Adoption of Electric Vehicles in Car Sharing Market

Hossein Abouee-Mehrizi
Department of Management Sciences, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada,
haboueemehrizi@uwaterloo.ca
Opher Baron, Oded Berman
Rotman School of Management, University of Toronto, Toronto M5S 3E6, Canada
opher.baron@rotman.utoronto.ca, berman@rotman.utoronto.ca
David Chen
School of Management and Economics, the Chinese University of Hong Kong, Shenzhen, China
davidchen@cuhk.edu.cn

Abstract

Problem description: We develop a model consisting of a profit-maximizing car sharing company

(CSC) and a population of utility-maximizing customers. The CSC sets the number of electric vehicles
(EVs), the number of fuel vehicles (FVs), and the rental price jointly to maximize its profit. Customers
decide whether to use EVs or FVs or public transportation to complete their trips based on the rental
price.
Practical relevance: Our work is motivated by the news that Car2go replaced all of its electric vehicle
fleet with gasoline-powered cars starting May 2016. We examine whether it is optimal to use EVs in
the car sharing market and determine the environmental impact of the optimal choice of EVs and FVs.
Methodology: We model the car sharing market as a queueing network and use the fixed population
mean approach to approximate it. This allows us to derive the optimal fraction of different types of
vehicles in the market and compare markets with different configurations.
Results: We show that it is optimal to use EVs only if the charging speed is high enough for EVs,
and both the number of charging stations and the range of EVs are large enough. Among these three
conditions, the recharging speed is the most important and the number of charging stations is more
important than the range of EVs. We also find that including EVs in the car sharing market may lead
to a higher total emission (due to a lower rental price that results in a higher usage rate).

Managerial implications: We find that the government should tax the CSC to induce a higher rental

price when the environmental impact is considered, and the government should subsidize the CSC to

reduce this price when the environmental impact is ignored. We apply our results to the case study of

Car2go in San Diego and observe that Car2go replaced EVs with FVs likely due to the slow recharging

speed rather than the lack of charging stations or the limited travel distance of EVs.

Keywords: electric vehicles, car sharing, sustainable operations, queueing networks

Electronic copy available at: https://ssrn.com/abstract=3191793

1 Introduction

Transportation is the second largest source of global greenhouse gas emissions from human ac-

tivities. According to U.S Environmental Protection Agency (2016), transportation accounts for

27 percent of greenhouse gas emissions in 2015 in the United States. Almost all of the emissions

from this sector involve fossil fuels burned for road, rail, air, and marine transportation. As a

result, global warming, urban air pollution, reduction in oil supplies, etc. become important global

problems in the near future. Electric vehicles (EVs) have been identified as one of the major oppor-

tunities to reduce greenhouse gas emissions in the transportation sector and fossil fuel consumption.

Compared to fossil fuel vehicles (FVs), EVs are more environmental friendly and their operating

cost is lower (2 cents per mile for EVs versus 12 cents plus per mile for FVs). However, there are

several major issues that hinder their widespread adoption. The relatively high price, the limited

driving range of EVs, and the lack of fast charging stations are the greatest concerns for consumers

among many others. The price of an EV is usually higher than the price of a FV (even with

tax credits or subsidies) and a typical EV on the market has a range of a few 100 miles at best.

Without enough fast charging infrastructures or battery switching stations, the recharge time for

EVs is not compatible with charging-while-travelling, which makes them inappropriate for long

trips. Moreover, the EV market faces a “chicken-and-egg” problem: people will not purchase EVs

without adequate charging infrastructure, and vast charging networks can only make profit if there

is enough demand.

To address these issues and to incentivize deployment of EVs, a number of policies have been

adopted. These policies generally fall into three categories: (1) vehicle tax credits or subsidies, (2)

investments in recharging stations and electric vehicle service equipments, and (3) research and

development (R&D). While tax credits have successfully increased the selling of EVs (for example,

the US Congressional Budget Office estimates that the current tax credits will be responsible for

30 percent of all EVs sold), other policies have limited impact on the adoption of EVs. According

to New York State Electric Vehicle Charging Station Report, the percentage of time that EVs were

connected to public charging stations, between July and September 2015, was 4.7%. Therefore,

investments in public charging infrastructure may offer marginal value in realizing the intended

benefits of EV adoption. Similarly, without successful mass production with new technology that

Electronic copy available at: https://ssrn.com/abstract=3191793

can increase the range of EVs and reduce their price, policies that encourage R&D are inefficient

in increasing their usage.

Compared to the slow adoption of EVs in the private market, the use of EVs seems highly

attractive in the car sharing market. First, car sharing can eliminate the purchase price burden of

EVs for consumers. It is a low-risk way for consumers to use EVs for their travel needs without

a large financial commitment. Second, it can address consumers’ “range anxiety” by allowing

them to choose a car that fits the trip (for example, an EV for a shorter trip and a FV for a

longer trip). Third, in shared vehicle environments, car sharing companies (CSCs) bear the costs

of infrastructure installation and the benefits from their usage. CSC can also control access to

the infrastructure to make sure that it is used effectively. This may resolve the “chicken-and-egg”

problem. Last, successful adoption of EVs in the car sharing market can increase the confidence

of consumers in the private market, which can lead to an increase in the adoption of EVs in the

private market.

Although the adoption of EVs in the car sharing market seems promising, there is still no clear

advantage in their usage in practice. While several CSCs introduce EVs into their fleet, others

replaced EVs by FVs. For examples, Zipcar launched the company’s first large-scale EV pilot

program in Chicago in March 2012 and BMW launched ActiveE DriveNow electric car sharing in

San Francisco in August 2012. In contrast, Car2go that started the car sharing service in San Diego,

California in 2011 with all electric cars, replaced all of its electric vehicle fleet with gasoline-powered

cars in May 2016 (see Car2go 2016). This is a symbolic stepback for the adoption of EVs.

Motivated by these examples, the goal of this paper is to examine how EVs can be deployed

in the car sharing market to improve consumers’ utility, CSC’s profit, and social welfare. More

specifically, we examine the following questions: (1) Is it optimal for CSCs to deploy EVs in the

car sharing market and if so, how many EVs should they deploy? (2) What factors may affect this

decision? (3) Is it environmentally wise to use EVs in the car sharing market? What policy should

the government follow to encourage the adoption of EVs in the car sharing market to increase social

welfare, or to reduce the environmental impact of greenhouse gases?

To address these questions, we develop a model consisting of a profit-maximizing CSC and a

population of utility-maximizing customers. The CSC can use both EVs and FVs, and therefore,

should decide how many of them acquire. The CSC charges customers a usage fee per unit distance

Electronic copy available at: https://ssrn.com/abstract=3191793

for each trip they make (for both EVs and FVs). Customers choose between renting a vehicle or

using public transportation, whichever is available and generates the largest net utility, to complete

a planned trip (we ignore customers who have their own vehicles as they are less likely to use car

sharing). The net utility of renting a vehicle is affected by the price. The utility of renting an EV

is also affected by the range of the EV and the environmental consciousness of the consumer. The

availability of vehicles is affected by the number of vehicles that are in the market. The CSC sets

the number of FVs, the number of EVs, and the rental price jointly to maximize its profit. We

consider two different business models: the Car2go Model, where the CSC serves the market using

either all EVs or all FVs, and the Zipcar Model, where the CSC may use both EVs and FVs in

the same market. Our main findings and contributions are:

(1) For both Car2go and Zipcar Models, we characterize conditions under which it is optimal

to use EVs in the market (see Theorems 1 and 2). In particular, we show that it is optimal for

the CSC to use EVs only if the charging speed is high enough for EVs and both the number of

charging stations and the range of EVs are large enough. Somewhat surprisingly, among these

three conditions, the recharging speed is the most important factor. If the condition about the

recharging speed is not satisfied, then it is optimal not to use any EVs no matter what the number

of charging stations and the range of EVs are. Moreover, the number of charging stations is more

important than the range of EVs.

We also show that the same conditions in the mixed usage Zipcar Model are weaker than in the

Car2go Model as it may be optimal not to use EVs in the Car2go model whereas it is optimal to

include some EVs in the Zipcar model. This may be one of the reasons that Car2go switched from

electric-powered cars back to gas-powered cars in San Francisco whereas Zipcar and Enterprise

started to include EVs in their fleets.

(2) We find that including EVs in the car sharing market may lead to a higher total emission

(see Theorem 3). This is particularly true when the charging speed for EVs is high enough and

both the number of charging stations and the range of EVs are large enough. This is because in this

case, the recharging speed, the limited range of EVs and the lack of fast charging stations are not

concerns for consumers. As a consequence, the CSC tends to charge lower prices for renting due to

the low operating cost of EVs to encourage more people to use the car sharing system. This will

induce more trips to be completed by the car sharing system. Thus, even if an EV generates less

Electronic copy available at: https://ssrn.com/abstract=3191793

emission than a FV does for the same distance, using EVs may generate a higher total emission.

(3) We also consider the problem from the perspective of a social welfare maximizer, where

social welfare is defined as the surplus of consumers plus the profit of the CSC minus the negative

environmental impact of the system. For the Car2go model, EVs are more efficient in increasing

the total social welfare than increasing the total profit of the CSC (see Theorem 4 and Proposition

5). That is, there are cases in which FVs are preferred to EVs in terms of the total profit, but EVs

are preferred to FVs in terms of social welfare. Moreover, we find that if the environmental impact

is important, the government should tax the CSC to induce a higher rental price in the market.

Otherwise, the government should subsidize the CSC to induce a lower rental price.

For the Zipcar model, we show that increasing the recharging rate of EVs or increasing the

number of charging stations can lead to a higher total social welfare. However, increasing the range

of EVs may reduce social welfare. This is the case if the the recharging speed or the number of

charging stations is low, resulting in a lower marginal social welfare if the trips are completed by

EVs. By numerical studies, using parameters estimated from the real cases, we find that social

welfare would increase with the range of EVs. This suggests that if the government can adopt

different policies to coordinate the system, increasing either the recharging speed, or the number

of charging stations, or the range of EVs can benefit the society.

2 Literature Review

The introduction of EVs has been considered in road logistics and transportation, with special

attentions to the environmental impact and the emerging strategic and operational issues. New

variants of the vehicle routing problems arise due to the introduction of EVs in the distribution fleet;

see the recent review by Juan et al. (2016). There are a few papers in the literature of operations

management that focus on sustainable operations related to the EV business models. Mak et al.

(2013) study the problem of deploying battery-swapping infrastructure for an EV charging service

provider. They develop two distributionally robust optimization models for the swapping station

location problem and examine the potential impact of battery standardization and technology

advancements on the optimal infrastructure deployment strategy. Lim et al. (2015) study the

impact of range anxiety and resale anxiety to mass adoption of EVs. They also examine how

Electronic copy available at: https://ssrn.com/abstract=3191793

different business models would affect the adoption of EVs. Avci et al. (2015) study the adoption

of EVs in a battery-swapping system and contrast it with conventional EVs (without battery-

swapping option). They examine the environmental impact of such a system and find that it

cannot effectively reduce emissions and oil dependence at the same time. All the papers discussed

above focus on the mass adoption of EVs or the adoption in the private market.

Our work is also related to the literature of the sharing economy. The car sharing business has

evolved from “round trip car sharing” to “station-based one way car sharing” and “free-floating one

way car sharing”. Al-Alawi and Bradley (2013) and Tanaka et al. (2014) provide a comprehensive

literature review on vehicle technology adoption studies. Stillwater et al. (2009), De Lorimier and

El-Geneidy (2013), and Le Vine et al. (2014) investigate the decision-making behavior of the car

sharing users. The papers by George and Xia (2011) and Fricker and Gast (2016) are the most

relevant to our study. Both papers model bike/car sharing systems as closed queueing networks

and use steady state approximations to evaluate system performance. The objective is to determine

the optimal fleet size and find principles to reduce the imbalances in the network, which is different

from our paper. Recently, there is a new type of car sharing operation in the emerging sharing

economy: the peer-to-peer sharing model. A number of papers have studied the emerging strategic

and operational issues in the peer-to-peer sharing market, including impact of sharing economy

on the welfare and consumption, matching mechanism, wage and price of on-demand services, and

etc.; see for example Benjaafar et al. (2018), Cachon et al. (2016), Hu and Zhou (2016), Tang et al.

(2016), and Taylor (2016).

There is a growing body of literature that focuses on servicizing business models. Servicizing

business models refer to models under which a firm sells the use of a product rather than the

product itself. Readers are referred to Orsdemir et al. (2015) and Agrawal and Bellos (2016) for

discussions and examples of servicization and the impact of servicization on ownership and usage,

profitability of the firm, and the environment. Bellos et al. (2017) consider an auto manufacturer

that offers car sharing in conjunction with conventional sales and examine the interaction between

business model choice, product line design, and environmental regulation by balancing the trade-off

between driving performance and fuel efficiency. They show that offering car sharing may not be

environmentally beneficial. In their paper, only FVs are considered.

To the best of our knowledge, there is very limited literature that examine the adoption of

Electronic copy available at: https://ssrn.com/abstract=3191793

electric vehicles in the car sharing market. Zoepf (2015) utilize mixed-integer programming and

simulation for the assignment of trips to vehicles, and a hybrid choice model for consumer pref-

erences to explore the feasibility of EV deployment in the car sharing market. The simulation

results suggest that a large fraction of round-trip car sharing fleets could be converted to EVs such

that the profit is increased and and gasoline consumption is reduced. Different from this paper,

we model the circulation of the vehicles as a queueing network and analytically derive conditions

under which it is optimal to use EVs in the car sharing market. We also analytically examine the

environmental impact of using EVs in the market. He et al. (2017) study the service region design

problem for a car sharing system with all electric vehicles. Different from our paper, their objective

is to determine the service region and the fleet size of the CSC and their approach is distributionally

robust optimization.

3 Model with Profit Maximization

We consider a car sharing market consisting of a profit-maximizing car sharing company (CSC)

and a population of utility-maximizing customers. The CSC can choose both EVs and fossil-fuel

vehicles (FVs) to serve the market and decides on how many EVs and FVs to use. Customers

choose between renting a vehicle or using public transportation to complete a planned trip (we

exclude customers who choose to have their own vehicles as they are significantly less likely to use

car sharing). The CSC charges customers a usage fee p per unit distance for each trip they make

(for both EVs and FVs). Note that in practice, the usage fee is usually charged per unit of time.

However, most of the trips are non-stop ones and the speed of each trip is almost the same in places

where car sharing happens (e.g., urban area). Thus, the usage fee can be approximately converted

to per unit distance. The CSC sets the number of FVs (NF ), the number of EVs (NE ), and the

usage fee (p) jointly to maximize its profit.

Customers are heterogeneous in their travel distance ξ. Note that the car sharing business is

usually restricted to a certain area and the maximum travel distance d¯ that someone can travel

using a vehicle in the care sharing system is limited. Without loss of generality, we assume d¯ = 1

and therefore ξ ∈ [0, 1] (if ξ > 1, then only public transportation, privately owned, or standard

car rental can be used). Suppose the arrival of each trip follows a Poisson process and let f (ξ) be

Electronic copy available at: https://ssrn.com/abstract=3191793

 R1
the arrival rate for trips with travel distance ξ. Then, Λ = 0 f (ξ)dξ is the total arrival rate and
f (ξ)
fˆ(ξ) = Λ is the density of customers whose travel distance is ξ.

When there are two types of vehicles (EVs and FVs) in the market, customers can choose to

rent an EV or a FV or to use public transportation. However, due to the short range of EVs

without enough battery swapping stations, customers cannot use EVs to complete long trips (in

car sharing market, it is impractical for customers to wait for recharging during a single trip). We

assume the range limit of an EV is r ∈ (0, 1]. Thus, customers with travel distance ξ > r can only

choose between renting a FV and using public transportation. We assume that the utility for a

customer with travel distance ξ is u(ξ) if she rents a vehicle (either an EV or a FV) and normalize

the utility that each customer gains using public transportation to be 0, where u(·) is an increasing

and concave function with u(0) ≥ 0. Moreover, customers derive an additional “green” utility

g(ξ) from renting an EV. Note that g(ξ) can also be viewed as the customers’ curiosity towards

the new technology. That is, when choosing between the new technology and the old technology

for the same price, customers are always willing to experience the new technology. We assume

g(ξ) is increasing, i.e., the longer the travel distance, the more they contribute to the environment

(compare to using FVs). Therefore, when both EVs and FVs are available, customers always prefer

EVs to FVs. We assume that g(ξ) = δξ, where δ measures the environmental consciousness of the

population. We note that our results and insights hold for the case with g(ξ) = δ1ξ>0 , which is

considered in Avci et al. (2015). Customers pay the usage fee p per unit distance for their trips

when using the car sharing system (with either an EV or a FV). The gas/electric charging cost is

considered as part of the CSC’s costs as detailed below. For each trip, customers choose the option

that is available and generates the largest net utility.

Note that according to the above assumptions, whenever there are EVs (FVs) available and using

EVs (FVs) generates the largest net utility, customers will always choose EVs (FVs). However, in

practice, unrented vehicles may be parked far from the customers and thus customers will choose

to not rent a vehicle. To address this issue, we assume that such cases occur with probability

ω. That is, when EVs (FVs) are available and using EVs (FVs) generates the largest net utility,

customers will choose to use EVs (FVs) with probability 1 − ω and choose to explore other options

with probability ω. We assume ω = 0 in the paper but all the results and the analyses continue to

hold for any other fixed ω ∈ [0, 1). In practice, ω may depend on the number of unrented vehicles

Electronic copy available at: https://ssrn.com/abstract=3191793

in the market. The larger it is, the large ω would be. However, taking into account this dependency

would significantly complicate the problem. Thus, for analytical tractability, we assume ω is fixed.

3.1 Benchmark: Market with Only FVs

We first consider the market with only FVs as a benchmark. In this case, customers will choose

to rent a FV if the net utility is larger than 0 and use public transportation otherwise. That is,

renting is preferred if and only if u(ξ) − pξ ≥ 0. Let xF be the unique solution to u(ξ) − pξ = 0

(the uniqueness of the solution is due to the concavity of u(ξ) and that u(0) ≥ 0). Then, a FV will

be chosen if and only if ξ ≤ xF .

The CSC sets the number of FVs (NF ) and the rental price per unit distance p jointly to

maximize its profit. It generates p units of revenue from renting the vehicles per unit distance,

incurs the operation cost cF per unit distance (for example, the refueling cost for gas or oil and the

maintenance cost), and also pays a fixed cost, MF , for purchasing each unit of FVs (MF is the fixed

cost per unit time, which equals the purchasing cost divided by the lifetime of the FV). A given

trip with travel distance ξ will be completed by renting if and only if (1) u(ξ) − p(ξ) ≥ 0, and (2)

there is a vehicle available. Let OF = {ξ : u(ξ) − pξ ≥ 0} = [0, xF ] denote the set of customers who

prefer renting a FV to using public transportation and αF denote the probability that a consumer

finds a FV available whenever she wants to rent one. Then, the CSC’s total profit is given by

Z
ΠF (NF , p) = (p − cF )αF ξf (ξ)dξ − MF NF .
OF

Note that the more FVs the CSC purchases, the higher the probability that a consumer finds

a FV available whenever she wants to rent one. Therefore, there is a one-to-one correspondence

between NF and αF and choosing NF is equivalent to choosing αF .

Given αF , to compute NF , we model the car sharing market as a closed queueing network,

which represents the circulation of FVs. Consider the circulation of a FV. First, this vehicle will be

in the parking lot waiting for customers. Without loss of generality, we assume that customers pick

vehicles based on the first-in first-out rule. That is, a particular FV will be chosen if and only if all

FVs that are in front of it have already been rented. Thus, we can model this part as a queueing

system with a single server where the service time is the time between two consecutive customer

Electronic copy available at: https://ssrn.com/abstract=3191793

arrivals, which is exponentially distributed. In practice, unrented vehicles may not be picked in the

same sequence as they are parked. Customers would likely pick the one closest to them. However,

the limiting behavior of the queue length is independent of the sequence.

Once a FV is picked by a customer, it will be returned to the parking lot after the trip. We

can then model this process as a queueing system with infinite servers since the usage duration

of each customer is independent of that of others. This service time is also independent of the

number of FVs that are being used. Thus, this network consists of two nodes, a •/M/1 node and a

•/G/∞ node. Note that for the •/M/1 node, the probability that a customer finds no FVs available

whenever she wants to rent one is equal to the fraction of time this node has an empty queue.

Similar to Bellos et al. (2013), we approximate this closed queueing network by using the fixed

population mean approach, see Whitt (1984) and Toktay et al. (2000). In particular, the •/M/1

node is approximated by an M/M/1 node and the •/G/∞ node is approximated by an M/G/∞

node. As mentioned in Bellos et al. (2013), this method is particularly accurate for queueing

networks with (1) a large population, and (2) a M/G/∞ node of slow service rate, both are

satisfied in our model.

R
Note that the demand for renting a FV follows a Poisson process with rate OF f (ξ)dξ. This
R
implies that the service rate of the M/M/1 node is µ = OF f (ξ)dξ. Let λF be the arrival rate of
λF
the M/M/1 node. By the definition of αF , the utilization of this node is αF , i.e., αF = µ . Thus,

the arrival rate of the M/M/1 node, which is the departure rate of the M/M/1 node and also the
R
arrival rate of the M/G/∞ node, is λF = αF µ = αF OF f (ξ)dξ. The expected number of FVs in
λF αF
this M/M/1 node is E(NM/M/1 ) = µ−λF = 1−αF .

Suppose that the average speed of all the trips is s. For a trip with travel distance ξ, the time
ξ
spent in the M/G/∞ node is s (we assume that the refueling time for FVs is negligible). The prob-
R f (ξ)
ability density that the distance of a trip is ξ given that this trip is completed by a FV is f (y)dy
.
OF
R ξ R
f (ξ)dξ s f (ξ)dξ
Therefore, the service rate of the M/G/∞ node is τ = 1/( ROF s
)= R OF
and the expected
f (ξ)dξ ξf (ξ)dξ
OF OF
R
s f (ξ)dξ
f (ξ)dξ/( R OF
R
number of EVs in this M/G/∞ node is E(NM/G/∞ ) = λF /τ = αF OF ξf (ξ)dξ
) =
OF
αF
R
s OF ξf (ξ)dξ.

In equilibrium, the number of FVs in the market should equal to the total number of vehicles
αF αF R
in both nodes. That is, NF = E(NM/M/1 ) + E(NM/G/∞ ) = 1−α F
+ s OF ξf (ξ)dξ. Therefore, the

Electronic copy available at: https://ssrn.com/abstract=3191793

CSC’s optimization problem is:

αF αF
Z Z
max (p − cF )αF ξf (ξ)dξ − MF ( + ξf (ξ)dξ). (1)
p≥0,0≤αF ≤1 OF 1 − αF s OF

u(xF )
Note that OF = [0, xF ] and by the definition of xF , p = xF . Thus, we can rewrite (1) as

xF xF
u(xF ) αF αF
Z Z
max ( − cF )αF ξf (ξ)dξ − MF ( + ξf (ξ)dξ). (2)
0≤xF ≤1,0≤αF ≤1 xF 0 1 − αF s 0

Let πF be the optimal profit of the CSC in this case and x∗F be the largest optimal solution.
u(x∗F )
Then p∗F = x∗F is the optimal price.

3.2 Car2go Model: Market with either All EVs or All FVs

We now consider the business model of Car2go in which the CSC serves the market using either

all EVs or all FVs. Car2go provides car sharing services using either EVs or FVs in many cities

in Europe and North America. For example, in Vancouver, Austin, and Berlin, Car2go offers only

gas-powered cars whereas in Madrid, Stuttgart, and Amsterdam, and in San Diego before May 1

2016, only EVs are provided in the market.

When deciding on whether to use EVs or FVs to serve the market, the CSC will choose the

one that generates a higher profit. If the CSC uses only FVs, the optimal profit is given by πF in

Section 3.1. Next, we compute the optimal profit if the CSC uses only EVs.

When there are only EVs in the car sharing market, customers with travel distance ξ > r

can only use public transportation. For customers with travel distance ξ ≤ r, if they choose to

rent an EV, the net utility is u(ξ) − pξ + δξ. If they choose to use public transportation, the net

utility is 0. Thus, customers with travel distance ξ ≤ r will choose to rent an EV if and only if

u(ξ) − pξ + δξ ≥ 0. Let xE be the unique solution to u(ξ) − pξ + δξ = 0. If xE > r, all customers

with travel distance ξ ≤ r will rent an EV. If xE ≤ r, then customers with travel distance ξ ≤ xE

will rent an EV and customers with travel distance ξ > xE will choose to use public transportation.

Let OE denote the set of customers who prefer renting an EV to using public transportation. Then,

OE = [0, min{r, xE }].

Similar to Section 3.1, the CSC sets the number of EVs (NE ) and the rental price per unit

10

Electronic copy available at: https://ssrn.com/abstract=3191793

distance p jointly to maximize its profit. Let cE be the operation cost per distance (for example,

the recharging cost for electricity and the maintenance cost), ME be the fixed cost for purchasing

each unit of EVs (per unit time). Then, the CSC’s total profit is given by

Z
ΠE (NE , p) = (p − cE )αE ξf (ξ)dξ − ME NE .
OE

Again, we model the car sharing market with only EVs as a queueing network. Differently from

FVs, EVs need to be recharged after each trip before it can be rented again. Thus, we add another

•/G/k node after the •/G/∞ into the network and approximate the two nodes by an M/G/k node

and a M/G/∞ node, where k is number of charging station. This M/G/k node represents the

recharging time for each EV after the trip.

In equilibrium, the number of EVs in the market should equal to the total number of vehicles

in the three nodes: NE = E(NM/M/1 ) + E(NM/G/∞ ) + E(NM/G/k ). Similar to Section 3.1, we have
αE αE
R
E(NM/M/1 ) = 1−α E
and E(N M/G/∞ ) = s OE ξf (ξ)dξ.

To compute E(NM/G/k ), suppose the average recharging speed is sr (here we assume the con-

sumption and recharge time of battery are both linear with respect to the travel distance). That
ξ
is, for a trip with travel distance ξ, it takes (units of time) to fully recharge the EV. Then, the
Rsr ξ R
f (ξ)dξ sr f (ξ)dξ
service rate of the M/G/k node is τr = 1/( OR E srf (ξ)dξ ) = R OEξf (ξ)dξ and the arrival rate for the
OE OE
q
R
M/G/k node is λE = αE OE f (ξ)dξ. Let L denote the expected queue length in this M/G/k

node. Clearly, Lq only depends on k and ρ = λE /τr = αsEr OE ξf (ξ)dξ. Thus, we write Lq as
R

Lq (k, ρ). We make the following assumption about Lq (k, ρ).

Assumption 1. Lq (k, ρ) is decreasing and convex in k, increasing and convex in ρ, and is sub-

modular in (k, ρ).

This assumption is intuitive and consistent with the numerical results.

Then, we have E(NM/G/k ) = λτEr + Lq (k, ρ) = αsEr OE ξf (ξ)dξ + Lq (k, ρ) and NE = 1−α αE
R
E
+
αE αE q
R R
s OE ξf (ξ)dξ + sr OE ξf (ξ)dξ + L (k, ρ). Therefore, the CSC’s optimization problem is:

αE 1 1
Z Z
max (p − cE )αE ξf (ξ)dξ − ME [ + αE ( + ) ξf (ξ)dξ + Lq (k, ρ))]. (3)
p≥0,0≤αE ≤1 OE 1 − αE s sr OE

u(r)
Note that when p ≤ r + δ, we have xE ≥ r. This implies that all customers with travel

11

Electronic copy available at: https://ssrn.com/abstract=3191793

distance ξ ≤ r will choose to rent an EV. That is, OE = [0, r]. Then,

r r
αE 1 1
Z Z
ΠE (NE , p) = (p − cE )αE ξf (ξ)dξ − ME [ + αE ( + ) ξf (ξ)dξ + Lq (k, ρ)],
0 1 − αE s sr 0

which is increasing in p. Therefore, for problem (3), we only need to consider the region where
u(r) u(xE )
p≥ r + δ. In this region, we have xE ≤ r, OE = [0, xE ], and p = xE + δ. Thus, we can rewrite

problem (3) as

xE xE
u(xE ) αE 1 1
Z Z
max ( + δ − cE )αE ξf (ξ)dξ − ME [ + αE ( + ) ξf (ξ)dξ + Lq (k, ρ)].
0≤xE ≤r,0≤αE ≤1 xE 0 1 − αE s sr 0
(4)

Let πE be the optimal profit and x∗E be the largest optimal solution in this case.

Theorem 1. Suppose ME = MF . Then,

ME
(1) if sr ≤ cF −cE +δ , πF ≥ πE ;

ME
(2) if sr > cF −cE +δ , there exists a threshold k∗ (sr ) such that

(i) if k ≤ k∗ (sr ), πF ≥ πE ;

(ii) if k > k ∗ (sr ), there exists a threshold r ∗ (sr , k) such that if r ≤ r ∗ (sr , k), πF ≥ πE and if

r > r ∗ (sr , k), πF ≤ πE .

Theorem 1 characterizes the conditions under which it is optimal to use only EVs and when it is

optimal to use only FVs. Specifically, when the recharging speed for EVs is low, FVs are preferred

to EVs. When the recharging speed is high enough, FVs are still preferred to EVs when the number

of charging stations is low. It is optimal for the CSC to use all EVs only if (a) the recharging speed

for EVs is high enough, (b) the number of charging stations is large enough, and (c) the range of

EVs is large enough. Among these three conditions, the recharging speed is the most important.

If the condition about the recharging speed is not satisfied, FVs are always preferred no matter

what the number of charging stations and the range of EVs are. Moreover, the number of charging

stations is more important than the range of EVs.

ME
Theorem 1 is illustrated in Figure 1 for k = ∞. As we can see, when sr ≤ cF −cE +δ , πF ≥ πE .
ME
When sr > cF −cE +δ , there is a threshold on r (the curve r ∗ (∞)) such that if r is less than the

12

Electronic copy available at: https://ssrn.com/abstract=3191793

 Figure 1: Comparison of the optimal profit between markets with only EVs and with only FVs

threshold, πF ≥ πE and if r is larger than this threshold, πF ≤ πE . As Theorem 1 suggests, a

similar structure holds for k < ∞.

In Theorem 1, we assume that ME = MF . In the private market, the price of an EV is

usually higher than the price of a conventional vehicle. However, governments provide tax credits

or subsidies to consumers who purchase an EV. The prices of the two types of vehicles are very

close after these benefits are taken into account. If ME ≥ MF , then FVs are even more preferred

when the recharging speed for EVs is low, or the number of charging stations is small, or the range

of EVs is small.

3.3 Zipcar Model: Market with Both EVs and FVs

Here, we consider the business model of Zipcar in which the CSC may use both EVs and FVs in

the same market. In this case, the CSC needs to decide on how many EVs and FVs to use.

Note that customers with travel distance ξ > r can only choose to rent a FV or use public trans-

portation. If u(ξ) − pξ ≥ 0, they will choose a FV. Otherwise, they will use public transportation.

For customers with travel distance ξ ≤ r, EVs are always preferred to FVs due to the “green” util-

ity. Thus, if u(ξ) − pξ + g(ξ) ≥ 0, they will first choose to rent an EV. When no EVs are available,

they will then choose between renting a FV or using public transportation, whichever provides a

higher utility. Again, if renting a FV is preferred to public transportation, she will rent a FV when

there are any FVs available and use public transportation otherwise. If u(ξ) − pξ + g(ξ) < 0, they

will use public transportation.

13

Electronic copy available at: https://ssrn.com/abstract=3191793

 Let OEF denote the set of customers whose first choice is to rent an EV and second choice

is to rent a FV, let OEP denote the set of customers whose first choice is to rent an EV and

second choice is to use public transportation, and OF P denote the set of customers whose first

choice is to rent a FV and second choice is to use public transportation. Let αE denote the

probability that a consumer finds an EV available whenever she wants to rent an EV and αF

denote the probability that a consumer finds a FV available whenever she wants to rent a FV.
R
Then, the total arrival rate for EVs is ΛE = OE ξf (ξ)dξ and the total arrival rate for FVs is
R R
ΛF = (1 − αE ) OEF f (ξ)dξ + OF P f (ξ)dξ. In this case, the CSC’s total profit is

ΠEF (NE , NF , p) = (p − cE )αE ΛE + (p − cF )αF ΛF − ME NE − MF NF .

Similar to Sections 3.1 and 3.2, we approximate the circulation of FVs using a queueing network

with two nodes and the circulation of EVs by a queueing network with three nodes. Thus, NF =
αF αF αE 1 1 q
R R R
1−αF + s [(1−αE ) OEF ξf (ξ)dξ+ OF P ξf (ξ)dξ] and NE = 1−αE +αE ( s + sr ) OE ξf (ξ)dξ+L (k, ρ).

The CSC’s problem becomes

Z Z Z
max (p − cE )αE ξf (ξ)dξ + (p − cF )αF [(1 − αE ) ξf (ξ)dξ + ξf (ξ)dξ]
p≥0,0≤αE ,αF ≤1 OE OEF OF P
αE 1 1
Z
− ME [ + αE ( + ) ξf (ξ)dξ + Lq (k, ρ)]
1 − αE s sr OE
αF αF
Z Z
− MF [ + [(1 − αE ) ξf (ξ)dξ + ξf (ξ)dξ]].
1 − αF s OEF OF P

(5)

Figure 2: Illustrations of OE , OEF , and OF P for different r

Note that OE , OEF , and OF P are determined by xE , xF , and r. We consider three cases:

14

Electronic copy available at: https://ssrn.com/abstract=3191793

(1) xF ≤ xE ≤ r, (2) xF ≤ r ≤ xE , and (3) r ≤ xF ≤ xE . These three cases are illustrated

in Figure 2. For the first case, OE = OEF + OEP = [0, xE ], OEF = [0, xF ], and OF P = ∅. For

the second case, OE = OEF + OEP = [0, r], OEF = [0, xF ], and OF P = ∅. For the last case,
u(xE ) u(xF )
OE = OEF + OEP = [0, r], OEF = [0, r], and OF P = [r, xF ]. Moreover, p = xE +δ = xF . The

detailed formulation for each case is provided in the proof of Proposition 2 in the appendix.

We first compare the optimal profits of the two different business models. Clearly, when the

CSC can use both EVs and FVs in the car sharing market, it will have higher profit than the profit

in the market with either all EVs or all FVs. That is, the Zipcar model has a higher profit than

the Car2go model. We state this obvious observation below (no formal proof is provided).

Observation 1. πEF ≥ max{πE , πF }.

In the Zipcar model, we assume that when using both EVs and FVs in the car sharing market,

the CSC does not incur the additional fixed cost for the two types of vehicles. This may not be

the case in practice. For example, the firm may require two different maintenance teams for EVs

and FVs because of the different technologies. This additional fixed cost can be significant, which

explains why both models coexist in the market.

Next, we examine when it is optimal for the CSC to use EVs in the market with the Zipcar

model.

u(xE )
Theorem 2. Let (x∗F , α∗F ) be the optimal solution to (1) and x̂FE be the unique solution to xE +δ =
u(x∗F )
x∗F . Then,

(1) when x̂FE < r, NE > 0 only if

x̂F x∗F
u(x∗F ) 1 1 ∂Lq u(x∗F ) MF ∗
Z Z
E
( ∗ −cE −ME ( + + (k, 0))) ξf (ξ)dξ−( −cF − )αF ξf (ξ)dξ−ME > 0;
xF s sr ∂ρ 0 x∗F s 0

(2) when x∗F < r ≤ x̂FE , NE > 0 only if

r x∗F
u(x∗F ) 1 1 ∂Lq u(x∗F ) MF ∗
Z Z
( −cE −ME ( + + (k, 0))) ξf (ξ)dξ −( −cF − )αF ξf (ξ)dξ −ME > 0;
x∗F s sr ∂ρ 0 xF∗ s 0

(3) when x∗F ≥ r, NE > 0 only if

r r
u(x∗ ) 1 1 ∂Lq u(x∗ ) MF ∗
Z Z
( ∗F − cE − ME ( + + (k, 0))) ξf (ξ)dξ − ( ∗F − cF − )αF ξf (ξ)dξ − ME > 0.
xF s sr ∂ρ 0 xF s 0

15

Electronic copy available at: https://ssrn.com/abstract=3191793

Moreover, if the objective function in (5) is concave in (xE , xF , αE , αF ), then the above statements

are if and only if statements.

Theorem 2 provides necessary conditions under which it is optimal for the CSC to include EVs

in the fleet. When the objective function in (5) is concave in (xE , xF , αE , αF ), these conditions are

also sufficient to have EVs in the market. Note that for general utility function u(ξ) and general

arrival rate f (ξ), we have no closed form solution for xE , xF , αE , and αF . Thus, we cannot

characterize conditions under which the objective function in (5) is concave in general. It can be

verified that when u(ξ) = 1b (aξ − 21 ξ 2 ) with a ≥ 1 and f (ξ) = λξ β (1 − ξ)γ with β ≥ γ ≥ 0 , the

objective function is concave (a similar utility specification and related discussions can be found in

Lambrecht et al., 2007 and Bellos et al., 2013). Note that f (ξ) denotes the beta distribution and

when β = γ = 0, it is the uniform distribution. In these cases, the conditions provided in Theorem

2 are also sufficient.

Note that the left hand sides of the inequalities in conditions given in Theorem 2 is increasing

in sr , k, and r. Thus, those conditions can be satisfied for large sr , k, and r. In other words, it

is optimal to adopt EVs in the car sharing system when the recharge speed is high, the number of

charging station is large, and the range of EVs is large.

Proposition 1. There exists three thresholds sZ Z Z

r , k (sr ) and r (k, sr ) such that NE > 0 if and
ME
only if sr > sZ Z Z Z
r , k > k (sr ) and r > r (k, sr ). These thresholds are bounded by sr ≤ cF −cE +δ ,

kZ (sr ) ≤ k∗ (sr ) and r Z (k, sr ) ≤ r ∗ (k, sr ).

The above proposition also implies that the conditions provided in Theorem 2 are weaker than

the conditions in Theorem 1. This suggests that it may be optimal to not include EVs in the

Car2go model when it is optimal to have EVs in the Zipcar model. Specifically, the manifestation

of Theorem 2 is that limiting the vehicles to either EVs or FVs removes the region where having

EVs is profitable. This also suggests one reason that might lead to Car2go to switch back from

electric-powered cars to gas-powered cars in San Francisco whereas Zipcar and Enterprise start to

include EVs in their fleets.

16

Electronic copy available at: https://ssrn.com/abstract=3191793

4 Environmental Impact for Car2go

In this section, we compare the benchmark model with the Car2go model in terms of total emission.

For the Zipcar model, due to the complexity of the problem, we cannot analytically make this

comparison, so we compare them numerically in Section 6, where we also examine how different

factors affect the total emission. This comparison is of particular interest because a primary reason

to encourage the adoption of EVs is to reduce carbon emission in the transportation sector. Let

eF and eE be the carbon footprints per unit distance for FVs and EVs, respectively, with eF ≥ eE

(according to U.S. Department of Energy 2016, the carbon footprint of EVs is around 25% of that

of FVs).

We assume that the impact of the car sharing users on the emission of the public transportation

is negligible. This is a reasonable assumption since each customer will only marginally change

the trips made by public transportation and marginally increase the travel distance of the public

transportation. Therefore, the total emission in the benchmark model is due to the FVs and can
R x∗
be characterized by EMF = eF α∗F 0 F ξf (ξ)dξ. The total emission in the Car2go model is EMF if
R x∗
πF ≥ πE and is EME = eE α∗E 0 E ξf (ξ)dξ if πE ≥ πF .

Theorem 3. Suppose ME = MF . Then, there exists a threshold s̃∗r such that

(1) if sr ≤ s̃∗r , EMF ≥ EME ;

(2) if sr > s̃∗r , there exists a threshold k̃∗ such that

(i) if k ≤ k̃∗ (sr ), EMF ≥ EME ;

(ii) if k > k̃∗ (sr ), there exists a threshold r̃ ∗ (sr , k) such that if r ≤ r̃ ∗ (sr , k), EMF∗ ≥ EME∗

and if r > r̃ ∗ (sr , k), EMF ≤ EME .

ME
Moreover, s̃∗r ≥ cF −cE +δ , k̃∗ (sr ) ≥ k∗ (sr ), and r̃ ∗ (sr , k) ≥ r ∗ (sr , k).

Theorem 3 shows that the market with only EVs may generate more emission than the market

with only FVs even if the carbon footprint of EVs is much lower than that of FVs. This is

particularly the case when the recharging speed for EVs, the number of charging stations, and the

range of EVs are all large. This observation is consistent with Zipcar (2015) and can be explained

as follows. Under these conditions, the low operating cost of EVs enables the CSC to charge a

17

Electronic copy available at: https://ssrn.com/abstract=3191793

lower rental price. This low price encourages more people to use the car sharing system. Thus,

even though an EV generates less emission than a FV for a given distance, overall the market

with only EVs will generate more emission since more trips will be completed by the car sharing

system. However, when the recharging speed for EVs is low, or the number of charging stations is

not sufficient, or the range of EVs is small, the market with only EVs will generate less emission.

Figure 3: Comparison of profit and emission between markets with only EVs and with only FVs

Figure 3 illustrates the comparison between the market with only EVs and the market with

only FVs with regards to the optimal profit and total emission for k = ∞ (for k < ∞ the borders

between the regions are different but the qualitative results are similar). As illustrated, the whole

region can be divided into three parts. In region I, FVs provide more profit to the CSC than

EVs, but also generate more emission. Thus, the benchmark and the Car2go model generates the

same emission. In region II, EVs provide more profit than FVs and also generate less emission.

Thus, the Car2go model generates less emission than the benchmark model. In region III, EVs

are preferred over FVs in terms of the total profits. However, the total emission in the market

with only EVs are larger than the emission in the market with only FVs. Thus, the Car2go model

generates more emission than the benchmark model. Therefore, in the Car2go model, region II is

the most desirable one since the CSC and the environment both benefit from the adoption of EVs.

Proposition 2. In the Car2go model, the total emission is first decreasing and then increasing in

sr , k, and r.

ME
This can also be observed in Figure 3. For example, if sr > cF −cE +δ , when r is small (in region

18

Electronic copy available at: https://ssrn.com/abstract=3191793

I), it is optimal to use FVs and the total emssion is EMF . When r is in its middle range (in region

II), it is optimal to use EVs and the total emission is EME , which is less that EMF . However,

when r continues to increase (in region III), the total emission of EVs will increase so that it will

be larger than EMF . Thus, the total emission is first decreasing ant then increasing in the range of

EVs. Therefore, policy makers should be cautious when considering policies that can increase the

recharging speed, the number of charging stataions, and the range of EVs. Too aggressive policies

may lead to more emission in the car sharing market if the CSC provides either EVs or FVs in the

market, which may harm the environment.

5 Social Welfare Maximization

In this section, we study the problem from the perspective of a social welfare maximizer. The social

welfare includes the surplus of customers, the profit of the CSC, and the environmental impact of

the system. In particular,

social welfare = total surplus of customers + profit of the CSC − η · total emission,

where η is the coefficient that measures the weight of the environmental impact. The larger η is,

the more the government cares about the environmental impact of the system. We are interested

in the optimal rental price and the optimal number of vehicles that maximize social welfare.

5.1 Social Welfare under the Car2go Model

We first examine the Car2go model. We compare the system with only FVs with the system with

only EVs. For the system with only FVs, given the rental price p and the number of FVs, NF (or

equivalently the probability that a consumer finds a FV available whenever she wants to rent one

αF ), the surplus of the customer with travel distance ξ is u(ξ)−pξ if she rents a FV and 0 if she uses
R
the public transportation. Thus, the total surplus of customers is αF OF (u(ξ) − pξ)f (ξ)dξ. The
αF
+ αsF OF ξf (ξ)dξ),
R R
profit of the CSC is given by (1), which is (p − cF )αF OF ξf (ξ)dξ − MF ( 1−α F
R
and the expected total emission is eF αF OF ξf (ξ)dξ. Therefore, in this case, the total social welfare

19

Electronic copy available at: https://ssrn.com/abstract=3191793

is

αF αF
Z Z Z
ΠSF (p, αF ) = αF (u(ξ) − pξ)f (ξ)dξ + (p − cF )αF ξf (ξ)dξ − MF ( + ξf (ξ)dξ)
OF OF 1 − αF s OF
Z
− ηeF αF ξf (ξ)dξ
OF
MF αF
Z
= αF [u(ξ) − (cF + + ηeF )ξ]f (ξ)dξ − MF .
OF s 1 − αF

To maximize the total social welfare, we need to solve the following problem:

MF αF
Z
max αF [u(ξ) − (cF + + ηeF )ξ]f (ξ)dξ − MF . (6)
p≥0,0≤αE ≤1 OF s 1 − αF

Proposition 3. For the system with only FV, the optimal rental price and the optimal number of

FVs that maximize social welfare are

MF
pS∗
F = cF + + ηeF ,
s
Z xS∗
αS∗ αS∗ F
NFS∗ = F
S∗
+ F
ξf (ξ)dξ,
1 − αF s 0

r
MF
where αS∗
F =1− R xS∗ MF
and xS∗
F is the solution to
F [u(ξ)−(cF + +ηeF )ξ]f (ξ)dξ
0 s

MF
u(ξ) − (cF + + ηeF )ξ = 0.
s

Proposition 3 shows that to maximize social welfare, the optimal rental price should equal the

operational cost, plus the fixed ordering cost (normalized to per unit distance) and the cost of

emission. It can be verified that pS∗ S∗

F is increasing in η and NF is decreasing in η. That is, when

the environmental impact of emission increases, so does the optimal rental price but the optimal

number of FVs in the market decreases.

Similarly, for the system with only EVs, given the rental price p and the number of EVs, NE

(or equivalently the probability that a consumer finds an EV available whenever she wants to rent

20

Electronic copy available at: https://ssrn.com/abstract=3191793

one, αE ), the optimization problem becomes

1 1 αE
Z
max αE [u(ξ) − (cE − δ + ME ( + ) + ηeE )ξ]f (ξ)dξ − ME − ME Lq (k, ρ). (7)
p≥0,0≤αE ≤1 OE s s r 1 − αE

Proposition 4. For the system with only EVs, the optimal rental price and the optimal number of

EVs that maximize the total social welfare are

ME ME ∂ q
pS∗
E = cE + + ηeE + (1 + L (k, ρ)),
s sr ∂ρ
Z min{r,xS∗ }
αS∗ S∗ 1 1 E
NES∗ = E
S∗
+ αE ( + ) f (ξ)dξ + Lq (k, ρ),
1 − αE s s r 0

where xS∗
E is the solution to

ME ME ∂ q
u(ξ) − (cE − δ + + ηeE + (1 + L (k, ρ))ξ = 0,
s sr ∂ρ

and αS∗
E is the solution to

min{r,xS∗
E } 1 1 1 ∂ q 1
Z
[u(ξ) − (cE − δ + ME ( + + L (k, ρ)) + ηeE )ξ]f (ξ)dξ − ME = 0.
0 s sr sr ∂ρ (1 − αE )2

From Proposition 4, we can see that for the system with only EVs, the optimal rental price

is affected by the fixed ordering cost, the cost of emission, and the operational cost that now

includes an extra term. This term reflects the additional fixed cost for vehicles that are either

being recharged or waiting for recharge. Clearly, this term is decreasing in the recharging rate

and the number of charging stations. Moreover, the optimal rental price is independent of the
u(r)
range limit r. When r is small (r ≤ xS∗ ∗
F ), any price that is smaller than or equal to p̃ = r +δ

would generate the same maximal social welfare. This is because the number of customers who are

willing to rent EVs is constrained by their range and reducing the rental price will not increase this

number. Thus, neither total emission nor social welfare would change. In such cases, reducing the

rental price will decrease the profit of the CSC, but will increase the surplus of the customers.

Similar to Theorem 1, we show that the optimal social welfare in the system with all EVs is

larger than that in the system with all FVs if and only if (1) the recharging speed for EVs is high

enough, (2) the number of charging stations is large enough, and (3) the range of EVs is large

21

Electronic copy available at: https://ssrn.com/abstract=3191793

enough.

Theorem 4. Suppose ME = MF . Then,

ME
(1) if sr ≤ cF −cE +δ+η(eF −eE ) , ΠS∗ S∗
F ≥ ΠE ;

ME
(2) if sr ≥ cF −cE +δ+η(eF −eE ) , there exists a threshold kS∗ such that

(i) if k ≤ kS∗ , ΠS∗ S∗

F ≥ ΠE ;

(ii) if k ≥ kS∗ , there exists a threshold rS∗ (k) such that if r ≤ rS∗ (k), ΠS∗ S∗
F ≥ ΠE and if

r ≥ rS∗ (k), ΠS∗ S∗

F ≤ ΠE .

Moreover, kS∗ ≤ k∗ and rS∗ (k) ≤ r ∗ (k).

Theorem 4 shows that the thresholds that maximize social welfare are smaller than the thresh-

olds that maximize the profit provided in Theorem 1. This suggests that in the Car2go model, EVs

are more efficient in increasing the total social welfare than in increasing the total profit. That is,

there are cases in which FVs are preferred to EVs in terms of the total profit but EVs are preferred

to FVs in terms of the total social welfare. Figure 4 illustrates the comparison between both models

for k = ∞ (for k < ∞ the qualitative results are similar). As illustrated, the whole region can

be divided into three parts. In region I, EVs are preferred to FVs in terms of both profit and

social welfare. In region III, FVs are preferred to EVs in terms of both profit and social welfare.

However, in region II, FVs are preferred to EVs in terms of the total profit, but EVs are preferred

to FVs in terms of the total social welfare. In these cases, government could subsidize the CSC to

encourage the use of EVs in order to increase social welfare.

We also compare the optimal rental prices between the profit maximizing model and the social

welfare maximizing model.

Proposition 5. There exists an η ∗ such that pS∗ ≤ pP ∗ when η ≤ η ∗ and pS∗ ≥ pP ∗ when η ≥ η ∗ ,

where pP ∗ is the optimal rental price in the the profit maximizing model and pS∗ is the optimal

price in the social welfare maximizing model.

Proposition 5 shows that when η is large, to maximize social welfare, the CSC should charge a

higher rental price. This is because with large η, reducing emission is more important. A higher

rental price can reduce the number of customers who are willing to use the car sharing system.

22

Electronic copy available at: https://ssrn.com/abstract=3191793

 Figure 4: Comparison between profit maximizing and social welfare maximizing

Therefore, total emission is reduced but at the expense of the convenience of customers and the

profit of the CSC. When η is small, the government should induce the CSC to reduce the price.

Even though the CSC will have less profit, the total surplus of customers will increase because

more people will use the car sharing system.

5.2 Social Welfare under the Zipcar Model

In this subsection, we examine social welfare under the Zipcar model – the system with both EVs

and FVs. Given p, NE , and NF (or equivalently p, αE , and αF ), the total social welfare is

1 1
Z
ΠSEF (p, αE , αF ) =αE [u(ξ) − (cE − δ + ME ( + ) + ηeE )ξ]f (ξ)dξ
OE s sr
MF
Z
+ (1 − αE )αF [u(ξ) − (cF + + ηeF )ξ]f (ξ)dξ
OEF s
MF
Z
+ αF [u(ξ) − (cF + + ηeF )ξ]f (ξ)dξ
OF P s
αE αF
− ME − MF − ME Lq (k, ρ).
1 − αE 1 − αF

The objective is to find the optimal p, αE , and αF that maximize social welfare ΠSEF (p, αE , αF ).

Again, we consider three cases: (1) xF ≤ xE ≤ r, (2) xF ≤ r ≤ xE , and (3) r ≤ xF ≤ xE .

For the first case, OE = [0, xE ], OEF = [0, xF ], and OF P = ∅. For the second case, OE = [0, r],

OEF = [0, xF ], and OF P = ∅. For the last case, OE = [0, r], OEF = [0, r], and OF P = [r, xF ].

23

Electronic copy available at: https://ssrn.com/abstract=3191793

Proposition 6. The optimal social welfare is increasing in sr and k. However, it is not always

increasing in r.

Proposition 6 shows that increasing the recharging rate of EVs or increasing the number of

charging stations can increase social welfare. However, increasing the range of EVs may sometimes

lead to lower social welfare. For example, when the recharging speed or the number of charging

stations is low, the CSC needs to pay a large fixed cost for EVs, which would reduce social welfare

significantly. In this case, increasing the range of EVs may allow more customers to use EVs instead

of using FVs. As a consequence, the total social welfare decreases.

6 Discussion and Insights

In this section, we present case studies and numerical experiments to provide more insights.

6.1 Case Study: Car2go in San Diego and in Europe

When Car2go was first launched in San Diego, all cars provided were EVs, making San Diego the

first city in America with a fully EV car sharing fleet. However, beginning of May 2016, Car2go

replaced all of its EV fleet with gasoline-powered cars. In contrast, Car2go still provides a full fleet

of EVs in European cities such as Madrid, Stuttgart, and Amsterdam.

ME
By Theorem 1, in the Car2go model, a full EV fleet is always preferred when cF −cE +δ ≤ sr .

We thus next estimate these parameters for both markets in San Diego and in Europe. As the

technology behind EVs is identical, the difference between the markets manifest itself in the number

of charging stations and their speed as discussed below.

For the fixed purchasing cost MF , we use the estimations in Bellos et al. (2013) which indicate,

on average, the fixed purchasing cost of a FV is $20, 000 and the vehicle useful life is 5 years.

We assume that the car sharing system works 12 hours a day from 9:00 to 21:00 and 365 days a

year. Thus, MF ≈ 20, 000/(5 · 365 · 12) ≈ 0.92 dollars/h. We then assume that ME = MF = 0.92

dollars/h. Also according to Bellos et al. (2013), the operation cost of a FV (gasoline-powered) is

approximately cF = 8.7 + 4.2 + 0.6 = 13.5 cents/mile (gas + maintenance + tires). The electricity

cost for EVs is approximately 4 cents/mile (see U.S. Department of Energy, 2014) and the mainte-

nance cost is approximately 3.8 cents/mile (Graham et al. 2001 shows that plug-in hybrid EVs can

24

Electronic copy available at: https://ssrn.com/abstract=3191793

save the maintenance cost by 10%). Thus, the operation cost of an EV is cE = 4 + 3.8 + 0.6 = 8.4

cents/mile. For simplicity, and since most customers are significantly more sensitive to price than

emission, we assume δ = 0.

In San Diego, according to Car2go (2016), “when fully charged, an EV Car2go can travel 65

miles max” and “an EV Car2go takes 6-8 hours to charge.” Thus, the charging speed sr in San
ME 0.92
Diego can be approximated by 65/7 ≈ 9.3 miles/h. As a consequence, cF −cE +δ = 0.135−0.084+0 ≈

18.1 > sr = 9.3. By Theorem 1, this suggests that it is optimal to use FVs instead of EVs, which

explains why Car2go switched from EVs to FVs in San Diego. This result is independent of the

number of charging stations and the range of EVs. In other words, with such a low recharging

speed, Car2go would always prefer FVs to EVs no matter what the number of charging stations

and the range of EVs are.

In Europe, most of the charging stations in these cities are fast charging stations. An EV

can be fully charged within 2 hours in a fast charging station. Thus, the recharging speed sr is
ME
65/2 = 32.5 miles/h, which is larger than cF −cE +δ = 18.1. Therefore, for Car2go in cities with

enough fast charging stations, EVs would be preferred to FVs. This explains why Car2go still use

an EV fleet in European cities.

According to Car2go’s announcement, it attributed the transition from EVs to FVs in San

Diego to the Department of Energy for the lack of charging stations: “(it) had a plan to install

1,000 charging stations in San Diego by the end of 2011. There are only 400 installed around the

city today, and only 126 are inside the Car2go Home Area.” Our study shows that Car2go’s blame

may not be accurate. This transition is likely due to the slow recharging speed or the lack of fast

charging stations.
ME
Note also that in San Diego, sr < cF −cE +δ ≤ s̃∗r . By Theorem 2, this implies that the total

emission generated if Car2go provides the EV fleet would be less than that if Car2go uses the FV

fleet. This leads to a conflict between the environmental impact and the profitability of Car2go

in San Diego: Car2go prefers FVs but FVs are more harmful to the environment. However, in

cities in Europe, sr = 32.5 < s̃∗r = 59.7 (s̃∗r is computed based on eE = 0.27 pounds CO2 /mile

and eF = 1.14 pounds CO2 /mile), which implies that EVs are better than FVs in terms of total

emission. Thus, in Europe, using EVs is both more profitable and more environmental friendly.

In order to encourage the use of EVs in San Diego, several policies can be proposed. Clearly, one

25

Electronic copy available at: https://ssrn.com/abstract=3191793

way is to replace all slow charging stations with fast charging stations, similar to the European cities.

Another way is to impose a tax on the use of FVs to increase the operation cost. When the tax is
ME
larger than 4.8 cents/mile, cF ≥ 0.183 and cF −cE +δ ≤ sr . However, as this tax is more than 35% of

the operation cost, it would significantly increase the cost of Car2go. Alternatively, the government

can subsidize Car2go or consumers who use EVs. Subsidizing Car2go is equivalent to reducing the

operation cost of EVs and subsidizing consumers is equivalent to increasing their environmental

consciousness. When the subsidy is larger than 4.8 cents/mile, or equivalently, cE ≤ 0.036 or
ME
δ ≥ 0.048, we have cF −cE +δ ≤ sr and EVs are preferred to FVs with enough charging stations.

According to He et al. (2017), the total travel distance using Car2go in San Diego is 13,919,178

miles per year. Thus, the government should subsidize approximately 13, 919, 178×0.048 = 668, 121

dollars per year. Therefore, while the policy that the government adopts depends on the cost, it

cannot be too aggressive since, by Proposition 2, it may lead to a low price and high number of

trips, which would increase emission.

6.2 Implication for the Zipcar model

In this subsection, we look at the Zipcar model and examine how the recharging speed, the number

of charging stations, and the range of EVs affect the CSC’s optimal profit, the optimal decision,

and the total emission of the system.

According to U.S. Department of Energy (2016), eE = 0.27 pounds CO2 /mile and eF = 1.14

pounds CO2 /mile. We normalize the total arrival rate Λ to be 1. Then, NE (NF ) represents

the ratio of the number of EVs (FVs) to the total arrival rate. We assume the utility function is

u(ξ) = 1b (aξ − 21 ξ 2 ) with a = 1.5 and b = 0.67, and demand is beta distributed with parameters

β = 3 and γ = 2. We examine three different types of the charging infrastructures, levels 1-3.

Level 1 charging stations are the regular charging stations, which account for a large fraction of the

charging stations in San Diego. It takes 6-8 hours to charge an EV using a level 1 charging station.

Level 2 charging stations are fast charging stations and it takes less than 1 hour to fully charge an

EV. Level 3 is a battery switching station, where EVs can be recharged within 5 minutes since the

old battery can be replaced by a new one. We also vary the range of the EVs from 0.3 to 1 (the

maximum travel distance d¯ is normalized to 1).

Figure 5 illustrate how the CSC’s optimal profit, optimal decisions, and total emission of the

26

Electronic copy available at: https://ssrn.com/abstract=3191793

 0.6 1.73
Level 1 charging stations Level 1 charging stations
0.5 Level 2 charging stations Level 2 charging stations
1.72
Battery Switching stations Battery switching stations
0.4
1.71
0.9
1.7
profit

price
0.8
1.69
0.7
1.68
0.6

0.5 1.67

0.4 1.66
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r r

(a) The optimal profit (b) The optimal price

0.26
Level 1 charging stations Level 1 charging stations
0.2 Level 2 charging stations Level 2 charging stations
0.24
Battery switching stations Battery switching stations
number of EVs (NE)

number of FVs (NF)

0.22
0.15

0.2

0.1
0.18

0.16
0.05
0.14

0 0.12
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r r

(c) The optimal number of EVs (d) The optimal number of FVs
0.08
Level 1 charging stations
Level 2 charging stations
0.06 Battery switching stations

0.04
emissions

0.1

0.08

0.06

0.04
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r

(e) The total emission

Figure 5: Impact of the charging speed and and the range of EVs (k = ∞)

system change with respect to r, the range of EVs, with different charging infrastructures.

From Figure 5c, we see that for any values of sr , k, and r, the optimal number of EVs is always

greater than 0, even though it may only account for a small fraction of the total vehicles in the

fleet. Thus, it is always optimal to include EVs in the fleet. This supports the inclusion of EVs

27

Electronic copy available at: https://ssrn.com/abstract=3191793

in the fleets of Zipcar, DriveNow, and other car sharing companies. Therefore, the different trends

towards the deployment of EVs are due to the two different business models. If the CSC decides

to use either all EVs or all FVs, then it may be optimal to use the FV fleet. However, if both EVs

and FVs can be used, then the CSC can benefit from the use of both vehicle types.

Impact on the optimal profit: Figure 5a shows that as expected, the optimal profit is increasing

in both the recharging speed and the range of EVs.

Impact on the optimal price: Interestingly, we find that while the optimal price is decreasing

in the recharging speed, it is not monotone in the the range of EVs. Figure 5b shows that the

optimal price is first increasing and then decreasing in the range of EVs. This can be explained as

follows. When r is small, most of the trips will be completed by FVs due to the limited range of

EVs. Thus, the CSC should set a price that balance the profit generated from FVs and the demand

of FVs. As r increases and when r is in the middle range, more customers will use EVs but there is

still a large fraction of the trips that will be completed by FVs. However, since EVs become more

attractive, the CSC will increase the number of EVs and reduce the number of FVs in the fleet.

With less FVs, the CSC would charge a higher price. This explains why the price first increases.

However, as r further increases, EVs can cover almost all trips. With their lower operation cost,

the CSC would charge a lower price to attract more customers. Thus, the price would decrease

when r is sufficiently large.

Similarly, when r is in the middle ranges, increasing the recharging speed will increase the

optimal price. This is because when the recharging speed increases, the CSC would use more EVs

and less FVs. However, since the range of EVs is not large enough, a large fraction of customers

will use FVs. Thus, the main focus of the CSC is to balance the profit generated by FVs and the

demand of FVs. With less FVs, the optimal price would be higher.

Impact on the optimal numbers of EVs and FVs: Clearly, as seen in in Figures 5c and 5d

when the recharging speed and the range of EVs are large, EVs are more attractive. Moreover,

because the operation cost of EVs is lower than that of FVs, as the recharging speed and the range

of EVs increases, the CSC would include more EVs (and reduce the number of FVs).

Impact on the total emission: From Figure 5e, we can see that the total emission is decreasing

in the recharging speed. When r is in the middle range, a higher recharging speed would lead to

a higher price. As a consequence, less people will use the car sharing system and the total travel

28

Electronic copy available at: https://ssrn.com/abstract=3191793

 2 0.4
profit maximization =0
=0 =1
1.9 =2 0.2 =2

1.8 0

Social welfare
price

1.7 0.6

1.6 0.4

1.5 0.2

1.4 0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r r

(a) Comparison of p in profit maximization and (b) Impact of r on the optimal social welfare for
social welfare maximization different η

Figure 6: Model with social welfare maximization

distance would be lower. Thus, the reduced emission results from the benefit of using EVs and the

lower travel distance caused by a higher price.

However, the total emission is not always decreasing in the range of EVs. When r is large,

increasing r may lead to higher emission. This is because a larger r would induce the CSC to

use more EVs and to charge a lower price. The emission reduction due to using EVs may not

compensate for the additional emission caused by the total distance increased by a lower price.

Thus, the total emission may increase.

Moreover, from Figure 5, we can see that the impact of changing level 1 charging stations to

level 2 charging stations is larger than the impact of changing level 2 charging stations to level 3,

i.e., battery switching stations. However, the benefit of battery switching stations may be more

significant than what is observed in Figure 5. With a battery switching station, EVs can be

recharged within 5 minutes. This allows consumers with longer trips (longer than the range of

EVs) to use EVs to complete the trips. Therefore, battery switching stations can not only increase

the recharging speed but also increase the range of EVs.

We also examine the impact of the number of charging stations. We find that it is similar to

the impact of the recharging speed, i.e, more charging stations will lead to a higher profit, possibly

a higher price (especially when r is in the middle range), more EVs and less FVs in the fleet, and

a lower total emission.

Next, we examine the optimal social welfare in the Zipcar model. As illustrated in Figure 6a,

29

Electronic copy available at: https://ssrn.com/abstract=3191793

(similar to the Car2go model), to maximize social welfare, the CSC should charge a higher rental

price when η is large and a lower price when η is small. Therefore, if the government focuses

more on the environmental impact, it should tax or penalize the CSC to increase the rental price.

Otherwise, it can provide subsidies to the CSC to induce it to charge a lower price. From Figure

6b, we find that social welfare is increasing with the range of EVs using the parameters estimated

above. This suggests that if the government can adopt different policies to coordinate the car

sharing system, then increasing either the recharging speed, or the number of charging stations, or

the range of EVs can benefit the society.

7 Conclusion

Motivated by the news that Car2go replaced all of its electric vehicle fleet with gasoline-powered

cars starting May 2016, we examined the questions of whether it is optimal to use EVs in the car

sharing market and what is the environmental impact of the optimal choice on the car mix. We

developed a model consisting of a profit-maximizing CSC and a population of utility-maximizing

customers and showed that it is optimal for the CSC to use EVs only if for EVs the charging speed

is high enough and both the number of charging stations and the range are large enough. Among

these three conditions, the recharging speed is the most important. If the condition about the

recharging speed is not satisfied, then it is optimal not to use any EVs no matter what the number

of charging stations and the range of EVs are. Moreover, the number of charging stations is more

important than the range of EVs.

We examined the environmental impact of using EVs. We found that including EVs in the

car sharing market may lead to a higher total emission. When the recharging speed for EVs, the

number of charging stations, and the range of EVs are large, the CSC tends to charge lower rental

prices due to the low operating cost of EVs to encourage more people to use the car sharing system.

This will induce more trips and thus more emission in the system.

We also provided guidelines to the government for policy making. To maximize social welfare,

the government should tax the CSC to use a higher rental price if the environmental impact is

important and subsidize the CSC to induce a lower rental price otherwise. The government may

also adopt policies that can increase either the recharging speed, or the number of charging stations,

30

Electronic copy available at: https://ssrn.com/abstract=3191793

or the range of EVs to increase social welfare.

Finally, we studied the case of Car2go in San Diego. We found that Car2go replaced EVs with

FVs likely due to the slow recharging speed rather than the lack of charging stations. However, if

the CSC can use both EVs and FVs, then it is optimal to include EVs in the fleet.

References
Agrawal, V. V. and Bellos, I. (2016). The potential of servicizing as a green business model.
Management Science.
Al-Alawi, B. M. and Bradley, T. H. (2013). Review of hybrid, plug-in hybrid, and electric vehicle
market modeling studies. Renewable and Sustainable Energy Reviews, 21:190–203.
Avci, B., Girotra, K., and Netessine, S. (2015). Electric vehicles with a battery switching station:
Adoption and environmental impact. Management Science, 61(4):772–794.
Bellos, I., Ferguson, M., and Toktay, L. B. (2013). To sell and to provide? the economic and
environmental implications of the auto manufacturers involvement in the car sharing business.
Available at SSRN 2372406.
Bellos, I., Ferguson, M., and Toktay, L. B. (2017). The car sharing economy: Interaction of
business model choice and product line design. Manufacturing & Service Operations Management,
19(2):185–201.
Benjaafar, S., Kong, G., Li, X., and Courcoubetis, C. (2018). Peer-to-peer product sharing: Impli-
cations for ownership, usage, and social welfare in the sharing economy. Management Science.
Cachon, G. P., Daniels, K. M., and Lobel, R. (2016). The role of surge pricing on a service platform
with self-scheduling capacity. Manufacturing & Service Operations Management.
Car2go (2016). Car2go kills all-EV fleet in san diego, replaces it with gas cars.
https://www.autoblog.com/2016/03/17/car2go-kills-all-ev-fleet-san-diego-gas-cars/.
De Lorimier, A. and El-Geneidy, A. M. (2013). Understanding the factors affecting vehicle usage
and availability in carsharing networks: A case study of communauto carsharing system from
montréal, canada. International Journal of Sustainable Transportation, 7(1):35–51.
Fricker, C. and Gast, N. (2016). Incentives and redistribution in homogeneous bike-sharing systems
with stations of finite capacity. Euro journal on transportation and logistics, 5(3):261–291.
George, D. K. and Xia, C. H. (2011). Fleet-sizing and service availability for a vehicle rental system
via closed queueing networks. European journal of operational research, 211(1):198–207.
Graham, R. et al. (2001). Comparing the benefits and impacts of hybrid electric vehicle options.
Electric Power Research Institute (EPRI), Palo Alto, CA, Report, 1000349.
He, L., Mak, H.-Y., Rong, Y., and Shen, Z.-J. M. (2017). Service region design for urban electric
vehicle sharing systems. Manufacturing & Service Operations Management, 19(2):309–327.
Hu, M. and Zhou, Y. (2016). Dynamic type matching. Working Paper, Rotman School of Manage-
ment, University of Toronto.

31

Electronic copy available at: https://ssrn.com/abstract=3191793

Juan, A. A., Mendez, C. A., Faulin, J., de Armas, J., and Grasman, S. E. (2016). Electric vehicles
in logistics and transportation: A survey on emerging environmental, strategic, and operational
challenges. Energies, 9(2):86.

Le Vine, S., Adamou, O., and Polak, J. (2014). Predicting new forms of activity/mobility patterns
enabled by shared-mobility services through a needs-based stated-response method: Case study
of grocery shopping. Transport Policy, 32:60–68.

Lim, M. K., Mak, H.-Y., and Rong, Y. (2015). Toward mass adoption of electric vehicles: impact of
the range and resale anxieties. Manufacturing & Service Operations Management, 17(1):101–119.

Mak, H.-Y., Rong, Y., and Shen, Z.-J. M. (2013). Infrastructure planning for electric vehicles with
battery swapping. Management Science, 59(7):1557–1575.

Orsdemir, A., Deshpande, V., and Parlakturk, A. K. (2015). Is servicization a win-win strategy?
profitability and environmental implications of servicization.

Stillwater, T., Mokhtarian, P., and Shaheen, S. (2009). Carsharing and the built environment:
Geographic information system-based study of one us operator. Transportation Research Record:
Journal of the Transportation Research Board, (2110):27–34.

Tanaka, M., Ida, T., Murakami, K., and Friedman, L. (2014). Consumers willingness to pay
for alternative fuel vehicles: A comparative discrete choice analysis between the us and japan.
Transportation Research Part A: Policy and Practice, 70:194–209.

Tang, C. S., Bai, J., So, K. C., Chen, X. M., and Wang, H. (2016). Coordinating supply and demand
on an on-demand platform: Price, wage, and payout ratio. Working Paper, UCLA Anderson
School of Management, University of California, Los Angeles.

Taylor, T. (2016). On-demand service platforms. Working Paper, Haas School of Business, Uni-
versity of California, Berkeley.

Toktay, L. B., Wein, L. M., and Zenios, S. A. (2000). Inventory management of remanufacturable
products. Management science, 46(11):1412–1426.

U.S. Department of Energy (2014). Charging plug-in electric vehicles at home.

https://www.afdc.energy.gov/fuels/electricity charging home.html.

U.S. Department of Energy (2016). Emissions from hybrid and plug-in electric vehicles.
https://www.afdc.energy.gov/vehicles/electric-emissions.php.

U.S Environmental Protection Agency (2016). Sources of greenhouse gas emissions.

https://www.epa.gov/ghgemissions/sources-greenhouse-gas-emissions.

Whitt, W. (1984). Open and closed models for networks of queues. Bell Labs Technical Journal,
63(9):1911–1979.

Zipcar (2015). Zipcar green: How car sharing helps the environment.
http://www.zipcar.com/ziptopia/inside-zipcar/zipcar-green-how-car-sharing-helps-the-
environment-infographic.

Zoepf, S. M. (2015). Plug-in vehicles and carsharing: user preferences, energy consumption and
potential for growth. PhD thesis, Massachusetts Institute of Technology.

32

Electronic copy available at: https://ssrn.com/abstract=3191793