Paper # E18

desertcockatooData Management

Nov 20, 2013 (3 years and 8 months ago)

101 views




Paper # E18



Fourth IFCIS Conference on Cooperative Information Systems (CoopIS'99)

(In cooperation with VLDB'99)

September 2
-
4, 1999

2 Edinburgh University Conference & Training Center, Edinburgh, Scotland


Distributed Empirical Modelling in Economic
and Finance

Soha Maad

(soha@dcs.warwick.ac.uk)

Department of Computer Science


University of Warwick, Coventry CV4 7AL UK




ABSTRACT

Distributed modelling is a powerful mean to capture, represent, and transmit knowledge to different viewers with different

backgrounds and interests. It allows multiple users to cooperatively interact with the model following security access and
different view levels. Associated to a distributed model are visual interfaces which allows multiple users to interact with t
he
dist
ributed model and share their knowledge and insights. Visualization is gaining importance in the financial analysis
process, a pioneer example is the use of maps in financial knowledge discovery, exploratory data analysis, and data mining
as applied to f
orecasting of financial indicators, selection and evaluation of different investment opportunities,
classification of firms, and risk assessment. Visual graphs are also useful in enhancing the understanding of different
spreadsheet financial models.

This

paper surveys the financial analysis and modelling tools in term of features, underlying data model, visual support,
and distributed interaction. The benefit gained from these tools in term of knowledge acquisition, transactions processing,
and support t
o human analysis, decision making and planning is assessed. The paper discusses the limitation of current
financial analysis and modelling tools in supporting the financial reasoning and decision making process and highlights the
need for enhanced financia
l analysis and visualization tools which can support the analysis of the cause/effect of financial
events such as crises and market crashes. The main feature of these tools is to simulate the occurrence of financial events
based on an underlying economic m
odel which is translated into a shared data model to which different visualizations are
associated, thus allowing open ended distributed interaction with the model. The potential use of the distributed Empirical
Modelling approach for enhanced financial an
alysis and visualization is investigated.




INTRODUCTION

Since markets first began, firms, investors, and financial
analysts have looked for ways of applying information
technology to gain competitive advantages. Managers
using clever investment tools,
quantitative investment
analysis and visual exploration tools claim their ability to
track the market and maintain profitable positions
[Gri98]. Many tools are developed to model, analyze,
and visually explore financial markets, however, relying
on these t
ools without further analysis is very dangerous.
Massive computer programs with huge statistical
capability is not the only sort of science applied to
investment. Although computer applications and graphs
can support the application of investment theory ho
wever
their potential benefit is minimal without appropriate
human vision and interpretation [Lan98].

To what extent can available tools support the financial
analysis and reasoning process and how distributed
modelling can be applied is the main theme of

the paper.

The following paper is divided into four sections. The
first section overviews the main features in financial
analysis, data mining, and visual exploration tools. The
second section highlights the need for tools to model
financial events. Distr
ibuted empirical modelling is
introduced in the third section, referencing to its potential
role in enriching our knowledge through interaction with
distributed financial models that support our
understanding of financial events such as crisis and
market c
rashes. Section four presents a case study of a
distributed Empirical Model simulating the 1980’s New
York Stock Exchange crash. The paper concludes with
the need to extend financial analysis beyond trend
detection, classification, and visual exploration w
hich are
being applied on samples of financial indicators taken
over a period of time. Crises and market crashes are
reshaping the financial sector and are affecting
investment decision, hence, a reliable financial analysis
tool should be able to incorpor
ate an appropriate
representation and interpretation of financial events.
Distributed Empirical Modelling can support
conventional financial analysis techniques with state
dependency and agency analysis viewed from different
perspectives.


1. OVERVIEW OF F
INANCIAL ANALYSIS AND
MODELLING TOOLS

The huge amount of data available to the business and
financial community contains a lot of valuable and
hidden information which requires enhanced exploratory
data analysis and mining tools to extract its embedded
kno
wledge and share it across the enterprise.
Visualization can support this knowledge extraction in
different ways. Visualization is not limited to the
graphical illustration of results but refers to the state
associated to the system at a given point in tim
e: it can be
an interface, a dynamic model, a static image, or a
graphic plotting.


1.1 survey of tools

Many tools are available to analyze, mine and model
financial data and concepts. The following sections
survey these tools in term of features, underly
ing data
model, visual support, and distributed interaction.


Spreadsheet tools:

spreadsheet models capture a wide
range of business and financial applications. Their ease
of use, functionality, and extended features make them
the most popular and basic to
ols for simple financial
analysis and graphics reporting. Spreadsheet data models
are the underlying models used in any intelligent tool
with extended analytical capability. Almost all financial
analysis and visual exploration tools import their data
sour
ces from spreadsheet applications.


Data mining tools:

Many databases contain valuable
information that is not readily obvious. An example of
unrevealed information might be patterns of high
-
risk
companies within a financial database. The search for
these

valuable, yet hidden, patterns and relationships
within a database is known as data mining. Data mining
infers rules that can guide decision making and forecasts
the results of the decision. A number of different types of
data mining tools are in use toda
y, including:



data visualization, which is data mining in its simplest
form, providing a picture of variables in the data



neural networks which are collections of connected
nodes with input, outputs and processing at each node.
The network uses a train
ing set of data which it corrects
according to the resulting outcome and applies it to the
processing at the nodes in the network.



decision trees which divide data into groups based on
values of the variables. At each node, a yes
-
no question
is asked, p
roducing a hierarchy of if then statements that
classify the data.



rule induction programs, which create non
-
hierarchical
sets of conditions. These programs are more accurate
than neural networks and decision trees, and infer rules
from overlapping set
conditions.

The source of data analyzed by data mining tools can be
loaded from ODBC compliant databases as well as
spreadsheet and statistical software packages. Some of
these tools can be also integrated into applications as
activeX components. Mining fe
atures include data
manipulation (sampling, selecting, and merging data
sets), modelling (classification, prediction, profiling,
clustering, and detection models), visual exploration
(visual presentation of the different stages of the data
mining process,
charts and graph plotting, and graphical
trees views), and online distributed reporting (reports and
the graphical visualization generated from the data
mining process can be viewed through internet
browsers).


Visual exploratory analysis tools
: tools for

building Self
Organized Maps are used in finance to perform credit
scoring, behavior modeling, knowledge discovery in data
bases, system state monitoring, process engineering,
quality control, and prediction. This range of tools
supports: dependency analy
sis, deviation detection,
unsupervised clustering, non
-
linear regression, data
association, pattern recognition, animated monitoring, as
well as other enhanced visualization techniques.


Classification tools:

Predicting corporate failure or
bankruptcy is a

classification problem which is
considered as one of the most important problems facing
business and government. The recent saving and loan
crisis is one example, where bankruptcies cost the United
States billions of dollars and became a national politica
l
issue [O’Le98]. Appropriate visualization can support
the understanding and interpretation of the results of a
classification model. There are several ways to approach
the phenomenon of corporate failure, the common
approach views the failure as a termi
nal disease of the
company that manifests itself in the financial statements
two or three years before the actual failure [Kiv98].
Financial indicators, such as the profitability, solvency
ratios, etc. are extracted from the financial statements of
a samp
le set of companies and are fed into neural
network models to predict corporate failure. Self
organized maps are used as a mean to visualize the
classification of failing and non failing companies.


Risk assessment tools:

The risk of a financial asset is
defined as the variation in its underlying value. Credit
granting, investing and trading involve risk. Hedging is a
way to protect against risk, it can be achieved by the use
of derivative financial instruments which are mainly
financial contracts that ca
n guarantee a non
-
loss position
if all possible adverse market conditions and price
movements are taken into consideration in this contract.
Risk management involves the detection of trends and
market surveillance. Self organized maps are used in
visualizi
ng some risk assessment and classification
problems, such as the classification of countries
according to their risk, where the level of risk is
evaluated in term of many economic and financial
factors.


Market prediction tools:

Analyzing time series data

in
order to recognize patterns or make prediction about
future values is important in many application area such
as investment, trading, etc.. The ability to predict future
values based upon past values and known future events is
implemented using statist
ical time series analysis or
neural networks[Bow90]. However, most research
revealed that neural networks outperform the time series
analysis.


Portfolio management tools:
A portfolio is the total
securities hold by an institution or a private individual.

Voluminous amounts of rapidly changing data in
financial markets create a challenging problem for
portfolio managers attempting to exploit such changes to
achieve their investment objectives. Changing market
conditions should be exploited to optimize the
value of
individual portfolios [Pat90]. Prices of financial assets
taken from reliable sources are fed into portfolio
management models which support the decision of the
buy and sell action to adjust the portfolio holding.
Portfolio management tools enable

the user to download
list of prices for stocks, options, bonds, mutual funds, or
other investments directly from the Internet. Based upon
the newly downloaded prices a portfolio is priced and
reports are generated to display the results in ways that
ena
ble the investor to make clear and precise investment
decisions based on the total portfolio behavior.
Appropriate v
isualization is used to highlight profitable
transaction based on a study of asset prices, and to show
the portfolio holding at any moment
in time.


Financial games:
Financial games simulate trading in
the market. They are a way to experience the thrill of
investing without having to risk any money. Financial
games have no practical financial use, the only gain is an
improved understanding of

the market behavior and an
educational knowledge. Simulation games allows the
user to choose notional investments and then track them
to see how well or poorly the choices perform [Hen98].


Security analysis tools:

The term securities is a
comprehensive
term referring to stocks, futures
(commodities), indices, mutual funds, bonds, options,
etc. Security analysis tools offer many functions
including: managing securities data (downloading from
the internet, importing, exporting, and presenting security
data
), building composite securities, and maintaining and
managing a portfolio. Charts are created to provide a
graphical view of the security data. These charts are
displayed in movable windows that may be easily
resized. Different types of charts can be plot
ted
including: price volume plots, moving averages,
indicators, candlestick Charting, point & figure charting,
regression trendline, etc.. Some of these tools support an
online research capability which allows the user to
retrieve a wide variety of informa
tion such as company
reports, current quotes, and late breaking news stories.



1.2 Evaluation of tools

The evaluation of financial analysis and visual
exploration tools is based on the evaluation of their
underlying data model and its ability to capture
r
equirements, their support for enhanced visualization
and distributed interaction, their capability to extract
hidden knowledge, and the range of functions and
features available in this tool. Most of the above
described tools take the spreadsheet model as

an
underlying data model, visualization in these tools is an
enhanced graphic capability giving the user the option to
choose a given predefined graph, shape, or plot. Graphs
are dynamically linked to the underlying spreadsheet
model and follow the modifi
cation of the content of the
spreadsheet cells. The distributed visualization in these
tools refers to a shared access through internet browsers
of the reports and graphs generated by the tools. The
knowledge gained from these tools is a documentary
knowl
edge which support to a loose extent the decision
support. Some tools are supported by online capability to
download relevant information and indicators from the
internet and to settle financial transactions online.



2. MODELLING FINANCIAL EVENTS

Investme
nt in financial markets requires an
understanding of markets micro
-
structures and the major
events occurring in these markets. Globalization,
integration, trade liberalization, monetary union,
deregulation, market micro
-
structures and financial crises
are
creating a highly state changing global environment.
The analysis of past trends might fail in predicting future
trends if it is not supported by an appropriate
interpretation of state changes and transitions in the
financial markets. The above surveyed to
ols, which cover
most of the currently available tools, operate on series of
financial indicators representing the market over a period
of time. A spreadsheet model captures in its rows and
columns these series of financial indicators. The
enhanced visual
exploration, trend detection and
classification tools operate on these indicators to predict
and reveal past and hidden information which might be a
valuable knowledge for future decision making if
appropriately linked to a state analysis of the financial
market. Current financial analysis tools, for example, can
classify companies in failing and non
-
failing categories
based on the analysis of a set of indicators taken from the
firms balance sheets over a period of time. However,
these tools cannot show the

transition of the state of the
company from a non
-
failing to a failing category or vice
-
versa. A similar example apply to the classification of
countries or investment opportunities according to their
risk profile.

The following paragraphs describe the ma
jor financial
events reshaping the financial industry and redirecting
enormously the investment decision.


2.1 Overview of major financial events

Over recent years, the world has faced many economic
crises. The most important one was the Asian crisis
whic
h had a large impact on the world economy. Recent
economic forecasts predict the development of further
crises around the world. Parallel to the economic crisis is
the birth of the single European currency, which is
accompanied by a high level of uncertain
ty and economic
slow down in the short term. The next years will be a
period of large
-
scale change. Players in the global market
have to learn, or rather re
-
learn, some very basic lessons,
for example about the pricing of risk or about the
appropriate valu
ation of assets in a zero inflation world
[Bey99]. The following section overviews the major
financial events:



The crash of the New York stock exchange in the
1980’s:

The largest stock
-
market drop in Wall Street
history occurred on "Black Monday"
-

Octob
er 19, 1987
-

when the Dow Jones Industrial Average plunged 508.32
points, losing 22.6% of its total value. That fall far
surpassed the one
-
day loss that began the great stock
market crash of 1929 and foreshadowed the Great
Depression. The Dow average was
computed from the
stock prices of 30 major companies, selected for their
broad level of public ownership and their total market
value. Unlike in 1929, the market soon rebounded after
the crash. The fallout from the '87 crash was remarkably
light, due in p
art to intervention by the central bank of
the U.S. and the Federal Reserve. The worst economic
losses occurred on Wall Street itself, where 15,000 jobs
were lost in the financial industry. In searching for the
cause of the crash, many analysts found fault

with
"program" trading by large institutional investing
companies. In program trading, computers were
programmed to automatically order large stock trades
when certain market trends prevailed. In response, the
New York Stock Exchange (NYSE) restricted som
e
forms of program trading. The NYSE and the Chicago
Mercantile Exchange also instituted a "circuit breaker"
mechanism in which trading would be halted on both
exchanges for one hour if the Dow Jones average fell
more than 250 points in a day, and for two
hours if it fell
more than 400 points [Key98].



The Asian crisis:

In June
-
July 1997 the currencies of
Asia started to fall. Without exception all Asian
currencies (including those of New Zealand, Australia
and Japan) have fallen
[Tho97]. The impact of the

Asian
crisis was felt worldwide. C
ombined with internal factors
and other external shocks, and a decline in commodity
prices, the crisis in Asia has led to a downward revision
of the projected real GDP growth rate for sub
-
Saharan
Africa. The impact of the

Asian crisis on European
countries has so far been muted, reflecting their relatively
limited direct trade with the countries in crisis and the
strong financial positions of most banks with Asian
exposure. Most important, but more indirectly, the crisis
c
ontributed to the sharp decline in oil prices, which has
had dramatic implications for macroeconomic balances
in many oil
-
exporting Arab countries
[IMF98].



Future crisis:

Future crises and market crashes are
likely to happen as predicted by many economis
ts, we
hear of the New York Stock Exchange crash in year
2000. The introduction of the Euro currency is
surrounded by a high level of uncertainty. The impact of
the Asian crisis is revolving. The question raised by us is
whether current technology can help

in better
understanding the cause and effect of crises and their
impact on investment decision.


2.2 Can information technology help in
understanding financial events?

In this paper we suggest the adoption of a distributed
Empirical Modelling approach wh
ich can complement
financial analysis, trend detection, classification, and
visual exploration tools, in capturing state change in the
financial markets starting from a bounded economic
model depicting a given aspect of the event, and
translating this econ
omic model into an Empirical Model
supported by interactive and distributed visual interfaces
which represents the views of different entities involved
in the model (cf. Figure 1).

The purpose of the distributed Empirical Model is to
capture and convey kn
owledge about state change in the
financial market by capturing the dynamic of the
underlying economic financial model. The potential
benefit of the construction of this type of model is to
enhance the understanding of the cause/effects of
financial events
, and to judge about the probability of a
successful projection over the future of past trends and
patterns. For example, a company classified as a non
failing one might be prone to failure in adverse market
conditions, irrelevant of past trends and patte
rns of
success in the analysis of its balance sheet indicators.
Similarly, the state of an international portfolio might be
highly affected by financial and economic events.


2.3 framing the challenges in modelling financial
events

The challenges faced in
modelling financial events can
be identified as follows:



the selection of an appropriate economic financial
model which explain one aspect of the financial
event



the identification of observables, their view, and
their agency role from the account of the e
conomic
financial model



the construction of a computer artefact which
capture the dynamic of the economic financial model



the association of distributed interactive visual
interfaces which represents the views of the different
entities involved in the mod
el and their agency in
changing the state of the model



the open ended interaction which gives the user the
potential to apply different scenarios in manipulating
the model



the investigation of the possibility of combining
models to gain a wider view of the

cause and effect
of financial events



the selection of appropriate enabling tools and
technologies which can support distributed and open
ended interaction with the dynamic model.


3. DISTRIBUTED EMPIRICAL MODELLING FOR
SIMULATING FINANCIAL EVENTS

3.1 Int
roduction to Empirical Modelling

Empirical Modelling (EM) is an approach to computer
-
based modelling that has been under development at the
University of Warwick for over ten years. It combines
agent
-
oriented modelling with state representation based
on sc
ripts of definitions that capture the dependencies
between observables. Unlike conventional modelling
methods, its focus is upon using the computer as a
physical artefact and modelling instrument to represent
speculative and provisional knowledge. More det
ails of
the EM project can be found at the EM website [Web].
The principal software tool used to develop Empirical
Models is the
tkeden

interpreter. In this model, the
viewpoint of each modeller is represented by a script of
definitions (a
definitive scrip
t
) resembling the system of
definitions used to connect the cells of a spreadsheet. The
variables on the LHS of such definitions are intended to
represent observables associated with an external
situation. There is typically some form of visualisation
att
ached to them, so that for example a variable can
denote a point, a text message or a window displayed on
the computer screen. Different
definitive notations

are
used to formulate scripts according to the semantics of
these visual elements.
Tkeden

incorpor
ates definitive
notations for line drawing and screen layout, and an
evaluator for definitive notations that allows definitive
scripts to be formulated over scalar types, non
-
homogeneous recursive lists and strings. A distributed
version of the
tkeden

inte
rpreter (
dtkeden
) is
implemented on a client
-
server architecture, in which the
viewpoints of individual modellers are represented by
independent definitive scripts executed on different client
workstations. State changes are communicated between
clients b
y sending re
-
definitions across the network via
the server. Different communication strategies can be
specified via the server to suit different purposes. In a
concurrent engineering context, for instance, the server
can play a role in negotiation betwee
n clients, resolving
conflicts between design decisions, or dictating the
Financial
Economic
Model

LSD account
capturing agency,
dependen
cy and
state change

A distributed Empirical Model

A computer
artefact








Distributed
Interactive
Interfaces

For
different
views




Analysis of state
change in financial
markets



Supporting
classification, trend
detection, and visual
exploration

tools in
decision making



Supporting
knowledge a
cquisition
and interpretation


Figure 1.

Distributed Empirical Modelling of Financial Events

pattern of interaction and privileges of design
participants [Adz94].

The approach adopted in developing a distributed
Empirical Model to simulate a railway accident can
potentially

serve as a good background to simulate
financial crises and market crashes. The railway accident
model simulate the crash of two trains passing through a
tunnel, due to the default in the security system which
synchronize the passage of the trains through

the tunnel
and due to a negligence from the personnel monitoring
and operating the security system. The agents involved in
the model, their view, and their role in inducing state
change to the model is described in an LSD template
which serve in the class
ification of observables from the
perspective of an observer, detailing where appropriate:



the observables whose values can act as stimuli for
an agent (its
oracles
);



which can be redefined by the agent in its responses
(its
handles
);



those observables who
se existence is intrinsically
associated with the agent (its
states
);



those indivisible relationships between observables
that are characteristic of the interface between the
agent and its environment (its
derivates
).



what privileges an agent has for state
-
changing
action (its
protocol
).

The agents in the railway accident model include the
drivers, the human operators, and elements of the
security system. Each agent view is represented on a
separate workstation in the distributed model. The
modeler on a giv
en workstation can play the role of the
agent in imposing state change on the system. Where
appropriate this state change is propagated through the
distributed model. The model developed allows open
ended interaction. All scenarios can be played, the cras
h,
the no
-
crash, the train delays, etc.. This open
-
ended
interaction with the model is experimental in nature and
enrich the user understanding of the cause and effect of
the railway accident event.

Although a railway accident is not a financial acciden
t,
however the approach adopted in simulating the railway
accident can prove useful in trying to simulate stock
market crashes and financial crises. It is clear that a
financial market crash or crisis is more complicated as
many agents (external and intern
al factors) intervene in a
way which cannot be mathematically formulated,
however economic models based on a set of assumptions
and framing the scope of the analysis of the cause and
effect of the crisis can be a good feed to a distributed
empirical model
.


3.2 The feedback conveyed from a distributed
empirical model

There are four types of output or feedback conveyed
through the interaction with a distributed empirical
model. The most obvious one is a cultivated
understanding of the real life situation,
event, or
phenomena depicted in the model. For example, the user
might be able to capture the sequence of states and their
chronological order, e.g. the detection of the
chronological order of state transition which lead to a
crash in simulating a railway

accident or a market crisis.
The second feedback gained through interaction with the
model is the identification of new insights, reasoning
and views of the cause/effect of an event. The third
output is an improvement, modification or extension of
the m
odel over time. Finally, new rules or theory can be
deduced to explain real life phenomena validated through
empiricism and experimentation. The following figure
first introduced by [Sun99] to highlight the importance of
a distributed model and the user co
llaborative interaction
with this model to cultivate their understanding of a
situation is extended to apply to the case of distributed
empirical model of financial events.














Figure 2.

Feedback gained from a distributed EM



3.3

Empirical Modeling for simulating financial
events

A distributed Empirical Model can potentially support
our understanding of the cause/effects and the agency
role of different entities involved in financial events
(crises, market crashes). Distributed mo
deling supported
by an appropriate visualization is paramount in this
context to scope and frame different agents views, to
simulate the occurrence of simultaneous events, to
account for multiple agents interacting with the system,
and to gain a wider know
ledge by closing the gap
between the virtual and the actual reality.

The use of the model as such described is not limited to a
financial educational game. The distributed open
-
ended
interaction can help in validating the underlying
economic financial mode
l adopted, and extending it
when necessary.







Initial

Economic
model

Distributed EM




New
theory

Model

validation

Cultivated

understanding

New
insight &
views


4. CASE STUDY: A DISTRIBUTED EMPIRICAL
MODEL OF THE 1980’S NYSE CRASH

In an attempt to model one aspect of the 1980’s NYSE
crash, a simple economic model presented by [Mil99]
and developed by [Kru87] and
[Gro87] is taken as the
basic underlying economic financial model to develop
the corresponding distributed Empirical Model. The
economic financial model depict the impact of the
behavior of stop
-
loss traders in causing the stock market
crash (cf. Figure 3)
.

In building a distributed Empirical Model, the LSD
The stock market crash of 1987


was it due to stop loss traders?

















Assumptions:

-

let earnings follow a random walk

-

let p, the share price, be the present discount
ed value of exposed earnings

Labels:

-

ON shows the share price with discount factor r, the riskless rate

-

OR shows the share price with discount factor r+

, where

-



is a risk premium reflecting variability of earnings when all shares are held by
portfolio in
vestors

-



is the fraction of share holding of risk neutral stop loss traders

-

OW is the share price if


were to remain constant

Scenario 1:

-

consider the entry and exit of a risk neutral stop loss traders holding a fraction


of the shares

-

let the stop l
oss traders exits at p
x

(a trigger for programmed trading)

-

if the exit was not predicted then there will be a crash from B to C

Scenario 2:

-

consider the entry and exit of a risk neutral stop loss traders holding a fraction


of the shares

-

let the stop lo
ss traders exits at p
x

(a trigger for programmed trading)

-

if the exit was predicted then the share price lies on AA

, steeper than OR












The eco
nomic model

of the stock market crash

running on the server computer

The view

of the market traders

The view of the

Computer systems

Share price p

Fundamental

(earnings)

N

W

B

R

C

A

A


1/r

1/r+


Stock p
x

Market Crash




‘excess volatility’

Figure 3.

Model i
ntroduced by [Mil99]

account capture the agency and dependency relationship
in the model. Reference to the account of the 1980’s
NYSE crash introduced in section 2.1, we can identify
the following observables in our model:

traders, share
prices, earnings, the computer trading system, expected
earning, the discount rate, the variability of earnings, and
the market risk premium rate.

Reference to section 3.1, an LSD account classifies
observables from the perspective of an o
bserver,
detailing where appropriate oracles, handles, states,
derivates, and protocol for each agent.

A template for the specification of the LSD agent takes
the form:

agent

agent_name

(
parameter_list
)

{

oracle

list_of_oracle_variables


state

list_of_state_variables


derivate

list_of_derivates


protocol

list_of_guarded_commands


}


In our case study the LSD template for each agent is
described as follows:



the trader agent:

For t
raders, stock prices act as a
stimuli (
oracles
). A trader can undertake buy and sell
actions which gives him a privilege to change the state of
the system. These buy and sell actions are the
protocols

of the trader. A trader can be a stop loss, risk neutra
l, or
risk avert, he can also hold a fraction


of shares. The
type of traders and their fraction of shares holding are the
states

of the traders.



the computer trading system agent:

Another agent
involved in our model is the computer trading system
whic
h monitors and analyzes sale prices (
oracles

of the
computer trading system agent), issues buy and sell
signals (
protocols

of the agent), and communicates these
buy and sell signal to its environment (
derivate
). The
computer trading system is a 24 hour tra
ding system
which performs some analysis over prices, this
information describe the
state

of this agent.



prices, earnings, discount factors, risk premium:
These are also considered as agents affecting the state of
the model. For example, prices have an a
gency role in
specifying the values of earnings. In turn prices are
affected by the discount factor and the risk premium. The
dependency and agency relationship between these
observables is captured from the mathematics of the
underlying economic financial

model.

The distributed Empirical model distribute the views of
the different agents in the model on different client
workstations, and the server receive commands and give
a global view of the model.

The dynamic of the model is as follows (cf. Figure 3):



In playing the crash scenario, we consider the entry and
exit of a number of risk neutral stop loss traders holding
a fraction


of the shares, we let the stop loss traders
exits at p
x

(a trigger for programmed trading). If the exit
was not predicted the
n there will be a sharp price drop
leading to market crash.



In playing the no crash scenario we consider the entry
and exit of a number of risk neutral stop loss traders
holding a fraction


of the shares, we let the stop loss
traders exits at p
x

(a trig
ger for programmed trading). If
the exit was predicted then the price volatility in the
market will increase.



5. CONCLUSION

The following paper presented the application of
advanced intelligent analysis and visual exploration in
the area of finance. Diff
erent tools serving this purpose
are surveyed highlighting their uses and limitations.

The use of computer technology in supporting
investment decision, financial planning, and analysis is
assessed. This support stays relatively weak if an
appropriate anal
ysis of market behavior explaining state
changes is not conducted. Past information contains
hidden and valuable patterns and trends which convey a
past and present unrevealed knowledge. However, the
use of this knowledge for future decision making should
be based on the analysis of the state change in financial
markets and the cause/effect of events taking place in
these markets.

Distributed Empirical Modelling is introduced as an
approach to analyze state change in financial markets and
as a mean to repr
esent and explain financial events in a
distributed environment where different modeller can
play the role of different agents initiating state changes in
the system. The 1980’s NYSE crash is taken as a case
study in modelling financial events.

The need f
or distributed interactive modelling is
emphasized to bridge the gap between the virtual and the
actual reality.

Future research aims at extending the features of EM
tools to support distributed visual interaction with the
model through internet browsers,
and to investigate the
potential of integrating financial analysis, data mining,
trend detection, classification, visual exploration, and
distributed empirical modelling tools to make more
effective use of technology in supporting investment
decision and p
lanning.



ACKNOWLEDGMENT

The author acknowledge the help of Prof. Marcus Miller
at Warwick Business School in supplying the
information about the economic model of the New York
Stock Exchange crash in 1987 and for encouraging the
application of distribute
d visual modelling to financial
crisis. The author would like to thank Dr. Meurig Beynon
and Dr. Steve Russ for their valuable comments on the
content of the paper and the ways to improve it.



REFERENCES

[O’Le98] D. E. O’Leary, "Using Neural Networks to

Predict Corporate Failure", International Journal
of Intelligent Systems in Accounting Finance
and Management, Volume 7 No.3, September
1998, pp187
-
197.

[Kiv98] K. Kiviluoto, P. Bergius, "Maps for analysising
failures of small and medium
-
size enterprises
",
Visual Explorations in Finance with self
-
Organizing Maps, Springer Finance 98, pp. 59
-
71.

[Pat90] T.M. Patel, G.E.Kaiser, "The SPLENDORS real
time portfolio management system". In Proc. of
the First International Conference on Artificial
Intelligence Ap
plications on Wall Street, pp.73

[Bow90] J.E. Bowen, "Using neural nets to predict
several sequential and subsequent future values
from time series data". In Proc. of the First
International Conference on Artificial
Intelligence Applications on Wall Street
, pp.30

[Adz94] V.D. Adzhiev, W.M.Beynon, A.J. Cartwright,
Y.P. Yung, 1994, "A Computational Model for
Multi
-
agent Interaction in Concurrent
Engineering",
Proceedings of

CEEDA'94
,
Bournemouth University, pp. 227
-
232.

[Web]
The Empirical Modelling Web site
,
http://www.dcs.warwick.ac.uk/research/modelli
ng

[Hen98] Hennessy K.,
"Planning your finances on a PC
can save time and money


if you buy the right
package investment software", Money wise,
February 1999, pp. 64
-
68.

[Bey99] Beynon W.M., Maad S., "Empiri
cal Modelling
of Real Life Financial Systems: The need for
Integration of Enabling Tools and
Technologies" , to appear in
Proceedings of the
Fourth World Conference On Integrated Design
& Process Technology,
June 1999, Turkey.

[Key98] "Key Event: Black Mo
nday Crash of 1987
Rocks Stock Markets", 1998


Facts on File
World News CD
-
ROM, January 1980.


[Tho97] Thompson A., "NewsRoom: Beginners Guide
To The Asian Economic Crisis", NewsRoom
Ltd, 1996
-
1997.

[IMF98] IMF Area Department Directors, "How Has the
Asi
an Crisis Affected Other Regions? " Finance
& Development, September 1998.

[Mil99] Marcus Miller, MSc lecture on New York Stock
Exchange crash at Warwick Business School.

[Gri98] Grimond M., "Fund management intuition or
science" Exchange, The magazine of

London
Stock Exchange, Issue 4, December 1998.

[Lan98] Lang J., "Fund management Maths isn’t
everything" Exchange, The magazine of London
Stock Exchange, Issue 4, December 1998.

[Gro87] Grossman S., "One Share/One Vote and the
Market for Corporate Control
", NBER Working
Paper No. W2347