Model-based decision support for future OEM power-train
portfolios: academic solutions for practical requirements
Karsten Kieckhafer ‘*, Katharina Wachter ', Joachim Axmann ?,
Thomas Stefan Spengler *
1: Technische Universitat Braunschweig,
Institute of Automotive Management and Industrial Production,
Chair of Production and Logistics,
Katharinenstrafe 3, 38106 Braunschweig, Germany
2: Volkswagen AG,
Products Division, Group Business Processes,
Mailbox 1441, 38436 Wolfsburg, Germany
* Corresponding author
Phone: +49-531-391-63050, Fax: +49-531-391-2203,
E-Mail: k.kieckhaefer@tu-braunschweig.de
Meeting 21% century’s challenges of climate change and scarcity of crude oil requires
the transition to alternatively powered vehicles, such as electric vehicles. As a conse-
quence, car manufacturers have to integrate these vehicles into their product porttoli-
os. Decisions have to be made about, for instance, the power-train to be oftered in spe-
cific vehicle models and their times of introduction. This is a complex decision mak-
ing task, especially due to high uncertainties about the future development of the mar-
ket demand for alternatively powered vehicles.
We here discuss how the application of system dynamics and agent-based simulation
can contribute to manage the transition to alternatively powered vehicles from a manu-
facturer’s perspective. To this end, we present practical requirements on a model-based
decision support and a scientifically novel simulation approach to filfill these re-
quirements. The simulation approach was developed in cooperation between universi-
ty and industry. It integrates a system dynamics model with an agent-based discrete
choice model to simulate aggregated system behavior and individual consumer choices
based on industrially proved data. We show that our novel approach meets users’ re-
quirements and can ofter multiple benefits for decision making in industry. We dis-
cuss how these benefits can be exploited in future.
Keywords: automotive industry, alternative power-trains, product portfolio decisions,
agent-based simulation, system dynamics
1 Introduction
Automobile manufacturers are currently focusing on alternatively powered vehicles to
reduce the environmental impact of their vehicle fleet. This is first due to stricter polit-
ical regulations such as the CO2-emission regulation of the European Union (Europe-
an Union 2010) or other countries, second, due to the approaching scarcity of crude oil
and gas, and, third, due to society’s expectations. Alternatively powered vehicles such as
sustainable natural gas vehicles, battery electric or fuel cell electric vehicles can be op-
erated independently of crude oil and gas. They thus seem promising to meet these
challenges. For this reason, automobile manufacturers are striving to integrate alterna-
tively powered vehicles into their vehicle portfolio.
This is typically a task of strategic product portfolio planning. Here the product portfo-
lio offered on the market has to be designed so that it is synchronized with the market
demand over time. For this, the question has to be answered, which vehicle models
should be introduced to the market at what point in time. On this strategic planning
level, a vehicle model is typically characterized by its size class and body style (Adelt
2003) and in future also by its power-train (Frick et al. 2011). Automobile manufacturers
thus have to decide about offering vehicle models with a specific size class, a specific
body style, and a specific power-train at a specific point in time (Kieckhafer et al. 2012).
In the automotive industry, strategic product portfolio decisions are mainly based on a
so called cycle plan or long range product plan (Adelt 2003, Hill et al. 2007). Here, mile-
stones for all vehicles of the current and future product portfolios are depicted. For
product portfolio planning, different cycle plans are developed and evaluated based on
their financial performance as well as on other aspects such as innovativeness and legal
compliance.
The development of the financial performance strongly depends on the future market
development of the different vehicle models included in the cycle plan. Forecasting
these developments is very challenging: First, the expected total market volume has to
be approximated. Afterwards, it has to be estimated which shares of this market volume
are attained by the vehicle models. By taking into account competition, sales volumes
for every vehicle model of a manufacturer in a market can be deduced. These sales vol-
umes are then used to calculate the financial performance of the vehicle models as well
as their technical and ecological consequences and thus to evaluate different cycle
plans (Hill et al. 2007).
In practice, the development of the market shares of different vehicle models is often
estimated using scenario technique (as done e. g. in ifmo 2002, Shell 2009). For this,
important influencing factors that determine the business environment are identified.
Based on qualitative statements from expert elicitations, the future development of
these factors and their interrelations are estimated (Mietzner and Reger 2005). These
estimates are then used to construct possible scenarios projecting the future develop-
ment of the automotive market. For each of the constructed scenarios, e. g. the opti-
2
mistic, the pessimistic, and the trend scenario, assumptions about the influence of the-
se developments on the market shares are made. Based on these assumptions, the mar-
ket shares of the different vehicle models are estimated.
Within this procedure some shortcomings can be identified. On the one hand, cause-
effect relationships and feedback loops are only regarded to a very small extent. This
holds true for the interdependencies between the different influencing factors as well
as the interdependencies between the influencing factors and the market shares of the
vehicle models. On the other hand, assumptions and mental models of the experts that
lead to the statements about the developments of the influencing factors and the mar-
ket shares are only implicitly included, but are not made explicit.
Against this background, a cooperative project between industry and university has
been initiated that aims at developing new methods to improve the scenario process
and thus to support strategic product portfolio planning. We here report how system
dynamics and agent-based modeling can help to support strategic product portfolio
planning for the transition to alternatively powered vehicles. Using these approaches
can be beneficial for decision making in industry in multiple ways. On the one hand,
the approaches allow for modeling the automotive market in detail and for taking into
account the important interrelationships therein. On the other hand, expert knowledge
and data from various departments (market research, product management, research
and development ...) can be incorporated into the model and can thus be made explic-
it. This way, developments of market shares of different vehicle models can be simulat-
ed and well founded recommendations for strategic product portfolio planning can be
derived. Confidence in the model outcomes and the resulting recommendations is en-
hanced by the broad institutional knowledge and data base. Beyond purely estimating
how the market shares could develop, also insights into why they develop this way can
be achieved. Thus, industrial decision makers can gain a deeper understanding of the
behavior of the automotive market.
The remainder of the paper is structured as follows. In Section 2 we derive the re-
quirements on a model based decision support for strategic product portfolio planning
and present a market simulation model that integrates system dynamics and agent-
based modeling to fulfill these requirements. In Section 3 we then discuss the potential
of this simulation model, the current limitations for its practical application and im-
plications for future work. Concluding remarks are given in Section 4.
2 Model-based decision support in theory and practice
2.1 Practical requirements
To follow the idea of supporting strategic product portfolio planning in the automo-
tive industry by means of a simulation model, different requirements have to be re-
garded. These requirements can be deduced from the planning task “strategic product
portfolio planning” and the procedure “scenario analysis”, which is currently often
used to support this planning task. The planning task itself requires simulating the
development of the market shares of different vehicle models subject to the vehicle
portfolio offered to the market. This means, the model has to allow for specifying vehi-
cle models as combinations of size class, body style, and power-train as well as their
time of introduction and concrete characteristics that influence the purchase decision
(e. g. price, fuel consumption, and brand) for different markets.
To estimate the development of the vehicle models’ market shares, various factors have
to be considered in the simulation model, whose development and influence on the
market development are analyzed in the scenario process. These factors stem from the
five fields society, technology, economy, ecology, and politics. Within the field society,
factors such as demography, urbanization, and changes in values are relevant. On a
more detailed level, consumer characteristics like preferences, technology awareness,
environmental awareness, income, mileage, holding period, life stage, age, or previous
experiences have to be taken into account. Technological factors include the develop-
ments of the cost parameters (e. g. costs of the engine, costs of the battery, and costs of
ownership) and performance parameters (e.g. fuel or energy consumption, cruising
range, engine power and characteristics) of the conventional and alternative power-
train technologies. With regard to the economy, factors such as the GDP, the availabil-
ity of resources, infrastructure (especially service stations for refueling or recharging)
or fuel prices are of interest. Ecological factors are used to describe the environmental
situation of the market in question thereby accounting for environmental problems
such as pollution, scarcity of resources, CO2 emissions etc. These factors often go to-
gether with political factors which include the regulations to diminish environmental
problems such as the CO: emission regulations set by the European Union (European
Union 2010).
Additional practical requirements exist on how the different factors are incorporated
into the simulation model. Factors from the economic, ecological, and political field
define the business environment. These factors can only be influenced by an OEM to a
very small extent, if at all. However, they have a significant influence on supply of and
demand for vehicles. Additionally, the development of these factors is highly uncertain.
This is why they should only be depicted as exogenous parameters on a high aggrega-
tion level. In contrast to this, the development of the power-train technologies is
strongly dependent on the actions of an OEM and the market share developments. It
thus has to be modeled endogenously thereby considering possible innovations and
leaps in technology. Also societal factors should be depicted in detail based on available
data to model consumer behavior as accurately as possible. The market share develop-
ment to be estimated is directly dependent on the consumer response to the vehicles
offered to the market. Thus, modeling realistic consumer behavior is necessary to en-
sure that industrial decision makers confide in the model results. Especially, the pur-
chase decisions have to be explained by means of a highly sophisticated purchase deci-
4
sion model which is accepted by an OEM. Usually, such a model should explain the
consumer choices taking into account the vehicle and the consumer characteristics.
Furthermore, the consumers should be divided into different consumer segments (e. g.
based on their life stage) to account for heterogeneous consumer behavior. At best, the
consumer characteristics and their resulting behavior should be considered on an in-
dividual level.
2.2 Literature review
With regard to the automotive market different system dynamics models exist that al-
low for simulating the development of the market shares of alternatively powered vehi-
cles. These models have been developed to analyze the market diffusion of alternatively
powered vehicles (e. g. Janssen et al. 2006; Meyer and Winebrake 2009; Sheperd et al.
2012; Struben and Sterman 2008; Walther et al. 2010; Wansart et al. 2008). To this end,
the main feedback loops between the development of power-train technologies, service
stations, consumer awareness, and consumer choice are modeled (cf. Figure 1). Often,
also regulatory measures are incorporated into the model to analyze their impact on
emission reductions in the transportation sector (e. g. BenDor and Ford 2006;
Bosshardt et al. 2008; Meyer 2009; Walther et al. 2010; Wansart et al. 2008).
Py
i Stock of vehicles
Sales of vehicles with powerstain
ith power-train i
- +
Costs of if
er-train i
Ease Consumer +
¥,
i . Consumer choice of awareness for
Performance o: vehicle with power-train i
power-train i power-train i
+ Service stations for
vehicles with
power-train i
+
Price of vehicle
with power-train i
Attractiveness of +
» vehicle with
+ power-train i
Figure 1: Main feedback loops considered in system dynamics models to simulate the development of the
market shares of alternatively powered vehicles
Overall, the system dynamics models meet a multitude of the practical requirements.
However, the models have two main shortcomings. (1) Homogeneous consumer behav-
ior is assumed, which is clearly in contrast to actual consumer behavior in the automo-
tive market (e. g. Achtnicht et al. 2008). A common approach is using a discrete choice
model (Train 2009) to model the purchase decision of an average consumer taking into
account the vehicle and the consumer characteristics. The choice model is combined
with adaptations of the Bass innovation-diffusion model (Bass 1969) to take into ac-
count consumers’ awareness of alternatively powered vehicles and their willingness to
consider them within a purchase decision. (2) The models usually do not depict strate-
gic product portfolio decisions. The vehicle models are only distinguished by their
power-trains and all vehicle models are offered to the market right from the beginning.
Obviously, these assumptions are not in line with reality. Only the simulation model
presented by Walther et al. (2010) and Wansart et al. (2008) allows for considering vehi-
cle models as combinations of power-trains and size classes as well as their time of in-
troduction.
In addition, agent-based models have been developed that allow for simulating the
market shares of alternatively powered vehicles and analyzing their market diffusion.
In contrast to the system dynamics models, the focus of these models lies on the indi-
vidual consumer behavior. Individual purchase decisions are considered here. This is
usually done by integrating discrete choice or conjoint-based choice models into the
agent-based simulation models (e. g. Giinther et al. 2011; Mueller and De Haan 2009;
Zhang et al. 2011). Furthermore, consumer awareness is modeled on an individual level,
either based on a stationary Markov process (Mueller and De Haan 2009) or on con-
sumer interaction on the micro level (Giinther et al. 2011; Zhang et al. 2011). Another
advantage of the agent-based simulation models lies in the consideration of highly dif-
ferentiated vehicle portfolios. However, important developments in the business envi-
ronment and the development of the power-train technologies as well as their interre-
lation with consumer behavior are largely neglected.
Overall, neither the system dynamics models nor the agent-based models that have
been developed so far meet all practical requirements. Thus, they cannot directly be
applied to support strategic product portfolio planning in the automotive industry.
This is why we developed a novel hybrid simulation model we will briefly present in
the following.
2.3 Modeling approach
To best meet the practical requirements, we developed a simulation model that inte-
grates a system dynamics model with an agent-based discrete choice model. The model
was developed at the university in close cooperation with the support of a large Ger-
man OEM. It allows for simulating the development of the market shares of alterna-
tively powered vehicles subject to the vehicle portfolio offered to the market. The mod-
el is implemented in the software AnyLogic from XJ Technologies, which allows for the
integration of the system dynamics and the agent-based model (Borshchev and
Filippov 2004). To support strategic product portfolio planning, the vehicle portfolio
offered to the market, the consumer behavior in the market, and scenarios of uncertain
developments in the business environment can be defined exogenously. Vehicle mod-
els to be considered in the portfolio can be specified by means of their power-train,
size class and time of introduction. Body styles are not subject of the investigation in
the first step.
The basis of the hybrid simulation approach is a system dynamics model. It draws on
experiences of modeling the automotive market with system dynamics gained within
longstanding research projects (Meyer 2009; Walther et al. 2008; 2010; Wansart 2008).
The model allows for considering the main feedback loops between consumer choice,
consumer awareness, development of power-train technologies, and service station
availability (cf Figure 1). To take into account heterogeneous consumer behavior, the
agent-based discrete choice model from Mueller and de Haan (2009) is adapted and
integrated into the system dynamics model (cf. Figure 2). This way, the model parts of
consumer choice and consumer awareness are refined.
System dynamics model
y ¥ Agent-based model
Purchase decision Market shares
[power-train, size class]
I
[ [ Vehicle stock
Vehicle model [power-train]
= Power-tra
= Size class
Consumer segment
in = Consumer choice |
= Characteristics
Endogenous
= Characteristics Service stations Technology
[power-train] [power-train]
| |
x 4. IX
A ini Wh
Vv
Vv
Vehicle portfolio
= Vehicle models offered
(power-train, size class)
ha) | = Time of introduction
(ogenous
Consumer behavior
= Region
= Consumer segments
= Purchase decision rule
Business environment
= Energy prices
= Regulatory measures
Figure 2: Concept of the integrated system dynamics and agent-based model to support strategic product
portfolio planning in the automotive industry (Kieckhafer et al. 2012)
The integration of the system dynamics and the agent-based model is based on the
vehicle portfolio offered to the market as well as on sales and stock figures resulting
from the purchase decisions of the agents. The characteristics of the vehicle models
that can be purchased by the agents in the agent-based simulation model are influ-
enced by the simulated developments in the system dynamics model. In turn, the de-
velopments in the system dynamics model are influenced by the sales and stock figures
that are calculated in the agent-based simulation model as a result of the simulated
purchase decisions of the agents. Roughly, from a process-oriented perspective the
simulation model is constructed as follows.
(1) Initialization
At the start of the simulation, the vehicle portfolio offered to the market as well as the
agents are initialized. To initialize the product portfolio, vehicle models are added to
the simulation model as objects. As already mentioned, these models differ in terms of
power-trains and size classes. In addition, vehicle characteristics that influence the
purchase decision (e. g. purchase price, cruising range) are specified. To initialize the
consumer agents, they are divided demographically into different consumer segments.
Each agent represents a certain number of consumers in a specific segment of the re-
garded market. The agents are characterized by means of socio-economic and socio-
demographic attributes (e. g. age, environmental awareness, kilometers travelled). Fur-
thermore, decision rules for carrying out the purchase decision are predefined. At the
end of the initialization step, one specific vehicle model is randomly assigned to every
agent. This is done by taking into account the actual power-train and size class split of
the current vehicle fleet that is operated in the regarded market. It allows for consider-
ing a realistic composition of the vehicle stock at the start of the simulation.
(2) Simulation of the purchase decisions in the agent-based simulation model
After the initialization step, the agents turn into new car buyers. To this end, the time
of purchase is determined first. This is done by defining the holding period of a vehi-
cle for each agent. The holding period is modeled as an exponentially distributed ran-
dom variable. Thus, every agent replaces the current vehicle at a certain point in time
and passes a multistage purchase decision process following Mueller and De Haan
(2009).
In the first step of this process, an agent specific choice set is determined. To this end,
the choice set size and composition are specified. The choice set size is drawn random-
ly. To compose the choice set of an agent, vehicle models from the product portfolio
offered on the market are selected. The selection process is modeled as a stationary
Markov process, which is based on transition matrices for the attributes size class and
power-train. These matrices describe the likelihoods that an agent includes a vehicle
model with a specific size class and power-train into the choice set given the power-
train and size class of the current vehicle model.
In the second step of the purchase decision process, the actual purchase decision is
modeled. To this end, discrete choice theory (Train 2009) is utilized to describe the
selection of one specific vehicle model from the choice set by the agent. First, the utili-
ty for every vehicle model in the choice set is computed. This utility is a function that
is deterministic and linear in the parameters. It is dependent on the vehicle character-
istics and the characteristics of the consumer/agent. Afterwards, the purchase probabil-
ities for the vehicle models in the choice set are determined. Here, a nested logit model
is used (Achtnicht et al. 2008). The nests are built with regard to the different power-
trains. Based on the estimated purchase probabilities, the purchase decision of the
agent is simulated using random wheel selection.
The purchase decisions of the agents lead to a recalculation of the vehicle sales and
stock figures in the agent-based simulation model. These figures are transferred to the
system dynamics model.
(3) Simulation of the developments in the system dynamics model
The system dynamics model serves to simulate the values of the vehicle characteristics
that have an influence on consumer choice. To this end, the development of the num-
ber of service stations and the development of the cost and performance parameters of
the power-trains as well as the corresponding development of the vehicle characteris-
tics are modeled endogenously (Walther et al. 2010). These developments are influ-
enced by the purchase decisions of the agents. For instance, the more agents purchase a
vehicle with a specific power-train, the higher is the demand for a certain energy carri-
er and the more service stations are required (Struben and Sterman 2008). To account
for these dependencies, the information about the sales and stock figures from the
agent-based simulation is used.
In the following, we will exemplarily illustrate this procedure by focusing on the char-
acteristics purchase price and cruising range of a battery electric vehicle. Their devel-
opment is modeled depending on the development of the energy density and the pro-
duction costs of the traction battery. The energy density (EnergyDensity) follows a lo-
gistic growth curve until a technical maximum (Max_EnergyDensity) is reached
(Wissema 1982). The development of the production costs per kilowatt hour battery
capacity (KWhUnitCost) is modeled by means of a standard experience curve (Hender-
son 1984). Both variables are dependent on the cumulated experience, which is approx-
imated by the battery capacity installed in the sold electric vehicles. The battery capaci-
ty installed is computed in the agent-based model and transferred to the system dy-
namics model. Its value is directly written in the stock KWhProduced (cf. Figure 3).
This way, the integration of the agent-based simulation model with the system dynam-
ics model becomes possible.
QP _Init_kwhProduced QBinit kwnunitCost Elasticity KWhExperience
Directly influenced by the
agents‘ purchase decisions
SURE iene t= see \ Influences purchase price
$$ ¥
KWhUnitCost
Max EnergyDensity @& @® Init EnergyDensity
Ww sii a Influences purchase price
———— and cruising range '
EnergyDensityimprovement EnergyDensity
G Fraction EnergyDensityimprovement
Figure 3: Computation of the production costs and the energy density of the traction battery in the system
dynamics model
To compute the purchase price of the battery electric vehicle (Price) the costs of the
traction battery (CostBattery) are calculated first. These costs are modeled in depend-
ence on the costs per kilowatt hour battery capacity (KWhUnitCost), the energy density
(KWhBattery) and the package weight of the traction battery (Package Weight) (cf. Fig-
ure 4). The cruising range of the battery electric vehicle (Range) is approximated in de-
pendence on the energy density (KWhABattery) and the electricity consumption
(ElectricityConsumption).
Price Calculation PackageWeight
<EnginePower> EnergyDensity Que @
= Lr -
O56 fran PomeeOrie ay, influence of energy density
KWhUnitCost ‘a
KWUnitCost O kwhBattery on purchase price
Q i a Influence of production
—.¢ OComponentcot producti
PriceBasie Cost Drive eet coston purchase price
@
Energy Cost Calculation Range Calculation
ElectricityConsumption
ElacticityPrice @
o. a
<KwhBattery>
Influence of energy density
on cruising range
EnergyCost —@
Figure 4: Computation of purchase price, energy cost, and cruising range of a battery electric vehicle in the
system dynamics model
10
(4) Update of the characteristics of the vehicle portfolio
The values of the vehicle characteristics that are simulated in the system dynamics
model are used to cyclically update the characteristics of the vehicle models the agents
can choose from in the agent-based simulation model. This way it is regarded that
OEMs do not adjust their offering continuously. The cycle time for updating the vehi-
cle characteristics (e. g. one year) has to be defined before the start of the simulation
run. Additionally, new vehicles can be introduced to the market at these points in time.
The adjustment is based on predefined information about the introduction of new ve-
hicle models (e. g. time of introduction, size class, power-train).
(5) Repeat
After the adjustment procedure, the agents can purchase vehicle models from the new
product portfolio (cf: Figure 5). The steps (2) to (4) are repeated until the simulation run
is terminated. Overall, this procedure allows for integrating the system dynamics mod-
el with the agent-based simulation model.
Purchase decision process | Current
v H i 1. Choice set size vehicle
Vehicle portfolio disposal 2. Choice set composition GF
* Distinguished by power-train, size class fi 3. Vehicle evaluation
4. Purchase decision
* Characterized by purchase price, cruising range, etc. t
_|
al
3B A @\ a
1 Utility vehicle = Parameter * Price +...
1
¥
Purchase probability based on
discrete choice model
i -L_. " 7
; Potential size Power-train technologies New car
1 Classes = Technology development buyers » | + FP --» Vehicle operation - -
z
= Production costs, etc.
5 eee Cy
WE“)
Service stations
. —
is a) = =~ Outside of system boundaries
Figure 5: Integration of the agent-based simulation model and the system dynamics model from a process-
oriented perspective (Kieckhéfer et al. 2012)
Ld
To enable the evaluation of simulation experiments and thus to support strategic
product portfolio planning, the model can be executed in two different ways. For a
“quick” evaluation of single simulation runs a user interface is provided (cf: Figure 6).
Here, the simulation runs can be customized by the user (e. g. definition of the times of
introduction of the vehicle models) and the resulting developments of the shares of
sales of the vehicle models and the power-trains are directly presented. However, the
information content of these results is limited due to the high aggregation level and
the stochastic nature of the agent-based simulation model. This is why also, for in-
11
stance, Monte Carlo experiments can be executed to allow for comprehensive evalua-
tions (e. g. development of the market shares in the consumer segments) of various
replications of a simulation run.
Customize simulation run Simulated annual vehicle sales in percent
Diese Fall ybrid Plug-in Hybrid Aateery Electr
runece +————
Coniston! tied athe tte ce
Figure 6: User interface to customize single simulation runs and quickly evaluate the development of the
shares of sales of the vehicle models and the power-trains
3 Potentials and limitations for practical use
3.1 Potentials
The simulation approach shows various potentials for practical use. Obviously, it ful-
fills many of the stated requirements and might thus be helpful for its original purpose
to support strategic product portfolio planning. The development of market shares of
various power-trains in different vehicle size classes could be simulated subject to the
vehicle portfolio offered to the market. Individual and heterogeneous consumer behav-
ior, technology development as well as aggregated developments in the business envi-
ronment can be accounted for. Furthermore, the model is founded on widely used sci-
entific theories and incorporates expert knowledge from industry. Based on data for
the German market that is partly provided by industry and partly taken from publicly
available sources, the model shows a quite reasonable behavior within validation tests.
On the one hand, the market shares of conventionally powered vehicles in the last few
years can be reproduced (Kieckhfer et al. 2012). On the other hand, model behavior,
specifically the development of the market shares of alternatively powered vehicles,
stands in accordance with diffusion theory. These results are achieved without calibrat-
ing the model by using the broad empirical data base. This provides a starting point for
12
building confidence in the model, which is a crucial point for its application in prac-
tice.
An additional potential of such a model is the possibility to store expert knowledge
that is normally spread over the company. During meetings and workshops, this
knowledge and the mental models of the experts from different divisions and disci-
plines can be brought together, discussed, and used to build the model. Thus, the
modeling process itself might support the interdisciplinary and interdivisional coop-
eration in the company. Once the model is built and the expert knowledge is incorpo-
rated, the model can easily be adjusted to execute simulation runs.
Another possibility to apply such a simulation model in a company might be to use it
as a tool for gaining a deeper understanding of the behavior of the automotive market
in management and in further steps as a training tool for strategic planners. The simu-
lation model allows analyzing the main feedback loops between the vehicle portfolio
offered to the market, consumer choice, consumer awareness, development of power-
train technologies, and service station availability. This way, a broader basis for the
strategic product portfolio decisions could be ensured.
With regard to external communication the simulation model could be used in the
discussion with other stakeholders like policy, NGOs, and journalists. Especially the
application of the simulation model in the field of regulatory impact assessment seems
to be promising. Here, for instance the model can be used to analyze the impact of dif-
ferent policy and technical measures on the reduction of CO: emissions in the automo-
tive sector (e. g. Herrmann et al. 2012, Meyer 2009, Walther et al. 2008; 2010). The re-
sults can then be discussed to improve accuracy in the design of regulatory measures
and raise their effectiveness.
3.2. Limitations
Despite its potentials, the presented simulation model is not yet used in practice. Up to
now, it remains an academic solution for practical requirements and is so far only in
use at the university. To support the application in practice, the confidence in the
model and the user acceptance still has to be raised to a large extent.
The confidence in the model depends strongly on choosing the right model scope and
structures as well as on using the right data. Even though the model was developed
based on various workshops and discussions with decision makers from industry there
is still no consensus about these aspects. The reasons for this are on the one hand het-
erogeneous opinions of the stakeholders. This heterogeneity is not only due to the fact
that the stakeholders stem from different disciplines, but also due to their specific ex-
pertise and previous experience in model building. On the other hand model scope,
structures, and data cannot be treated independently of each other. For instance, it is
still an open question how to model consumer behavior adequately. As in science, a lot
of opinions exist in practice which consumer behavior model is most suitable to ex-
133
plain the purchase decisions of the consumers. The discussion on this topic ranges
from selecting the best model from various existing discrete choice and conjoint mod-
els to incorporating completely new insights from neuroscience into the model. From
this discussion the main data issues originate. At best, a widely trusted consumer be-
havior model has to be found that explains the individual purchase decisions on the
basis of company internal data. However, even if agreement on one consumer behavior
model is reached, it is not always guaranteed that internal data for this model exists.
Furthermore this data is usually confidential, so that, if at all, only aggregated internal
data can be incorporated into the model as long as it remains an academic solution.
To ensure user acceptance in industry, the applicability of the simulation model is of
great importance. This is a big challenge due to the complex model structure as well as
the variety of methods and data incorporated in the model. Expert knowledge is re-
quired to adjust the model structure, the data base, or the evaluation options of the
simulation runs. This is why such a model can currently be handled nearly exclusively
by the researchers from university. It cannot be granted that the designated users of the
simulation model also possess the necessary knowledge.
3-3. Implications for future work
The presented simulation model can be considered as an innovation with regard to
strategic product portfolio planning in industry. Future work has to ensure that the
model is adopted by industrial decision makers and diffuses into the company to in-
crease the possibility that its discussed benefits are exploited in industry. Analogously
to the Bass model, effective advertising and word-of-mouth are required for this. The
key to success is to identify innovators in the company that can act as disseminators
and ensure their confidence in the model and user acceptance. Once they are con-
vinced of the new approach, they can promote it within the company.
Thus, the question remains how to build confidence and raise user acceptance in a way
that industrial decision makers are convinced of such an innovative and complex simu-
lation model. Obviously, further comprehensive validation tests have to be carried out
to build confidence. Within these tests only industrially collected and proved data
should be used. Based on this data, the simulation model, of course, has to provide re-
liable and verifiable results that are easily traceable for industry. However, this seems
to be nearly impossible. Reliability and verifiability are hard to obtain because of mani-
fold opinions in the company about the right model and the uncertain future devel-
opment of the automotive market. The complexity of the model impedes traceability of
its results.
To nevertheless build confidence into the model, future work should focus on the fol-
lowing issues. (1) It is of great importance to avoid false expectations of the industrial
decision makers. It has to be clearly communicated, that the simulation model, just
like the scenario technique, does not strive for point predictions. (2) The model has to
14
be successfully applied to exemplary real world case studies, like in Walther et al. (2010)
for the Zero-Emission Vehicle Regulation of the state of California. (3) Sensitivity anal-
yses with regard to complete sub-modules of the simulation model seem to be very
promising, as done e. g. in Whitefoot et al. (2011). For instance, different consumer be-
havior models could be incorporated in the simulation model to demonstrate the im-
pact of these models and compare the simulation results. Based on this comparison
one or maybe more widely accepted consumer behavior model(s) could be agreed on
and used in the simulation model to actually support strategic product portfolio plan-
ning.
As stated, a further prerequisite for the user acceptance is the usability of the simula-
tion model. To make the model applicable for non-experts, an interactive and hands-
on user-interface has to be provided. Here, the design of a management flight simula-
tor that allows for a user-friendly manipulation of selected model parameters as well as
the evaluation and visualization of selected results seems to be promising. At best, the
flight simulator also allows for easily changing complete sub-modules (as discussed
above) that are predefined by the experts. This way not only the execution of the simu-
lation model would be simplified, but also the adjustment of the model structure.
In any case, educating students in university to become experts in modeling and ana-
lyzing complex and dynamic socioeconomic systems and diffuse into the company
could be advantageous.
4 Conclusion
In this contribution, we report how agent-based modeling and system dynamics can be
used to support strategic product portfolio planning in the automotive industry. These
approaches provide the opportunity to estimate the development of the market shares
of different vehicle models which are one important input for the evaluation of differ-
ent product portfolios. Especially the integration of system dynamics and agent-based
modeling enables meeting various practical requirements. We show this potential by
presenting a simulation model that was developed in cooperation between university
and industry. The simulation model integrates a system dynamics model to consider
aggregated system behavior with an agent-based discrete choice model to consider in-
dividual consumer behavior. Herewith it is possible to support strategic product port-
folio planning by simulating the market share developments of alternatively powered
vehicles taking into account different vehicle portfolios. Thereby various factors from
business environment, technology, and consumer behavior, whose developments and
influences on the market environment are analyzed in the scenario process, as well as
their interdependencies can be taken into account.
The presented simulation approach shows various potentials for practical use, such as
supporting strategic product portfolio planning, enhancing the understanding of the
automotive market, and storing knowledge of experts from various divisions. However,
15
some limitations for its practical application have been identified. Here, the main tasks
lie in building confidence in the model and raising user acceptance, which thus has to
be trageted by future work. Further comprehensive validation tests and sensitivity anal-
yses can be used for building confidence. To enhance the applicability of the simula-
tion model and thus raise user acceptance, a hands-on user interface e. g. in form of a
management flight simulator should be developed and implemented.
Acknowledgements
This research project is partially funded by the German Federal Ministry of Education
and Research (BMBF). The authors would like to acknowledge the support of the BMBF
for funding the research project "STROM - Strategic options of the automobile indus-
try for the migration towards sustainable powertrains in established and emergent
markets" under the reference 01UN1006A. The authors would also like to acknowledge
the support of the Volkswagen AG for providing additional funding, data, and expert
knowledge.
Literature
Achtnicht, Martin, Georg Biihler, and Claudia Hermeling. “Impact of service station
networks on purchase decisions of alternative-fuel vehicles.” ZEW Discussion Paper
No. 08-088, Zentrum ftir Europaische Wirtschaftsforschung GmbH, Mannheim.
Adelt, Bruno. ,Uberlegungen zur Weiterentwicklung der Unternehmensplanung bei
Volkswagen“. In Neugestaltung der Unternehmensplanung: Innovative Konzepte
und ertolgreiche Praxislésungen, edited by Péter Horvath, Ronald Gleich, 451-467.
Stuttgart: Schaffer-Poeschel, 2002.
Bass, Frank M. “A new product growth for model consumer durables.” Management
Science 5 (1969): 215-227.
BenDor, Todd and Andrew Ford. “Simulating a combination of feebates and scrappage
incentives to reduce automobile emissions.” Energy 31 (2006): 1197-1214.
Bosshardt, Mathias, Silvia Ulli-Beer, Fritz Gassmann, and Alexander Wokaun. ,,The
effect of multi-incentive policies on the competition of drivetrain technologies.” In
Proceedings of the 26th International Conference of the System Dynamics Society,
edited by Brian C. Dangerfield. Athens: System Dynamics Society, 2008.
Borshchev, Andrei and Alexei Filippov. “From system dynamics and discrete event to
practical agent based modeling: reasons, techniques, tools.” In Proceedings of the
22nd International Conference of the System Dynamics Society, edited by Michael
16
Kennedy, Graham W. Winch, Robin S. Langer, Jennifer I. Rowe and Joan M Yanni.
Oxford: System Dynamics Society, 2004.
European Union. Regulation (EC) No 443/2009 of the European Parliament and of the
Council of 23 April 2009 setting emission performance standards for new passenger
cars as part of the Community's integrated approach to reduce COz emissions from
light-duty vehicles. Available: http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri
=OJ:L:2009:140:0001:0015:EN:PDF.
Frick, Lutz, Nicolai Miiller, Wolfgang Pointner, and Andreas Tschiesner. Boost! Trans-
forming the powertrain value chain - a portfolio challenge. McKinsey&Company,
2011. Available: —_ http://autoassembly.mckinsey.com/html/resources/publication/
b_Boost_Transforming_powertrain_2011-02.asp
Giinther, Markus, Christian Stummer, Lea M. Wakolbinger, and Michael Wildpaner.
“An agent-based simulation approach for the new product diffusion of a novel bio-
mass fuel.” Journal of the Operational Research Society 62 (2011): 12-20.
Henderson, Bruce D. “The application and misapplication of the experience curve”.
Journal of Business Strategy 4 (1984): 3-9.
Herrmann, Christoph, Karsten Kieckhafer, Mark Mennenga, Steven Skerlos, Thomas S.
Spengler, Julian Stehr, Vineet Raichur, and Grit Walther. “A framework to analyze
the reduction potential of life cycle carbon dioxide emissions of passenger cars.” In
Leveraging technology for a sustainable world. Proceedings of the 19th CIRP Con-
ference on Life Cycle Engineering, edited by David A. Dornfeld and Barbara S.
Linke. Berlin: Springer (2012): 55-60.
Hill, Kim, Morgan Edwards, and Steven Szakaly. How automakers plan their products:
A primer for policy makers on automotive industry business planning. Center for
Automotive Research: 2007. Available: http://www.cargroup.org/?module=
Publications&event=View&pubID=32.
ifmo. Zukunft der Mobilitat Szenarien ftir das Jahr 2030. Miinchen: Institut fiir Mobili-
tatsforschung, 2010. Available: http://www.ifmo.de/basif/pdf/publikationen/2010/
100531_Szenarien_2030.pdf.
Janssen, Arthur, Stephan F. Lienin, Fritz Gassmann, and Alexander Wokaun. “Model
aided policy development for the market penetration of natural gas vehicles in Swit-
zerland.” Transportation Research A, Vol. 40, 4 (2006): 316-333.
17
Kieckhafer, Karsten, Thomas Volling and Thomas Stefan Spengler. “Supporting strate-
gic product portfolio planning by market simulation: The case of the future power-
train portfolio in the automotive industry.” In Quantitative marketing and market-
ing management, edited by Adamantios Diamantopoulos, Wolfgang Fritz, and Lutz
Hildebrandt. Wiesbaden: Springer Gabler (2012): 123-147.
Meyer, Grischa. ,,Analyse und technisch-ékonomische Bewertung von Gesetzesfolgen
im Individualverkehr: dargestellt am Beispiel der Automobilindustrie Japans und
Deutschlands “PhD diss., Technische Universitat Braunschweig, 2009.
Meyer, Patrick E. and James J. Winebrake. “Modeling technology diffusion of comple-
mentary goods: The case of hydrogen vehicles and refueling infrastructure.”
Technovation 29 (2009): 77-91.
Mietzner, Dana and Guido Reger. “Advantages and disadvantages of scenario ap-
proaches for strategic foresight.” International Journal of Technology Intelligence
and Planning Vol. 1, 2. (2005): 220-239.
Mueller, Michael G. and Peter De Haan. “How much do incentives affect car purchase?
Agent-based microsimulation of consumer choice of new cars: Part I: Model struc-
ture, simulation of bounded rationality, and model validation.” Energy Policy 37
(2009): 1072-1082.
Shell. Shell PKW-Szenarien bis 2030. Fakten, Trends und Handlungsoptionen
fiir nachhaltige Auto-Mobilitat. Hamburg: Shell Deutschland Oil GmbH,
2009. Available: —_http://www-static.shell.com/static/deu/downloads/aboutshell/
our_strategy/mobility_scenarios/shell_mobility_scenarios.pdf
Sheperd, Simon, Peter Bonsall, and Gillian Harrison. “Factors affecting future demand
for electric vehicles: A model based study.” Transport Policy 20 (2012): 62-74.
Struben, Jeroen and John D. Sterman. “Transition challenges for alternative fuel vehi-
cle and transportation systems.” Environment and Planning B: Planning and De-
sign 35 (2008): 1070-1097.
Train, Kenneth E. Discrete choice methods with simulation, 2nd edition. Cambridge:
Cambridge University Press, 2009.
Walther, Grit, Jorg Wansart, Karsten Kieckhafer, Eckehard Schnieder, and Thomas S.
Spengler. “Impact assessment in the automotive industry: Mandatory market intro-
duction of alternative powertrain technologies.” System Dynamics Review 26 (2010):
239-261.
18
Walther, Grit, Grischa Meyer, Thomas S. Spengler, and Jorg Wansart. “Regulatory Im-
pact Assessment for the Transportation Sector — Case Study Germany.” In Proceed-
ings of the 26th International Conference of the System Dynamics Society, edited by
Brian C. Dangerfield. Athens: System Dynamics Society, 2008.
Wansart, Jorg, Grit Walther, and Thomas S. Spengler. “Limiting motor vehicles’ CO2
emissions - a manufacturer's challenge.” In Proceedings of the 26th International
Conférence of the System Dynamics Society, edited by Brian C. Dangerfield. Athens:
System Dynamics Society, 2008.
Whitefoot, Kate S., Hilary G. Grimes-Casey, Carol E. Girata, W. Ross Morrow, James J.
Winebrake, Gregory A. Keoleian, and Steven J. Skerlos. “Consequential life cycle as-
sessment with market-driven design.” Journal of Industrial Ecology 15 (2011): 726-
742.
Wissema, Johan G. “Trends in technology forecasting.” R&D Management 12 (1982): 27-
36.
Zhang, Ting, Sonja Gensler, and Rosanna Garcia. “A study of the diffusion of alterna-
tive vehicles: An agent-based modeling approach.” Journal of Product Innovation
Management 28 (2011): 152-168.
19
