TEACHING MANAGEMENT AND FINANCE THROUGH

SIMULATION

Choosing the Proper Paradigm

Marco Remondino, Anna Maria Bruno and Nicola Miglietta

e-business L@B, University of Turin, Torino, Italy

Keywords: Simulation, Discrete Events, Agent Based, System Dynamics, Knowledge Transfer.

Abstract: Compared to analytical modelling, simulation has sometimes a greater expressive and computational power,

especially for a behavioural study of an activity, system, organisation and other topics from the Social

Sciences, which an analytic study cannot adequately achieve. In this paper simulation is discussed as a

support for teaching and knowledge transfer: a model can be built and used to dynamically show and

explain a particular phenomenon through direct experiments, though contributing to a “maieutical” way of

learning by the students (learning by doing). Also team work is triggered by using simulation models as a

educational tool. In particular, three simulation paradigms are described, along with their potential

applications and points of strength and weakness. Management and Finance are the focus of the work, but

the same considerations may be extended to other social disciplines and sciences. Note: Although the article

is the result of a joint research project, the paragraphs are divided among the authors as follows: paragraphs

1 and 5 are jointly written and equally divided among all the authors; paragraph 2 is by Nicola Miglietta;

paragraph 3 is by Anna Maria Bruno and Marco Remondino; paragraph 4 is by Marco Remondino.

1 INTRODUCTION

This work presents an analysis of modelling and

simulation applied to the Social Sciences, as a

supporting methodology for teaching purposes. A

model is a scaled down representation of a target

system in the real world; it wouldn't be useful to

create a one-to-one representation of the reality, so

it's very important to identify which are the main

features of the studied system and bring them in the

model (this process is called abstraction). Modelling

is applied when prototyping or experimenting with

the real system is expensive or impossible, and thus

it seems a perfect tool for researching in the social

field. In Ostrom (1988), simulation is described as a

third way to represent social models, being a

powerful alternative to other two symbol systems:

the verbal argumentation and the mathematical one.

The former, which uses natural language, is a non

computable way of modelling though a highly

descriptive one; in the latter, while everything can be

done with equations, the complexity of differential

systems rises exponentially as the complexity of

behaviour grows, so that describing complex

individual behaviour with equations often becomes

an intractable task. Simulation has some advantages

over the other two: it can easily be run on a

computer, through a program or a particular tool;

besides it has a highly descriptive power, since it is

usually built using a high level computer language,

and, with few efforts, can even represent non-linear

relationships, which are tough problems for the

mathematical approach. According to Gilbert and

Terna (1999), the logic of developing models using

computer simulation is not very different from the

logic used for the more familiar statistical models. In

either case, there is some phenomenon that the

researchers want to understand better, that is the

target, and so a model is built, through a

theoretically motivated process of abstraction. The

model can be a set of mathematical equations, a

statistical equation, such as a regression equation, or

a computer program. The behaviour of the model is

then observed, and compared with observations of

the real world; this is used as evidence in favour of

the validity of the model or its rejection. Computer

programs can be used to model either quantitative

theories or qualitative ones; simulation has been

successfully applied to many fields, and in Social

464

Remondino M., Bruno A. and Miglietta N. (2010).

TEACHING MANAGEMENT AND FINANCE THROUGH SIMULATION - Choosing the Proper Paradigm.

In Proceedings of the 2nd International Conference on Computer Supported Education, pages 464-468

 SciTePress

Sciences it allows to verify theories and create

virtual societies. Three different approaches to

simulation are analyzed: System Dynamic (SD),

Discrete Events (DE) and Agent Based simulation

(AB). Qualitative and quantitative methodologies

are then examined, and points of strength and

weakness of each methodology, when used as a

teaching support, are analyzed.

2 THREE SIMULATION

PARADIGMS

SD deals mostly with continuous processes whereas

“DE” and AB work mostly in discrete time, i.e.

jump from one event to another. Consider how

approaches correspond to abstraction. SD dealing

with aggregates is located at the highest abstraction

level. DE modelling is used at low to middle

abstraction. As for AB modelling, this technology is

being used across all abstraction levels.

2.1 System Dynamics

In (Sterman, 2000) we read that: System dynamics is

a method to enhance learning in complex systems.

Just as an airline uses flight simulators to help pilots

learn, systems dynamics is, partly, a method for

developing management flight simulators, often

computer simulation models, to help us learn about

dynamic complexity, understand the sources of

policy resistance, and design more effective policies.

Developed by Jay W. Forrester, System Dynamics is

“the study of information-feedback characteristics of

industrial activity to show how organizational

structure, amplification (in policies), and time delays

(in decisions and actions) interact to influence the

success of the enterprise” (Forrester). In SD the real-

world processes are represented in terms of stocks

(e.g. of material, knowledge, people, money), flows

between these stocks, and information that

determines the values of the flows. SD abstracts

from single events and entities and takes an

aggregate view concentrating on policies. To

approach the problem in SD style one has to

describe the system behaviour as a number of

interacting feedback loops, balancing or reinforcing

and delay structures. Mathematically, an SD model

is a system of differential equations. The model

works with aggregates, the items in that same stock

are indistinguishable, they do not have individuality.

2.2 Discrete Event Modelling

The term “discrete event modelling” applies to the

modelling approach based on the concept of entities,

resources and block charts describing entity flow

and resource sharing. This approach roots to 1960s

when Geoffrey Gordon conceived and evolved the

idea for GPSS and brought about its IBM

implementations (Gordon 1961). Entities

(transactions in GPSS) are passive objects that

represent people, parts, documents, tasks, messages,

etc. They travel through the blocks of the flowchart

where they stay in queues, are delayed, processed,

seize and release resources, split, combined, etc.

For the purpose of this investigation we would

like to underline that DE modelling may be

considered as definition of a global entity processing

algorithm, typically with stochastic elements. DE

simulation is usually applied to process modelling.

2.3 Agent Based Simulation

The main feature of this approach, when compared

to the other two examined is that it's essentially

decentralized. The modeller defines behaviour at

individual level, and the global behaviour emerges

as a result of many individuals, each following its

own rules, living together in some environment and

communicating with each other and with the

environment. That is why AB modelling is also

called bottom-up modelling. Instead of creating a

mathematical model, the underlying model is based

on a system comprised of various interacting agents.

Therefore, its structure and behaviour have potential

to resemble the actual economic theory and reality

better than simple mathematical models. While in

DE the stress is on the function of the single parts,

that are deeply modelled as resembling the reality,

and is SD the focus is on the sole aggregate

behaviour (high abstraction) in agent based

simulation the most important side is interaction

among entities, which creates the aggregate

emergent behaviour. The single agents can be very

simple, with few rules and directives. For example, a

somewhat realistic artificial stock market can be

simulated by creating different types of intelligent

agents, which follow inner rules; some of them will

simply act randomly, while others will “study” the

trend before acting. Some of them, on the contrary,

could use advanced techniques, such as stop loss. By

observing the general trend of an artificial stock

market created with these rules, one can be amazed,

by seeing that it resembles in many ways a real one.

On the other side, agents can be modelled with inner

TEACHING MANAGEMENT AND FINANCE THROUGH SIMULATION - Choosing the Proper Paradigm

465

reasoning and learning capabilities, for example

using neural networks, genetic algorithms, classifier

systems or other learning paradigm (e.g.:

reinforcement learning), which create an

evolutionary environment. Each agent has the

capacity to reason on the global effects of local

actions, or even to create its own forecasts on the

actions that will be performed by other agents. The

agents built using this approach can decide on which

action to perform, according to the stimuli coming

from the environment, and not only according to

their internal rules.

3 A SUPPORT FOR TEACHING

Simulation, in its various forms, can be a great

support for teaching Social Sciences, and in

particular Management and Finance. No matter what

paradigm is used, the model is an operative and

dynamic example for the theories usually explained

with natural language or case study. While the

theoretical background is fundamental, sometimes

this is regarded as something far from applicability

or, even, something that can’t be applied in real

world. A simulation model could thus serve three

main purposes in the educational field:

1. It can be used to dynamically show theories

previously explained, with an increased

expressive power. A metaphor for this is the

following: theory is a picture of the real

situation. A simulation model is the movie for

this. A picture can be looked at, and the

dynamics behind it must be imagined by the

observer; the simulation model gives a clear

insight on this background dynamics, thus

helping the explanation of social interrelations

and theories.

2. It can serve as an experimental desk, a sort of

laboratory for the Social Sciences. Using a model

for this purpose, students can experiment cause-

effect relations among some parameters (the

input variables of the simulation) and the results

(the output, be it quantitative or qualitative). This

virtual laboratory allows for example to simulate

enterprises in business games, so that students

“learn by doing” and try to make decisions about

managing and driving a firm.

3. It can be used as a substitute (or a dynamic

assistance) for use cases. Use cases are

interesting and very useful, but often static. They

are synthetic examples for general situations and

they could generate theory. Simulation models, if

correctly built, can also be the basis of new

theories and explain the behaviour of a system.

Thanks to computational power of today’s

computers, simulation models can be used to test

social theories, such as individual perception,

diffusion issues, aggregation, segregation,

reputation and so on.

All these three modes require a big interactivity

among the users and the model. Interactivity is,

itself, something that trigger learning; something

which is directly tried (and not just listened to) is

surely easier to understand and remember. Besides,

in this way, group learning is also motivated. For

example, in business simulations, a group of

students can decide who’s managing a particular

enterprise function, needing a coordination among

them. Simulations can be the basis for learning team

collaboration and appreciating team work.

4 WHICH PARADIGM?

All the mentioned paradigms are efficient and

potentially interesting as an educational support;

though, they have particular features that make them

more or less effective, when used for simulating

Economical and Financial situations. Generally

speaking, it can be noted that DE simulation can be

used to model and dynamically show processes that

can be easily decomposed into basic activities,

acting as building blocks. This enables a micro-level

analysis for thus systems like plants, machineries, or

such processes defined by standards – production,

informative systems and so on. SD, on the other

hand, allows to carry on an analysis at the system

level, i.e.: aggregate macro-level. Instead of

designing the single parts composing a process,

when dealing with SD it’s necessary to individuate

some macro-classes, and the interrelations among

them. For example, SD is optimal for carrying on

analyses of diffusion phenomena (e.g.: innovation,

new technologies) or the evolution of a network of

firms. AB simulation is possibly the more “general”

approach, in the sense of its potentially wider

application, when compared to the others. It’s

particularly powerful when the behaviour of the

system at a macro level is not known a priori, or at

least the roles and interrelations of the single parts

are not explicit or directly expressible through

equations. By creating the simulated world with a

bottom-up approach, the aggregate behaviour is an

emergent property of the individual actions of the

agents, i.e.: the micro parts composing the system.

This makes AB optimal for studying the human

factor in Management and Finance, or – if cognitive

CSEDU 2010 - 2nd International Conference on Computer Supported Education

466

agents are employed, the adaptive behaviour of

societies. AB is for example the most used approach

for representing Game Theory, (e.g.: Power, 2009;

Remondino and Cappellini, 2005), spontaneous

aggregation phenomena, and so on.

4.1 Pros and Cons

Besides having different application fields, as

discussed, each of the reviewed paradigms has some

points of strength and weakness. DE modelling has,

as a first fundamental strength when used as a

teaching support, to be able to show, through a flow

chart (which is something very straightforward and

cognitively powerful) how a process works and

operates. Each activity can be explained in details,

through completion time, required resources,

preconditions, local outputs and so on. Besides, after

defining the framework defining each part of the

process, the output could be measured by means of

simulation; the average time for process completion

can be calculated, as well as the costs, possible

bottle necks, redundant resources. This is a very

strong way of showing how deterministic or

stochastic processes work, and can be the basis for

scenario and what-if analysis. Another point of

strength for DE modelling is the easiness of data

interpretation; if structural values are used correctly,

the results are directly comparable with the real

world situation that has been modelled. Besides, DE

models are not too difficult to build and modify,

thanks to well known and widespread standards

(e.g.: UML) and easy-to-use visual tools. Last but

not least, when building a DE model, the real world

situation has to be studied in detail and decomposed,

first. This requires a theoretical and deep pre-

modelling study, which is very useful for students.

On the other hand, DE modelling has some

weaknesses; first, it’s not always easy or even

possible to divide a process (especially if complex

and social) into basic activities, without severe

approximations: for example a Financial market.

Besides, the model has a deterministic or stochastic

nature, thus making it impossible to realistically

emulate human-like behaviour and human factor in

the most general extent. Another limitation is that

the whole schema is built a priori (the flow) and is

not subject to structural changes during the

simulation. Eventually, it’s worth mentioning that

it’s quite difficult to have more big processing to

cross interact among them, unless loosing detail or

expressive power, which is something fundamental

for educational purposes. As to SD, its biggest point

of strength for educational purposes is the easiness

to transmit the connections and interactions among

macro systems. This is a fundamental part for

Economic systems (markets) and Finance. Feedback

loops are highlighted and so cause/effect relations

among the components. Another important and

natural pro of SD is the fact that stock variables and

flow variables are identified and represented. This is

exactly the way in which many accounting and

Economic systems work; e.g.: the flow of new

investment (positive flow) and the depreciation or

depletion (negative flow) determine the stock of

capital. Last but not least, as for DE modelling, there

are many visual tools to build SD simulations, thus

overcoming possible technical difficulties. A first

drawback of SD technique is that it usually gives

general “average” results for the macro classes

represented. For example, in a market model, it will

represents the average investments for enterprises, or

the average value added for customers and so on.

The modeller has to think in terms of global

structural dependencies and has to provide accurate

quantitative data for them. Since SD is at an high

abreaction level, it’s difficult – or even impossible –

to explore lower levels, thus showing to students and

learners how individual sub-parts actually influence

the overall behaviour of the class that comprises it.

When building a SD model, it’s necessary to

define relations among the parts through

mathematical formalisms. This is not always

possible, especially in social and financial systems;

while a general and theoretical market can be well

described, some investment and negotiation

strategies based on perception cannot be represented.

The relations among the macro classes are then

static and immutable during the simulation process.

Essentially, as for DE simulation, the PC calculates

the overall result given a mathematical framework

and some initial parameters, but cannot change the

way in which these macro classes interact among

them. According to (Bonabeau, 2002), AB

modelling has three main benefits, over other

approaches: it captures emergent phenomena; it

provides a natural description of a system; and it is

flexible. In fact, as already noted, AB simulation is

the widest applicable approach, in the sense that it

can be used for exploring problems at different

levels (from an higher abstraction, like markets or

societies, to lower levels, like individual behaviours

for the agents). Besides this is possible in the same

model, making the different level interconnected

among them. The bottom-up approach used for

building AB models is most natural for educational

purposes, since it examine a situation from the

perspective of the single entities involved. The

TEACHING MANAGEMENT AND FINANCE THROUGH SIMULATION - Choosing the Proper Paradigm

467

overall behaviour is thus not prescribed or defined

by the designer of the system, but rather an emergent

property, coming out from the aggregation of many

individual – possibly simple – behaviours. As noted,

the agents can also be cognitive and this fact brings

in the possibility of studying human factor in

organizations, and to study consumer choices when

facing the market, or behavioural finance biases. As

a first counter altar to the wide applicability of this

paradigm, it must be noted that there is not a general

prescription for building an agent based model.

Usually, a base framework has to be built, with some

general rules (environment) and then the properties

at the agent level must be defined and implemented.

Since the aggregate behaviour is emergent,

quantitative results are often difficult to validate and

to directly compare to real situations. The results

gets more and more difficult to interpret when there

are many parameters and wide environments.

Anyway, especially as a teaching support,

sometimes qualitative results are more than enough,

especially when they are supported by graphical

views (e.g.: network representation, 2D/3D views of

the underlying environment and so on). Scaling

issues often arise in AB models, meaning that the

overall behaviour is influenced by the quantity of

agents involved, or better by the order of magnitude.

While for both DE and SD paradigms there are

many visual tools that can be used to build models,

in order to implement AB simulation some

programming skills should be possessed (e.g.: Java,

C++ or others). Some languages have been

conceived with the purpose of building AB models

(e.g.: Starlogo and others), which have some built in

primitives and routines to facilitate the

implementation of such simulations, and many

support libraries exist for high level languages (the

most known being Swarm, Ascape, Repast, Mason),

but even so building an AB model is much more

technically heavy than using the other paradigms.

5 CONCLUSIONS

A simulation model is essentially a set of rules that

define how the system being modelled will change

in the future, given its present state. In the paper

three simulation paradigms are explored and

reviewed in detail, in order to show their

applicability in the educational field, as a support for

teaching. After discussing the expressive power of

simulation in general when coming to knowledge

transfer and representation, each paradigm is

described, and so are the fields in which they could

be more naturally employed, with particular

attention to Management and Finance. As a

conclusion, in the paper it is debated and maintained

that all simulation models can be a perfect support

for teaching these subjects, since they can be used to

dynamically show theories previously explained,

they can serve as an experimental desk, a sort of

laboratory for the Social Sciences and can be used as

a substitute (or a dynamic assistance) for use cases.

If Discrete Event modelling is perfect for

describing and representing situations at a micro

level, System Dynamics is optimal for designing

markets and financial systems at an high abstraction

level, where the links among the macro classes can

be expressed through equations. Agent Based

modelling has a wider scope, and is optimal when

human factors and social issues must be taken into

accounts, thanks to its bottom-up modelling, but is

also the most technically difficult to implement, due

to the lack of visual tools, which exist for the other

paradigms. Though, when the system to be

simulated has a complex aggregate behaviour, not

easy to describe just studying and modelling the

single basic activities, and at the same time

impossible to describe formally through equations,

agent based simulation seems to be the best usable

approach. In complex systems the sum of the parts is

often not enough to describe the whole, and usually

from the interaction of many simple entities a

complex behaviour emerges.

ACKNOWLEDGEMENTS

The authors wish to thank prof. Gianpiero Bussolin,

their unforgivable mentor.

REFERENCES

Bonabeau E., 2002. Agent-based modeling: Methods and

techniques for simulating human systems, PNAS 99

Gilbert N., Terna P., 2000. How to build and use agent-

based models in social science. Mind & Society, n. 1

Ostrom T., 1988. Computer simulation: The third symbol

system. Journ. of Experimental Social Psychology, 24

Power, C., 2009. A Spatial Agent-Based Model of N-

Person Prisoner's Dilemma Cooperation in a Socio-

Geographic Community. JASSS n. 12.

Remondino M., Cappellini A., 2005. Introducing Social

Issues into a Minority Game by Using an Agent Based

Model, in Social Simulation: Technologies, Advances

and New Discoveries, Edmonds B. et al. eds. Idea

Group Publishing, ISBN: 978-1-59904-522-1.

CSEDU 2010 - 2nd International Conference on Computer Supported Education

468