Cost-based analysis of autonomous mobility services

Abstract

Fast advances in autonomous driving technology trigger the question of suitable operational models for future autonomous vehicles. A key determinant of such operational models’ viability is the competitiveness of their cost structures. Using a comprehensive analysis of the respective cost structures, this research shows that public transportation (in its current form) will only remain economically competitive where demand can be bundled to larger units. In particular, this applies to dense urban areas, where public transportation can be offered at lower prices than autonomous taxis (even if pooled) and private cars. Wherever substantial bundling is not possible, shared and pooled vehicles serve travel demand more efficiently. Yet, in contrast to current wisdom, shared fleets may not be the most efficient alternative. Higher costs and more effort for vehicle cleaning could change the equation. Moreover, the results suggest that a substantial share of vehicles may remain in private possession and use due to their low variable costs. Even more than today, high fixed costs of private vehicles will continue to be accepted, given the various benefits of a private mobility robot.

Full text

Cost-based analysis of autonomous mobility services
Patrick M. B€
osch
*
, Felix Becker, Henrik Becker, Kay W. Axhausen
IVT, ETH Zürich, CH-8093 Zürich, Switzerland
ABSTRACT
Fast advances in autonomous driving technology trigger the question of suitable operational models for future autonomous vehicles. A key determinant of such
operational models’viability is the competitiveness of their cost structures. Using a comprehensive analysis of the respective cost structures, this research shows that
public transportation (in its current form) will only remain economically competitive where demand can be bundled to larger units. In particular, this applies to dense
urban areas, where public transportation can be offered at lower prices than autonomous taxis (even if pooled) and private cars. Wherever substantial bundling is not
possible, shared and pooled vehicles serve travel demand more efficiently. Yet, in contrast to current wisdom, shared fleets may not be the most efficient alternative.
Higher costs and more effort for vehicle cleaning could change the equation. Moreover, the results suggest that a substantial share of vehicles may remain in private
possession and use due to their low variable costs. Even more than today, high fixed costs of private vehicles will continue to be accepted, given the various benefits of
a private mobility robot.
1. Introduction
Autonomous vehicles (AVs) are expected to revolutionize mobility by
turning cars into mobility robots and allowing more dynamic and intel-
ligent forms of public transportation. A multitude of transport services
are conceivable with AVs, yet it is largely unclear which ones will prevail.
Besides travel time, reliability and comfort, price is the key attribute of a
transport service. Therefore, predicting level of acceptance and resulting
competitiveness of future AV operational models requires knowledge
about their cost structures. The validity of scenarios, simulations and
conclusions of such studies relies heavily on accuracy of assumptions
about the absolute and relative competitiveness of new transport services
compared to current offerings. Better estimates of absolute competi-
tiveness thus allow better estimates of mode choice, induced demand and
spatial distribution of travel demand - in short: future travel behavior.
First cost estimates of future transport services with AVs were pro-
posed by Burns et al. (2013). For three different cases (small to medium
town, suburban and urban), they calculated the cost, per trip, of a cen-
trally organized system of shared AVs (medium sedans with AV tech-
nology), which would replace existing transport services. Their estimates
are based on different cost categories, which capture fixed and variable
costs. They concluded that such systems could provide “better mobility
experiences at radically [up to ten times] lower cost”. In the case of a
shared AV system for a small to medium town, they found the cost of
driverless, purpose-built vehicles to be 0.15 US$ per trip-mile.
In a second approach, Fagnant and Kockelman (2014) considered the
external costs (e.g. crash or congestion cost) of today's private transport
system to calculate AVs' potential benefits, which they found to be sub-
stantial. In a following paper, Fagnant and Kockelman (2015) focused on
possible prices for users of a centrally organized, shared AV system. By
assuming an investment cost of 70000 US$ and operating costs of 0.50
US$ per mile for AVs only, they found that a fare of 1.00 US$ per trip-mile
for an AV taxi could still produce a profit for the operator. This is a higher
price level than in Burns et al. (2013), but still very competitive
compared to today's transport options.
Litman (2015) introduced additional factors into the discussion, like
cleaning costs of shared vehicles. He estimates costs based on different
categories. For some values, however, the paper remains unclear about
sources, or uses ballpark estimates. For example, it assumes shared
autonomous vehicles cost more than car-sharing (0.60 US$ - 1.00 US$ per
mile), but less than driver-operated taxis (2.00 US$ - 3.00 US$ per mile).
Building on the work above, Johnson (2015) estimated the price of
shared AVs to be 0.44 US$ per trip-mile (operating cost plus 30% profit
margin). For purpose-built shared AVs used as pooled taxis, they estimate
the price per trip-mile as only 0.16 US$. They use detailed cost categories
to estimate the total cost, but do not fully specify the sources of the
numbers. It is, therefore, difficult to reproduce and understand their es-
timates. In contrast to earlier studies, however, they compare and vali-
date their calculations against today's private cars.
Less rigorous and detailed, but more transparent estimates are pro-
vided by Stephens et al. (2016) and Friedrich and Hartl (2016).Stephens
et al. (2016) find the lower-bound cost of fully autonomous vehicles used
* Corresponding author.
E-mail addresses: patrick.boesch@ivt.baug.ethz.ch (P.M. B€
osch), felix.becker@ivt.baug.ethz.ch (F. Becker), henrik.becker@ivt.baug.ethz.ch (H. Becker), axhausen@ivt.baug.ethz.ch
(K.W. Axhausen).
Contents lists available at ScienceDirect
Transport Policy
journal homepage: www.elsevier.com/locate/tranpol
https://doi.org/10.1016/j.tranpol.2017.09.005
Received 2 February 2017; Received in revised form 14 August 2017; Accepted 12 September 2017
Available online xxxx
0967-070X/©2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Transport Policy xxx (2017) 1–16
Please cite this article in press as: B€
osch, P.M., et al., Cost-based analysis of autonomous mobility services, Transport Policy (2017), https://doi.org/
10.1016/j.tranpol.2017.09.005

with ride-sharing to be less than 0.20 US$ per passenger-mile and the
upper bound to be 0.30 US$ per passenger-mile. This encompasses a
range similar to Friedrich and Hartl (2016), who assume 0.15 €per
passenger-km for a ride-sharing scheme in an urban area in Germany.
Stephens et al. (2016), however, do not differentiate between private and
commercially offered vehicles and Friedrich and Hartl (2016) focus their
cost analysis on the ride-sharing service only. Costs of US$0.30 per
passenger-mile are also estimated by Johnson and Walker (2016) in a less
rigorous, but more detailed approach. With a less detailed approach for
the Netherlands, Hazan et al. (2016) estimate that fully-autonomous
vehicles in a ride-sharing scheme can be operated at costs as low as
0.09 €per passenger-km - i.e. at lower cost than rail services.
1
Overhead costs of shared services were neglected in all cases, which is
a major limitation given the new service market in the transport sector;
for example, Lyft or Uber, which - in their definitions of their services -
provide only the overhead of shared transport services, but no actual
transport service.
As outlined above, earlier approaches to determining the cost struc-
tures of operational models for AVs were incomplete for both the di-
versity of possible operational models and cost components. This
research addresses this gap by conducting a comprehensive, bottom-up
calculation of the respective cost structures of fully autonomous (level
5(SAE International, 2014)) vehicles for various operational models,
such as dynamic ride-sharing, taxi, shared vehicle fleets or line-based
mass transit. The chosen methodology allows determination of different
cost components’importance and differentiation of vehicle automation
effect on individual cost components. This research focuses on passenger
transportation. Freight transport, where AVs will undoubtedly also cause
major disruptions, cannot be investigated in this paper.
The remainder of this paper is structured as follows: Section 2pre-
sents a bottom-up determination of operating cost for a variety of vehicle
systems in various situations, while Section 3studies their respective
utilization for different use-cases. Based on this, the cost structures of
different operational models are calculated. The results, including a
robustness analysis against different assumptions of key variables, as well
as the impact of autonomous vehicle technology on future transport
systems are presented in Section 5. Section 6then goes one step further
by assuming a future with autonomous-electric vehicles and studying the
prospects of different modes under these circumstances. Finally, in Sec-
tions 7 and 8, insights gained through this research are discussed and
suggestions for further research are given.
2. Cost structures
This research covers three generic operational models:
line-based mass transit (public transport),
taxi (pooled or individual),
private car.
In this context, line-based mass transit uses full-size buses or trains
running along predefined lines on a fixed schedule. Taxi represents a taxi
or ride-hailing scheme, as it is known today, where transport may be
offered as individual service providing private ride, or pooled services in
which multiple travelers may be bundled into one vehicle.
2
Private cars
are owned by private persons and are solely used by themselves, or their
family and friends. As detailed below, for the generic operational models
taxi and private cars, different vehicle types were considered. Although
many further variations of operational models can be hypothesized, their
cost structures are assumed to be close to one of those three
generic models.
Various indicators can be used to represent the competitiveness of a
service. The most important dimensions are:
cost of production vs. prices,
vehicle kilometers vs. passenger kilometers,
full cost of a trip vs. direct cost only.
While the cost of production is relevant for fleet operators to meet
demand most efficiently, prices can be assumed to be a key attribute of
customer mode choice. The two indicators can be converted into each
other by considering taxes, payment fees and profit margins. Similarly,
vehicle kilometers can be planned by the operator, whereas passenger
kilometers take demand reaction to the service (i.e. occupancy, empty
travel, etc.) into account. Finally, the direct cost of a trip is the operating
cost for a ride from point B to point C, while the full cost of a trip also
includes a possible empty access trip from point A to B. While the first
measure determines the customer's willingness to pay, an operator must
cover the full cost of the trip. Pursuing a bottom-up approach; in this
research, first, individual cost components are determined for different
vehicle types, then operating costs are determined from an operator's
perspective and after that, travel behavior impact is estimated
using prices.
The respective cost components (overview in Appendix B) are ob-
tained in two steps: First, based on manufacturer data and additional
sources, fixed and variable vehicle costs are determined for the case of
private ownership and use of the vehicle. In a second step, variations are
introduced into the calculation to cover the case of commercial owner-
ship and shared use of vehicles. Then, using a separate approach, the cost
components of today's line-based mass transit are established. Eventu-
ally, the effects of vehicle electrification and automation are estimated
for individual cost components, allowing calculation of overall operating
costs and required minimum charges.
2.1. Cars
2.1.1. Fixed cost
Fixed vehicle costs depend substantially on the vehicle type. In this
research, four general vehicle categories are considered:
Solo: One-seat urban vehicle (Example: Renault Twizy
3
),
Midsize: Standard four-seat all-purpose car (Example: VW Golf
4
),
Van: Large eight-seat all-purpose car (Example: VW Multivan
5
),
Minibus: Minibus with 20-seats with small trunk (Example:
Mercedes-Benz Sprinter
6
).
For each of these categories, example vehicle acquisition cost was
obtained from the car manufacturer's website for a model with a medium
level of optional equipment. It should be noted that the costs mentioned
in this and the next section include Swiss VAT of 8%. Depreciation is split
into a fixed part, attributable to aging of the vehicle and a variable part
from its usage. For the fixed part, it is assumed that the vehicle de-
preciates one tenth of its acquisition cost every year, independent of
mileage. The variable part is explained at the beginning of the next
section. It should be noted, however, that the calculations do not reflect
costs for private owners who prefer to drive relatively new cars.
Furthermore, as it is the purpose of the paper to derive an internal cost
1
2016 exchange rates (ER) (Organisation for Economic Co-operation and Development
(OECD), 2017a) and purchasing power parities (PPP) (Organisation for Economic
Co-operation and Development (OECD), 2017b): CHF/US$: ER 1.0, PPP 1.2; €/US$: ER
0.9, PPP 0.7.
2
In this research, shared and pooled use are differentiated. Shared use of vehicles refers
to sequential sharing of a vehicle (e.g. taxi), while pooled use describes simultaneous
sharing of a vehicle, where different customers travel in the same vehicle at the same time
(e.g. UberPool).
3
Webpage: http://www.renault.ch/, last accessed 01.02.2017.
4
Webpage: http://www.volkswagen.ch/, last accessed 01.02.2017.
5
Webpage: http://www.volkswagen-nutzfahrzeuge.ch/, last accessed 01.02.2017.
6
Webpage: http://www.mercedes-benz.ch/, last accessed 01.02.2017.
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
2

calculation, cash flow calculations, such as the repayments of the
necessary loans, are not explicitly considered.
The interest amount was determined based on an annuity loan with
an interest rate of 7.9% and a five-year credit period (Migrosbank, 2017).
Processing fees for the borrower are ignored. Insurance rates were
determined using the cheapest fully comprehensive insurance according
to the internet comparison service Comparis (Comparis AG, 2016) for
vehicles registered in the Canton of Zurich. The rates reflect the cheapest
offer for non-business customers with 25 years of driving experience
without accidents and 30000 driven kilometers per year. It should be
noted that the policy with 15000 km per year is only 10 CHF cheaper.
Taxes were obtained using a web-tool provided by the Canton of Zurich
(Strassenverkehrsamt Kanton Zürich, 2017). For parking costs, the
average for private cars in Switzerland was used (TCS, 2016). For tolls,
the price of the Swiss motorway permit sticker was used for solo, midsize
and van vehicles (Eidgen€
ossische Zollverwaltung, 2017b). Minibuses are
subject to a lump-sum heavy vehicle charge (Eidgen€
ossische Zollver-
waltung, 2017a). The resulting monetary values are presented in Table 1.
2.1.2. Variable cost
Variable costs of private vehicles were estimated for the four vehicle
types. For depreciation and maintenance cost, the average for the Swiss
car fleet was used and scaled by price of the car (TCS, 2016). For a
midsized car, this results in a fixed depreciation of 3500 CHF per year and
an additional 11.67 CHF for every 100 km driven. It is clear that this
procedure underestimates value losses in the first years and over-
estimates them in later years. As this paper compares average cost fig-
ures, this nonlinear nature of value losses is ignored. It is further assumed
that a private car is cleaned eight times a year, based on a median value
between 6 and 10 in Germany (abh Market Research, 2007). The asso-
ciated cost was estimated based on the price list of a self-service car-wash
facility in Zurich (Autop, 2016). Concerning tires, it was assumed that
two sets of tires (each 300 CHF) and two annual tire changes (each 100
CHF) would allow for 50000 km of driving. Fuel costs were given by
urban fuel consumption as reported by the manufacturer at a fuel price of
1.40 CHF per liter (Shell, 2016). Given that there are no distance-based
tolls for passenger cars in Switzerland at this time, a zero toll has been
used here. The resulting monetary values are presented in Table 2.
2.1.3. Fleet effects
Commercial fleet operators benefit from various discounts on fixed
and variable vehicle costs due to scale effects. By reviewing their plat-
form transactions, Blens (2015) assessed the average discount granted to
commercial customers for thirty popular company cars. The discounts
range between 8.5% and 30.5%, with a median of 21%. As the number of
vehicles bought by fleet operators should be substantially larger than for
an average company, a general discount on the vehicle price of 30% is
assumed. Due to more intense use, commercially used vehicles are
further assumed to be written off over 300000 km (Klose, 2012), rather
than ten years. Moreover, it is assumed that insurance rates for fleet
operators are 20% lower, reflecting discounts typically available for
group insurances (Helvetia, 2016). Given that the German car rental
company Sixt SE issues bonds at 1.125% p.a. with six years duration (Sixt
SE, 2016), the corporate interest rate is set at 1.5%; credit period is
assumed to be three years. In addition, maintenance and tire costs are
assumed to be 25% lower due to better conditions for bulk buyers. For
fuel costs, a 5% reduction is assumed based on typical group discounts
(Migrol, 2016). Based on Schlesiger (2014), parking costs for fleet op-
erators are assumed to be 133% higher than for private drivers. In
addition, VAT is deducted where appropriate, as costs in the previous
section are based on gross prices of products and services.
It is assumed that customers pay less attention to third parties’
property (tragedy of the commons), leading to substantially higher
cleaning costs. Based on experiences of a car-sharing operator,
7
it was
assumed that vehicles need to be cleaned after every 40th trip. If a car is
not automated, the costs per driver hour are estimated at 35 CHF, based
on the average yearly salary of Swiss taxi-drivers (Lohnanalyse, 2016)
and the calculation tool of Braendle (2013), which helps determine labor
costs for a company based on gross income.
Further cost components include overhead costs (consisting of man-
agement, human resources, fleet-coordination, advertising, etc.) and
vehicle operations costs per vehicle and day for commercially operated
on-demand services. These figures are assumed to depend on the fleet
size and composition. An analysis based on US data (Federal Transit
Administration, 2016) is presented in Appendix A. It suggests that, for a
case in Switzerland, approximately 14 CHF per vehicle-day can be
assumed for overhead and 10 CHF per vehicle-day for operations costs.
In addition to operating costs, user prices take into account a profit
margin of 3%, the Swiss VAT of 8%, and a payment transaction fee of
0.44% (Wettbewerbskommission WEKO, 2014). The profit margin esti-
mation is based on the study of SCI Verkehr (2016) that reports that the
median profit margin for German logistic companies is between 2%
and 4%.
2.2. Public (mass) transportation
For a fair comparison of different modes, the full production costs of
today's public transportation services were estimated before direct sub-
sidies. Given the large fleet sizes of most public transportation operators,
it was assumed that administrative overhead share in the full costs is
independent of fleet size. Therefore, overhead costs are assumed to be
already incorporated in the full production cost per kilometer, not treated
separately.
Operating costs for passenger rail are based on the annual report of
the SBB (Swiss national railway company), for the year 2015 (SBB AG,
2016). The cost per train kilometer is stated as 31.40 CHF/km. This
number should serve as a ballpark estimate, given that it only reflects the
average across various train types and routes. Although the figure in-
cludes track fees, it should be highlighted that track fees in Switzerland
are subsidized.
Among local public transportation providers, tight competition hin-
ders transparency in reported business results. To attempt an accurate an
estimate as possible, the framework introduced by Frank et al. (2008)
was adapted, using Swiss salaries (Verkehrsbetriebe Zürich, 2016); this
was then applied to an urban and a regional transit provider. The results
were cross-validated, with data on average kilometer costs of 98 urban
and 787 regional public transport lines across Switzerland from 2012
Table 1
Fixed vehicle costs.
Solo Midsize Van Minibus
Capacity 1 4 8 20
Acquisition [CHF] 13000 35000 66000 70000
Interest [CHF/a] 260 700 1320 1400
Insurance [CHF/a] 500 1000 1200 1400
Tax [CHF/a] 120 250 700 1100
Parking [CHF/a] 1500 1500 1500 1500
Toll [CHF/a] 40 40 40 2200
Table 2
Variable vehicle costs.
Solo Midsize Van Minibus
Depreciation [CHF/100 km] 2.60 7.00 13.20 14.00
Maintenance [CHF/100 km] 2.38 6.40 12.07 12.80
Cleaning [CHF/100 km] 2.00 3.00 4.00 5.00
Tires [CHF/100 km] 2.00 2.00 2.00 2.00
Fuel [CHF/100 km] 5.60 7.98 12.88 20.30
Toll [CHF/100 km] 0.00 0.00 0.00 0.00
7
Private communication.
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
3

(Bundesamt für Verkehr, 2011). It should be noted that trams are not part
of the current analysis.
For bus and rail transportation, no scale effects are assumed because
such effects are already incorporated in public transport providers’re-
ported cost structures.
2.3. Cost effects of new technologies
In this research, two technological advances are expected to have a
substantial impact on the cost structures of vehicles and transport ser-
vices: electric propulsion technology and vehicle automation.
2.3.1. Electric propulsion
The battery is one of the main cost drivers of electric vehicles. As
multiple batteries may be needed during the lifetime of a vehicle, it was
decided to add the depreciation of the battery to the maintenance costs. It
is thus assumed that the purchase price of an electric vehicle is similar to
its conventional counterpart. Saxton (2013) analyzed the battery ca-
pacity of used Tesla Roadsters in line with their age, mileage and climate
conditions. He found that only mileage would have a significant corre-
lation and that most Tesla Roadsters would have a battery capacity of
80–85% after 160000 km. As fleet vehicles are written off over
300000 km, it is assumed that a battery needs to be replaced every
150000 km. Furthermore, a McKinsey (2017) analysis concludes that
average production costs of an electric car battery are 227 US$ (227 CHF)
per kwh. Including a profit margin of 3% and taking into account the
Swiss VAT of 8%, this amounts to 252 CHF/kwh per customer. Taking the
Volkswagen E-Golf
8
with a battery of 24.2 kwh as a reference, additional
maintenance costs of 0.04 CHF/km are calculated. However, Diez (2016)
report that the remaining maintenance costs for electric vehicles were
found to be 35% less than for conventional vehicles. In total, mainte-
nance costs therefore increase by 28%. Given that maintenance costs are
adjusted to the different vehicle types, it is assumed that this increase
covers the different battery capacities. Nevertheless, it needs to be
emphasized that prices for batteries are likely to decline in the future. In
fact, Nykvist and Nilsson (2015) even highlight that past predictions
about today's costs of battery packs have been too pessimistic.
Furthermore, according to an internet comparison service (Comparis
AG, 2016), insurance fees are 35% lower for the e-Golf than for a com-
parable gasoline Golf. This ratio has thus been assumed for all vehicle
categories. In addition, electric vehicles are exempted from road tax in
the Canton of Zurich (100% reduction). The fuel costs are 50% lower
based on a comparison of the Golf and e-Golf energy consumption at
current fuel/electricity prices.
2.3.2. Vehicle automation
Impacts of AV technology are less clear. It was assumed that the
necessary technology would increase vehicle price by an average of 20%
(IHS, 2014), leading to higher acquisition cost, interest cost and depre-
ciation. Due to more balanced driving, it is further assumed that auto-
mation lowers fuel costs by 10% (Stephens et al., 2016). Due to more
considerate automatic driving, it is expected that autonomous vehicles
will need less maintenance for traditional car components. However,
since it can be expected that the new sensors themselves need periodic
maintenance, we do not assume different cost figures for the total
maintenance costs. Based on earlier research, it was assumed that safer
driving would lower insurance rates by 50%. This is regarded as con-
servative, as today's Tesla Autopilot is reported to have already decreased
accident rates by 40% (NHTSA, 2017). The authors acknowledge, how-
ever, that this estimate is highly uncertain, given the profound changes
ahead for the insurance industry (Morgan Stanley and Boston Consulting
Group, 2016), which are beyond the scope of this research.
2.3.3. Automation and electrification of public transportation
To estimate the effect of vehicle automation on trains, the number of
full time equivalents of active train drivers in passenger rail (2485 (SBB
AG, 2016),) was multiplied by the average personnel cost of train drivers
(87,946 CHF, including the average salary (Lohnanalyse, 2017) and
employer outlay (Braendle, 2013)). The resulting total salary sum was
then divided by the total annual operating cost of SBB passenger rail in
2015 (SBB AG, 2016). Assuming train drivers will not be needed any
longer, this corresponds to a 4.7% decrease in cost per kilometer. Given
that in Switzerland, railway lines already operate on electrified tracks, no
further impact through electric propulsion is assumed.
To estimate the impact of electric propulsion and automated vehicle
systems on bus operations, the framework by Frank et al. (2008) was
used. Following assumptions on private and taxi operations, it was
assumed that electric propulsion halves the fuel cost (5.5% decrease in
total kilometer cost). Based on information by the Swiss Federal Office
for Transport (Bundesamt für Verkehr, 2011) a bus driver's salary
amounts to 55% of the total cost. As with trains, it is assumed that costs
decrease by this share through automation.
Automation technology and electric propulsion are not expected to
have substantial impacts on the fixed and variable cost of public bus and
train services because automation technology is already pre-installed (in
trains) or would not represent a substantial increase in the purchase price
of a vehicle (for buses). Moreover, it is assumed that systems will
continue to be operated in the same manner as today, so that impact on
administration costs will be minimal.
2.4. Cost and price calculation
Based on cost factors provided above, comparable costs and prices of
vehicles and transport services are calculated. The primary results are
costs per vehicle-kilometer, cost per seat-kilometer, cost per passenger-
kilometer, and price per passenger-kilometer. Cost thus represents the
production costs for the fleet operator, while price also includes a profit
margin for the provider, the VAT, and a payment transaction fee.
The cost-calculation framework allows for specification of the vehicle
fleet, expected usage of this fleet and the operation model, including
expected revenue.
Vehicle fleet: The vehicle fleet is specified by type of vehicles,
number of those vehicles and their features (electrified, automated).
Expected usage: Usage is differentiated between peak, non-peak and
night usage. For all three cases, average number of operating hours,
relative active time (share of operating hours during which passen-
gers are on board), average occupancy, average speed, average pas-
senger trip length, relative empty rides, relative maintenance rides
and relative maintenance hours are required. Expected usage is
further specified in Section 3.
Operational model: Besides private vehicle ownership, two forms of
public transport are differentiated here: dynamic shared fleet opera-
tion and line-based mass transit.
Expected usage parameters (Section 3) result in an average number of
kilometers travelled per day, plus number of passengers, including pas-
senger kilometers and passenger hours per day.
From the fleet, usage, and operation specifications, different cost and
price estimates are calculated:
The price per passenger-kilometer P
pkm
is the production cost per
passenger-kilometer C
pkm
, plus expected profit margin r, payment
transaction fee p, and VAT:
Ppkm ¼Cpkm
ð1rÞð1pÞ
⋅ð1þVATÞ
8
Webpage: http://www.volkswagen.ch/, last accessed 14.08.2017.
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
4

The cost per passenger-kilometer C
pkm
is calculated as the average
total cost per vehicle and day C
vd
divided by average passenger ki-
lometers per vehicle and day.
Similarly, the cost per vehicle-kilometer C
vkm
is calculated as the
average total cost per vehicle and day C
vd
divided by average total
kilometers per vehicle and day vkm
tot/d
.
Average total kilometers per vehicle and day vkm
tot/d
is the sum of
average service kilometers skm and average empty kilometers driven
ekm.
Average total cost per vehicle and day C
vd
is the sum of fixed costs
per vehicle and day (fix cost of the vehicle FC
vd
, plus service provider
overhead costs OHC
vd
), plus total variable cost per vehicle and day
(variable cost of the vehicle per kilometer VC
vkm
, plus variable service
costs per hour VSC
vh
(e.g. the driver)):
Cvd ¼ðFCvd þOHCvdÞþVCvkm ⋅vkmtot=dþVSCvh ⋅DTh=d
With DT
h/d
being the total time the vehicle is on duty per day and with
vehicle costs, FC
vd
and VC
vd
, being sums of respective cost factors
adjusted by scaling factors representing technologies and applicable
fleet discounts.
The cost per seat-kilometer C
skm
is calculated by dividing the cost
per vehicle-kilometer C
vkm
by the number of seats per vehicle seats
v
.
The cost calculation framework is implemented in the programming
language R with an input interface in Microsoft Excel.
9
The cost calcu-
lation software framework is available from the authors on request.
Fellow researchers are encouraged to reproduce the cost estimates pre-
sented in this paper, but also to estimate costs for different situations, use
cases and other possible future transport services with AVs.
3. Utilization parameters
Average cost per user depends not only on vehicle characteristics, but
also on service efficiency, i.e. vehicle utilization, empty travel and
overhead. To approximate these values, this section first presents the
utilization cases differentiated in this paper and then describes their
average usage as assumed for the Zurich, Switzerland region. A summary
of the complete parameter set can be found in Appendix C.
In this paper, three spatial and three temporal cases were differenti-
ated for private and shared services including taxis. The three spatial
cases are:
Urban: Trips starting and ending in an urban area and which are
shorter than 10 km.
Regional: Trips starting and ending outside of an urban area and
which are shorter than 50 km.
Overall: Any trips shorter than 200 km, independent of the area.
For each of these spatial cases, the following three temporal cases
were defined:
Peak: Trips beginning or ending between 7am and 8am or 5pm and
6pm.
Off-peak: Not peak-trips, which begin or end between 8am and 5pm.
Night: Neither peak, nor off-peak trips.
For each of the above use cases, the following parameters were
assumed for the Zurich, Switzerland region:
Average operation hours: The time a taxi or a service is available.
For private vehicles in Switzerland, the average usage time was
determined for each of the above cases based on the Swiss trans-
portation microcensus (Swiss Federal Statistical Office (BFS) and
Swiss Federal Office for Spatial Development (ARE), 2012). This
suggests 0.32 h during peak, 0.7 h during off-peak, and 0.3 h during
night.
Conventional mobility services were assumed to stay online for 20 h
per day (minus maintenance). The live time would span the peak
hours, the off-peak hours and a part of the night. To account for
maintenance issues, 5% of the live time was subtracted during peak
and off-peak times and 20% during the night. In contrast, professional
AV services would be operated throughout the day, except for
maintenance (as above).
Relative active time: The share of operation hours, during which the
vehicle serves customers.
For private vehicles, this is 100% of the operation time.
For commercial services, these numbers were calculated from B€
osch
et al. (2016). This represents relative active time if a substantial part
of the population were using the service for all of their trips. Ac-
cording to their results, relative active time for taxis and shared ser-
vices is 57% during peak, 59% during off-peak, and 30% during night.
The lower relative active time during peak hours than during off-peak
is due to the more uni-directional demand in peak hours (com-
muters). No spatial differentiation was made here.
Average occupancy: Average number of passengers in the vehicle when
on duty.
For private cars in Switzerland, average occupancy was calculated
from the Swiss transportation microcensus (Swiss Federal Statistical
Office (BFS) and Swiss Federal Office for Spatial Development (ARE),
2012). It was found to be 1.34 for peak, 1.46 for offpeak, and 1.43 for
night. For commercial services, taxi and ride sharing (pooled) was
differentiated. For taxi services, the same occupancy values as found
for private cars were assumed. For ride sharing (pooled) services,
occupancy values found by the International Transport Forum (2015)
were used; 2.6 for peak, 2.4 for off-peak, and 2.3 for night. The same
values were used for urban and regional settings.
Average speed: The overall average speed of the vehicle (service and
passenger trips).
Average speed of cars in Switzerland was calculated for each spatio-
temporal case, as found in the Swiss transportation microcensus
(Swiss Federal Statistical Office (BFS) and Swiss Federal Office for
Spatial Development (ARE), 2012); overall peak 32.4 km/h, overall
off-peak 34.0 km/h, overall night 36.2 km/h, urban peak 20.6 km/h,
urban off-peak 21.9 km/h, urban night 23.7, regional peak
31.3 km/h, regional off-peak 32.2 km/h and regional night
34.0 km/h.
Average passenger trip length: The average in-vehicle trip length of
passengers using the service.
Average Swiss car trip length was calculated based on the Swiss
transportation microcensus (Swiss Federal Statistical Office (BFS) and
Swiss Federal Office for Spatial Development (ARE), 2012) for each
spatio-temporal case; overall peak 15.0 km, overall off-peak 10.5 km,
overall night 12.6 km, urban peak 3.4 km, urban off-peak 3.0 km,
urban night 3.4 km, regional peak 8.0 km, regional off-peak 5.9 km
and regional night 7.1 km.
Relative empty rides: Percentage of empty vehicle trips due to, for
example, relocation, pick-up drives, or return-to-base drives.
As empty kilometers for private vehicles would be the result of a
substantially different user behavior than today (e.g. several people
privately coordinating for a vehicle), assumptions here would imply
many far-reaching assumptions about user behavior. As this is beyond
9
The detailed input values, as well as the resulting cost structures and prices of
different services, are presented in the appendix of this publication.
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
5

the scope of this publication, empty rides for private vehicles are
excluded here. For commercial services, based on (Fagnant et al.,
2015), number of empty rides was assumed to be 8% for urban set-
tings, and based on (B€
osch et al., 2016), 15% for regional settings.
Relative maintenance rides: Share of vehicle trips due to maintenance,
cleaning and repair.
For private vehicles, no maintenance rides were assumed (given the
dense network of gas stations in the Zurich area, additional distance
through refilling and/or cleaning is negligible). For commercial ser-
vices, 5% was assumed, based on experience with traditional car
sharing services.
Given the limited information available about public transportation,
average daily values were determined without a temporal differentiation.
In average, 19 h of operation were assumed for bus and rail services to
account for the official public transport night break from 00:30 to 05:30
in the Zurich, Switzerland region. This probably overestimates vehicle
operation hours, as each individual vehicle is not operated for the full
19 h. It is thus a conservative estimate for public transportation. Relative
active time was estimated at 85%, based on the official schedule (Zürcher
Verkehrsverbund, 2017). Average occupancy and speed of public trans-
port vehicles was derived from Bundesamt für Verkehr (2011); for urban
buses an average occupancy of 22.42% and a speed of 21.31 km/h were
reported, for regional buses 12.6% and 20.89 km/h, and for regional
trains a speed of 37.82 km/h. Average regional train occupancy was
22.7% according to the SBB annual report (SBB AG, 2016). Empty rides
and maintenance rides were assumed to be already accounted for in the
average occupancies and thus set to 0% (SBB AG, 2016).
A full table presenting the various cases and corresponding values is
included in Appendix C. Current usage values as presented here are
average values based on current market situation and price structure. If
lower costs were assumed, these values would probably be too high for
occupancies and too low for trip lengths.
4. Validation with current services
Combining the cost structures (Section 2) with utilization parameters
(Section 3) allows for the calculation of average cost values for each use
case (Section 2.4). In a first step, results were validated against data on
current transportation services.
4.1. Private car
An average private car in Switzerland costs 0.71 CHF per kilometer
(TCS, 2016). Our own cost calculations resulted in a cost of 0.69 CHF per
kilometer for a conventional private midsize car with average usage as in
(TCS, 2016). Also considering that shares of different cost components
show only minimal differences, the framework is considered to correctly
calculate cost structures for private vehicles.
4.2. Taxi
In Zurich, taxi prices are regulated, with a base price of 8.00 CHF plus
5.00 CHF per kilometer. (Stadt Zürich, 2014). UberPop fares consist of a
base price of 3.00 CHF plus 0.30 CHF per minute and 1.35 CHF per
kilometer (Uber, 2017). For a conventional midsize car used as taxi in an
urban setting (usage as defined in Section 2), the framework returns costs
of 3.38 CHF per kilometer, which can be assumed to be in the correct
range, given taxi and UberPop charges cited above.
4.3. Public transport
Calculation of cost values for city and regional buses was calibrated to
the full cost per vehicle kilometer, as stated in (Bundesamt für Verkehr,
2011). Original reference values are 7.14 CHF per kilometer for city
buses and 6.70 CHF per kilometer for regional buses (Bundesamt für
Verkehr, 2011). Cost of rail services was derived from SBB AG (2016) as
31.40 CHF per kilometer for trains. In both cases, given the lack of an
additional, independent source of information, no actual validation
is possible.
5. Results
After validation in Section 4, the framework presented above was
used to estimate the cost of future transport services. Given the large
number of operational model combinations, vehicle types, and spatial
classifications, only a selection is presented below, representing the use
cases offering the most important insights. A more complete list with
other combinations can be found in Appendix D.
In a first step, the framework was used to analyze the impact of
vehicle automation on the cost structures of various mobility services.
5.1. Cost structures
To better understand the impact of autonomous vehicle technology,
the cost structures of different operational models were compared. First,
conventional private vehicles (Private Car) and conventional taxi fleet
vehicles without pooling (Ind. Taxi) were studied. Fig. 1 presents the cost
components for the case of conventional midsize vehicles in an urban
setting with (Autonomous) and without (Conv) vehicle automation (for
the complete list of values see Table 3).
As shown in Fig. 1, there is a general difference between the two
operational models. While the operating cost of a private vehicle is
mostly determined by its fixed costs related to the car (such as depreci-
ation, insurance or interest), the operating cost of current taxi fleets is
mostly determined by the driver's salary. Administration cost, as well as
car related costs, play a minor role.
Vehicle automation drastically changes cost structures of the different
services. In general, three effects are at work; on one hand, autonomous
vehicle technologies raise the vehicle purchase price, but on the other
hand, reduce operating cost through lower insurance fees, maintenance
and fuel costs. In addition, they allow taxi fleets to operate without
drivers, thus cutting their main cost component.
It is interesting to note that for privately used cars, the first two effects
cancel out, so that running costs remain largely unchanged (around 0.49
CHF/km, see Table 3). The third effect does not apply for private vehi-
cles, because there is no direct monetary gain through their automation.
Due to substantially higher utilization, reductions in variable and
labor costs more than outweigh the increases in fixed costs for ride-
hailing or taxi vehicles. In particular, driverless technology is the key
factor in substantially lower production costs for such fleets. It is assumed
that the operating costs of a ride-hailing or taxi vehicle would plummet
from 2.73 CHF/km to 0.41 CHF/km (see Table 3). Given the absence of a
driver and fellow passengers, however, customers of such taxi services
are expected to show more irresponsible behavior in the vehicle (e.g. by
eating) resulting in a faster soiling of the vehicle. To estimate this effect,
minimum cleaning efforts of current car-sharing schemes were used. As
shown in Fig. 1, even minimum assumptions result in substantial clean-
ing efforts, which would rapidly account for almost one-third of auto-
mated taxi's operating costs. Combined with an estimated share of 20%
due to overhead cost, this means that more than half of autonomous
vehicle fleets' operating costs will be service and management costs.
Hence, by optimizing their operations processes, providers may realize
substantial efficiency gains, allowing them to increase their respective
market share. However, even without such additional measures, auton-
omous vehicle technologies would allow fleets to operate at lower costs
than private vehicles.
5.2. Impact of vehicle automation
As outlined above, the effect of vehicle automation on the cost
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
6

structure of future transport services is substantial. Fig. 2 shows the effect
of automation on today's main operational modes for the urban and the
regional (suburban and exurban) case. Given the dominant role of
midsize cars in today's transport system, private and taxi operational
models are assumed to be conducted using midsize vehicles. Up to this
point, no electrification of vehicles is considered (except for trains which
are already electrified today) to isolate the effect of automation.
Fig. 2 shows that, without automation, the private car has the lowest
12%
8%
42%
9%
14%
12%
11%
9%
48%
13%
10% 88%
11%
18%
8%
29%
13%
20%
0%
25%
50%
75%
100%
Private Car
Conv
Private Car
Autonomous
Ind. Taxi
Conv
Ind. Taxi
Autonomous
Type of Car
Share of cost per passenger km
Cost Type
Overhead and Vehicle Operations
Salaries
Fuel
Cleaning
Parking and Tolls
Tax
Insurance
Depreciation
Interest
Maintenance and Wear
Fig. 1. Cost structure comparison with (Autonomous) and without (Conv) vehicle automation for private vehicles (Private Car) and taxi fleet vehicles without pooling (Ind. Taxi).
Table 3
Cost structure comparison with (Autonomous) and without (Conv) vehicle automation for private vehicles (Private Car) and taxi fleet vehicles without pooling (Ind. Taxi).
Private Car Conv Private Car Autonomous Ind. Taxi Conv Ind. Taxi Autonomous
CPKM
[CHF]
Share of CPKM CPKM
[CHF]
Share of CPKM CPKM
[CHF]
Share of CPKM CPKM
[CHF]
Share of CPKM
Overhead and Vehicle Operations 0.083 3.0% 0.08 19.6%
Salaries 2.409 88.3%
Fuel 0.056 11.6% 0.051 10.0% 0.057 2.1% 0.051 12.5%
Cleaning 0.005 1.1% 0.005 1.0% 0.026 0.9% 0.117 28.8%
Parking and Tolls 0.068 13.9% 0.068 13.4% 0.033 1.2% 0.032 7.9%
Tax 0.011 2.3% 0.011 2.2% 0.002 0.1% 0.002 0.6%
Insurance 0.044 9.1% 0.022 4.4% 0.008 0.3% 0.004 0.9%
Depreciation 0.203 41.9% 0.243 48.3% 0.061 2.2% 0.073 18.0%
Interest 0.039 8.0% 0.046 9.2% 0.002 0.1% 0.002 0.5%
Maintenance and Wear 0.059 12.2% 0.058 11.4% 0.047 1.7% 0.046 11.3%
Sum 0.485 100% 0.504 100% 2.728 100% 0.407 100%
CPKM: Costs per passenger-kilometer.
0.5
0.29
0.41
0.24
0.48
1.61
2.73
0.53
0.5
0.21
0.32
0.4
0.44
0.48
1.13
1.92
0.89
0.47
Urban Regional
202 202
Private Car
Pooled Taxi
Ind. Taxi
Bus
Rail
CHF per passenger kilometer
Steering
Autonomous
Not autonomous
Fig. 2. Cost comparison of different modes with and without autonomous vehicle technology.
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
7

operating cost per passenger kilometer (except regional rail services).
Because of the paid driver, taxi services are substantially more expensive.
In the current transportation system, they are used for convenience or in
situations without alternatives, not because of their cost competitiveness.
Non-automated urban buses and regional rail lines operate at similar
costs per passenger kilometer as private cars.
The picture changes substantially with the automation of vehicles.
While the cost of private cars and rail services changes only marginally,
autonomous driving technology allows taxi services and buses to be
operated at substantially lower cost, even more cheaply than private cars.
In an urban setting, taxis become cheaper than conventional buses,
yet they remain more expensive than automated buses. The absolute cost
difference between buses and taxis, however, is reduced substantially
through automation from 2.20 CHF/pkm to 0.17 CHF/pkm for individual
taxis. Even in relative terms, automated taxis will be only 71% more
expensive for individual and 21% more expensive for pooled use than
automated buses (compared to 415% and 204% before automation).
In regional settings, defined as suburban and exurban trips, auto-
mated taxis and buses become cheaper than private vehicles and rail
services. Here, pooled taxis are the cheapest mode (0.21 CHF/pkm),
followed by individual taxis (0.32 CHF/pkm). In a regional setting, based
on operating cost, automated buses and trains no longer seem to be
competitive (0.40 and 0.44 CHF/pkm).
5.3. Robustness of the estimates
Calculation of the presented cost structures relies on a number of
assumptions, which - given the limited state of knowledge today - have
different degrees of certainty. To analyze how robust the results are to
changes in assumed cost components, three of the variables with highest
uncertainty were varied in the calculation. The selected variables are:
vehicle sticker price, overhead costs for fleet operators and relative active
time of the vehicles.
To reduce the complexity of the problem, changes in each of the three
variables were analyzed uni-dimensionally only. The resulting elasticities
for a 10% increase in the three variables are presented in Table 4. They
are in line with Fig. 1 in that salaries are the key cost driver for con-
ventional services, whereas for autonomous vehicles, hardware is also
important (because vehicle price relative effect is substantially higher).
As shown in the table, changes in sticker price of a vehicle have a sub-
stantial impact only for privately used vehicles. For fleet vehicles, in
contrast, neither the vehicle price nor changes in the overhead cost show
a substantial impact. Increases in the relative active time of a vehicle
show a substantial effect on the operating cost. Although actual effect of
changes in the relative active time are non-linear because fixed costs are
distributed over relative active time (hence an inverse relation), the
linear approximation can be assumed valid between values of 30% and
90% active time. It is therefore argued that for this range, combined
changes in the three variables can be approximated by adding up the
individual effects.
When comparing the different modes, results show that autonomous
buses and solo vehicles benefit most from increased active times and
reduced overhead costs, whereas in the conventional scheme, all systems
benefit comparably. Yet, when compared to Fig. 2, substantial disrup-
tions in the three variables would be required to actually change the
ratios of the costs of the different vehicle types. It can thus be assumed
that cost structures presented above also robustly represent the different
vehicle types’relative order.
6. The new competitive situation
Based on the cost structures described above, suitable market niches
are studied for different operational models with automated vehicles.
Besides private vehicle ownership (midsize, Private aCar), midsize taxis
(individual aTaxi) and line-based public transport (aCityBus in urban
settings and aRegBus for regional settings), a fleet of shared solo vehicles
(Shared aSolo) is considered. Taking into account ongoing developments,
it is assumed that by the time autonomous driving technology is intro-
duced, all vehicles will be equipped with electric propulsion.
Here, as opposed to the previous two sections Section 4and Section 5,
the perspective changes from supplier to user, represented by a change
from cost analysis to price analysis. The price consists of the operator cost
plus the expected profit margin, VAT and fees (see Section 2.4). Obvi-
ously, this does not apply to private vehicles, as there the user is also
the operator.
6.1. The global view
Fig. 3 summarizes the prices per passenger kilometer for the above
modes in an urban and a regional (sub- and exurban) setting with average
usages as described in Section 3. It indicates the future competitive sit-
uation. The cost of private vehicles is differentiated between fixed and
variable cost, as private users often consider only immediate out-of-
pocket cost in their short-term mode choice.
In both settings, it can be observed that a service with shared aSolo
vehicles is not substantially cheaper than aTaxis, because in Switzerland
today, the average occupancy of a midsize vehicle ranges between 1.34
and 1.46 persons per ride (see Section 3). This is enough to make - per
passenger-kilometer - aTaxis competitive with shared aSolos. It can also
be observed that, while city buses remain substantially cheaper than
aTaxis (see Fig. 3(a)), aTaxis can provide services for around 80% of the
regional bus service price (see Fig. 3(b)). In summary, aTaxis are very
competitive, especially for regional settings, if the full cost of private
vehicles is considered. Together with aSolos, they are the cheapest mode
Table 4
Robustness of estimated cost structures (cost per passenger kilometer) to a 10% increase in selected variables.
Private Urban Regional
Pooled Taxi Line-based Pooled Taxi Line-based
Midsize Solo Midsize Van Bus Solo Midsize Van Bus
Vehicle price Conventional 0.024 0.003 0.004 0.007 0.010 0.004 0.004 0.008 0.018
5.0% 0.1% 0.2% 0.4% 1.9% 0.3% 0.8% 0.8% 2.0%
Autonomous 0.029 0.004 0.004 0.008 0.010 0.004 0.005 0.009 0.018
5.8% 1.0% 1.5% 2.3% 4.2% 2.3% 3.6% 4.2% 4.4%
Overhead Conventional –0.006 0.003 0.003 0.004 0.002 0.001 0.001 0.005
–0.2% 0.2% 0.2% 0.7% 0.2% 0.1% 0.1% 0.6%
Autonomous –0.007 0.003 0.003 0.004 0.002 0.001 0.001 0.005
–1.6% 1.0% 0.7% 1.7% 0.9% 0.5% 0.3% 1.3%
Active time Conventional –0.331 0.137 0.137 0.038 0.088 0.036 0.036 0.065
–8.8% 8.5% 8.2% 7.2% 7.9% 7.2% 6.2% 7.3%
Autonomous –0.015 0.006 0.007 0.012 0.004 0.002 0.002 0.017
–3.6% 2.2% 1.8% 4.9% 2.1% 2.0% 1.3% 4.3%
The upper line shows the absolute change in the cost [CHF] per passenger kilometer, the lower line shows relative change.
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
8

in a regional setting, ranking among the cheapest individual service
options in an urban setting.
Quite like today, private vehicles represent the cheapest option if only
immediate out-of-pocket costs (0.17 CHF) are considered. In particular,
in an urban setting (Fig. 3(a)), they incur about two thirds of what has to
be charged for line-based city-bus services and about 40% of autono-
mous taxis.
It should be mentioned here that the above refers to relative differ-
ences in prices. Even if these differences could be substantial, in absolute
terms they are still small (see Fig. 2), which makes the value of travel
time savings a substantially more important factor in mode choice.
6.2. Demand-based view
Having shown that fleet vehicles can be offered at competitive prices,
in a next step, different vehicle sizes are compared. To that end, mini-
mum prices per passenger kilometer were calculated for different vehicle
types and demand levels. In this context, demand levels are defined as
passengers per vehicle per main load direction. It is further assumed that
the vehicles operate as part of larger fleets. Therefore, no additional scale
effects arise when adding more vehicles to a specific route and the
operational model from above is used (see Section 3).
The results are presented in Fig. 4. It becomes immediately clear that
electric propulsion and self-driving technologies allow a substantial
decrease in prices for all modes. This decrease, however, does not affect
all modes in the same way as shown in Section 5. In fact, the most sub-
stantial gains are achieved for shared midsize vehicles, for which the
price per passenger kilometer falls precipitously by 78% to 0.24 CHF/
Pkm at full load. This way, the price gap between a city bus and a midsize
taxi at full load decreases from 0.95 CHF/pkm to 0.18 CHF/pkm. Inter-
estingly, autonomous-electric vans (0.21 CHF/pkm) and minibuses (0.17
CHF/pkm) are not substantially less expensive than midsize vehicles
when operated at full load. The detailed cost values can predict viability
of different autonomous vehicle operational models at different levels of
demand. For example, the data shows that autonomous solo vehicles are
the cheapest mode only for low-demand origin-destination relations.
With an average occupancy of two on this route, a midsize car would
already be more efficient. On the high-demand side of the spectrum, for
demand levels of more than fifteen simultaneous passengers per main
load direction, a city-bus is more economical than a midsize car. The
threshold between a minibus and a city bus is at 21 passengers. The
average occupancy of a city bus today is 22.42 passengers (Bundesamt für
Verkehr (2011)).
7. Discussion
7.1. A new world?
The results of this research help to understand the roles that the
various modes may play in a future transport system, encouraging future
research on AVs and the study of possible implementations in a more
efficient way.
In particular, this research shows that private cars still represent an
attractive option in the era of autonomous vehicles, as out-of-pocket costs
for the user (0.17 CHF/km, c.f. Fig. 3) are lower than for most other
modes (and could be even lower if the user e.g. produced its own elec-
tricity). This is in the range of the $0.15/mile Burns et al. (2013) found
for shared AVs and the $0.16/mile Johnson (2015) found for
purpose-built pooled aTaxis, but they neglected important cost factors -
for example cleaning. Compared to the assumption by Fagnant and
Kockelman (2015) of a $1.00/mile price for a shared AVs, even the full
cost of private vehicle ownership might be competitive.
In fact, buying an autonomous vehicle could be regarded as an in-
vestment in a private mobility robot, which can be used both for chauf-
feur services and for errands; this venture will therefore be even more
Fig. 3. Future competitive situation with autonomous vehicle technology in an urban and
a regional setting.
0.0
0.5
1.0
1.5
0204060
Number of passengers
Price per passenger km [CHF]
VehicleType
CityBus
Midsize
Minibus
Solo
Van
Conventional
0.0
0.5
1.0
1.5
0204060
Number of passengers
Price per passenger km [CHF]
VehicleType
CityBus
Midsize
Minibus
Solo
Van
Autonomous−Electric
Fig. 4. Prices per passenger km versus number of passengers.
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
9

attractive than a conventional vehicle. Hence, it can be expected that a
substantial number of people would value the private use of a mobility
robot and will agree to pay the associated premium. Additionally,
traditional car manufacturers are strongly motivated to maintain the
current emotional connection many people have to their cars. In
conclusion, even as low costs for shared AVs, as estimated by Burns et al.
(2013) and Johnson (2015), might not be low enough to end the reign of
the private car.
In contrast to private vehicle ownership, results of this research
suggest that current line-based public (mass) transportation will prob-
ably be subject to adjustments beyond its automation. Although its
operation will become cheaper, pooled taxi schemes and other new forms
of public transportation will emerge as new and serious competition. The
results of this paper (see Figs. 3 and 4) suggest, however, that the situ-
ation is not as clear as suggested by for example Hazan et al. (2016).If
full cost of shared vehicles is considered, combined with the low average
occupancy achievable with pooling even in an urban setting (Interna-
tional Transport Forum, 2015), mass transit public transport is still
competitive - especially for urban settings and high-demand relationships
- and even more, if unprofitable low-demand relationships can be served
with more flexible services, thus increasing average occupancy. The sit-
uation gets even more interesting if one considers that today's form of
public transportation in Switzerland receives subsidies for approximately
50% of its operating costs (Laesser and Reinhold, 2013). In principle,
with autonomous driving technologies and constant levels of subsidies
and demand, operators would be able to offer line-based mass-transit
public transportation for free. However, public money - if available -
would be much more wisely spent on new forms of public transportation
promising not only lower operating costs, but also higher customer value
by offering more comfort, faster travel times and fewer transfers. Hence,
it is likely that particularly rural or tangential relations will be served by
new forms of AV-based public transportation.
Shared solo vehicles are seen by some researchers as an ideal first-
and-last-mile complement of mass transit public transportation. Offer-
ing direct point-to-point service for single travelers, they promise short
access and travel times. However, they may not be designed to carry
baggage and are not much less expensive to operate than shared midsize
vehicles (see Fig. 3). Policies aside, it is therefore more likely that fleet
operators opt for a homogeneous midsize or van fleet to serve individual
travelers, as well as smaller groups. One vehicle size functions well,
particularly because - assuming acceptably low waiting times - few re-
lations see a demand as low as only one traveler per time interval.
Fleets of (pooled) aTaxis are another mode often proposed as the new
‘jack of all trades’in transport: offered at such low prices that they will
replace every known mode. As this study shows, however, this picture
changes if full costs, including overhead, parking, maintenance and
cleaning, are considered. Just these factors, neglected in most studies on
the topic so far (see Section 1), contribute two thirds of the total cost of
aTaxis (see Section 5). It is thus no surprise that the costs of shared ser-
vices determined here are substantially higher than those in previous
work (e.g. Burns et al. (2013); Johnson (2015); Stephens et al. (2016);
Hazan et al. (2016)). Because shared AVs are still very economical to
operate and OEMs might also have an interest in shared services
(Abhishek et al., 2016), these business models might still have a bright
future. Shared fleets of aTaxis also have the advantage that the usage can
be controlled and (geographically) restricted - especially compared to
privately owned vehicles. This minimizes the liability risk for OEMs and
mobility providers as long as AVs are not yet an established and proven
technology.
Car-pooling might lower prices if high average occupancies can be
achieved. High occupancies, however, come with more detours, longer
individual travel times and more strangers in the same intimate envi-
ronment of a car. Especially the latter factor might be an obstacle to the
success of pooled midsize or van fleets. Although people usually prefer
the privacy of their own vehicle to the anonymity of a bus ride, many find
sharing a vehicle with strangers burdensome (Schmid et al., 2016). In
this respect, more research is required to better understand customer
preferences and design pooled vehicles accordingly, because, if success-
ful, ride-sharing with AVs will definitely have a number of benefits
(Friedrich and Hartl, 2016; International Transport Forum, 2015).
A further challenge for shared services will be to find a solution to
maintain vehicle cleanliness. The analysis above revealed that even with
low cleaning frequencies and costs, cleaning is the single largest contri-
bution to the operating cost of autonomous (individual) taxi schemes.
Any higher level of soiling may even endanger their competitiveness
through high costs or low service standards.
7.2. Limitations
The analysis presented in this paper is - to date - the most compre-
hensive approach to estimating operating costs of future autonomous
transportation services. It includes several aspects previous studies have
overlooked, or assumed negligible (e.g. overhead costs or cleaning) and it
draws a clearer and/or different picture of the future transportation
system than earlier works (Burns, 2013; Fagnant and Kockelman, 2015;
Litman, 2015; Johnson, 2015; Stephens et al., 2016; Hazan et al., 2016).
While this detailed approach reveals new insights, it is also comes with
various limitations.
First, it is likely that vehicle automation will change the demand for
certain kinds of infrastructure, like parking, which could also have an
effect on assumed prices. It is also possible that automated vehicle
introduction induces changes in overall mobility behavior, which will
require the government to take measures to counterbalance negative
effects. The impact of governmental policy measures (e.g. road pricing),
external subsidies, company-internal cross-subsidies and special pricing
strategies, however, have been ignored in this paper; attention to policies
was restricted to those already implemented today. Extended services,
such as for example non-driving personnel in the vehicles or premium
vehicles have been ignored too. Such extras might be part of an improved
customer experience (with a matching price adjustment), or due to legal
requirements (e.g. personnel assisting users with disabilities).
While such measures, policies, and extra services are expected to
substantially influence price structure and thus the competitive situation,
their cost will probably be passed along to the user. It follows that they
would not substantially influence the basic cost structure of different
services for the provider. Given their unpredictable nature and that they
are often designed based on analyses like the one presented in this paper,
it was decided to leave these questions open for future research.
Scenario-based analyses of possible measures, e.g. to achieve socially
optimal prices, or to incorporate externalities in the prices, as well as to
investigate different price vs. service combinations, could further be a
topic for future work.
It should be mentioned that bus and train capacities are likely to be
adjusted after automation. While the influence of different occupancy
rates has been analyzed in Section 6.2, that study does not account for
different vehicle purchase prices and increased vehicle management
costs for growing fleets. It is anticipated that increased resource sched-
uling flexibility in a system without drivers will lead to financial savings.
As extensive analysis would be required to quantify these savings, these
effects would exceed the scope of this paper.
While the cost structures of cars, whether private or as shared vehi-
cles, could be analyzed with a high degree of detail (Section 2.1), the
overhead costs of shared mobility services are well-kept secrets within
the respective companies. Furthermore (Federal Transit Administration,
2016), data suggests that these figures vary substantially among com-
panies of similar size, indicating that further factors like internal orga-
nisation and detailed business case play an important role. Accordingly,
estimates of the overhead cost of shared services and total costs of public
transport services must be treated with caution.
The same applies for the cost effects of new technologies. Electric cars
are already on the market and thus estimates of the cost effects are
reliable and well grounded. The effects of automation, however, are
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
10

uncertain. Admitting this, the approach presented in this paper identifies
different factors resulting in the observed total cost. Then, the effect of
autonomy is estimated for each cost factor separately, resulting in more
precise and reliable estimates. Where available, these estimates were
based on values reported in literature. The authors are aware, however,
that even such detailed estimates’accuracy is questionable. Corre-
spondingly, usage values in this study are based on current market sit-
uation and price structures. If radically lower costs are assumed, these
values are likely to change substantially (e.g. lower occupancy or longer
trips). Once travel behavior impacts and usage patterns of such schemes
become clearer, the framework introduced in this research can be used
for a scenario-based analysis of their cost structures and will probably
yield more accurate results. Until then, the reader should be aware of
these limitations when interpreting the results presented in this paper.
8. Conclusion
This paper presents a detailed cost estimation for current and future
transport modes, with special consideration of automated vehicles. It is
based on a detailed cost structure reconstruction of different transport
services as far as available and best knowledge estimates otherwise. This
analysis goes beyond earlier assessments of future modes’cost structures
in both its level of detail and its rigor.
The framework was validated and delivers new insights into auto-
mation's impacts on different transport services and vehicle categories.
For example, it is clear that fleets of shared autonomous vehicles may
become cheaper than other modes in relative terms, but in absolute
numbers, the difference will be small. Thus, there will still be competi-
tion by other modes and even by private car ownership, which may well
persist beyond the dawn of autonomous vehicle technologies by offering
the luxury and convenience of a personal mobility robot. On the other
hand, this research was able to confirm expectations (e.g Meyer et al.
(2017)) that conventional forms of public transportation may face fierce
competition in the new era.
Importantly, this research also revealed that the success of shared AV
fleets may well depend on a factor which has been previously ignored -
cleaning efforts. According to our findings, developing viable business
models for shared AV fleets will entail solutions to require that customers
behave appropriately while on board (e.g. video observation of passen-
gers, or a confirmation check by the next user on the condition of the car
to identify irresponsible passengers) and/or to clean and repair vehicles
efficiently and at low costs.
Based on an exclusively cost-based approach, this research was able
to clarify use cases of future modes of transportation, although many
open questions remain and require further research. For example, the
actual mode choice is determined not only by cost, but also substantially
by travel time and comfort (value of time), as well as other factors like the
perception of transfers, waiting times, etc. - all of which were not
investigated in this paper. Therefore, actual implementations of the
proposed schemes need to be implemented in a field trial or simulation
approach to better understand the size of the respective market segments
in a realistic environment. Moreover, due to its complexity, long-distance
travel could not be covered in this research, but it is known to play an
important role in mobility tool choices (Canzler and Knie, 1994).
Another future research area is the investigation of a re-sized, line-
based transit resulting from the automation of buses. If driver wage is no
longer part of the cost structure, it might be worthwhile to operate buses
with smaller capacities and higher frequencies. Not only is demand
bundling, when possible, more economic than point-to-point service (see
Section 5), there is also a user preference for high-frequency, line-based
service over dynamic services (for a recent review, see (White, 2016)).
Current passenger statistics of the City of Zurich (Verkehrsbetriebe Zür-
ich, 2014) make a first approximation of the utilization possible. None-
theless, cost ramifications of legal requirements, as well as infrastructure
necessary for mass transit need further consideration that would exceed
the scope of this paper.
Acknowledgement
This research is part of two projects funded by the Swiss National
Science Foundation; the SNF-project number 200021_159234, Autono-
mous Cars, and the SNF-project number 407140_153807, Sharing is
Saving, which is part of the National Research Program 71: Managing
Energy Consumption, as well as the SVI project 2016/001, Induced De-
mand by Autonomous Vehicles.
The authors would like to thank (in alphabetical order) Hans-J€
org
Althaus, Nick Beglinger, Ralf Bosch, Peter Brandl, Frank Bruns, Walter
Casazza, Kilian Constantin, Hauke Fehlberg, Martin Fellendorf, Markus
Friedrich, Christian Graf von Normann-Ehrenfels, J€
org Jermann, Simon
Kettner, Scott Le Vine, Don MacKenzie, Peter White, and the anonymous
reviewers of Transport Policy for their valuable feedback and inputs.
Their reviews were a great help in improving the paper. All remaining
errors are our own.
A. Determination of Overhead Costs
Given their competitive environment, ride-hailing and car sharing services treat any information on their overhead and vehicle management cost as
confidential. One source, however, provides data on the financial, operating, and asset condition of 660 transit providers in the US (Federal Transit
Administration, 2016). Overhead costs are denoted as General Administration and include: Transit service development, injuries and damages, safety,
personnel administration, legal services, insurance, data processing, finance and accounting, purchasing and stores, engineering, real estate manage-
ment, office management and services, customer services, promotion, market research, and planning. Vehicle management is denoted as vehicle op-
erations and includes: Transportation administration and support, revenue vehicle movement control, scheduling of transportation operations, revenue
vehicle operation, ticketing and fare collection, and system security. Detailed information is only available for a subset of all agencies (so called “full
reporters”). In order to derive cost figures for on-demand fleet services, the analysis focused only on full reporters offering van pooling and demand
responsive transport services, with the latter being defined as transit services with a flexible route and schedule. The resulting sample consists of 527
services of 430 agencies. Fleet size (number of revenue vehicles operated in the annual maximum service) ranges from 1 to 1840.
The analysis started by plotting overhead costs per vehicle and day versus the fleet size. It was observed that these costs per vehicle and day vary
substantially (between 0.60 US$ and 311.44 US$) below and equal to the fleet size of 150. Focusing on this data and regressing fleet size on overhead
costs per vehicle and day reveals that fleet size does not have a significant effect in that range (p-value equals 0.14). Above a threshold of 150 vehicles,
overhead costs per vehicle and day range between 1.39 US$ and 15.47 US$, except for two agencies, and data show neither a positive nor a negative
trend (p-value of fleet size equals 0.53). It was therefore hypothesized that fleet composition plays a key role in explaining these variations. When,
however, regressing average vehicle capacity and fleet size on overhead costs per vehicle and day using the whole dataset, the average vehicle capacity
does not have a significant effect (p-value equals 0.62). Interestingly, a similar picture emerges when analyzing dependence of vehicle operations cost
per vehicle and day on the fleet size, as well as average vehicle seat capacity. Again, vehicle operations cost per vehicle and day vary substantially
(between 4.70 US$ and 623.06 US$) below and equal to the fleet size of 150. With the exception of two agencies, this figure is between 1.79 US$ and
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
11

23.97 US$ for fleets with a size above 150.
To conclude, results indicate that either: important data is missing needed to determine the relationship between fleet composition and overhead
and operations cost, or that agencies operate with very different cost structures. As mentioned previously, however, companies with fleet sizes of 150
vehicles and more show stable figures for both cost structures. Therefore, corresponding median values are used as estimates for overhead (10.24 US$
per vehicle-day) and vehicle operation costs per vehicle and day (8.39 US$ per vehicle-day). Adjusting these figures to unit labor costs in Switzerland
(Neff et al., 2015) results in approximately 14 CHF per vehicle-day for overhead and 10 CHF per vehicle-day for operations costs.
B. Cost parameters
B.1
Costs vehicles
Car Type Solo Midsize Van Minibus
Capacity [Seats for Pass.] 1 4 8 20
Acquisition [CHF] 13000
a
35000
b
66000
c
70000
d
Yearly insurance [CHF/a] 500
e
1000
e
1200
e
1400
e
Yearly tax [CHF/a] 120
f
250
f
700
f
1100
f
Yearly parking [CHF/a] 1500
g
1500
g
1500
g
1500
g
Yearly toll [CHF/a] 40
h
40
h
40
h
2200
i
Maintenance [CHF/km] 0.02
j
0.06
j
0.12
j
0.13
j
Cleaning [CHF/km] 0.02
k
0.03
k
0.04
k
0.05
k
Tires [CHF/km] 0.02
l
0.02
l
0.02
l
0.02
l
Fuel [CHF/km] 0.06
m
0.08
m
0.13
m
0.2
m
a
Data: http://www.renault.ch/, last accessed 01.02.2017.
b
Data: http://www.volkswagen.ch/, last accessed 01.02.2017.
c
Data: http://www.volkswagen-nutzfahrzeuge.ch/, last accessed 01.02.2017.
d
Data: http://www.mercedes-benz.ch/, last accessed 01.02.2017.
e
Source: Comparis AG (2016).
f
Source: Strassenverkehrsamt Kanton Zürich (2017).
g
Source: TCS (2016).
h
Source: Eidgen€
ossische Zollverwaltung (2017b).
i
Source: Eidgen€
ossische Zollverwaltung (2017a).
j
Data: TCS (2016).
k
Data: Autop (2016).
l
Assumption: Two sets of tires (each 300 CHF) and two annual tire changes (each 100
CHF) would allow for 50000 km of driving.
m
Sources: Fuel price - (Shell, 2016); Consumption - Respective manufacturers website -
Solo: 4.3 l/100 km; Midsize 5.7 l/100 km, Van 9.3 l/100 km, Minibus 14.2 l/100 km.
B.2
Cost adjustments
Position Electric Automated Fleet VAT deductible
Acquisition –+20%
a
30%
b
Yes
Yearly insurance 35%
c
50%
d
20%
e
Yes
Yearly tax 100%
f
––No
Yearly parking –– +133%
g
Yes
Yearly toll –– – Yes
Maintenance +28%
h
–25%
i
Yes
Cleaning –See Cost parameters No
Tires –10%
j
25%
i
Yes
Fuel 50%
k
10%
d
5%
l
Yes
a
Source: IHS (2014).
b
Data: Blens (2015).
c
Data: Comparis AG (2016).
d
Source: Stephens et al. (2016).
e
Source: Helvetia (2016).
f
Source: http://www.stva.zh.ch/internet/sicherheitsdirektion/stva/de/StVAgeb/
GEBva14.html, last accessed 01.02.2017.
g
Data: Schlesiger (2014).
h
Data: Diez (2016), Saxton (2013), McKinsey (2017).
i
Assumption: In-house maintenance.
j
Assumption: Similar effect as fuel consumption.
k
Data: Fuel price - (Shell, 2016); Electricity price - https://www.ewz.ch/de/private.
html, last accessed 01.02.2017; Consumption - http://www.volkswagen.ch/, last
accessed 01.02.2017.
l
Source: Migrol (2016).
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
12

B.3
Costs mass transit
Transportation Type CityBus RegBus Rail
Capacity 60
a
60
a
297
b
Costs per vehicle-kilometer [CHF/km] 7.14
c
6.7
c
31.4
d
Effect electric 5.5%
e
5.5%
e
–
Effect automated 55%
e
55%
e
2.5%
d
a
Data: https://www.stadt-zuerich.ch/vbz/de/index/die_vbz/fahrzeuge/autobusse.
html, last accessed 01.02.2017.
b
Source: SBB AG (2016).
c
Data: Frank et al. (2008), Verkehrsbetriebe Zürich (2016), Bundesamt für Verkehr
(2011).
d
Data: SBB AG (2016).
e
Data: Frank et al. (2008).
C. Usage parameters
C.1
Average usages private ownership and dynamic fleet public transport
Transport Type
Area * * * Urban Urban Urban Urban Regional Regional Regional Regional
VehicleType Solo Midsize Van Solo Midsize Midsize Van Solo Midsize Midsize Van
Operation Priv Priv Priv PT-NP PT-NP PT-P PT-P PT-NP PT-NP PT-P PT-P
Peak Hours
operationHours_AV [h] 0.32
a
0.32
a
0.32
a
3.80
b
3.80
b
3.80
b
3.80
b
3.80
b
3.80
b
3.80
b
3.80
b
operationHours [h] 0.32
a
0.32
a
0.32
a
3.80
b
3.80
b
3.80
b
3.80
b
3.80
b
3.80
b
3.80
b
3.80
b
relActiveTime [%] 100.00 100.00 100.00 57.00
c
57.00
c
57.00
c
57.00
c
57.00
c
57.00
c
57.00
c
57.00
c
avOccupancy [%] 100.00 33.50
a
16.75
a
100.00 33.50
a
65.00
d
32.50
d
100.00 33.50
a
65.00
d
32.50
d
avSpeed [km/h] 32.40
a
32.40
a
32.40
a
20.60
a
20.60
a
20.60
a
20.60
a
31.30
a
31.30
a
31.30
a
31.30
a
avTripLengthPass [km] 15.00
a
15.00
a
15.00
a
3.40
a
3.40
a
3.40
a
3.40
a
8.00
a
8.00
a
8.00
a
8.00
a
relEmptyRides [%] 0.00 0.00 0.00 8.00
e
8.00
e
8.00
e
8.00
e
15.00
c
15.00
c
15.00
c
15.00
c
relMaintenanceRides [%] 0.00 0.00 0.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00
Off-Peak Hours
operationHours_Av [h] 0.67
a
0.67
a
0.67
a
7.60
b
7.60
b
7.60
b
7.60
b
7.60
b
7.60
b
7.60
b
7.60
b
operationHours [h] 0.67
a
0.67
a
0.67
a
7.60
b
7.60
b
7.60
b
7.60
b
7.60
b
7.60
b
7.60
b
7.60
b
relActiveTime [%] 100.00 100.00 100.00 59.00
c
59.00
c
59.00
c
59.00
c
59.00
c
59.00
c
59.00
c
59.00
c
avOccupancy [%] 100.00 36.50
a
18.25
a
100.00 36.50
a
60.00
d
30.00
d
100.00 36.50
a
60.00
d
30.00
d
avSpeed [km/h] 34.00
a
34.00
a
34.00
a
21.90
a
21.90
a
21.90
a
21.90
a
32.20
a
32.20
a
32.20
a
32.20
a
avTripLengthPass [km] 10.50
a
10.50
a
10.50
a
3.00
a
3.00
a
3.00
a
3.00
a
5.90
a
5.90
a
5.90
a
5.90
a
relEmptyRides [%] 0.00 0.00 0.00 8.00
e
8.00
e
8.00
a
8.00
e
15.00
c
15.00
c
15.00
c
15.00
c
relMaintenanceRides [%] 0.00 0.00 0.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00
B.4
General cost parameters
Cost parameter Value
Vehicle lifetime professional 300000 km
a
Vehicle lifetime private 10 a
Cleaning price 15 CHF
b
Yearly cleaning frequency private 8
c
Yearly cleaning frequency professional conventional 183
d
Cleaning frequency commercial AV per trip 0.025
e
Driver's hourly wage 35 CHF
f
Daily overhead costs per vehicle 14 CHF/d
g
Daily vehicle management costs per vehicle 10 CHF/d
g
Annual interest private customer 0.079
h
Annual interest commercial customers 0.015
i
Credit period private customers 5a
h
Credit period commercial customers 3a
VAT in Switzerland 0.08
j
Profit margin 0.03
k
Payment transaction fee 0.0044
l
a
Data: Klose (2012).
b
Source: Autop (2016).
c
Data: abh Market Research (2007).
d
Assumption: Cleaned every second day.
e
Source: Private communication with car sharing operator.
f
Data: Lohnanalyse (2016); Braendle (2013).
g
Data: Federal Transit Administration (2016).
h
Source: Migrosbank (2017).
i
Data: Sixt SE (2016).
j
Source: https://www.ch.ch/en/vat-rates-switzerland/, last accessed 01.02.2017.
k
Data: SCI Verkehr (2016).
l
Source: Wettbewerbskommission WEKO (2014).
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
13

C.1 (continued )
Night Hours
operationHours_AV [h] 0.30
a
0.30
a
0.30
a
9.60
b
9.60
b
9.60
b
9.60
b
9.60
b
9.60
b
9.60
b
9.60
b
operationHours [h] 0.30
a
0.30
a
0.30
a
8.60
b
8.60
b
8.60
b
8.60
b
8.60
b
8.60
b
8.60
b
8.60
b
relActiveTime [%] 100.00 100.00 100.00 30.00
c
30.00
c
30.00
c
30.00
c
30.00
c
30.00
c
30.00
c
30.00
c
avOccupancy [%] 100.00 35.75
a
17.88
a
100.00 35.75
a
57.50
d
28.75
d
100.00 35.75
a
57.50
d
28.75
d
avSpeed [km/h] 36.20
a
36.20
a
36.20
a
23.70
a
23.70
a
23.70
a
23.70
a
34.00
a
34.00
a
34.00
a
34.00
a
avTripLengthPass [km] 12.60
a
12.60
a
12.60
a
3.40
a
3.40
a
3.40
a
3.40
a
7.10
a
7.10
a
7.10
a
7.10
a
relEmptyRides [%] 0.00 0.00 0.00 8.00
e
8.00
e
8.00
e
8.00
e
15.00
c
15.00
c
15.00
c
15.00
c
relMaintenanceRides [%] 0.00 0.00 0.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00
Abbreviations.
*: Overall.
Priv: Private ownership.
PT-P: PT with Pooling.
PT-NP: PT Non-Pooling, especially relevant for midsize taxis.
a
Data: Swiss Federal Statistical Office (BFS) and Swiss Federal Office for Spatial
Development (ARE) (2012).
b
Assumption: Defined duration minus maintenance.
c
Data: B€
osch et al. (2016).
d
Data: International Transport Forum (2015).
e
Source: Fagnant et al. (2015).
C.2
Average usages line-based mass transit
Transport Type
Area Urban Regional Regional
VehicleType CityBus RegBus Rail
Operation PT PT PT
operationHours_AV [h] 19
a
19
a
19
a
operationHours [h] 19
a
19
a
19
a
relActiveTime [%] 85
b
85
b
85
b
avOccupancy [%] 22.42
c
12.6
c
22.7
d
avSpeed [km/h] 21.31
c
20.89
c
37.82
c
avTripLengthPass [km] –– –
relEmptyRides [%] 0 0 0
relMaintenanceRides [%] 5 5 5
a
Assumption: 24 h minus night break.
b
Data: Zürcher Verkehrsverbund (2017).
c
Data: Bundesamt für Verkehr (2011).
d
Source: SBB AG (2016).
D. Detailed cost figures for different modes
VehicleType Area Operation Steering Propulsion CostVehKM [CHF] CostSeatKM [CHF] CostPassKM [CHF]
Bus Reg PT-P Aut Elec –0.05 0.38
Bus Reg PT-P N. aut Elec –0.11 0.84
Bus Reg PT-P Aut N. elec –0.05 0.40
Bus Reg PT-P N. aut N. elec –0.11 0.89
Bus Urb PT-P Aut Elec –0.05 0.23
Bus Urb PT-P N. aut Elec –0.11 0.50
Bus Urb PT-P Aut N. elec –0.05 0.24
Bus Urb PT-P N. aut N. elec –0.12 0.53
Midsize * Priv Aut Elec 0.67 0.17 0.47
Midsize * Priv N. aut Elec 0.63 0.16 0.44
Midsize * Priv Aut N. elec 0.72 0.18 0.50
Midsize * Priv N. aut N. elec 0.69 0.17 0.48
Midsize Reg PT-NP Aut Elec 0.35 0.09 0.31
Midsize Reg PT-NP N. aut Elec 2.16 0.54 1.90
Midsize Reg PT-NP Aut N. elec 0.37 0.09 0.32
Midsize Reg PT-NP N. aut N. elec 2.19 0.55 1.92
Midsize Reg PT-P Aut Elec 0.39 0.10 0.20
Midsize Reg PT-P N. aut Elec 2.16 0.54 1.12
Midsize Reg PT-P Aut N. elec 0.41 0.10 0.21
Midsize Reg PT-P N. aut N. elec 2.19 0.55 1.13
Midsize Urb PT-NP Aut Elec 0.48 0.12 0.39
Midsize Urb PT-NP N. aut Elec 3.35 0.84 2.70
Midsize Urb PT-NP Aut N. elec 0.51 0.13 0.41
Midsize Urb PT-NP N. aut N. elec 3.38 0.85 2.73
Midsize Urb PT-P Aut Elec 0.58 0.15 0.28
Midsize Urb PT-P N. aut Elec 3.35 0.84 1.60
Midsize Urb PT-P Aut N. elec 0.61 0.15 0.29
Midsize Urb PT-P N. aut N. elec 3.38 0.85 1.61
Minibus Reg PT-P Aut Elec 0.61 0.03 0.31
Minibus Reg PT-P N. aut Elec 2.37 0.12 1.23
Minibus Reg PT-P Aut N. elec 0.67 0.03 0.35
Minibus Reg PT-P N. aut N. elec 2.44 0.12 1.26
Minibus Urb PT-P Aut Elec 0.98 0.05 0.24
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
14

(continued )
VehicleType Area Operation Steering Propulsion CostVehKM [CHF] CostSeatKM [CHF] CostPassKM [CHF]
Minibus Urb PT-P N. aut Elec 1.89 0.09 0.46
Minibus Urb PT-P Aut N. elec 1.04 0.05 0.25
Minibus Urb PT-P N. aut N. elec 1.96 0.10 0.47
Rail Reg PT-P Aut Elec –0.10 0.44
Rail Reg PT-P N. aut Elec –0.11 0.47
Rail Reg PT-P Aut N. elec –0.10 0.44
Rail Reg PT-P N. aut N. elec –0.11 0.47
Solo * Priv Aut Elec 0.35 0.35 0.35
Solo * Priv N. aut Elec 0.33 0.33 0.33
Solo * Priv Aut N. elec 0.38 0.38 0.38
Solo * Priv N. aut N. elec 0.38 0.38 0.38
Solo Reg PT-NP Aut Elec 0.23 0.23 0.28
Solo Reg PT-NP N. aut Elec 2.07 2.07 2.59
Solo Reg PT-NP Aut N. elec 0.25 0.25 0.31
Solo Reg PT-NP N. aut N. elec 2.09 2.09 2.62
Solo Urb PT-NP Aut Elec 0.34 0.34 0.39
Solo Urb PT-NP N. aut Elec 3.26 3.26 3.74
Solo Urb PT-NP Aut N. elec 0.36 0.36 0.41
Solo Urb PT-NP N. aut N. elec 3.28 3.28 3.77
Van * Priv Aut Elec 1.14 0.14 0.80
Van * Priv N. aut Elec 1.04 0.13 0.73
Van * Priv Aut N. elec 1.22 0.15 0.86
Van * Priv N. aut N. elec 1.14 0.14 0.80
Van Reg PT-P Aut Elec 0.55 0.07 0.28
Van Reg PT-P N. aut Elec 2.31 0.29 1.19
Van Reg PT-P Aut N. elec 0.58 0.07 0.30
Van Reg PT-P N. aut N. elec 2.35 0.29 1.21
Van Urb PT-P Aut Elec 0.74 0.09 0.35
Van Urb PT-P N. aut Elec 3.49 0.44 1.66
Van Urb PT-P Aut N. elec 0.77 0.10 0.37
Van Urb PT-P N. aut N. elec 3.54 0.44 1.69
Abbreviations.
Reg: Regional (suburban and exurban).
Urb: Urban.
*: Overall.
Priv: Private ownership (for private ownership the area was not differentiated (see Aver-
ageusages) resulting in the same values for both regions).
PT-P: PT with Pooling.
PT-NP: PT Non-Pooling, especially relevant for midsize taxis.
Aut: Autonomous.
N. aut: Not autonomous.
Elec: Electrified.
N. elec: Not electrified.
References
abh Market Research, 2007. Autow€
asche in Deutschland, Ein Markt mit Potenzial.
Technical Report.
Abhishek, V., Guajardo, J.A., Zhang, Z., 2016. Business Models in the Sharing Economy:
Manufacturing Durable Goods in the Presence of Peer-to-peer Rental Markets. Online
article (last accessed 10.03.2017), available at: SSRN:. Dec. 2016. https://ssrn.com/
abstract¼2891908.
Autop, 2016. Preisliste Richterswil. Webpage (last accessed 08.11.2016). http://autop.
ch/dienstleistungen/waschstrassen.
Blens, S., 2015. 30 beliebte Firmenwagen im Rabattcheck, Press Information,
MeinAuto.de, K€
oln, Jul. 2015.
B€
osch, P.M., Ciari, F., Axhausen, K.W., 2016. Required autonomous vehicle fleet sizes to
serve different levels of demand. Transp. Res. Rec. 2542, 111–119.
Braendle, T., 2013. Wie viel kostet ein Mitarbeiter wirklich? Webpage (last accessed
12.10.2016). https://www.runmyaccounts.ch/2013/07/wie-viel-kostet-ein-
mitarbeiter-wirklich/.
Bundesamt für Verkehr, 2011. Kennzahlen Regionaler Personenverkehr mit
Erschliessungsfunktion. Webpage (last accessed 08.11.2016). https://www.bav.
admin.ch/bav/de/home/das-bav/aufgaben-des-amtes/finanzierung/finanzierung-
verkehr/personenverkehr/rpv-mit-erschliessungsfunktion/rpv-kennzahlen.html.
Burns, L.D., 2013. Sustainable mobility: a vision of our transport future. Nature 497,
181–182.
Burns, L.D., Jordan, W., Scarborough, B., 2013. Transforming Personal Mobility,
Technical Report. The Earth Institute, Columbia University.
Canzler, W., Knie, A., 1994. Das Ende des Automobils: Fakten und Trends zum Umbau der
Autogesellschaft, 1 edn., C.F. Müller, Heidelberg.
Comparis AG, 2016. Webpage (last accessed 08.11.2016). http://www.comparis.ch/.
Diez, W., 2016. Elektroautos mit deutlich niedrigeren Unterhaltskosten. Webpage (last
accessed 08.11.2016). http://www.ifa-info.de/downloads/3241%
2FPresseinformation%2520Elektromobilit%25C3%25A4t%252020.11.12.pdf.
Eidgen€
ossische Zollverwaltung, 2017a. Pauschale Schwerverkehrsabgabe (PSVA) für
Schweizer Fahrzeuge. Webpage (last accessed 26.01.2017). https://www.ezv.admin.
ch/ezv/de/home/information-firmen/transport–reisedokument–strassenabgaben/
schwerverkehrsabgaben–lsva-und-psva-/pauschale-schwerverkehrsabgabe–
phtml#681554958.
Eidgen€
ossische Zollverwaltung, 2017b. Vignette (Autobahngebühren). Webpage (last
accessed 01.02.2017). https://www.ezv.admin.ch/ezv/de/home/information-
private/reisedokumente-und-strassenabgaben/vignette–autobahngebuehren-.html.
Fagnant, D.J., Kockelman, K.M., 2014. Preparing a Nation for Autonomous vehicles:
opportunities, barriers and policy recommendations for capitalizing on self-driven
vehicles. In: Paper Presented at the 93rd Annual Meeting of the Transportation
Research Board, Washington DC, January 2014.
Fagnant, D., Kockelman, K., 2015. Dynamic ride-sharing and optimal fleet sizing for a
system of shared autonomous vehicles. In: Paper Presented at the 94th Annual
Meeting of the Transportation Research Board, Washington DC, January 2015.
Fagnant, D., Kockelman, K., Bansal, P., 2015. Operations of shared autonomous vehicle
fleet for Austin, Texas market. Transp. Res. Rec. J. Transp. Res. Board 2536, 98–106.
Federal Transit Administration, 2016. National Transit Database - Urbanized Area.
Webpage (last accessed 23.01.2017). http://www.apta.com/resources/statistics/
Pages/NTDDataTables.aspx.
Frank, P., Friedrich, M., Schlaich, J., 2008. Betriebskosten von Busverkehren schnell und
genau ermitteln. Der Nahverk. 26 (11), 15–22.
Friedrich, M., Hartl, M., Dec. 2016. MEGAFON - Modellergebnisse geteilter autonomer
Fahrzeugflotten des oeffentlichen Nahverkehrs. Research report. Universit€
at Stuttgart,
Institut für Strassen- und Verkehrswesen, Stuttgart.
Hazan, J., Lang, N., Ulrich, P., Chua, J., Doubara, X., Steffens, T., 2016. Will autonomous
Vehicles Derail Trains?, bcg.Perspectives. September 2016.
Helvetia, 2016. Die Mitarbeitendenrabatte auf Helvetia Versicherungen. Webpage (last
accessed 08.11.2016). https://www.helvetia.com/ch/content/de/landingpages/
rabatt/mitarbeitendenrabatte.html.
IHS, 2014. Emerging Technologies: Autonomous Cars –Not if, but when. Technical
Report, Englewood, Colorado.
International Transport Forum, 2015. Urban Mobility System Upgrade: How Shared Self-
driving Cars Could Change City Traffic. Technical Report. OECD.
Johnson, B., 2015. Disruptive Mobility, Research Report, Barclays.
Johnson, C., Walker, J., 2016. Peak Car Ownership, Research Report. Rocky Mountain
Institute.
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
15

Klose, R., 2012. BMW gibt Seinen Autos Nur 150’000 Kilometer. Online Article (last
accessed 23.01.2017), Sep. 2012. http://www.spiegel.de/auto/aktuell/bmw-setzt-
maximal-laufleistung-von-150-000-km-voraus-a-855355.html.
Laesser, C., Reinhold, S., 2013. Finanzierung des OeV in der Schweiz: Was zahlt der
Nutzer, was die Allgemeinheit?, Arbeitsbericht. Schriftenreihe SBB Lab, St. Gallen:
SBB Lab.
Litman, T., 2015. Autonomous Vehicle Implementation Predictions. Technical Report.
Victoria Transport Policy Institute. December 2015.
Lohnanalyse, 2016. Lohnanalyse für Taxifahrer/in in der Schweiz. Webpage (last
accessed 12.10.2016). http://www.lohnanalyse.ch/ch/loehne/details/taxifahrerin.
html.
Lohnanalyse, 2017. Lohnanalyse für Lokführer/in in der Schweiz. http://www.
lohnanalyse.ch/ch/loehne/details/lokfuehrerin/sql-administrator-it-operations-
konzern-it-385943.html.
McKinsey, 2017. Electrifying Insights: How Automakers Can Drive Electrified Vehicle
Sales and Profitability. Technical Report.
Meyer, J., Becker, H., B€
osch, P.M., Axhausen, K.W., 2017. Autonomous Vehicles: the Next
Jump in accessibilities?, Research in Transportation Economics. https://doi.org/10.
1016/j.retrec.2017.03.005.
Migrol, 2016. Treibstoffbon Rabatt 3 Rp./l. Webpage (last accessed 08.11.2016). http://
www.migrol.ch/de/Tankstellen/Treibstoffbon-Rabatt-3-Rp-l.aspx.
Migrosbank, 2017. Privatkredit –Pers€
onliche Pl€
ane günstig Verwirklichen. Webpage (last
accessed 16.01.2017). https://www.migrosbank.ch/de/privatpersonen/kredit/
privatkredit.html.
Morgan Stanley and Boston Consulting Group, 2016. Motor Insurance 2.0, Bluepaper,
September 2016.
Neff, M., Kl€
ager, R., Brivio, S., Hauser, A., Koch, A., Arapovic, D., Vonwiller, C., Seger, L.,
Kunz, M., 2015. Der starke Franken: Fluch und Segen für die Industrie. Technical
Report. Raiffeisen Economic Research, Zurich.
NHTSA, 2017. Automatic Vehicle Control Systems. Technical Report.
Nykvist, B., Nilsson, M., 2015. Rapidly falling costs of battery packs for electric vehicles.
Nat. Clim. Change 5 (4), 329–332.
Organisation for Economic Co-operation and Development (OECD), 2017a. OECD Data-
Exchange Rates. Webpage (last accessed 18.04.2017). https://data.oecd.org/
conversion/exchange-rates.htm.
Organisation for Economic Co-operation and Development (OECD), 2017b. OECD Data
Purchasing Power Parities (PPP). Webpage (last accessed 18.04.2017). https://data.
oecd.org/conversion/purchasing-power-parities-ppp.htm.
SAE International, 2014. Taxonomy and Definitions for Terms Related to On-road Motor
Vehicle Automated Driving Systems, Warrendale, PA j3016 edn., Jan. 2014. http://
standards.sae.org/j3016_201401/.
Saxton, T., 2013. Plug in America's Tesla Roadster Battery Study. Technical Report.
SBB AG, 2016. Zahlen und Fakten 2015. Webpage (last accessed: 08.11.2016). http://
www.sbb.ch/zahlen-und-fakten.
Schlesiger, C., 2014. Der Neue Straßenkampf. Online Article (last accessed 01.02.2017),
Aug. 2014. http://www.wiwo.de/politik/deutschland/mobilitaet-in-grossstaedten-
carsharing-verschaerft-parkplatznot/10218492–10218493.html.
SCI Verkehr, 2016. Deutsche Logistik-konjunktur Bleibt Weiterhin Stabil. Technical
Report, Hamburg.
Schmid, B., Schmutz, S., Axhausen, K.W., 2016. Explaining mode choice, taste heterogeneity
and cost sensitivity in a post-car world. In: Paper Presented at the 95th Annual Meeting
of the Transportation Research Board, Washington, D.C., January 2016.
Shell, 2016. Durchschnittliche S€
aulenpreise - Schweiz. Webpage (last accessed
08.11.2016). http://www.shell.ch/de/products-services/on-the-road/fuels/fuel-
pricing.html.
Sixt SE, 2016. Sixt SE emittiert Anleiheüber 250 Mio. Euro zur weiteren
Wachstumsfinanzierung. Webpage (last accessed 08.11.2016), Oct. 2016. https://
about.sixt.com/websites/sixt_cc/German/2999/news-details.html?
newsID¼1585535#news-EQS.
Stadt Zürich, 2014. Taxitarif der Stadt Zürich, Stadtratsbeschluss, Zurich.
Stephens, T., Gonder, J., Chen, Y., Lin, Z., Liu, C., Gohlke, D., 2016. Estimated Bounds and
Important Factors for Fuel Use and Consumer Costs of Connected and automated
Vehicles. Technical Report. National Renewable Energy Laboratory, U.S. Department
of Energy, Golden, CO. Nov. 2016.
Strassenverkehrsamt Kanton Zürich, 2017. Verkehrsabgabenrechner. Webpage (last
accessed 01.02.2017). http://www.stva.zh.ch/internet/sicherheitsdirektion/stva/
de/StVAgeb/GEBva14/VArechner.html.
Swiss Federal Statistical Office (BFS) and Swiss Federal Office for Spatial Development
(ARE), 2012. Mobilit€
at in der Schweiz - Ergebnisse des Mikrozensus Mobilit€
at und
Verkehr 2010. Swiss Federal Statistical Office (BFS) and Swiss Federal Office for
Spatial Development (ARE), Neuchatel and Berne.
TCS, 2016. Kosteneines Musterautos. Webpage (last accessed 08.11.2016). https://www.
tcs.ch/de/auto-mobilitaet/autokosten/kosten-eines-musterautos.php.
Uber, 2017. Uber Bewegt Zürich. Webpage (last accessed 04.01.2017). https://www.
uber.com/de/cities/zurich/.
Verkehrsbetriebe Zürich, 2014. Fahrgastzahlen VBZ. Webpage (last accessed
23.01.2017). https://data.stadt-zuerich.ch/dataset/vbz-fahrgastzahlen-ogd.
Verkehrsbetriebe Zürich, 2016. BusfahrerIn - Die Bewerbung. Webpage (last accessed
08.11.2017). https://www.stadt-zuerich.ch/vbz/de/index/die_vbz/arbeiten_bei_
den_vbz/berufswelten/busfahrer_in/busfahren_ details.html.
Wettbewerbskommission WEKO, 2014. Kreditkarten Domestische Interchange Fees II
(KKDMIF II) (Technical Report).
White, P., 2016. The roles of ‘conventional’and demand-responsive bus services. In:
Mulley, C., Nelson, J.D. (Eds.), Paratransit: Shaping the Flexible Transport Future
(Transport and Sustainability Volume 8). Emerald Group Publishing Limited, Bingley,
U.K, pp. 307–330.
Zürcher Verkehrsverbund, 2017. Main Page. Webpage (last accessed 01.02.2017). http://
www.zvv.ch/zvv/de/home.html.
P.M. B€
osch et al. Transport Policy xxx (2017) 1–16
16