quite large. The VW
ble 2), presented and discussed in the review of literature section,
Polo (gasoline) would loose its competitive advantage and loose
our findings confirm that the CV (gasoline and diesel) are going in
about 24% market share. The Ford Fiesta (diesel), that in Italy en-
the near future to maintain their strong position in the market. We
joys a state subsidy of €2000, would gain 9 points. The Fiat Punto
also confirm that the pricing strategy per se, either in a form of a
Evo (bi-fuel – CNG) and the Natural Power Alfa Romeo Mito subsidy, a price cut or a fuel tax, is not sufficient to alter sig-
(bi-fuel – LPG) would lose about 8 and 3 points, respectively. On nificantly the relative market power between the conventional
the contrary, the Toyota Yaris (Hybrid – gasoline) would rise to and AFVs. Only, the HEVs appear to benefit from a reduction in the
7.3% of the market and the BEVs would jointly reach a market purchase price whereas the BEVs would not. Similarly, the per-
share of about 21%, with the Renault Zoe (electric – leased battery) formance strategy (increased battery range) appears in our simu-
enjoying two-thirds of the BEVs' share. lations ineffective while previous studies found different results
 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56 55

(e.g. Horne et al., 2005). This difference might be due to the fact Acknowledgments
that our research does not find a strong enough evidence on the
effect of the charging network density as some previously studies The first author thanks the European Social Fund of the Friuli
do (Achtnicht et al., 2008; Daziano and Bolduc, 2013; Daziano and Venezia Giulia region for funding to her the S.H.A.R.M. (Supporting
Achtnicht, 2013; Hackbarth and Madlener, 2013). Consequently, Human Assets of Research and Mobility) Project “Modelli per la
the performance strategy associated to the increase in the char- stima degli scenari di penetrazione di mercato delle automobili con
ging infrastructure density is not tested in our simulations. How- differenti tipologie di alimentazione”, in which both the University
of Trieste and the Denmark Technical University – Department of
ever, when the pricing and the performance strategies are si-
Transport are involved.
multaneously used by the policy makers (car purchase price sub-
We again thank the Friuli Venezia Giulia region for funding the
sidies and fuel taxes) and car manufacturers (battery improvement
research project “Un Electric Car Club for the Friuli Venezia Giulia”
and car price reduction) the AFVs, and the BEVs specifically, would (http://www.ecc.units.it/) on which these data are collected. Many
penetrate significantly the market. The obvious policy implication thanks to Lucia Rotaris, Concepción Román and Juan Carlos Martin
is that in order to spur the development of the AFVs in Italy, a for their useful suggestions. A special thanks to Lorenzo Anto-
combined effort from the policy makers and the car manufactures naglia, Tin Pofuk and Andrea Rusich for administering the
is need. interviews.
This paper represents an initial effort to estimate and simulate
the future changes in the Italian car market as a result of a set of
potential technological, economic or policy changes. Several im- Appendix A
provements are possible and desirable. An increase in the sample
size and the geographical coverage is urgently needed. Testing the See Table 16.
robustness of the econometric estimates and of the simulations
comparing them with similar studies is also required. As some of
these engine technologies are quite new, such as the hybrid engine References
and the BEVs, their knowledge by the Italian drivers is still quite
Achtnicht, M., Bühler, G., Hermeling, C.C., 2008. Impact of Service Station Networks
limited and, hence, their preference structures might be unstable.
on Purchase Decisions of Alternative-fuel Vehicles. ZEW-Zentrum für Euro-
Further elements that can dramatically change the choice pre- päische Wirtschaftsforschung/Center for European Economic Research, ZEW
ference structure are: the development of the BEVs' charging in- Discussion Papers, No. 08-088.
Anderson, S., Palma, A.D., Thiesse, J.F., 1992. Discrete Choice Theory of Product
frastructure; the appearance of new car types and models both for Differentiation. MIT Press, Cambridge.
the hybrid cars and for the BEVs; and the regulations enacted by Ben-Akiva, M., Lerman, S., 1985. Discrete Choice Analysis: Theory and Application
to Travel Demand. MIT Press, Cambridge.
the national and local authorities with regards to the city center
Berkovec, J., Rust, J., 1985. A nested logit model of automobile holdings for one
and reserved lane access, as the Norwegian case demonstrates vehicle households. Transp. Res. B: Methodol. 19 (4), 275–285.
(Nayum et al., 2013). Since, the policy and regulatory framework Bliemer, M.C., Rose, J.M., 2010. Construction of experimental designs for mixed logit
models allowing for correlation across choice observations. Transp. Res. B:
and the level of knowledge of the AFVs (and BEVs) is gradually Methodol. 44 (6), 720–734.
changing also in Italy, a further empirical research will be able to Bliemer, M.C., Rose, J.M., 2011. Experimental design influences on stated choice
outputs: an empirical study in air travel choice. Transp. Res. Part A: Policy Pract.
test whether changes occurred in the preference structure of the
45 (1), 63–79.
Italian car users. Daziano, R.A., Chiew, E., 2012. Electric vehicles rising from the dead: data needs for
forecasting consumer response toward sustainable energy sources in personal
transportation. Energy Policy 51, 876–894.
Daziano, R.A., Bolduc, D., 2013. Incorporating pro-environmental preferences to-
wards green automobile technologies through a Bayesian hybrid choice model.
Table 16 Transp. A: Transp. Sci. 9 (1), 74–106.
Italian population by specific segments (2011). Daziano, R.A., Achtnicht, M., 2013. Forecasting adoption of ultra-low-emission ve-
hicles using Bayes estimates of a multinomial probit model and the GHK si-
Source: Our elaboration on National Statistical Institute (2012).
mulator. Transp. Sci. 0 (0), 1–13. http://dx.doi.org/10.1287/trsc.2013.0464.
Dimitropoulos, A., Rietveld, P., van Ommeren, J.N., 2012. Consumer Valuation of
Population segments % Absolute values
Driving Range: A Meta-Analysis. Tinbergen Institute Discussion Paper, TI 2011-
133/3, VU University, Amsterdam.
a1.b1.c1.d1 1.3 775,792 Ewing, G.O., Sarigöllü, E., 1998. Car fuel-type choice under travel demand man-
a1.b2.c1.d1 3.5 2,097,512 agement and economic incentives. Transp. Res. Part D: Transp. Environ. 3(6),
a1.b1.c2.d1 2.8 1,687,040 429–444.
a1.b1.c1.d2 2.8 1,725,807 Glerum, A., Themans, M., 2011. Modeling Demand for Electric Vehicles: The Effect of
a2.b2.c2.d2 17.8 10,767,770 Car Users' Attitudes and Perceptions. Second International Choice Modeling
a2.b1.c1.d1 1.4 823,265 Conference.
a2.b2.c1.d1 3.7 2,225,866 Hackbarth, A., Madlener, R., 2013. Consumer preferences for alternative fuel ve-
a2.b1.c2.d1 3.0 1,790,276 hicles: a discrete choice analysis. Transp. Res. Part D: Transp. Environ. 25 (0),
a2.b1.c1.d2 3.0 1,831,415 5–17.
a1.b2.c2.d1 7.5 4,561,257 Horne, M., Jaccard M., Tiedemann, K., 2005. Improving behavioural realism in hy-
a1.b2.c1.d2 7.7 4,666,070 brid energy-economy models using discrete choice studies of personal trans-
a1.b1.c2.d2 6.2 3,752,945 portation decisions. Energy Econ. 27(1), 59–77.
Huber, J., Zwerina, K., 1996. The importance of utility balance and efficient choice
a2.b2.c2.d1 8.0 4,840,375
designs. J. Market. Res. 33, 307–317.
a1.b2.c2.d2 16.7 10,146,851
Jensen, A.F., Cherchi, E., Mabit S.L., 2013. On the stability of preferences and atti-
a2.b1.c2.d2 6.6 3,982,600
tudes before and after experiencing an electric vehicle. Transp. Res. Part D:
a2.b2.c1.d2 8.2 4,951,602
Transp. Environ. 25(0), 24–32.
Total: 100.0 60,626,442 Jensen, A.F., Cherchi, E., Mabit, S.L., Ortúzar, J.D., 2014. Predicting the Potential
Market for Electric Vehicles. TRB 93rd Annual Meeting Compendium of Papers,
Notes: Whashington.
a ¼ gender (1¼ M, 2¼ F), Lave, C.A., Train, K., 1979. A disaggregate model of auto-type choice. Transp. Res. A:
b¼ family income (1 ¼ high level ( 451,000 yearly), 2¼ low level), Policy Pract. 13(1), 1–9.
c ¼ children (1 ¼ family with at least a child, 2¼ family without children), Link, C., Sammer, G., Stark, J., 2012. Modeling demand for electric cars – a metho-
d ¼age (1 ¼ o 30 years old, 2 430 years old). dical approach. Proc. – Soc. Behav. Sci. 48, 1958–1970.
56 E. Valeri, R. Danielis / Transport Policy 37 (2015) 44–56

Louviere, J.J., Hensher, D.A., 1983. Using discrete choice models with experimental Qian, L., Soopramanien, D., 2011. Heterogeneous consumer preferences for alter-
design data to forecast consumer demand for a unique cultural event. J. Con- native fuel cars in China. Transp. Res. Part D: Transp. Environ. 16, 607–613.
sum. Res. 10(3), 348–381. Rusich, A., Danielis, R., 2013. The private and social monetary costs and the energy
Mabit, S.L., Fosgerau, M., 2011. Demand for alternative-fuel vehicles when regis- consumption of a car. an estimate for seven cars with different vehicle tech-
tration taxes are high. Transp. Res. Part D: Transp. Environ. 16(3), 225–231. nologies on sale in Italy. Società Italiana di Economia dei Trasporti e della Lo-
Manski, C.F., Sherman, L., 1980. An empirical analysis of household choice among gistica (SIET), Working paper no. 13.01.
motor vehicles. Transp. Res. A: Policy Pract. 14(5/6), 349–366. Train, K.E., 2002. Discrete Choice Methods with Simulations. Cambridge University
McFadden, D., 1981. Econometrics Models of Probabilistic Choice, Structural Ana- Press, Cambridge.
lysis of Discrete Data. MIT Press, Cambridge. Unione Nazionale Rappresentanti Autoveicoli Esteri, 2013. La struttura del mercato
McFadden, D., Train, K., 2000. Mixed MNL models for discrete response. J. Appl. italiano dell'automobile immatricolazioni-Ottobre. Link: 〈http://www.unrae.it/
Econ. 15, 447–470. studi-e-statistiche/categorie/dati-statistici/item/
Nayum, A., Klöckner, C.A., Prugsamatz, S., 2013. Influences of car type class and 2635-struttura-del-mercato-ottobre-2013〉.
carbon dioxide emission levels on purchases of new cars: a retrospective Yu, J., Goos, P., Vandebroek, M., 2009. Efficient Choice-based Designs for Estimating
analysis of car purchases in Norway'. Transp. Res. Part A: Policy Pract. 48, 96– Panel Mixed Logit Models. Presented at the Leuven Statistical Day.
108.