1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Der Riitli Manag i Simulator - a new pt in system dy i
based management flight simulators
Pal I Davidsen Magne Myrtveit
Department of Information Science ModellData AS
University of Bergen Bergen
Norway Norway
Tel: +47 55 544134 Tel: +47 56 374009
Fax: +47 55 544108 Fax: +47 55 373500
E-mail: davidsen @ifi.uib.no E-mail: magmy @modeld.no
Abstract
A new management institute, Der Rutli Fuh it was ished in Germany this
year. Its leadership is committed to the teaching of management based on the principles of System
Dynamics. As an element in the promotion of these ideas, a new System Dynamics management
simulator has been developed. There are several ways in which this simulator constitutes a new
concept in management training and research. In particular, we have addressed the issue of teaching
policy design, supplemented by decision making.
The properties of the simulator originate from: the characteristics of the underlying software.
Consequently they can be applied gi ly to any system based In
particular, the simulator;
(1) the simul. is a multi-group si utilizing any personal computer network that supports
MS Windows;
(2) The simulator allows the user not only to submit decisions to the server that calculates and retums
a new state, but also to formulate and try out, on their local computer, their own strategies over any
strategic horizon;
(3) the simulator gives the users access to;
-the underlying simulation model (to an extent determined by the administrator);
-The tool by which the simulation model was built, allowing the users to formulate and test their own
dynamic hypotheses.
In this paper, we describe these ch istics of a new ion of simul: for
training, i by Der Rutli i and we discuss the potential implications of
taking advantage of these characteristics.
Microworlds, page 9
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Der Riitli Management Simulator - a new concept in system
dynamics based management flight simulators
ModellData AS, the D of ion Science, University of Bergen, and Stord
College of Education, Norway, have developed a number of uuiteracttve, computer-1 -based
learning environments for system and i range
from a relatively general training simulators for Der Riitli, a German management institute,
and the Norwegian Naval Academy, via p ion and market for
Norwegian enterprises in a variety of domains ranging from the aluminum industry, via fish
farming, and on to tanker, helicop and airplane si for pilot training. These
ications have been ped using the iated MS Windows™-based modeling and
simulation package Powersim™.
Powersim™ by itself is a general purpose learning i for the i igation of
complex, dynamic systems. In this paper we will focus on the general principles of its
application to specific domains and, for now, we reserve the concept "simulator" for that
purpose. The system dynamics approach implied a series of requirements that are generally
not satisfies by current, ‘based learning often called "
flight simulators" (Morecroft et al. 1992). In this paper we outline some of these
requirements and document how we use Powersim™ in the case of Der Riitli Management
Simulator to meet these requirements.
The system dynamics approach supports the management of dynamic systems at all these
levels by helping us und d the intimate i ip between and behavior in
such systems. The higher levels of administration on which we operate, the wider is the time
horizon, the more complex are the systems in question, and the larger is the need for an
approach of this kind.
Although system dynamics based learning environments can be used to learn domain-specific
decision procedures, the purpose of developing and using such environments is more general.
When managing dynamic systems, the impl of di should be
thought of as a result of a strategy formulation (Forrester 1980, 1986). A strategy is
d of a well-bal: d set of domain-specific policies, such as production, inventory,
sales, p 1, and capital policies. Mi , there is the tactical
selection and adaptation of policies leading up to operative decision procedures. The
challenge we face is to design learning environments that allow us to design and test policies
before they are implemented. In this paper, we describe how this challenge has been met in
the case of Der Riitli Management Simulator.
Before we focus on the novelties of the concept " based learning
supported by Powersim™, we will briefly describe the model underlying this simulator and
the seminar which constitutes the context in which the simulator is being utilized.
The model
Sectors
The model consists of four parallel production-supply-lines satisfying orders that originate in
a market and is adjusted in accordance with the level of saturation in the backlog- and
supply-lines. There is one supplier of raw material, four producers, four groups of retailers,
Microworlds, page 10
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
each associated with a producer, serving two market segments, a small special market
segment that prefers quality to low price, and a large mass market segment that prefers low
price to high quality Exhibit 1.
When used as a management training simulator, the trainees act as one of the four producers.
An example of the kind of product sold is mid-sized cars used in the mass market as the
consumers’ only car and in the special market as the.consumers ' second car.
Special Market
Orders
4 Producers 4 Retailers
Lr)
Supplier
>
Shipments Mass Market
Exhibit 1. Inventories and supply-lines of raw material and finished goods.
The volume produced and shipped downstream is determined by the order flow upstream,
restricted by production capacity, and subject to inventory and backlog adjustments. Capacity
(capital and work force) is adjusted with a delay to satisfy demand. There is a full financial
sector. The effective value of the firm (growth and profit taken into consideration)
determines the availability of loans and the unit gain from issuing shares. The liquidity feeds
back to impact p marketing, quality upgrade programs,
and payments.
The product is ch ized by its ionality, life exp , image (added value through
packaging, advertising etc. that position the product in the market), and price. To increase
quality, capital may need to be retrofitted and the work force may require training. This
quality upgrade program lly results in di img, and supply of better
products. The goal of this program is determined by the target quality of the product to be
produced. There is a delay, dererrained by the investment in the program. The effect of those
is Therefore, to speed up the program, implies
exponentially increasing posta To adjust more quickly to the goal, capital can be depreciated
at a higher rate than normal. There is co-flow of product quality originating from production
that eventually causes the installed base in the market to attain the produced quality.
The market responds to its ion of posted, effective prices of the product. It is
ized by a constant elasticity of demand and a constant elasticity of substitution. The
effective price is a measure of the market value of the product, relative to its posted price.
Microworlds, page 11
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
The perception of prices and product characteristics takes time. The perceived functionality
translates the perceived: price into a perceived net value, onto which we may add a value of
perceived image. When the perceived life expectancy is taken into consideration, this
transforms into a perceived value-rate, i.e. value per time unit, which constitutes the basis for
product comparison and substitution.
Subject to di ies in the elasticity of demand to ionality, life (interest
rate), and image, the two market segments may respond differently to changes in price and
modifications in product quality.
The market can be affected by delivery delays as well as advertising. The sensitivity of the
market segments to advertising may vary. A no ig, the ion of
lity and life drifts towards the average real characteristics of the
products that constitute the installed base and the perceived image erodes. Advertising may
change these perceptions and the speed with which they are obtained. The effect of
in ig is i g. Therefore, to speed up the effect of
k implies exp costs. , there is a measure of credi-
bility, depending on the advertised product cake as pared to their real ch istics
If a product is advertised as better than it actually is, then the credibility of the company falls.
In the long run, this reduces the effectiveness of investments in advertising. To restore
credibility, a company must advertise the true product characteristics for a relatively long
period of time.
The model typically generates business cycles and provides the producers with limited
control over the market, partly due to significant delays and partly due to the intermediate
layer of retailers, represented by the model, that adjust backlog, inventories, and prices
simultaneously in competition with each other.
Management; policies to be designed
There are altogether 6 policies to be designed -- ch ized by 26 p The policies
focus on relative volume, quality, and rate of change of various processes. Associated with
the panels applied to set these parameters is information of immediate relevance to the policy
(Exhibit 2).
For the product, target attributes are set by the producers. So is the time spent on upgrading
capital and work force to produce that quality. The trainees obtain an immediate estimate of
the associated current cost per year and can adjust the policy accordingly. The time to
depreciate capital can also be set by the prodi e.g. to sub quality capital with
new capital expediently.
The volume of production is determined implicitly by the producers through their inventory
policies. The producer determines the target inventory coverage of raw and finished goods.
The market is influenced by the producer, indirectly through pricing and directly through
advertising. To d the diffe between operative and strategic pricing, we offer
two ways to set the price (Exhibit 2). The price can be set unconditionally (upper panel).
Current costs (unit raw material and production costs) are indicated as a platform for
decisions. So is the is the average competitors price. However, these circumstances may
change over a period of time and it is far more convenient and effective for the trainee to
develop a general price policy.
Microworlds, page 12
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
The price policy available is determined as a weighed average between a profit-oriented
policy and a competition-oriented policy (lower panel). On the one hand, the desired and the
minimal profit fractions are set. On the other hand, the price, relative to the weighed average
of competing prices is set. Then there is a possibility to negotiate between the two policies by
assigning a relative weight to the competitive policy.
As for advertising, target levels for the percei ionality, life exp and image,
and the time taken to establish this perception is set. Again, the consequent current costs are
indicated immediately to aliow for policy adjustments.
[ses|f ssa]
Exhibit 2: Operative and strategic pricing. Note the information supplied to support the
identification. of a policy.
Financially, the producer must determine the fraction of profit to be paid as dividends and
must identify the borrowing fraction of any financing. Alternatively it must be determined
what fraction to finance by issuing shares.
A typical scenario to be evaluated would be this: Suppose you plan to cut prices relative to
the competitors, in order to increase demand and production volume. The would
be, in view of the expected reaction from retailers and producing competitors, to identify
effective policies for inventory, quality, marketing, and finances that work in harmony with
that price cut. Typically, a price cut will increase demand and, unless inventory coverage is
cut, a significant oscillation that amplifies along the order- and supply-lines will result.
Moreover, the liquidity will deteriorate, partly due to reduced income and partly due to
as well as i and ing costs. An
policy will aggravate the situation even if it grossly exaggerates product quality, because
credibility is lost only gradually as the installed base increases. Quality modifications
complicate the picture.
Microworlds, page 13
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Clearly, such a price-move requires careful planning. Inventories should be built up in
~ advance under sound financial circumstances, and the volume and timing of marketing
initiatives should be weli-adapted:
The seminar
Der Riitli M: imul, is i complex. The model contains about 2.500
elements, in part distributed along 4 dimensions. For this purpose, the Powersim™ array
capabilities are being crane) utilized. The model can eaaily be adopted to a larger
number of products, supply-lines (p and retailer-gi product ch: istics, and
market segments (refer to the description of the model below). Hoes are 137 parameters and
graph i i for the ion of the si to various contexts. In view of
this complexity, the key to success is a five days seminar associated with the simulator. The
seminar takes this form:
The two first two days, there is an introduction to system dynamics utilizing a series of repli-
ca of the submodels of the simulator. These models are open for inspection and we encourage
the participants to familiarize Ives with these sub- and their resulting
behavior. The panels used to run these replica, correspond to those of the simulator. In
addition, there are panels for setting and ipulating the ions under which
the replica aré being mun -- conditions otherwise provided by the simulator itself. These
submodels are;
1. Retailer and supplier
2. Producer
3. Order-line and deliveries
4, Supply-line and quality diffusion
5. The management of the quality in capital and work force
6. The price mechanisms
7. The market's perception of effective price
8. Marketing
9. Financing
Each submodel session includes an introduction to the practical and methodological issues
addressed: by the model, assignments to be solved using the models, free experimentation,
and a summary of the lessons learned utilizing system to und d the underlyi
model structure and design matching strategies.
Having completed these sessions, the participants engage in a session on Der Riitli
Management Simulator itself, run in a single user mode, i.e. against their own computer. This
is to familiarize the trainees with the complete interface and the complexity involved when
told to combine effectively a number of different policies.
The following two days, the simulator is run in a network whereby the participants can
compete against each other (exhibit 3). We can set up four teams of competitors or simulate
specific ies onan idle The participants can enter a session at any
point of time allowing us, for instance, to study the building of barriers against entry.
Microworlds, page 14
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Exhibit 3: Two human competitors, a controlled strategy run on an idle computer, and one
computer ready to allow a fourth competitor to enter.
The simulator
Transparency and modifiability; representing the integration of structure and
behavior.
The learning-environment must effectively support strategic investigation, formulation, and
evaluation. This implies that the model, dded in the learning. must be
available for inspection and modification seed Cc , the learning.
environment must offer a smooth transition from tools used to model and set up simulations
in general to the kind of " flight simul: i ly where the model
is often hidden and where there is minimal support for anaylysis.
Users investigate the underlying simulation model of the learning environment by inspecting
the structural assumptions embodied in that model and by utilizing a test mode to reveal their
behavioral consequences. Using PowerSim™, we can open the model for inspection. We do
so by iconizing views of the model to be opened by the user, the result of which is potrayed
in Exhibit 4.
These model views can represent the model in the form of a stock-and-flow diagram
(Davidsen 1994, Exhibit 2a) where the user is allowed to inspect the underlying equations.
Also, the model can be represented in the form of a feedback loop diagram (Davidsen 1994,
Exhibit 2b-c) and in the form of texts. Moreover, there are icons that can be opened to
specify input or define plicies using slid-bars (Exhibit 2) or opened to study the resulting mo-
el behavior.
To facilitate investigation. we integrate into the ion of the simulati
model, whether in the form of stock-and-flow diagrams or feed-back-loop di (“live"
graphs and/or numbers represenung the model behavior. This is illustrated in exhibit 5, where
we dynamically map the behavior of the supply-line onto a stock-and-flow diagram.
Microworlds, page 15
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
PowerSim
e
Advertising Advertising Price and
Functionality Image (High -- Market Share
{High Market) Market)
* © @ 8
Advetising Advestising Advertising Capital
Lifetime (Low Functionality Image (Low
Market) {Low Markel) Market]
! ? ' & gqQ
Product Shates and Inventory Labor
Functionally Dividends Coverage Maskel Vaio
and Lifetime (Veais}
cu Ww ‘ Supply Line
§ Cash and Equity and Stocks and
“erses Liquidity Liguidation Flows
Exhibit 4: Der Riitli Management Simulator.
We can also embed such miniature graphs directly in texts (Davidsen 1994). Thus, using the
graphical editing tools of P im™, we let the I d ion in the learning
environment “come alive" during simulation. In that way, behavior is linked closely to the
underlying structure and becomes self-explanatory,
Strategic formulations typically require strutural‘re-formulations of the model. To facilitate
that, we can open windows on the underlying model that focus on policies to be designed by
participants as parts of a new strategy (Davidsen 1994, Exhibit 6).
Production Flow
glioma? Backlog: inversions Shipments
Ship
_
PRawinventory 5 G ‘ Z
PProduction T Shipments 4
om aor
Exhibit 5: A structure/behavior graph.
Most importantly, Powersim™* is all the time available during the sessions. So are the replica
of sub-models used in the introductory sessions. Thus the trainees have available to them the
Microworlds, page 16
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
tool applied to create the simulator, and may experiment with tum-key models as well as with
their own, self-made models.
Facilitating the ion between op and ic modes of
Although presented as strategic, most simulation-based learning envi do not support
the design and evaluation of policies and strategies. For an environment to do so, three
requirements must be satisfied. The learner must be allowed to (1) investigate the underlying
assumptions, (2) formulate alternative policies, and (3) evaluate these policies before
deciding to continue. Moreover, this. must be done repeatedly so that the experience gained
from evaluating ofa strategy, can form the basis for re-investigations, and the formulation
and of al ies. Current flight si have not
opened for an option whereby hypothetical strategies can be made subject to simulation-
based eval the simulator reverts to current. time. Using PowerSim™, we
have built in this capability in the Riitli Management Simulator. Thus we facilitate double
loop learning (Senge et. al. 1991): d in the form in Exhibit 6.
1 2
Strategy
Formulation
& Testing
Decision
6 Making
5 4
| Game Control — Player 1 v
Strategic
. Horizon
Operative
Horizon
Exhibit 6: The two modes of strategic policy design and operative decision making.
There are, in other words, two modes in which the simul can be run it 2)
an operative mode in which final decisions are being made and submitted to a model that
responds with a state projection over the assigned operative horizon (e.g. one year); (2) a
stategic mode in which stategies may be developed and itted along with defined
assumptions about the environment (e.g. competitors) to a model that responds with a state
projection over the assigned strategic horizon (c.g. five years).
Computer-based learning envi should support off-the-shelf computer networks. The
PowerSim™ software has been designed to do so. Consequently, the simulator may be run in
a single-user mode, i.e. against a local computer, or in a computer network against other
participants.
Microworlds, page 17
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
The procedure for running a imul designed in P im™, is summerized
in Exhibit 1, and is the same whether participants run solely against a local computer or in a
network of computers. In the latter case, identical simulations models are stored on each of
the local computers participating as well as on the server. The phases in the two modes of the
simulator are as follows:
1, Make ions about the
One of the major challenges in the design of ‘based learning envi intended
for strategic purposes, is the fact that, during strategy i there is no peti
environment available to provide input to the model. The participants are operating into the
future, and must each make their own i about the envi in which they are
operating. In the Riitli M Simulator, ions are made ing the price,
functionality, lifetime, and image of the products supplied by competitors.in a market, and
their corresponding delivery delays. A specific panel is designed for the participants to make
these assumptions
A technique has beed developed that allows the simulator to choose between "real input”
from a computer model or other actors, via the server, while in operative mode, and user-
defined assumptions while in test-mode.
2-3. Formulate and submit a strategy to the computer
Secondly, the participants must formulate their policies. That is, in Der Riidi Management
Simulator, set the 26 parameters that make up the price-, inventory-, quality-, personnel-,
capital-, advertising-, and finance-policies.
Having done so, the participants press “Try Decisions” in a control panel to submit their
ions and ies to their local
4. Let local computer compute the behavioral consequences
That local computer is now ready to project the model across the strategic horizon. The
simulation can be made to pause at a certain frequency, under specific conditions or at the
will of the participant so as to allow for the modification of assumptions or strategy.
5. Evaluate effects of strategy
Using a number of si ion reports, the partici can then evaluate the effects of their
ions about the envi and their H i
6. Revert to current time
If desired, they can modify their assumptions and/or their policies, run another test, and
through a systematic analysis of the results, decide upon a combination of policies. Or the
participants may decide to go ahead and implement their strategy. In any case must their local
model be brought back to current state at current time. To do so, the participants press
"Revert" in a control panel.
7. Formulate and submit an operative decision to the computer
Having found a policy to be applied and reverted the model, the participants are ready to
enter the operative mode. In accordance with their strategy, they then formulate the
corresponding operative decisions that apply to the current operative time horizon.
This time they press "Make Decisions” in a control panel to submit their operative decisions
to their local computer.
9. Let compute the b ioral
Microworlds, page 18
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
In case the participants runs solely against their local P that ip i the
of all decisions across the operative decision horizon. In case a network is set
up, the decisions are being transmitted to the server which awaits decisions from all
participants and projects the model based upon these inputs. In that case, all decisions are
being ded to ali local ip to form the basis for updating the state of the
simulation models stored locally.
10. Evaluate effects of decisions
Using the available simulation reports, the participants can now evaluate the effects of their
collective actions and, when desired, reenter the strategic mode.
Networks of co-models
In most cases, we analyze systems into subsystems and synthesize models from submodels
‘ing these sub . Using P im™, we define these models in the form of co-
models to be i in any sup del. Ci dels are being i d using “chains”
that indicate the transfer of variable values from one model to variables in any of its co-
models in the course of the simulation.
We specify the synchronized simulation of all co-models in a main model - just another co-
model. For each co-model, we may define individual start and stop times as well as an
integration interval synchronized with the one of the main model. In our design of learning
envir we make ive use of dels, beyond the plane analysis of systems and
synthesis of models, Here are some examples:
We usually define the control panels and all reports in one or more co-models -- the most
essential of which are often contained as part of the main model. Other co-models may be
used to emphasize various aspects of the relationship between structure and behavior
(Davidsen 1994). Note that we can define views that encompass a variety of controls and
reports, Thus we.can supply information, relevant to the formulation of a policy, in the form
of numerical or graphical reports related to the control panel for that policy. In figure 2, for
example, we indicated how information about current costs, historic competitors’ prices, and
historic price components are being used as a background for price setting.
We can also use co-models to initiate and phase out main models. a Navy project may consist
of a 5 years construction and acquisition phase that partly overlaps with a 25 years operation
phase, which ageing overlaps with a 5 year phase-out. The first five years period may be
Tepresented by one or several co-models that supply the main model with a fleet, initial
supplies, infrastructure, trained personnel etc. of a certain quality that, in part, determines the
operative ct istics of the fleet, d by the main model. The final 5 years period
_ is portrayed in a second co-model that gradually takes over and phases out all of the
components listed above.
Moreover, co-models with longer integration intervals can be used to sample variable values
from any model, and can typically be used to support or represent information systems with
its biases and delays, ¢.g. a financial sector or a user interface. A model can also supply any
of its co-models with empirical information -- one of which may compare this information
with the sampled behavior of the simulation model.
Carefully defined and individually ran del it an imp ducational
resource. As in the case of Der Riitli i they can
supplement an introduction to the model as a whole.
Microworlds, page 19
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
In the case of applying computer networks, co-models are also being used to create an
interface between the i of such a multi learining and the server
model so that she can set up the envi with a well-initialized model and i ‘igate the
policies of all participants and their dynamic Other dels are devel
to form an interface between each of the participants and their local models. These interfaces
filter out information that the participants should not share involuntarily, e.g. financial
information.
Some features of PowerSim™ that facilitates an integrated design of models and
learning environments.
The PowerSim™ software utilizes MS Windows™ facilities extensively. Consequently, we
can easily custom design and organize our system dynamics models and learning
environments using a variety of graphical tools supported by MS Windows™, A precise and
consistent use of colors contributes significantly in that respect. For educational purposes, we
can hide parts of the model by assigning the backgroud color to the symbols we want our
students to identify on their own. Alternatively, we can use a "show what" menu item to hide
any class of symbols in the model (levels, rates etc.) and in the learning environment (graphs,
slide-bars etc.) (Davidsen 1994).
In some cases, we may want to prevent anyone from inspecting certain model- or interface-
components or to modify such components, In that case, we can use a passworkd to lock the
model and the i in ion mode so that no editing can take place.
The same password can be used to gain access to the model, to disclose hidden components,
and to modify the result.
PowerSim™ allows us to define models or model components as arrays. To do so is relatively
straight-forward. Arrays are indicated by model iP ipped with double border
lines. Range variables with sub-ranges can be defined and utilized as index variables. A well-
formulated grammar and a library of special functions help modelers define complex
operations on arrays. An extensive series of consistency checks wam against and prohibit
dimensional inconsistencies.
The library of special functions, consists of a set of sub-libraries and are open for user-
supplied additions written in the form of dynamics link libraries (DLLs). Each simulation
generates a new generation of results. These results can be saved automatically, enumerated,
plotted and compared. Moreover the MS Windows™ technologies dynamic data exchange
(DDE) and object linking and embedding (OLE) significantly enhances the possibilities to
include easily retrievable references to externally stored and processed reference material
(say, historic material) including interactive video footage and the activation of Powersim™
from other programs so as to ¢.g. create dynamic reports based upon simulations running in
the background.
References
Davidsen, P.I.:" Perspectives on teaching System Dynamics. Coupling structure and behavior,
annotating simulations, and supporting Just In Time Open Learning (JITOL).", Proceedings
from the 1994 System Dynamics Conference, Stirling, 1994.
Forrester, J.W.: "System Dynamics - Future Opportunities", in Legasto, pp. 7 - 22, 1980.
Microworlds, page 20
