=Paper=
{{Paper
|id=Vol-1420/dc-paper1
|storemode=property
|title=Knowledge Based Modeling of Financial Decision Support Systems
|pdfUrl=https://ceur-ws.org/Vol-1420/dc-paper1.pdf
|volume=Vol-1420
|dblpUrl=https://dblp.org/rec/conf/bis/LiutvinaviciusL15
}}
==Knowledge Based Modeling of Financial Decision Support Systems==
https://ceur-ws.org/Vol-1420/dc-paper1.pdf
Knowledge Based Modeling of Financial Decision
Support Systems
Marius Liutvinavičius, Audrius Lopata
Vilnius University, Kaunas Faculty of Humanities
{marius.liutvinavicius, audrius.lopata}@khf.vu.lt
Abstract. Financial markets are perceived as dynamic systems of interacting
agents whose modeling and prediction are based on computer technology.
Decision support systems analyze the state parameters of controlled systems
and external factors in order to make the appropriate decisions. The aim of this
paper is to analyze the techniques for system dynamics and decision-making
process modeling. The simulation of controlled systems’ models, modeling the
decision-making processes and goals modeling are connected in a whole. All
different types of models are analyzed in one process area. Also the conceptual
example of controlled financial process is presented.
Keywords: decision support, system dynamics, knowledge base
1 Introduction
The research area of computational economics combines economics and computer
science. Financial market is perceived as a dynamic system of interacting agents
whose modeling and prediction are based on computer technology.
In particular, the main effort is to model complex economic processes, the
influence of different factors to the final result and the interdependencies of these
factors. First of all the models of real life processes are created and then they are
simulated in order to experimentally analyze the various possible scenarios. Decision
support systems (DSS) analyze the state parameters of controlled systems and
external factors in order to make the appropriate decisions.
The main research question is how to increase effectiveness of identification of
anomalous situations in financial markets. Combining intelligent data analysis
methods, multi-criteria modeling and knowledge based decision making can solve this
problem. It allows more accurately evaluate the markets efficiency and better forecast
upcoming crisis.
But first of all we need effective methods for easier modeling of complex
mechanisms of financial systems. Different types of factors that affect the financial
markets must be integrated in models. Turning this to knowledge base which is used
by decision support systems can increase the efficiency of such systems. The
simulation of controlled systems’ dynamics, modeling the processes of decision
making and goals versus risks modeling must be connected in a whole.
Copyright © 2015 by the authors. Copying permitted for private and academic purposes.
This volume is published and copyrighted by its editors.
148
The aim of this paper is to analyze the techniques for systems dynamics and
decision-making process modeling. In order to achieve this, we present different
types of models in one process area (PA). In the section 2 we explain how these
models can be connected in the formalized way and present the conceptual example
of controlled financial process. In section 3 we present the review of methods for
modeling system dynamics, decision-making process, goals and risks.
This paper serves as theoretical background for past authors’ works (e.g. “Dynamic
simulation of pension funds’ portfolio” [12], “Research of customer behavior
anomalies in big financial data” [11], “Multi-criteria model for choosing long term
investments” [13]) and for future works. In the future researches we will try to
increase modeling and prediction effectiveness by incorporating irrational factors that
affect financial markets in addition to traditional fundamental and technical analysis
indicators.
2 DSS modeling methods in one research area
First of all we need to have the methodology for connecting different types of models
and their elements in a whole. The main aim is to correctly identify hierarchical
structures and relationships between system elements. There are different
methodologies for systems modeling, such as ontology based modeling or
metamodeling.
Ontology is an explicit specification of a conceptualization (T.R. Gruber 1993) [3].
Ontology defines the concepts of investigated domain, types of entities (events,
objects), conceptual hierarchies, the interrelations of entities types and their
interdependencies, axioms, rules, patterns of entities’ types and relationships, case
studies [9]. According to J. Dietz (2006), as a branch of philosophy, ontology
investigates and explains the nature and essential properties and relations of all
beings, as such, or the principles and causes of being. [5]. We distinguish ontology
based modeling methods offered by J. Dietz, because they are based on process
modeling. This can be applied to modeling the processes of financial markets.
Assuming our goal is to create a model of real word (a simplified representation of a
certain reality), the importance of metamodeling must be understood.
According to S. Mellor (2004) a meta-model is a model of a modeling language.
The meta-model defines the structure, semantics and constraints for a family of
models [14]. S. Clark (2008) states that a meta-model is a model of a language that
captures its essential properties and features. These include the language concepts it
supports, its textual and/or graphical syntax and its semantics (what the models and
programs written in the language mean and how they behave). Correct meta-model is
ontology, but not all ontologies are explicitly modeled as meta-models [3].
Further in this paper we analyze process area and elementary management cycle,
proposed by S. Gudas [7]. In original work it was applied for enterprise modeling. In
this approach we apply it to the decision support systems (DSS) of financial markets.
The main objective is to interconnect process simulation of controlled system
(financial market domain), decision-making process modeling and objectives versus
risk modeling.
149
Fig. 1. EMC and three types of models in one Process area (based on S. Gudas. Theory
fundamentals of information systems engineering [7])
The diagram of Elementary management cycle (Figure 1) shows how the managing
system (DSS) interprets the parameters of the controlled system and implements
decisions in accordance with the goals model.
All three types of models (controlled system, decision-making process, goals and
risks) can be displayed in one research area.
Process Area (PE) has three axes: aggregation (AG), generalization (GE) and time
(T). They express the processes that form the hierarchy of the modeled systems
entities [7]:
• AG axis forms the hierarchies of material entities. Usually it means the
technological processes, roles and data of enterprise. In our case it means the
processes of investment, market data, shares etc. i is the index of aggregation
hierarchy level.
• GE axis forms the hierarchies of conceptual entities (only informational processes
and objects). j is the index of generalization hierarchy level.
• T axis forms the sequences in time of material and conceptual entities. It also
forms causality of processes and objects. t is the index of time hierarchy level.
Process area (PE) is designed to investigate the management processes of
controlled system. Modeled objects of real world are called entities:
• Material entities are modeled in plane (AG, T). The term „material“ comes from
enterprise modeling and usually is connected with technological processes. But in
our case it is perceived slightly different. In this plane we present models of
investment processes, factors affecting these processes and actors involved in
them. Object oriented modeling and system dynamics are combined to investigate
operating principles and causalities of controlled financial domain. These
modeling methods are analyzed more deeply in section 3.2.
• Conceptual entities are modeled in plane (GE, T). Models of managing system
(DSS) cover data processing and decision making processes, as well as data and
knowledge structures. Architecture of DSS is presented in section 3.1.
• Goals and risks are modeled in plane (AG, GE). All decisions are made according
to them. Methods for goals modeling are presented in section 3.3
150
During the process simulation of controlled system its’ entities and factors
affecting them gain various values. Every moment of time (t) managing system
processes the data, makes decisions according to goals and risk management rules.
Fig. 2. Components of portfolio management EMC
Figure 2 presents the author’s example of elementary management cycle of
financial market domain. Managed object is the investment portfolio.
Head of investment strategy development makes strategic decisions in time
moment ti(1) and formulates strategic goals (i+1, j+1). Investment strategy is actual
for the entire period of time ti. Strategic goals are detailed and investment tactic is
prepared. In accordance with tactical goals the fund manager analyzes the market and
makes asset allocation decisions in time moment ti(2) (i, j). Broker finds best time and
performs buy and sell operations in accordance to tactical goals (i-1, j-1).
Figure 3 is also designed by author and shows these three separate EMC in one
coordination plane. ECC* corresponds to strategic investment planning. ECC stands
for fund management process, ECC' corresponds to the lowest level – buy and sale
operations management.
This method enables to connect models in a formalized way. It also allows
analyzing the hierarchical structures of entities and establishing relationships and
informational transactions between them.
There is a wide variety of possible coordination cases. They are classified by S.
Gudas [7]. Our example is consistent with the coordination type A3.1.
Coordination type A3 (i≠i`; j≠j`; p = p`; r= r`; t= t`) theoretical basis. The two
(in general may be more) EMC belong to the same type of management function (r)
and the same period of time (t). But their hierarchy levels of aggregation (..., i, i +
1,...) and generalization (... , j; j + 1, ...) differ.
They both manage the same object, but „see“ it from different angle. More detail,
coordination type A3.1. means that coordinated managing process is at higher
151
aggregation and generalization levels than controlled processes. Such modeling type
reveals specification and concretization aspects of managing and controlled systems
[7]. Higher-level EMC can be described as (i + 1, j + 1, t, r, p), when controlled EMC
has indexes (i, j, t, r, p). Managing information (orders, limitations) is transferred
from higher to lower level EMC. In our case that means strategic goals and tactics of
asset allocation. Some system elements can be modeled very deeply, other can have
only brief description. But the main objective is to gain understanding of what system
consists of and what elements should be modeled.
Fig. 3. Coordination of EMC
3 Modeling methods review
3.1 Decision making process modeling
One of the most popular models of decision making was developed by Herbert A.
Simon (1960) [15]. The model consists of three steps: intelligence, design, and choice
[19]. Figure 4 presents the architecture of decision support system. One of its essential
components is models subsystem. It contains all types of models discussed in
preceding section of this paper. This subsystem communicates with data subsystem,
which process internal data of controlled system and values of external factors
affecting it. All decisions are made in knowledge management subsystem. Users
communicate with DSS through interface subsystem. Best decisions are saved in
knowledge base and can be reused later [4] (Fig. 4). Such architecture allows
separating data from decision making mechanisms and developing their components
independently.
152
Fig. 4. Architecture of Decision support system(based on Decision making.T120B120 [4])
3.2 Process simulation of controlled system
Model is an abstract construct that attempts to replicate some properties of the real
system. The main objectives are deeper insight and better knowledge of the system or
its elements [18]. Dynamic simulation of systems processes are closely related to the
knowledge-based IS engineering. Both uses models of controlled domain and
knowledge database [8]. Internal modeling and process simulation allows looking
inside the system: monitor the changes of the status of system components, analyze
the impact that various factors make to the final results and to each other. For
example, when simulating the process of investing to funds, we can see not only the
final portfolio value, but also monitor its changes over time and analyze what factors
has the most significant influence.
System dynamics is the methodology intended for modeling and analysis of real
world systems. It is an applied discipline which allows understanding the behavior of
such systems [16]. Term system means an interdependent group of items forming a
unified pattern. Generally in system dynamics internal structure of the system is often
more important than external events [17]. But modeling the financial market domain
requires broader approach – external factors affecting the system are also involved.
Two main types of diagrams are used in system dynamics: causal loop diagrams
and stock and flow diagrams. The causal loop diagram shows the circular chains of
cause-and-effect in the actual system. [17]. The chain of the causes can be extended
us much as analytic can understand the system. As mentioned above, most causal loop
diagram involve not only internal elements of system, but also external factors that
are outside the system and makes influence to it.
As with a causal loop diagram, the stock and flow diagram shows relationships
among variables which have the potential to change over time. But unlike a causal
loop diagram, a stock and flow diagram distinguishes between different types of
variables: stocks, flows, and information [17]. Stock and flow diagrams are widely
used in process simulation software. The most popular of them are iThink®,
STELLA, Powersim Studio, Vensim, Anylogic, Insight Maker.
UML diagrams can be used beside causal loops or stock and flow diagrams. For
example a class diagrams can represent the static structure and relationships between
153
system elements or factors affecting it. Activity diagrams can show the possible
scenarios of system behavior. Sequence diagrams can represent the users interaction
with decision support system. For UML modeling we can distinguish Magic Draw
(commercial), SmartDraw and Visual Paradigm (both free) tools.
3.3 Goals modeling
KAOS methodology refines goals into requirements and includes 4 model types:
Goal, Agent, Operationalization and Object. KAOS Goal model has 5 elements:
Agent, Object, Operation, Requirement, Expectation [6]. Agent models depict agent
responsibilities and can be inferred from the goal models. Object Model specifies
objects used in the goal model. The syntax is similar to that of a UML class diagram
[6].
URN is the first international standard for business goals, scenarios and
relationships modeling in a graphical way. It includes GRL (Goal-oriented
Requirement Language), which is a graphical notation designed for modeling the
goals and requirements of different stakeholders.
The focus of GRL is to design the why and the what aspects of a model. GRL
divides its modeling elements into three main categories [1]:
• Intentional elements are the constructs which are used to model the system.
• An actor represents a stakeholder or a system.
• Links are used to connect intentional elements and actors
It‘s important that GRL can be successfully integrated to UML. For this purpose a
special UML profile is used for goal modeling. In such way goal diagrams (prepared
using GRL methodology) can be connected with other UML diagrams. [1]. We can’t
state, that UML doesn’t allow to model goals. The problem is that UML doesn’t
specify goals as separate element. There is no unique element class for it. But GRL
integration to UML solves this problem.
4 Conclusions
Financial markets are perceived as dynamic systems of interacting agents whose
modeling and prediction are based on computer technology. The main effort is to
model complex economic processes, the influence of different factors to the final
result and the interdependencies of these factors. All such information is used in
decision support systems in order to provide profitable solutions.
First of all the models of real life processes are created ant then they are simulated
in order to experimentally analyze the various possible scenarios. Decision support
systems (DSS) analyze the state parameters of controlled systems and external factors
in order to make the appropriate decisions. This area covers different types of models
and different types of factors that affect the financial markets must be integrated in
models. There is a need to have a formalized way to analyze complex processes of
financial systems and their management.
154
In this paper we presented different types of models in one process area (PA). The
dynamic controlled systems’ modeling, modeling the processes of decision making
and goals versus risks modeling are connected in a whole. The conceptual example of
funds’ portfolio management explains how complex systems and processes can be
described in such formalized way. Knowledge of controlled systems behaviour is
turned to knowledge base which is used by decision support systems. All types of
models are also saved in models subsystem and used for decision making.
The analysis of techniques for systems dynamics and decision-making process
modeling was made. For dynamic simulation of system model causal loop and stock
and flow diagrams can be used. UML diagrams can be used in conjunction with them.
KAOS and GRL methodologies can be used for goals modeling.
References
1. Abid, M. R. UML Profile for Goal-oriented Modelling. Lecture Notes in Computer
Science. Volume 5719, pp 133-148 (2009)
2. Blazyte, I. Analysis of business rules modeling using UML. Master thesis (2005)
3. Clark, Sammut etc. Applied Metamodelling. A Foundation for Language Driven
Development. Second edition. Ceteva (2008)
4. Decision making. T120B120. Enterprise Information technologies (2010)
5. Dietz, J. Enterprise Ontology: Theory and Methodology. Springer, ISBN:3540291695 (2006)
6. Goal modeling. Michigan state university. Lecture notes (2006).
http://www.cse.msu.edu/~chengb/RE-491/Lectures/06-goal-modeling.pdf
7. Gudas S. Informacijos sistemu inzinerijos teorijos pagrindai. Monografija, Vilnius, Vilniaus
universiteto leidykla, 382 p. ISBN978-609-459-075-7 (2012)
8. Gudas S. Ziniomis grindžiamu sistemu kurimo problemos. Lecture notes (2012)
9. Klimas S. Automated gathering of marketing information from the internet. Master thesis,
Kaunas (2010)
10.Maskeliunas S. Ontologijos ir semantinis pasaulinis tinklas. Lecture notes (2015)
11.Liutvinavicius M., Kriksciuniene D. Research of customer behavior anomalies in big
financial data. Proceedings of HIS 2014: 91-96; ISBN: 978-1-4799-7633-1 (2014)
12.Liutvinavicius M., Sakalauskas V. Dynamic simulation of pension funds’ portfolio. BIS
2012. Lecture Notes in BIP, Vol. 127, Springer. ISBN 9783642342271. p. 69-80.
13.Liutvinavicius M., Sakalauskas V. Multi-criteria model for choosing long term investments.
„Informacinės technologijos 2015”: Conference proceedings. p. 131-334. ISSN
2029 249X (2015)
14.Modeling and metamodeling in Model driven development, Universidad Carlos III de
Madrid (2009). http://www.ie.inf.uc3m.es/ggenova/Warsaw/Part3.pdf
15.MIT5312: Systems Analysis and Design. Simon's model of decision making (2007)
16.Norvaisas S., Sruogis V. Basics of System dynamics. Lecture notes.
http://www.culture.lt/science/SD/sd.htm
17.System dynamics.http://www.public.asu.edu/~kirkwood/sysdyn/SDIntro/SDIntro.htm
18.Sutiene K. Imitacinis modeliavimas – sistemoms pazinti, analizuoti ir sprendimams priimti
(2011)
19.Svanidzaite S. A Methodology for Capturing and Managing Non-Functional Requirements
for Enterprise Service-oriented Systems. Baltic J. Modern Computing, Vol. 2, No. 3, 117-
131 (2014)
155