The contents are reproduced below. The last one or two issues may be missing from the list, but are included in the auction.

Journal of Computational Intelligence in Finance (formerly NeuroVest Journal)

A list of the table of contents for back issues of the Journal of
Computational Intelligence in Finance (formerly NeuroVest Journal) is
provided, covering Vol.1, No.1 (September/October 1993) to the present.


See "http://ourworld.compuserve.com/homepages/ftpub/order.htm"
for details on ordering back issue volumes (Vols. 1 and 2 are out of print,
Vols. 3, 4, 5, 6 and 7 currently available).


***
September/October 1993
Vol.1, No.1


A Primer on Market Forecasting with Neural Networks (Part1) 6
Mark Jurik
The first part of this primer presents a basic neural network example,
covers backpropagation, back-percolation, a market forecasting overview,
and preprocessing data.


A Fuzzy Expert System and Market Psychology: A Primer (Part 1) 10
James F. Derry
The first part of this primer describes a market psychology example, and
looks at fuzzifying the data, making decisions, and evaluating and/or
connectives.


Fuzzy Systems and Trading 13
(the editors)
A brief overview of fuzzy logic and variables, investing and trading, and
neural networks.


Predicting Stock Price Performance: A Neural Network Approach 14
Youngohc Yoon and George Swales
This study looks at neural network (NN) learning in a comparison of NN
techniques with multiple discriminant analysis (MDA) methods with regard
to the predictability of stock price performance. Evidence indicates that
the network can improve an investor's decision-making capability.
Selecting the Right Neural Network Tool 19
(the editors)
The pros, cons, user type and cost for various forms of neural network
tools: from programming languages to development shells.
Product Review: Brainmaker Professional, version 2.53 20
Mark R. Thomason
The journal begins the first of its highly-acclaimed product reviews,
beginning with an early commercial neural network development program.
FROM THE EDITOR 2
INFORMATION EXCHANGE forums, bulletin board systems and networks
NEXT-GENERATION TOOLS product announcements and news
QUESTIONNAIRE 26
4
23

***
November/December 1993
Vol.1, No.2

Guest Editorial: Performance Evaluation of Automated Investment Systems 3
Yuval Lirov
The author addresses the issue of quantitative systems performance evaluation.

Performance Evaluation Overview 4
(the editors)

A Primer on Market Forecasting with Neural Networks (Part2) 7
Mark Jurik
The second part of this primer covers data preprocessing and brings all of


the components together for a financial forecasting example.


A Fuzzy Expert System and Market Psychology: A Primer (Part 2) 12
James F. Derry
The second part of this primer describes several decision-making methods
using an example of market psychology based on bullish and bearish market
sentiment indicators.


Selecting Indicators for Improved Financial Prediction 16
Manoel Tenorio and William Hsu
This paper deals with the problem of parameter significance estimation,
and its application to predicting next-day returns for the DM-US currency
exhange rate. The authors propose a novel neural architecture called SupNet
for estimating the significance of various parameters.


Selecting the Right Neural Network Tool (expanded) 21
(the editors)
A comprehensive list of neural network products, from programming language
libraries to complete development systems.


Product Review: NeuroShell 2 25
Robert D. Flori
An early look at this popular neural network development system, with support
for multiple network architectures and training algorithms.


FROM THE EDITOR 2
NEXT-GENERATION TOOLS product announcements and news
QUESTIONNAIRE 31


***
January/February 1994
Vol.2, No.1
Title: Chaos in the Markets


Guest Editorial: Distributed Intelligence Systems 5
James Bowen
Addresses some of the issues relevant to hybrid approaches to
capital market decision support systems.


Designing Back Propagation Neural Networks:
A Financial Predictor Example 8
Jeannette Lawrence
This paper first answers some of the fundamental design questions regarding
neural network design, focusing on back propagation networks. Rules are
proposed for a five-step design process, illustrated by a simple example
of a neural network design for a financial predictor.


Estimating Optimal Distance using Chaos Analysis 14
Mark Jurik
This article considers the application of chaotic analysis toward estimating
the optimal forecast distance of futures closing prices in models that
process only closing prices.


Sidebar on Chaos Theory and the Financial Markets 19
(the editors) [included in above article]


A Fuzzy Expert System and Market Psychology (Part 3) 20
James Derry
In the third and final part of this introductory level article, the author
discusses an application using four market indicators, and discusses
rule separation, perturbations affecting rule validity, and other relational
operators.


Book Review: Neural Networks in Finance and Investing 23
Randall Caldwell
A review of a recent title edited by Robert Trippi and Efraim Turban.



Product Review: Genetic Training Option 25
Mark Thomason
Review of a product that works with BrainMaker Professional.

FROM THE EDITOR 2
OPEN EXCHANGE letters, comments, questions 3
CONVERGENCE news, announcements, errata 4
NEXT-GENERATION TOOLS product announcements and news 28
QUESTIONNAIRE 31

***
March/April 1994
Vol.2, No.2
Title: A Framework

IJCNN '93 8
Francis Wong
A review of the International Joint Conference on Neural Networks recently
held in Nagoya, Japan on matters of interest to our readers.

Guest Editorial: A Framework of Issues: Tools, Tasks and Topics 9
Mark Thomason
Issues relevant to the subject of the journal are extensive. Our guest
editorial proposes a means of classifying and organizing them for the purpose
of gaining perspective.

Lexicon and Beyond: A Definition of Terms 12
Randall Caldwell
To assist readers new to certain technologies and theories, we present a
collection of definitions for certain technologies and theories that have become
a part of the language of investors and traders.

A Method for Determining Optimal Performance Error in Neural Networks 15
Mark Jurik
The popular approach to optimizing neural network performance solely on its
ability to generalize on new data is challenged. A new method is proposed.

Feedforward Neural Network and Canonical Correlation Models as
Approximators with an Application to One-Year Ahead Forecasting 18
Petier Otter
How do neural networks compare with two classical forecasting techniques
based on time-series modeling and canonical correlation? Structure and
forecasting results are presented from a statistical perspective.

A Fuzzy Expert System and Market Psychology: (Listings for Part 3) 23
James Derry
Source code for the last part of the author's primer is provided.

Book Review: State-of-the-Art Portfolio Selection 25
Randall Caldwell
A review of a new book by Robert Trippi and Jae Lee that addresses "using
knowledge-based systems to enhance investment performance," which includes
neural networks, fuzzy logic, expert systems, and machine learning
technologies.

Product Review: Braincel version 2.0 28
John Payne
A new version of a low-cost neural network product is reviewed with an eye on
applying it in the financial arena.

FROM THE EDITOR 5
OPEN EXCHANGE letters, comments, questions 6
CONVERGENCE news, announcements, errata 7
NEXT-GENERATION TOOLS product announcements and news 32
QUESTIONNAIRE 35


***
May/June 1994
Vol.2, No.3
Title: Special Topic: Neural and Fuzzy Systems

Guest Editorial: Neurofuzzy Computing Technology

8
Francis Wong
The author presents an example neural network and fuzzy logic hybrid system,
and explains how integrating these two technologies can help overcome the
drawbacks of the other.

Neurofuzzy Hybrid Systems 11
James Derry
A large number of systems have been developed using the combination of
neural network and fuzzy logic technologies. Here is an overview on several
such systems.


Interpretation of Neural Network Outputs using Fuzzy Logic 15
Randall Caldwell
Using basic spreadsheet formulas, a fuzzy expert system is applied to the
task of interpreting multiple outputs from a neural network designed to
generate signals for trading the S&P 500 index.


Thoughts on Desirable Features for a Neural Network-based
Financial Trading System 19
Howard Bandy
The authors covers some of the fundamental issues faced by those planning
to develop a neural network-based financial trading system, and offers a list
of features that you might want to look for when purchasing a neural network
product.


Selecting the Right Fuzzy Logic Tool 23
(the editors)
Adding to our earlier selection guide on neural networks, we provide a list of
fuzzy logic products along with a few hints on which ones might most
interest you.


A Suggested Reference List: Recent Books of Interest 25
(the editors)
In response to readers' requests, we present a list of books, some of which
you will want to have for reference.


Product Review: CubiCalc Professional 2.0 28
Mark Thomason
A popular, fuzzy logic tool is reviewed. Is the product ready for investors


and traders? The answer may be somewhat fuzzy itself.
FROM THE EDITOR 5
OPEN EXCHANGE letters, comments, questions
CONVERGENCE news, announcements, errata
NEXT-GENERATION TOOLS product announcements and news
6
7
31

***
July/August 1994
Vol.2, No.4
Title: Special Topic: Neural and Genetic Systems

Guest Editorial: Neurogenetics and its use in Trading System Development
Jeffrey Katz and Donna McCormick
The authors discuss some of the differences between standard neural technology
and neurogenetics, and present a basic example of how an S&P trading system
might be developed using neurogenetics.

Neurogenetic Computing Technology 12
Francis Wong
Genetic algorithms can be usefully applied to the optimization of neural
networks for forecasting and classification problems. The author discusses


this general application area along with a specific financial application.

An Introduction to Genetic Algorithms: A Mutual Fund Screening Example 16
Richard J. Bauer, Jr.
The basic mechanics of genetic algorithms are covered. A mutual fund
screening example is used to illustrate the process and to suggest ways in
which the technology might be used to explore various trading strategies.

Selecting the Right Genetic Algorithm Tool 20
(the editors)
Adding to our earlier selection guides on neural networks and fuzzy logic
products, we provide a list of genetic algorithm products along with a few
hints on which ones might most interest you.

Nonlinear Trading System Costs: Dollars and Time 22
Mark Thomason
The Neurophyte Column: Elements of Interest to the Novice

Book Reviews: Three Books on Time Series Forecasting 26

M. Edward Borasky
A look at three books dedicated to the task of forecasting, with discussion on
the important but often-overlooked matter of statistical significance. "The
Forecasting Accuracy of Major Time Series Methods", Spyros Makridakis,
et al., " Nonlinear Modeling and Forecasting", Martin Casdagli and Stephen
Eubank (editors), and "Time Series Prediction," Andreas Weigend and Neil
Gershenfeld (editors).
Product Review: MicroGA 30
Steven Swernofsky
A genetic algorithm product for developing applications in C++ is reviewed.
It runs on the PC and the Mac, and includes an interesting C++ code generator.


FROM THE EDITOR 5
OPEN EXCHANGE letters, comments, questions 6
CONVERGENCE news, announcements, errata 7
MUSINGS OF NOTE 24
NEXT-GENERATION TOOLS product announcements and news 32
COMING UP IN FUTURE ISSUES Back Cover


***
September/October 1994
Vol.2, No.5
Title: Special Topic: Neural Network Design


Guest Editorial: A Neural Network Project Roadmap 7
James E. Bowen
An overview of relevant issues and considerations of neural network
development projects at a systems level.


Design of Neural Network-based Financial Forecasting Systems:
Data Selection and Data Processing 12
Randall Caldwell
An in-depth review of two tasks critical to neural network design are
presented, including metrics, parameters, and concerns of interest to
investors and traders.


Design Issues in Neural Network Development 21
Peter C. Davies
The author addresses several fundamental considerations to be made, during
the design phase of a project, when developing a neural network application.


Neural Network-based Trading System Design:
Prediction and Measurement Tasks 26
Howard B. Bandy
A discussion of four closely-related tasks fundamental to the design of a
neural network-based trading system is provided, along with a spreadsheet
implementation of a profitability tester.



Applying Nonlinear Financial Tools: Getting Started 33
Mark R. Thomason
The Neurophyte Column: Elements of Interest to the Novice

Book Review: Trading on the Edge 35
Sandy Warrick
The reviewer looks at a new, ambitious book, edited by Guido Deboeck, that
describes a wide variety of new technologies and techniques currently being
applied to the world's markets.

FROM THE EDITOR 4
OPEN EXCHANGE letters, comments, questions 5
CONVERGENCE news, announcements, errata 6
MUSINGS OF NOTE 25
NEXT-GENERATION TOOLS product announcements and news 39
COMING UP IN FUTURE ISSUES Back Cover

***
November/December 1994
Vol.2, No.6
Title: Neural Network Implementation

Implementation Issues in Neural Network Development 7
Peter C. Davies
This article addresses several fundamental considerations to be made, during
the implementation phase of a project, when developing a neural network
application.

Discriminant Analysis Versus Neural Networks in Credit Scoring 11

R.J. van Eyden and J.J.L. Cronje
The authors implementation of a neural network for comparison with multiple
discriminant analylsis, using a financial application.
Selecting the Right Neural Network Tool 16
(the editors)
In support of the task of implementation, we provide an update to our popular
guide to neural network products, along with comment on how readers might
select the one most appropriate for them.

A Basic Neural Network-based Trading System Development Project #1 23
Mark Thomason
The Neurophyte Column: Elements of Interest to the Novice

Product Review: AIM for Windows 28
Howard B. Bandy
The Reviewer looks at a modeling product that often compares itself with
neural networks, and provides an analysis of its performance using a
financial time series.

Book Reviews: Two recent book on finance and advanced technologies 32
Mark R. Thomason
Two recent single-author titles of particular interest to most readers are
reviewed, one on neural networks and one on genetic algorithsm. "Neural
Network Time Series Forecasting of Financial Markets" by E. Michael Azoff,
and "Genetic Algorithms and Investment Strategies" by Richard J. Bauer, Jr.

FROM THE EDITOR 4
OPEN EXCHANGE letters, comments, questions 5
CONVERGENCE news, announcements, errata 6
MUSINGS OF NOTE reflections on the literature 22
INFFC update on the First International Nonlinear Financial
Forecasting Competition 36
NEXT-GENERATION TOOLS product announcements and news 37
BACK ISSUES 39
COMING UP IN FUTURE ISSUES Back Cover


***
January/February 1995
Vol.3, No.1
Title: A Competitive Task

Extracting Meaning from a Neural Network 7

E. Michael Azoff
A method for performing weight perturbation ifferential analysis as a more
accurate approach than simple weight magnitude analysis is provided. This
supports the claim that the black box tag often attached to neural networks
by newcomers to the field is not an inherent property of the technology.
Induction: Learning Rules From Data (Part 1) 11
James F. Derry
The author embarks on the task of extracting expertise from databases for
the purpose of market analysis and forecasting, offering insight into a
useful tool that may have been overlooked by many investors.

Secondary Pre-processing 17
John Payne
A method for performing a second stage of neural network pre-processing is
suggested, using a group of standard functions (squares, square roots and
logarithms). The results are tested for empirical evidence of the usefulness
of the method.

A Basic Neural Network-based Trading System Development Project #2 23
Mark Thomason
The Neurophyte Column: Elements of Interest to the Novice

An Index to NEUROVEST JOURNAL: September 1993 to December 1994 16
Indexed by general topics is the material published in the Journal to date.

Product Review: Propagator for Windows 28
Howard B. Bandy
An inexpensive, stand alone, neural network development system based on the
backpropagation algorithm is reviewed. What are the high and low points of
this addition to the commercial neural network product base?

Book Review: An Introduction to the Bootstrap 32
Mark R. Thomason
A book by the inventor of a method for estimating distributions, parameters
and error rates is reviewed, including suggestions as to why the method is
relevant to investors and traders.

FROM THE EDITOR a competitive task 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
MUSINGS OF NOTE reflections on the literature 22
INFFC update on the First International Nonlinear Financial Forecasting
Competition 36
NEXT-GENERATION TOOLS product announcements and news 37
BACK ISSUES 39
COMING UP IN FUTURE ISSUES Back Cover

***
March/April 1995
Vol.3, No.2
Title: Special Focus: Performance Metrics

NNCM-94 7
Ypke Hiemstra
A review of the Neural Networks in the Capital Markets workshop held last
year in Pasedena, California on matters of interest to our readers.

Monitoring Forecast Performance Using the Breakeven Locus 8

E. Michael Azoff
A method for visual, qualitative analysis of trading system performance is

presented, including a practical example of its application.

Performance Metrics for Neural Network-based
Trading System Development 13
Randall B. Caldwell
An overview of prediction, neural network, and financial forecasting
performance metrics and methods is presented, along with strategies for their
application to neural network-based trading system development.

A Basic Neural Network-based Trading System Development Project #3 25
Mark R. Thomason
The Neurophyte Column: Elements of Interest to the Novice

The Stochastics Indicator: A New Perspective Using Neural Networks 31
Randall B. Caldwell
Technical Analytica: Technical Market Analysis and Insight

Product Review: Neuralyst for Windows 36
Howard B. Bandy
The latest version of this popular neural network development system, which
functions as an Excel add-in, is reviewed. New features include genetic
algorithms and trading system tools.

Book Review: Design, Testing, and Optimization of Trading Systems 39
Randall B. Caldwell
One of the few books to focus exclusively on the design and testing of trading
systems is reviewed in light of this issue’s special focus on the subject of
performance metrics.

FROM THE EDITOR performance metrics 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
MUSINGS OF NOTE reflections on the literature 24
REVIEWS IN BRIEF NeuroForecaster 4.0, The New Technical Trader 40
NEXT-GENERATION TOOLS product announcements and news 41
BACK ISSUES 43
COMING UP IN FUTURE ISSUES Back Cover

***
May/June 1995
Vol.3, No.3
Title: Special Topic: Chaos in the Markets

Meeting of the Society for Nonlinear Dynamics and Econometrics 7
Robert McClelland
A review of the 1995 SNDE meeting in New York on matters of interest
to our readers.

A Direct Approach to Forecasting Stock Equities
using Nonlinear Dynamics Modeling 8
Bernard V. Kessler
The first part of an introductory overview of the application of nonlinear
dynamics (NLD) and chaos theory to the prediction of stock market equity
prices is presented.

A Neural Network Supports the Chaotic Paradigm
for the S&P 500 Index 16
Mary E. Malliaris
Challenging the efficient market hypothesis and supporting those who claim
that they have found statistical evidence that a chaotic dynamics structure
underlies the market, this paper constructs a neural network which lends
support to the deterministic paradigm.

Chaos and Prediction Horizons in Silver Futures Trading 22
Ted W. Frison
The dynamical structure of a silver futures contract is determined. The
system has chaotic like behavior, the evidence coming from the Lyapunov
exponents.


A Basic Neural Network-based Trading System Development Project #4 30
Mark R. Thomason
The Neurophyte Column: Elements of Interest to the Novice

Product Reviews: Three Products on Chaos from the
Academic Software Library 35
Mark R. Thomason
Three introductory-level software tools on chaos analysis and demonstrations
are reviewed from a supplier of educational physics-related software. Chaos
Data Analyzer, Dynamics Workbench, and Chaos Demonstrations

Book Review: Fractal Market Analysis 38
Sandy Warrick
The follow-up to the popular title Chaos and Order in the Capital Markets is
reviewed in light of its author’s continuing study of chaotic financial market
behavior.

FROM THE EDITOR chaos in the markets 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
MUSINGS OF NOTE reflections on the literature 29
REVIEWS IN BRIEF Profiting from Chaos, Chaos Theory in the Financial
Markets 40
NEXT-GENERATION TOOLS product announcements and news 41
BACK ISSUES 43
COMING UP IN FUTURE ISSUES Back Cover

***
July/August 1995
Vol.3, No.4
Title: Going on Three

Supervised Evolution of the Neural Trader Component
of a Stock Portfolio Trading System (Part 1) 7
David L. March
A method is described for adjusting neural network weights in situations
where there is no advance knowledge about the correspondence between the
network input and output, and where the target objective or profit function
is stepwise instead of continuous.

Induction: Learning Rules From Data (Part 2) 13
James F. Derry
The author completes his report on extracting expertise from databases for
the purpose of market analysis and forecasting, offering insight into a
potentially useful tool that may have been overlooked by many investors.

The Adaptive Moving Average 18
Howard B. Bandy
In this issue of Technical Analytica the details of constructing and applying
adaptive moving averages to trading are described, along with explicit
mathematical and spreadsheet formulas.

A Basic Neural Network-based Trading System Development
Project #5 26
Mark R. Thomason
The Neurophyte column continues with details on training and selecting best
networks, this time completing the prediction component of the system.

User Survey '95: Results 30
A summary of the results to our first survey of readers regarding
commercial neural network products for financial applications is presented.

Product Review: Pattern Recognition Workbench 32
Howard B. Bandy
A new high-end neural network development system for the professional
is reviewed with an eye towards its application to finance.


Book Reviews: Two Books on Neural Networks and C++ 35
Mark R. Thomason and Sandy Warrick
Two recent titles on neural networks, both of which include
C++ software on disk, are separately reviewed. "Advanced Algorithms
for Neural Networks" by Timothy Masters, and "Neural Network and
Fuzzy Logic Applications in C/C++" by Stephen Welstead.


FROM THE EDITOR going on three 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
MUSINGS OF NOTE reflections on the literature 25
REVIEWS IN BRIEF NeuroShell 2 v.2, Momentum Data on CD-ROM 38
INFFC an update on the First International Nonlinear Financial
Forecasting Competition 39
NEXT-GENERATION TOOLS product announcements and news 41
BACK ISSUES 43
COMING UP IN FUTURE ISSUES Back Cover


***
September/October 1995
Vol.3, No.5
Title: Special Topic: Anything but Backpropagation


Neural Networks in Finance : Design and Applications,
Louvain-la-Neuve, Belgium 7
Eric de Bodt
An overview of a recent seminar on neural networks in finance is presented.


Backpropagation versus Conjugate Gradient Training Methods 8
Paul A. Billings
An alternative to backpropagation, with less critical "user-tunable"
parameters, is discussed. Benchmarks are generated to compare these two
algorithms for training multilayer perceptrons.


The General Regression Neural Network 13
Timothy Masters
An objective and intuitive look at the details of a neural network, as a
modification to probabilistic networks to allow for function mapping.
Both its strengths and weaknesses are discussed.


Supervised Evolution of the Neural Trader Component
of a Stock Portfolio Trading System (Part 2) 18
David L. March
The author completes his 2-part report on neural network traders with detailed
examples on applying the methods presented earlier to stock portfolio trading.


Improved Prediction Performance Metrics for
Neural Network-based Financial Forecasting Systems 22
Randall B. Caldwell
This paper presents a study of traditional and new measures for comparing the
prediction performance of neural network-based trading systems. Results
reported will be of significant interest to trading system developers using
neural networks.


Generating Principal Components using TimeStat 27
James Hampton
In this issue of Technical Analytica, the details for using a new freeware
product to generate principal components are described.


A Basic Neural Network-based Trading System Development
Project #6 29
Mark R. Thomason
The Neurophyte column continues with details on training and selecting best
networks, this time completing the prediction component of the system.


Product Review: Neural Network Tutor 36
Randall B. Caldwell
A new, unique product on learning neural networks is reviewed, including a



look at its built-in neural network simulator.


FROM THE EDITOR anything but backpropagation 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
MUSINGS OF NOTE reflections on the literature 28
REVIEWS IN BRIEF the new science of technical analysis, virtual trading 38
ESSAY AWARD CONTEST congratulating the winner of our first award 39
NEXT-GENERATION TOOLS product announcements and news 41
BACK ISSUES


***
November/December 1995
Vol.3, No.6
Title: Special Topic: Data Selection and Preprocessing


A Resource List:
Software, Books and Articles on Principal Components Analysis 7
(the editors)
In response to reader requests, we provide a resource on the subject.


Input Variable Set Diversity and a Neural Network’s
Financial Forecasting Ability 8
Andrew A. Kramer
This paper explores the ability of three different sets of input variables to
predict a biotechnology stock index, and compares the results using both
multilayer feedforward and generalized regression neural networks.


An Explicit Feature Selection Strategy for
Predictive Models of the S&P 500 Index 14
Tim Chenoweth and Zoran Obradovic
This paper focuses on the selection of an appropriate set of features for a
feedforward neural network model used to predict both future market direction
and future returns for the S&P 500 Index. Daily and monthly predictions of
returns and market direction are analyzed.


Three Methods of Neural Network Sensitivity Analysis for
Input Variable Reduction: A Case Study in Forecasting the
S&P 500 Index (Part 1) 22
Randall B. Caldwell
This paper examines three commonly-applied sensitivity analysis methods using
a financial forecasting problem for the S&P 500 index as an example.
Preliminary results indicate that financial practitioners and researchers
should consider the use of alternative sensitivity metrics to those commonly
employed.


The Fast Fourier Transform for Analyzing
Financial Time Series 26
James Hampton
In this issue of Technical Analytica, the author addresses the processing
steps necessary to apply FFTs to time series analysis by financial
practitioners. Issues regarding the stationarity and persistence of market
cycles are addressed. An approach to using FFTs and cyclic market
information as part of a data selection strategy for neural networks applied
to forecasting the Dow Jones 20-Bond Average index is presented.


Product Review: GeneHunter 34
Howard B. Bandy
A new, genetic algorithm add-in product for Excel is reviewed with the
financial practitioner in mind.


Book Reviews: Bayesian Forecasting and Artificial Life 38
James Hampton
Two books on entirely different subjects are reviewed for their relevance to
trading and investing. "Applied Bayesian Forecasting and Time Series
Analysis" by Andy Pole et al., and "Artificial Life: An Overview" by
Christopher Langton.



FROM THE EDITOR data selection and preprocessing 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
NEXT-GENERATION TOOLS product announcements and news 41
MUSINGS OF NOTE reflections on the literature 41
BACK ISSUES 43
COMING UP IN FUTURE ISSUES Back Cover


***
January/February 1996
Vol.4, No.1
Title: Addressing a Framework of Issues


INFFC Update 7
A summary of preliminary results of the first International Nonlinear
Financial Forecasting Competition is presented.


Forecasting the 30-year U.S. Treasury Bond with a
System of Neural Networks 10
Wei Cheng, Lorry Wagner, and Chien-Hua Lin
A forecasting model based on a system of artificial neural networks is
used to predict the direction of the 30-Year U.S. Treasury Bond on a weekly
basis. This paper describes the methods used for data selection, training
and testing, the basic system architecture, and how the decision model
improved the total system accuracy as compared to individual networks.


Three Methods of Neural Network Sensitivity Analysis for
Input Variable Reduction: A Case Study in Forecasting the
S&P 500 Index (Part 2) 16
Randall B. Caldwell
This paper concludes an examination of three commonly-applied sensitivity
analysis methods using a financial forecasting problem for the S&P 500 index
as an example. Preliminary results indicate that financial practitioners
should consider the use of alternative sensitivity metrics to those commonly
employed.


Rescaled Range Analysis:
Approaches for the Financial Practitioner (Part 1) 23
James Hampton
Technical Analytica: This paper begins an investigation of the application of
rescaled range (R/S) analysis techniques to analyzing financial time series.
An example using the S&P 500 daily index is utilized to illustrate the
material presented.


Principal Components Analysis for Neural Network Input
Variable Reduction and Financial Forecasting (Part 1) 29
Mark R. Thomason
The Neurophyte: Principal components analysis (PCA) has been successfully
applied to neural network-based systems in finance, particularly in the area
of dimension reduction and input variable selection. This paper presents an
objective analysis of PCA accessible to the financial practitioner and to
the applied researcher interested in exploring financial applications,
providing a foundation for future work on the subject.


Book Review: The Fuzzy Systems Handbook 33
James F. Derry
A popular and very readable tutorial on the subject of fuzzy systems,
complete with practical examples and C++ code, is reviewed.


An Index to the NEUROVEST JOURNAL:
September 1993 to December 1995 38
An index for locating past articles and reviews in volumes 1, 2 and 3.


FROM THE EDITOR addressing a framework of issues 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
REVIEWS IN BRIEF 36



chaos data analyzer professional
chaos and nonlinear dynamics in the financial markets
MUSINGS OF NOTE reflections on the literature 37
NEXT-GENERATION TOOLS product announcements and newsCOMING UP IN FUTURE ISSUES Back Cover

***
March/April 1996
Vol.4, No.2
Title: Special Topic: Visualization Tools for Complexity and Finance

Visualization Tools for Complexity and Finance
(or Looking Before We Leap) 7
James Hampton and Randall Caldwell
Before embarking on the development of new visualization tools for
implementation on advanced visualization platforms, it is important to
briefly review some of the tools and techniques currently available to us,
as practitioners and applied researchers in finance.

An Overview of Data Dimensions and Visualization 14
Brand Fortner
An introductory overview of data dimensions and visualization is presented.
The purpose is to illustrate the following: for all kinds of data, even
financial data, being fully aware of its dimensionality can be very helpful
to visualization and analysis tasks.

Visualization and Neural Network Tools under Linux 21
Kenneth Lin
Hundreds of powerful and useful software programs are available which run
under Linux, a public-domain version of Unix for the Intel x86 platform.
This introductory paper presents information on a few of those programs
which support visualization applications and neural network development.

A Visualization Technique for Selecting
Neural Network Trading Thresholds 25
James Hampton
Trading systems which use neural networks trained to predict future price
variances are often based upon a single pair of crossover thresholds as part
of a trading strategy. This paper proposes a visualization method that can
be used to greatly simplify the task of selecting the most robust yet
profitable trading thresholds based on common risk and reward measures.

Principal Components Analysis for Neural Network Input
Variable Reduction and Financial Forecasting (Part 2) 30
Mark R. Thomason
The Neurophyte: Principal components analysis (PCA) has been successfully
applied to neural network-based systems in finance, particularly in the area
of dimension reduction and input variable selection. This paper presents an
objective analysis of PCA accessible to the financial practitioner and to
the applied researcher interested in exploring financial applications,
providing a foundation for future work on the subject.

Product Review: MatLab and Neural Network Toolbox 35
Mark Thomason
The latest version of a popular program which integrates matrix computation,
numerical analysis, signal processing, data analysis, and graphics into a
common interactive environment is reviewed along with its neural network
add-on toolbox.

FROM THE EDITOR visualization tools for complexity and finance 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 5
REVIEWS IN BRIEF 38
the elements of graphing data
visualizing data
the visual display of quantitative information
PSI-Plot
MUSINGS OF NOTE reflections on the literature 40


NEXT-GENERATION TOOLS product announcements and news 41
COMING UP IN FUTURE ISSUES Back Cover

***
May/June 1996
Vol.4, No.3
Title: Back to Basics

Conference Report: CIFEr '96 7
Mario Bortoli
A report on the second conference on “Computational Intelligence in
Financial Engineering.

Applying Neural Networks and Genetic Algorithms to
Tactical Asset Allocation 8
Ypke Hiemstra
Tactical Asset Allocation (TAA) involves the prediction of asset class
returns and adjustment of the strategic portfolio. The paper claims that by
its very nature the return generating process is nonlinear, and presents a
neural network that applies a fundamental approach to predict the S&P500.
An optimization model using genetic algorithms exploits the predictions to
adjust the strategic portfolio.

Using a Fuzzy Logic Model for Portfolio Insurance of
Japanese Stocks 16
Kay-Hwang and Woon-Seng Gan
In this paper a portfolio insurance strategy based on Nikkei Stock Index
Futures is used to insure a portfolio of Japanese Stocks which has the
same component stocks as in the Nikkei 225 Stock Index. A new approach
using fuzzy logic is developed to decide when to rebalance the replicating
portfolio, and is compared with the conventional method which rebalances the
portfolio daily.

Rescaled Range Analysis:
Approaches for the Financial Practitioner (Part 2) 23
James Hampton
Technical Analytica: The second part of a paper which investigates the
application of rescaled range (R/S) analysis techniques to analyzing
financial time series. An example using the S&P 500 daily index is utilized
to illustrate the material presented.

Neural Network Input Variable Selection (Revisited) 30
Mark R. Thomason
The Neurophyte: Methods for selecting variables as inputs to neural
networks for financial forecasting purposes represent a subject of
considerable interest. This paper briefly elaborates on the topic and
discusses the related topics of multicolinearity, degrees of freedom and
performance metrics.

Product Review: NeuroGenetic Optimizer 35
Mark Thomason
The latest version of a neural network development system which uses genetic
algorithms to optimize network architectures and input variables is reviewed.
The software includes special features for time series prediction.

FROM THE EDITOR back to basics 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
REVIEWS IN BRIEF 38
the new money management
neural networks in the capital markets
MUSINGS OF NOTE reflections on the literature 39
NEXT-GENERATION TOOLS product announcements and news 40
COMING UP IN FUTURE ISSUES Back Cover

***
July/August 1996
Vol.4, No.4


Title: Special Topic: Nonstationary Analysis and Finance

INFFC Update 7
Manoel F. Tenorio and Randall B. Caldwell
The latest on the first International Nonlinear Financial Forecasting
Competition, along with final results for the Prediction Strategy entries.

Nonstationary Time-Series Forecasting within a
Neural Network Framework 9
Sara M. Abecasis and Evangelina S. Lapenta
Modeling and forecasting the behavior of univariate time series with the
back-propagation learning algorithm is presented in this paper.
Nonstationary time series were mapped to stationary ones by the use of the
power transformation. Some success was achieved regarding predictions based
on the validation data samples.

Nonstationary State Space Models for
Multivariate Financial Time Series: An Introduction 17
Mario Bortoli
A simple class of State Space Models is presented, as black-box, parametric,
stochastic and dynamic models that can be effectively used to describe the
dynamics of nonstationary multivariate time series. A brief comparison
between State Space Models and other techniques (Auto Regressive Integrated
Moving Average, Error Correction Models and Neural Networks) is proposed.
An example for predicting financial time series is presented.

Rescaled Range Analysis:
Approaches for the Financial Practitioner (Part 3) 27
James Hampton
Technical Analytica: The third part of a paper which investigates the
application of rescaled range (R/S) analysis techniques to analyzing
financial time series. An example using the S&P 500 daily index is utilized
to illustrate the material presented.

An Introduction to Nonstationary Analysis and
Financial Time Series Preprocessing 31
Mark R. Thomason
The Neurophyte: Reviewed is the procedure of price differencing for
financial time series, its use in conjunction with other preprocessing
techniques, its association with data stationarity in the context of the
time-frequency relationship, along with the autocorrelation function for
analysis. Discussion addresses practical considerations to be made when
applying filtered data as inputs to neural network predictors.

Product Review: Market Skill-Builder 37
James Hampton
A new tool for developing trading skills within a familiar spreadsheet
environment is reviewed.

FROM THE EDITOR nonstationary analysis and finance 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
REVIEWS IN BRIEF 39
a friendly guide to wavelets
resampling stats
MUSINGS OF NOTE reflections on the literature 40
NEXT-GENERATION TOOLS product announcements and news
COMING UP IN FUTURE ISSUES Back Cover

***
September/October 1996
Vol.4, No.5
Title: On Non-Traditional Tools

Regime Signaling Techniques for
Non-Stationary Time-Series Forecasting 7
Radu Drossu and Zoran Obradovic
An accuracy-based signaling technique as an alternative to statistics-based


signaling for detecting changes in a time series distribution is proposed.
The validity of the proposed technique is evaluated in the context of either
low-noise or high-noise, non-stationary time series.

Comparing Conventional and Artificial Neural Network Models
for the Pricing of Options on Futures 16
Paul Lajbcygier, Christopher Boek, Andrew Flitman and
Marimuthu Palaniswami
Pricing of American-style options on futures is compared using conventional
models and artificial neural networks. The conventional models used in the
evaluation are the Black-Scholes, the modified Black and the
Barone-Adesi/Whaley models, while the alternative considered are feedforward
artificial neural networks.

Rescaled Range Analysis:
Approaches for the Financial Practitioner (Part 4) 24
James Hampton
Technical Analytica: The final part of a paper which investigates the
application of rescaled range (R/S) analysis techniques to analyzing
financial time series. The series concludes with a discussion on applying
local Hurst estimates as inputs to neural network-based financial
forecasters.

Selecting the Right Neural Network Tool — Third Edition 33
(the editors)
Our first update on commercial neural network products in almost two years is
presented. Results from our survey of vendors indicate that, though there
are a few new players, the overall number has substantially decreased.

FROM THE EDITOR on non-traditional tools 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
REVIEWS IN BRIEF 38
artificial intelligence in finance & investing
neural networks in finance and investing
modelquest
neuroclassifier
MUSINGS OF NOTE reflections on the literature 40
NEXT-GENERATION TOOLS product announcements and news
COMING UP IN FUTURE ISSUES Back Cover

***
November/December 1996
Vol.4, No.6
Title: Predictors Anonymous

Neural Network Model Development and Optimization 7
Costas Siriopoulos and Raphael N. Markellos
This paper is concerned with applying artificial neural network (ANN) models
in forecasting financial time series. The methodology includes a model
development and optimization stage and the translation of forecasts into
investment timing decisions. Performance is evaluated in terms of both
statistical and economic significance. The use of BDS and R/S analysis
results for ANN modelling is explored.

Qualitative Information in Finance:
Natural Language Processing and Information Extraction 14
Marco Costantino, Russell J. Collingham and Richard G. Morgan
This article describes the importance of qualitative information in the
financial operators' investment decision-making process and how natural
language processing can be successfully used for processing and analyzing
such information. Natural language processing is briefly compared to other
artificial intelligence techniques which are widely employed in finance:
neural networks and expert systems.

Rough Sets Help Time the OEX 20
Chris Skalkos
Technical Analytica: This paper describes an application of rough sets


analysis to trading the OEX. Using rough sets techniques, a set of rules
for short-term trading the OEX based on the Hines indicator is extracted.
A system is then developed, encompassing all of the derived rules, in order
to evaluate trading system performance.

Application of Wavelet Filters to
Non-Stationary Financial Time Series 29
Mark R. Thomason
The Neurophyte: This paper proposes an application of the discrete wavelet
transform to the processing of nonstationary data within the context of
financial time series analysis and prediction. Shortcomings, limitations
and advantages of wavelets, with respect to filtering financial time series
for prediction applications, are discussed.

Product Review: ThinksPro 38
Mark R. Thomason
A new comprehensive neural network development system, with features for time
series analysis and an abundance of network options and parameters, is
reviewed.

FROM THE EDITOR predictors anonymous 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
REVIEWS IN BRIEF 41
pattern recognition and prediction with applications to signal
characterization
wavelet toolbox for matlab
an introduction to neural networks
an introduction to genetic algorithms
MUSINGS OF NOTE reflections on the literature 43
NEXT-GENERATION TOOLS product announcements and news 44
COMING UP IN FUTURE ISSUES Back Cover

***
January/February 1997
Vol.5, No.1
Title: Special Topic: Hybrid Neural Networks for Financial Forecasting

Guest Editorial: Hybrid Intelligence for Financial Forecasting 4
Zoran Obradovic
A look at the papers and the topic of this special issue by our guest editor.

A Neural-Fuzzy System for Financial Forecasting 7
Zuohong Pan, Xiaodi Liu and Olugbenga Mejabi
This paper introduces a hybrid Neural-Fuzzy system for financial modeling
and forecasting. The model’s performance is compared with a random walk
model, an ARIMA model, a regression model corrected for autocorrelation, a
regression corrected for autoregressive conditional heteroskedasticity, and
a regression model corrected for both autocorrelation and ARCH. The power
and predictive ability of the models are evaluated on the basis of mean
absolute error, root mean squared error, turning point prediction, pattern
recognition, correlation between output pattern and actual pattern, and
conditional efficiency.

A New Neural Network for Nonlinear Time-Series Modeling

16
Amir Hussain, John J.Soraghan and Tariq S. Durrani
This paper describes a new two-layer linear-in-the-parameters feedforward
network termed the Functionally Expanded Neural Network (FENN). The new
structure can be considered to be a hybrid neural network incorporating to a
variable extent the combined modeling capabilities of the conventional
Multi-Layered Perceptron (MLP), Radial Basis Function (RBF) and Volterra
Neural Networks (VNN) structures. Simulated chaotic Mackey-Glass time
series and real-world noisy, highly non-stationary sunspot and actual stock
market time series data are used to illustrate the superior modeling and
prediction performance of the FENN compared with other recently reported,
more complex feedforward and recurrent neural network based predictor
models.


The Pricing and Trading of Options using a
Hybrid Neural Network Model with Historical Volatility 27
Paul Lajbcygier, Andrew Flitman, Anthony Swan and Rob Hyndman
The residuals between conventional option pricing models and market prices
have persistent patterns or biases. The "hybrid" method models the residuals
using an artificial neural network. The pricing accuracy of the hybrid method
is demonstrated on real data using the Australian All Ordinaries Share Price
Index options on futures and is compared with all major competing
conventional models. The hybrid method is found to be both statistically and
economically superior to the conventional models alone.

A First Multi-Network Hybrid for Financial Forecasting 41
Mark R. Thomason
The Neurophyte: A basic approach to designing and analyzing multi-network
hybrids for financial forecasting is presented. The hybrid consists of the
combination of three MLP neural networks using simple linear combining
techniques. Several market indicators are used as network input variables
to forecast weekly S&P 500 prices at different horizons. The prediction and
trading performance of the hybrid network is compared with that of the
individual networks and a buy-and-hold trading strategy.

Product Review: The Financial Toolbox 45
Mark R. Thomason
A new toolbox from the maker of MatLab is reviewed with the interests of
practitioners and applied researchers in finance in mind.

FROM THE EDITOR hybrid neural networks for financial forecasting 5
OPEN EXCHANGE letters, comments and questions 6
CONVERGENCE news, announcements, addenda, errata 6
MUSINGS OF NOTE reflections on the literature 41
REVIEWS IN BRIEF 49
advances in knowledge discovery and data mining
NEXT-GENERATION TOOLS product announcements and news
COMING UP IN FUTURE ISSUES Back Cover

***
March/April 1997
Vol.5, No.2
Title: In Search of a Discipline

Forecasting the CHF-USD Exchange Rates
using Neural Networks 7
Jingtao Yao, Yili Li and Chew Lim Tan
A study of using neural networks to predict the exchange rates between Swiss
Francs and American Dollars. Results show that a simple backpropagationtrained
network with efficient learning and a simple set of technical
indicators as inputs serves well as a predictive model. Issues on the
frequency of sampling, choice of network architecture, forecasting periods,
and measures for evaluating the model's predictive power are discussed.

Improving Decision-Making in the Financial Markets
with the Probabilistic Neural Network Paradigm 14
Mike P. Foscolos and Sujinda Nilchan
This paper demonstrates the probabilistic neural network to be theoretically
and practically the most suitable neural network algorithm for financial
decision-making. The authors compare the decision-making ability of the
probabilistic algorithm with the commonly-applied standard backpropagation
algorithm and decisions formulated by fundamental financial analysts.

Time Synchronization of Technical Indicators
as Model Inputs 22
James Hampton
Technical Analytica: Published reports on techniques applied in support of
cycle analysis, technical analysis, and leading/lagging indicators often
rely on the frequently subjective interpretation of charts and charting
methods, and the existence of persistent, periodic market characteristics.
Here, several issues are reviewed regarding phase and the time


synchronization of variables applied to data-driven financial systems.

Financial Forecasting with Wavelet Filters
and Neural Networks 27
Mark R. Thomason
The Neurophyte: Band-pass filters based on wavelets for pre-processing
inputs to neural network-based financial forecasters are studied. Results
are compared with simple high-pass and low-pass filters. Results indicate
that, for the dataset and test period studied, the wavelet filters provide
improvement over the benchmark filters when used with neural networks for
forecasting the S&P 500 Index.

Product Review: WAVEWI$E Market Spreadsheet and Data Server
Edward Weiss
The new version of this spreadsheet application, with features specifically
designed for data manipulation, market analysis and testing trading systems,
is considered.

FROM THE EDITOR in search of a discipline 4
OPEN EXCHANGE letters, comments and questions 5
CONVERGENCE news, announcements, addenda, errata 6
REVIEWS IN BRIEF 39
analysis of observed chaotic data
neural network design
NEXT-GENERATION TOOLS product announcements and news
MUSINGS OF NOTE reflections on the literature 40
COMING UP IN FUTURE ISSUES Back Cover

***
May/June 1997
Vol.5, No.3
Title: Special Topic: Data Mining for Financial Applications

Database Mining/Knowledge Discovery in
Financial Databases: An Overview 5
James F. Derry
An overview of database mining methods and implementations is presented. Of
interest are financial applications in the areas of investing, trading,
stock selection and portfolio optimization. The use of software agents for
searching financial information on the Internet are addressed.

Sidebar: Rough Sets, Rough Neurons, Induction and
Data Mining #2 10
Edward Weiss

Self-Organizing Data Mining for a Portfolio Trading System 12
Frank Lemke and Johann-Adolf Mueller
This paper describes the application of data mining algorithms for a
portfolio trading system. Parametric models are adaptively created from
data by the Group Method of Data Handling (GMDH) in the form of networks of
optimized transfer functions. Nonparametric models are selected from a given
variable set by analog complexing, representing one or more patterns of a
trajectory of past behavior which are analogous to a chosen reference
pattern. The trading system simulates trading a portfolio of diverse stocks
using daily out-of-sample price data.

A Qualitative Approach to Pattern Identification for
Financial Data Mining 27
Mirko Dohnal
This paper considers Interest Rate (IR) and Purchasing Rate (PR) models as two
methods for forecasting exchange rates. An integrated model is created by
merging IR and PR models using qualitatively-degraded conventional equations.
Lists of all possible qualitative scenarios are generated as part of the case
study presented in this paper. Qualitative scenarios result in the
development of transition graphs, which capture all possible transitions
between the scenarios. Since transition graphs provide insight into possible
future market behavior, qualitative modelling can provide a tool for
financial forecasting.


Data Mining with the ADX Indicator using Neural Neteworks 37
James Hampton
Technical Analytica

Data Mining and Financial Forecasting with a
Probabilistic Neural Network 39
Mark R. Thomason
The Neurophyte

Product Review: Generator: The Financial Version 43
Edward Weiss

FROM THE EDITOR data mining for financial applications 4
MUSINGS OF NOTE reflections on the literature 26
REVIEWS IN BRIEF 45
rough sets and data mining: analysis of imprecise data
data mining with neural networks
data mining
OPEN EXCHANGE letters, comments and questions 47
CONVERGENCE news, announcements, addenda, errata 47
NEXT-GENERATION TOOLS product announcements and news
COMING UP IN FUTURE ISSUES Back Cover

***
July/August 1997
Vol.5, No.4
Title: Practice versus Research

Company Viability Prediction using Neural Networks
with Sparse Data 5
Jeroen van Bussel and Leo P.J. Veelenturf
In research related to neural networks, the quantity of data is often
restricted with respect to its dimension. This shortage of data occurs in
problems such as the forecasting of the financial condition of a company.
This paper describes the prediction of company viability using neural
networks. Because of the large dimensional input space and limited datasets,
three methods were examined for reducing the dimension of the input.

Predicting Deterministic Chaotic Time Series 14
Tim S. Hatamian
Auto-regressive (AR) linear prediction is a method commonly used to forecast
the future values for a time series generated by a linear-stationary system.
Extension of the method to (stationary) nonlinear systems requires a bit of
non-trivial work. The basic derivation of these methods and several examples
are discussed in the context of forecasting stock or futures prices.

The ABC's of BDS 22
Kenneth Lin
The BDS statistic represents a widely-used modern tool for testing serial
dependence in a time series. It has demonstrated capabilities for detecting
serial correlation even in difficult chaotic time series where others methods
fail. It can thus be an effective tool for determining the forecastability
of a time series. A very brief guide to the BDS test along with examples
and software is presented.

Market Volatility as a Leading Indicator 27
James Hampton
Technical Analytica

Residual Analysis for Neural Network Financial Predictors:
An Introduction 30
Mark R. Thomason
The Neurophyte

Product Review: Neural Connection 36
Edward Weiss


FROM THE EDITOR practice versus research 4
REVIEWS IN BRIEF 39
nonlinear dynamics and time series
chaos and order in the capital markets, 2nd edition
fixed income securities: tools for today's markets
OPEN EXCHANGE letters, comments and questions 41
CONVERGENCE news, announcements, addenda, errata 41
MUSINGS OF NOTE reflections on the literature 42

NEXT-GENERATION TOOLS product announcements and news

43
COMING UP IN FUTURE ISSUES Back Cover

***
September/October 1997
Vol.5, No.5
Title: On Qualitative versus Quantitative Methods

An Appraisal of Various Linear and Nonlinear Methods
Utilized for Combining Neural Network Predictors:
A Practitioner's Perspective 5
Robert J. Van Eyden
Seeing a phenomenon once does not mean it is de facto or de rigour. However,
multiple observations can lead to the conclusion that it is to be expected
or may be considered a natural consequence. The same notions hold for time
series forecasting especially in the financial markets where the data is
inherently noisy. In this arena, reliance on a single neural network result
could be deemed unacceptable. This study seeks to combine the results of a
collection of neural network forecasters in various manners to determine the
best combination method for forecasting the South African long bond rates.

Modelling the Merval Index with Neural Networks
and the Discrete Wavelet Transform 15
Sara M. Abecasis and Evangelina S. Lapenta
The crux of this research is the evaluation of the effectiveness of a neural
network implementation for modelling the share prices of the Argentine stock
index named MERVAL, taking into account the influence of different indices
of the New York Stock Exchange. Four methods of sensitivity analysis and the
Discrete Wavelet Transform are considered. Different metrics were applied
for the purpose of determining the performance of the neural networks
implemented.

Neural Network Approximation of Option-Pricing
Formulas for Analytically Intractable Option-Pricing Models 20
Michael Hanke
A new method which combines numerical approximation techniques and artificial
neural networks is used to approximate formulas for option prices and
derivatives. Using this method, highly precise analytical formulas can be
derived for option types (American, Asian, binaries,...) and models (GARCH,
stochastic volatility,...) that are otherwise analytically intractable.
Using the formulas derived according to this new approach, option prices and
greeks under these models can be computed instantaneously.

The Construction of Risk-Adjusted Returns as Target Variables 28
James Hampton
Technical Analytica

A Primer on Radial Basis Function Networks for
Financial Forecasting 32
Mark R. Thomason
The Neurophyte

Product Reviews: The Options Toolbox
The Black-Scholes and Beyond Interactive Toolkit 36
Mark R. Thomason

FROM THE EDITOR on quantitative versus qualitative methods

4
REVIEWS IN BRIEF 39


computational intelligence: a dynamic system perspective
modelquest miner, version 1.0
getting started in options, third edition
CONVERGENCE news, announcements, addenda, errata 40
OPEN EXCHANGE letters, comments and questions 41
MUSINGS OF NOTE reflections on the literature 42
NEXT-GENERATION TOOLS product announcements and news
COMING UP IN FUTURE ISSUES Back Cover

***
November/December 1997
Vol.5, No.6
Title: On the Science of Finance

Backtesting Trading Systems 5
Raphael N. Markellos
Several procedures are described that can be used to assess the historical
performance of trading systems on the basis of statistical and financial
criteria. These procedures range from informal graphical analysis to
sophisticated statistical techniques that employ GARCH modelling,
cointegration analysis and bootstrapping simulation.

Adaptive Supervised Learning Decision Networks
for Trading and Portfolio Management 11
Lei Xu and Yiu-Ming Cheung
A trading and portfolio management system is proposed, based on an Adaptive
Supervised Learning Decision Network, which learns the best past investment
decisions directly instead of making predictions first and then making
investment decisions based on the predictions. Without any additional effort,
this network can be realized directly utilizing any existing adaptive
supervised-learning neural network.

Multivariate Embedding Methods: Forecasting
High-Frequency Financial Data in the First INFFC 17
Carol Alexander and Ian Giblin
A forecasting method is described, where each point to be forecast is embedded
in an m-dimensional library made from historic data. The approach is based on
the well-known 'nearest neighbor' algorithm but there are important
differences, including the facility for multivariate embedding, the use of
predictor variables which may be different from the
embedding variables, and the ‘rolling library’ which is of a constant size
but is continuously updated as each successive point is forecast.

Rough Set Theory: The Basics (Part 1)
James Hampton
Technical Analytica
25
Multicollinearity Revisited
Mark R. Thomason
The Neurophyte
30
An Index to the Journal
Covering all issues of the Journal to present.
34
Product Review: ModelQuest Expert
Mark R. Thomason
38

FROM THE EDITOR on the science of finance 4
ESSAY AWARD recognizing the winning paper for 1997 33
REVIEWS IN BRIEF 41
an introduction to kolmogorov complexity and its applications, 2nd ed.
pattern recognition and neural networks
cyber investing, second edition
OPEN EXCHANGE letters, comments and questions 43
CONVERGENCE news, announcements, addenda, errata 44
MUSINGS OF NOTE reflections on the literature 45
NEXT-GENERATION TOOLS product announcements and news


COMING UP IN FUTURE ISSUES Back Cover

***
January/February 1998
Vol.6, No.1
Title: Special Topic: Improving Generalization for Nonlinear Financial
Forecasting Models

Wavelet-based Density Estimator Model for Forecasting

6
Zuohong Pan and Xiaodi Wang
A nonparametric model for financial time series forecasting is presented. To
address the issue of generalization in estimation, a density estimator based
on wavelets is first established. Then, information in the given data is
denoised through wavelet shrinkage to extract the true pattern, while
ignoring the disturbing noises.

Exploiting Local Relations as Soft Constraints to
Improve Forecasting 14
Andreas S. Weigend and Hans Georg Zimmermann
This paper introduces a new architecture for the development of predictive
models for financial data. On the output side, we predict dynamical variables
such as first derivatives and curvatures on different time spans. On the
input side, we propose a new internal preprocessing layer connected with a
diagonal matrix of positive weights to a layer of squashing functions.

The Use of Parsimonious Neural Networks for
Forecasting Financial Time Series 24
Robert Dorsey and Randall Sexton
A genetic algorithm is used for global search and, by modifying the
objective function, is used to simultaneously select a parsimonious
structure. The chosen structure often eliminates all connections to
unnecessary variables and thus identifies irrelevant variables. Models with
the complete architecture are compared to those with the reduced structure.
Based on the preliminary model analysis a composite model is constructed.

Adaptive Local Linear Models for Financial Time Series 32
Claudio Pizzi and Paolo Pellizzari
An adaptive local linear approach to model and forecast financial time
series is developed. Local Linear Approximation (LLA) is estimated by a
fuzzy weighted regression, where weights are essentially similarities
between vectors of lagged observations (patterns). Hence, forecasts are
mainly due to patterns that most resemble the vector containing the current
observation. The method represents a flexible tool both in modeling
nonlinearities and in coping with weak non-stationarities.

Rough Set Theory: The Basics (Part 2) 40
James Hampton
The Practitioner: Technology

Dynamic Normalization: Outliers and Time 43
Mark R. Thomason
The Practitioner: Method

Product Review: ROSETTA 45
James Hampton

GUEST EDITORIAL
improving generalization for nonlinear financial
forecasting models 4
FROM THE EDITOR on special issues 5
REVIEWS IN BRIEF 47
intelligent systems for finance and business
neuroshell easy predictor
financial modeling
cybernetic trading strategies
OPEN EXCHANGE letters, comments and questions 49
CONVERGENCE news, announcements, addenda, errata 50
MUSINGS OF NOTE reflections on the literature

50


NEXT-GENERATION TOOLS product announcements and news

***
March/April 1998
Vol.6, No.2
Title: Special Topic: New Directions in Financial Research

Wavelet-Based Feature Extraction and Decomposition Strategies
for Financial Forecasting 5
Alex Aussem, Jonathan Campbell, and Fionn Murtagh
A wavelet decomposition of the original time series, with an adaptation
accounting for the time-varying nature of the data, is first carried out
to decompose the data into varying scales of temporal resolution. In
transform space, a dynamic recurrent neural network (DRNN) is trained to
provide five-day ahead forecasts for the S&P500 closing prices.

A Genetic Adaptive Neural Network Approach to Pricing Options:
A Simulation Analysis 13

A. Jay White
This study examines a Genetic Adaptive Neural Network's (GANN) ability to
approximate a pre-specified option-pricing function. It is shown that the
GANN is able to approximate, to a high degree of accuracy, the complex,
nonlinear option-pricing function used to produce the simulated call
and put option prices.
Intelligent Stock Trading Decision Support System through the Integration
of Artificial Neural Networks and Fuzzy Delphi Models 24

R.J. Kuo, L.C.Lee and C.F.Lee
Most research on the stock market is limited to the study of quantitative
factors, such as price and volume data, instead of qualitative factors,
such as political effects. However, qualitative factors play a critical
role in the stock market environment. The proposed system consists of
four parts: (1) factors collection, (2) a quantitative model, (3) a
qualitative model, and (4) a decision integration.
Rough Set Theory: The Basics (Part 3) 35
James Hampton
The Practitioner: Technology

Predicting and Trading the Sharpe Ratio 38
Mark R. Thomason
The Practitioner: Method

Product Review: BrainMaker Professional/MMX 42
James Hampton
FROM THE EDITOR new directions in financial research 4
REVIEWS IN BRIEF 45
econometric inference using simulation techniques
pattern classification: a unified view of statistical
and neural approaches
intelligent systems and financial forecasting
OPEN EXCHANGE letters, comments and questions 47
CONVERGENCE news, announcements, addenda, errata 47
MUSINGS OF NOTE reflections on the literature 48

May/June 1998
Vol.6, No.3
Title: On Market Efficiency and the Internet

Hierarchical and Feed-Forward Fuzzy Logic Systems for Interest
Rate Prediction 5
Masoud Mohammadian, Mark Kingham and Bob Bignall
The development of novel hierarchical and feed-forward fuzzy logic systems
using genetic algorithms is discussed. The systems developed are used for
the prediction and modelling of fluctuations in interest rates in Australia.
A genetic algorithm is proposed as a method for learning the fuzzy rules. The


results from the hierarchical and feed-forward fuzzy logic systems are
compared.

Discovering Lawlike Regularities in Financial Time Series 12
Boris Kovalerchuk and Evgenii Vityaev
This paper seeks to discover regularities in financial time series using
Machine Methods for Discovering Regularities (MMDR) and a related "discovery"
software system. This is accomplished by combining mathematical logic and
probability theory in data mining. Discovered regularities were used for
forecasting a target variable, represented by the relative difference in
percent between today's closing price for the S&P 500 daily index and the
price five days ahead.

Application of Reasoning Neural Networks to Financial Swaps 27
Ray Tsaih, Wei-Kuang Chen and Yi-Ping Lin
This study investigated two learning procedures to see which is better at
extracting the trend of asset price movements. One is the Back Propagation
learning algorithm, the other is a learning procedure call Reasoning Networks
using Back Propagation. For this investigation, the application of these two
learning procedures to forecasting the trends of interest-rate swap midrates
is considered.

Model Validation by the Bootstrap 38
James Hampton and Edward Weiss
The Practitioner: Method and Tools

Product Review: S-PLUS 4.0 44
Mark R. Thomason

FROM THE EDITOR on market efficiency and the internet 4
OPEN EXCHANGE letters, comments and questions 48
CONVERGENCE news, announcements, addenda, errata 48
REVIEWS IN BRIEF 47
neural network data analysis using simulnet
constructing intelligent agents with java
MUSINGS OF NOTE reflections on the literature 49
NEXT-GENERATION TOOLS product announcements and news

***
July/August 1998
Vol.6, No.4
Title: Complexity and Dimensionality Reduction in Finance - Part 1

Complexity Reduction for Co-Trending Variables 6
Raphael N. Markellos and Terence C. Mills
Complexity reduction techniques for systems comprising co-trending variables
are commonly used by financial practitioners in the form of simple ratios.
The construction of ratios and ratio-based forecasting models and be
formalized and improved upon using cointegration analysis and error-
correction modeling, respectively. This paper reviews these methods and
discusses a complexity reduction example in finance.

Forecasting Financial Time Series Using Stacked Generalization
James V. Hansen and Ray D. Nelson
This paper explores the efficacy of stacking models that in tandem
accomplish data filtering and feature extraction, utilizing methods from
both the statistics and machine learning communities. A meta-algorithm
is provided along with evidence on reduction in the dimensionality of
the search presented to the highest-level generalizer.

Applying Quantitative Representations to Data Mining in Financial
Time-Series Databases 25
Xuemei Shi and Man-Chung Chan
One quantitative approach to data mining involves the extraction of
general patterns from massive original data in terms of qualitative and
linguistic variables. A critical problem associated with applying
qualitative representations to time-series data is maintaining linguistic
variables which are consistent over time. A new technique is proposed in


this paper for solving this problem.

Non-Traditional PCA for Dimensionality Reduction of Financial
Forecasting Models 34
Mark R. Thomason
The Practitioner: Method and Tools

Product Review: SIMSTAT for Windows 40
Mark R. Thomason

GUEST EDITORIAL complexity and dimensionality reduction

in finance - part 1 4
FROM THE EDITOR 5
REVIEWS IN BRIEF 42

evolutionary algorithms in engineering applications
psi-plot and pro-stat
OPEN EXCHANGE letters, comments and questions 43
CONVERGENCE news, announcements, addenda, errata 43
MUSINGS OF NOTE reflections on the literature 44
NEXT-GENERATION TOOLS product announcements and news

***
September/October 1998
Vol.6, No.5
Title: Complexity and Dimensionality Reduction in Finance - Part 2

Bayesian Ying-Yang Dimension Reduction and Determination 6
Lei Xu
A new general theory is proposed for dimension reduction and determination
(DRD), based on the so-called Bayesian Ying-Yang (BYY) learning theory
developed in recent years. Examples presented include (a) a new algorithm
for factor analysis in both batch and adaptive modes, (b) criteria for
determining the number of factors and the dimension of the PCA subspace,

(c) a procedure for implementing a specific nonlinear BYY DRD based on
gaussian mixtures, and (d) extensions for auto-association and LMSER-based
nonlinear PCA. Some experimental results are provided.
Time Deformation: Definition and Comparisons 19
Gaelle Le Fol and Ludovic Mercier
The practical importance of time deformation is to give a preprocessing
technique to obtain a regularly spaced grid of data. A new trading strategy
in which the trading timepoints are endogenous to prices is presented. It
is shown that a changing timescale can improve daily gains.

Identifying Irrelevant Input Variables in Chaotic Time Series Problems:
Using a Genetic Algorithm for Training Neural Networks 34
Randall S. Sexton
Because gradient search techniques are incapable of identifying unneeded
weights in a solution, researchers have not been able to distinguish
contributing inputs from those that are irrelevant. By using a global
search technique (the genetic algorithm) for neural network optimization,
it is possible to identify unneeded network weights and, thus, irrelevant
input variables. This paper demonstrates, through an intensive Monte Carlo
study, that the genetic algorithm can be utilized to automatically reduce
the dimensionality of neural network models during network optimization.

Reducing Serial Bias of Direction-Oriented Forecasting Metrics 42
Mark R. Thomason and Randall B. Caldwell
The Practitioner: Method
Most financial forecasting performance criteria of practical benefit are
functions of market direction. However, performance criteria that reward
correct forecasts of market direction will naturally over-estimate
performance on datasets that exhibit significant serial dependency in
market direction. Here, two of many possible approaches for working with
performance measures that are inherently biased in trending markets are
considered.

Product Review: NeuroShell Trader 48


Mark R. Thomason
GUEST EDITORIAL complexity and dimensionality reduction

in finance - part 2 4
FROM THE EDITOR 5
OPEN EXCHANGE letters, comments and questions 53
REVIEWS IN BRIEF 52
s+garch and s+wavelets (software)
MUSINGS OF NOTE reflections on the literature 53

***
November/December 1998
Vol.6, No.6
Title: On Intra-Market Analysis

Self-Organizing Maps for Data Analysis: An Application to the Belgian
Leasing Market 5
Eric de Bodt, Emmanuel-Frederic Henrion, Marie Cottrell, and
Charles Van Wymeersch
Self-Organizing Maps (SOM) have been used a great deal for data analysis in
recent years. Here, we propose an application to a large real dataset,
composed of the finanical ratios of more than 12,000 Belgian companies.
The objective of the study is to understand the role of leasing as a
financing tool at the disposal of companies. The results clearly emphasize
that the nonlinear and robust properties of SOM make this tool very useful
for gaining a deeper understanding of the financing behavior of firms
through the analysis of their accounting data.

Building a Warrant Trading System using Hierarchical Neural Networks 25
Kwok-fai Cheung and Kin-hong Wong
In this paper, a warrant trading system based on the warrant sensitivity
formula is proposed. The estimation of parameter functions of the warrant
sensitivity model is carried out by two methods: (1) computed analytically,

(2) estimated by a hierarchical Correlation Basis Function (CBF) network.
From our simulation results using 43 warrants, both the hierarchical CBF
network trading system and the CBF network valuation trading system can
outperform the analytical Black-Scholes formula and the warrant sensitivity
formula respectively with regard to profitability.
Optimization of a Trading System using Global Search Techniques and
Local Optimization 36
Donald L. Iglehart and Siegfried Voessner
In this paper, we present a Hybrid Algorithm (HA) that combines a robust
genetic algorithm (GA) with a local optimization technique (LOT). The LOT
uses a quasi-Newton algorithm (QNA) for continuous variables and a
hill-climbing algorithm (HCA) for discrete variables. HA is applied to a
rule-based system for trading the S&P500 Index using daily closing prices.
The HA, which we compare to other algorithms, is shown to improve the
performance of this trading system in a reasonable amount of computer time
without using any previous knowledge of good parameter values.

Risk Management: The Equity Curve Revisited 47
James Hampton
The Practitioner: Method and Tools
Parameters that describe risk management criteria will naturally vary
among investors and traders. One popular indicator for measuring risk is
maximum equity drawdown. This article takes a new look at equity curves and
drawdown as part of an investigation that encompasses equity variance and
trendlines.

Product Review: BioComp Profit 51
Mark R. Thomason

FROM THE EDITOR on intra-market relationships 4
REVIEWS IN BRIEF 56
a guide to econometrics, 4th edition (book)
MUSINGS OF NOTE reflections on the literature 56


***
January/February 1999
Vol.7, No.1
Title: On Frequentist versus Bayesian Inferencing


Predicting Real Estate Returns using Neural Networks 5
Rakesh Bharati, Vijay S. Desai, and Manoj Gupta
Examined are the predictability of returns on real estate assets by
employing variables used in the finance and economics literature. Rather
than using conventional linear regression models to predict returns, a class
of nonlinear models, namely neural networks, are used. The use of neural
networks is motivated by the statistical evidence of neglected nonlinearity
reported in this paper. A variety of methods for testing nonlinearity are
employed.


Chaotic Prediction Applied to Financial Time Series 16
Carlos. A. Thompson, Claudio F. Silva, and Fabio Hochleitner
This paper deals with the development of a nonlinear Chaotic Prediction
Method (CPM) to calculate the one-day-ahead forecasts for several values of
the learning set size s, the maximum memory p and the retained dominant
modes d. A software package especially developed for this work demonstrates,
throughout computer experiments, that the predicted values strongly depend
on the variation of these parameters. Artificial Neural Networks (ANN) are
also used as an independent tool to estimate the time series data.


Neural Networks vs. Black-Scholes: An Empirical Comparison
of the Pricing Accuracy of Two Fundamentally Different Option
Pricing Methods 26
Michael Hanke
The aim of this paper is to empirically compare the pricing accuracy of the
Black-Scholes formula to that of option pricing formulas approximated by
neural networks. After demonstrating that previous comparisons found in the
literature do not distinguish between forecasting and pricing capabilities of
neural networks, it is shown that even in a framework that is advantageous
for the Black-Scholes model, neural networks prove superior in terms of
pricing accuracy.


Cointegration 101 35
James Hampton
The Practitioner: Method and Tools
Cointegration, as a tool for removing nonstationarity and reducing model
dimensionality, can be perplexing to any new user. This article provides a
brief overview of cointegration, unit root tests and error correction models
for the purpose of preparing practitioners for the effort that may be
required of them should they want to explore this tool in detail.


Product Review: Time Series and Forecasting for SimStat 39
Mark R. Thomason


Journal Index 42
An index to journal articles and reviews for Volumes 1 - 6.


FROM THE EDITOR on frequentist versus bayesian inferencing


4
ADDENDA AND ERRATA 41
REVIEWS IN BRIEF 47
industrial applications of neural networks (two books of the same title)
resampling stats for windows 95
MUSINGS OF NOTE reflections on the literature

49
ESSAY AWARD ACKNOWLEDGMENTS

51

***
March/April 1999
Vol.7, No.2
Title: Financial News Analysis using Distributed Data Mining

Text Processing for Classification 6
Vincent Cho, Beat Wüthrich and Jian Zhang


These days textual information becomes increasingly available through the
Web. This makes text an attractive resource from which to mine knowledge.
The major difficulty in mining textual data is that the information is
unstructured. Hence the data has to be preprocessed first so as to obtain
some form of structured data which is amenable to data mining techniques.
This paper focuses on this preprocessing step. The prediction accuracy
achieved by the best text processing method is very close to what can be
expected by human experts.

Analysis of Dealers' Processing Financial News
Based on an Artificial Market Approach 23
Kiyoshi Izumi and Kazuhiro Ueda
In this study we used a new agent-based approach to analyze the ways that
dealers in a foreign exchange market process the information in financial
news. An artificial market model is constructed using a Genetic Algorithm.
Using the simulation results, we classified, according to the ways that
agents regard the news, three categories of news data. We conclude that
emergent phenomena can be explained by the phase transition of forecast
variety, which is due to the interaction of agent forecasts and the
demand-supply balance.

IE-Expert: Integrating Natural Language Processing and Expert System
Techniques For Real-Time Equity Derivatives Trading 34
Marco Costantino
Quantitative data are today largely analyzed by automatic computer programs
based on traditional or artificial intelligent techniques, which provide
traders with quantitative information that helps them hedge their risks.
Qualitative data and, in particular, articles from on-line news agencies
are instead not yet successfully processed. As a result, financial operators,
notably traders, suffer from qualitative data-overload. This paper describes
how Natural Language Processing, Information Extraction and Expert Systems
can be used for reducing the traders' qualitative information overload.

Mining Financial News 53
James Hampton
The Practitioner: Method and Tools
Data-driven market-forecasting tools primarily rely on quantitative
information. Such information conforms well to forecasting models that are
developed using algorithms which sequence through explicit, discrete samples
of numbers. However, because of this, an abundance of potentially beneficial
market information in the form of textual and non-periodic financial news
is largely overlooked by most active investors and traders. Here, we take a
look at how we might utilize this information for financial forecast
modeling.

Product Review: Matlab and the Financial Toolbox 55
Mark R. Thomason

GUEST EDITORIAL financial news analysis using distributed data mining 4

FROM THE EDITOR 5
REVIEWS IN BRIEF 59
wordstat v1.2
MUSINGS OF NOTE reflections on the literature 59

***
May/June 1999
Vol.7, No.3
Title: On Global Minds and Markets

Utilization of Vector Autoregressions and Neural Networks
in Identifying the Return Interaction among Global,
Asia-Pacific Regional and Local Stock Markets 5
Chih-Chou Chiu and Yin-Hua Yeh
This study investigates the interactions in the returns of the Global, Asia
Pacific regional, and Local stock markets using the vector autoregressions
(VAR) and artificial neural networks. As the results reveal, influences on
the Hong Kong and Singapore stock markets by the stock market of South East


Asia do exist prior to the financial crisis in July 1997. This finding may
explain why the stock market in South East Asia affected the stock markets
in other Asian countries after the financial crisis.

The Wavelet Transform for Filtering Financial Data Streams 18
Zheng Gonghui, Jean-Luc Starck, Jonathan Campbell, and Fionn Murtagh
Relating this work to earlier results, the authors introduce a new wavelet
transform, the Haar à trous transform. Its advantages for modeling and
predicting financial data streams are described. The basic principles of
of decomposing the financial signal into scale-related components and fusing
the forecasts at each scale remain the same. The denoising of time series
data is also discussed. A multilayer perceptron is used to provide
predictions, and to demonstrate the advantages of the new wavelet transform
and wavelet-based denoising.

A Basic Neural Network-based Trading System
Project Revisited (Parts 1 and 2) 36
Mark R. Thomason
Due to reader interest, we revisit The Neurophyte, one of the most popular
series ever published on the application of neural networks in finance for
the novice. This included a neural network-based trading system project
published in 6 parts. In this issue, we present an updated version of
parts 1 and 2 of that project, published in November 1994 and January 1995,
respectively.

Product Review: e Professional version 1.3 46
Mark R. Thomason

FROM THE EDITOR 4
REVIEWS IN BRIEF 49
specifying and diagnostically testing econometric models, second ed.
MUSINGS OF NOTE reflections on the literature 49

***
July/August 1999
Vol.7, No.4
Title: On Performance Metrics

Performance Metrics for Financial Time Series Forecasting 5
Sara M. Abecasis, Evangelina S. Lapenta and Carlos E. Pedreira
In this paper the state of the art of performance metrics for financial time
series forecasting is presented. The focus of interest is centered on
prediction performance. However, part of the paper addresses the relevance of
metrics to trading performance. Characteristics of prediction performance
metrics are described. After we present the nomenclature, we describe each of
the performance metrics in detail. Characteristics of interest to financial
time series forecasting are noted. Finally, a survey on univariate and
multivariate financial time series is presented. Our purpose is to provide a
review of published research in this area as well as an opening for future
research.

Multi-Agent Approach as a Catalyst to a
Dynamic Financial Knowledge Discovery Process 24
Soe-Tsyr Yuan
Currently, most KDD research is focused on the automation of data mining,
although users still setup up and integrate other processes (such as data
collection and data engineering) manually. When confronting dynamic KDD
extensive manual effort. Basically, dynamic KDD applications are
characterized by dynamic data hunting and dynamic mining. Therefore, in the
search for a generation of flexible KDD applications, what should the KDD
flexible KDD applications? Our hypothesis is that the multi-agent approach
fills this role perfectly. We support this hypothesis through our
demonstration here of a cooperative information system for automating
dynamic KDD applications from a large amount of stock data using multi-agent
technology.

A Basic Neural Network-based Trading System


Project Revisited (Parts 3 and 4) 35
Mark R. Thomason
Due to reader interest, we revisit The Neurophyte, one of the most popular
series ever published on the application of neural networks in finance for
the novice. This included a neural network-based trading system project
published in 6 parts. In this issue, we present an updated version of
parts 3 and 4 of that project, published in March and May 1995, respectively.

Product Review: NeuroShell Predictor, Classifier and
Run-Time Server 46
Mark R. Thomason

FROM THE EDITOR on performance metrics 4
REVIEWS IN BRIEF 49
neural smithing
MUSINGS OF NOTE reflections on the literature 49

***
September/October 1999
Vol.7, No.5
Title: Advancements in Option Pricing Using Computational Intelligence

Literature Review:
The Problem With Modern Parametric Option Pricing 6
Paul Lajbcygier
Conventional parametric option-pricing models based on the Black-Scholes
have been generalized to form a new class of models referred to as the
modern parametric option-pricing models. The aim of this literature review
is to introduce and critique the modern parametric option-pricing models.
Conventional option pricing, although very accurate, has been shown to be
persistently, systematically and significantly biased. In the hope of
rectifying these biases, the assumptions of the conventional parametric
option-pricing models (OPMs) have been generalized to produce the modern
parametric OPMs.

Extraction of Intraday Implied Probability Distributions
in Illiquid Option Markets 24
Fernando Gonzalez and Neil Burgess
This paper describes a method for recovering the risk neutral market's
perceived probability distribution (RND) of European options on the FTSE100
Index in an hourly time basis. A nonparametric procedure is used to choose
probabilities that minimize an objective function subject to requiring that
the obtained probabilities comply with observed option prices. The
optimization technique for estimating probability distributions incorporates
a 'smoothness' and a 'variability' factor in the objective function to
account for situations where little smoothness and high variability in the
posterior distributions are plausible due to problems in the data.

Adaptive Hybrid Neural Network Option Pricing 33
Michael Hanke
Standard option pricing models show well-known deviations when compared to
market prices. The best-known of these phenomena is the smile in implied
market volatilities calculated from the Black/Scholes formula. In this paper,
feedforward networks are used in an adaptive fashion to fit the smile on a
day-to-day basis. This approach has some advantages compared to designs
previously used in the literature, e.g. drastically reduced training times
through a smaller number of parameters, resulting from the reduction of the
input space dimension to one and smaller network sizes.

GUEST EDITORIAL advancements in option pricing using
computational intelligence 4
FROM THE EDITOR 5
REVIEWS IN BRIEF 40
computation, causation and discovery
MUSINGS OF NOTE reflections on the literature 40

***


November/December 1999
Vol.7, No.6
Title: Advancements in Option Pricing Using Computational Intelligence
(Part 2)

Literature Review:
The non-parametric models 6
Paul Lajbcygier
Empirical option pricing is going through a crisis. Once, the seminal
Black-Scholes model was thought to be the last word on option pricing:
all that was needed, it was thought, was some adjustments and it could then
be applied to any new instrument: futures, foreign exchange and bonds.
However, in the past decade, increases in the bias of the Black-Scholes
model (and the conventional parametric option pricing models derived
using similar approaches) have led researchers to develop new models (coined
modern parametric option pricing models).

An Artificial Neural Network Approach to the Valuation
of Options and Forecasting of Volatility 19
David S. Geigle and Jay E. Aronson
Using data from the S&P 500 futures options from 1991 through 1996,
artificial neural networks were trained to estimate the value of an option
and forecast volatility of the underlying futures contract. Using the same
variables as are used in the Black-Scholes and ISD formulas, ten artificial
neural networks were trained in the valuation of an option and three
artificial neural networks were trained in the forecasting of future
volatility. The results of the artificial neural networks were compared to
actual prices and the Black-Scholes results for the valuation analysis and
to realized volatility, historical volatility and ISD for the volatility
forecast analysis. The artificial neural networks performed well in both
evaluations.

Option Pricing with the Genetic Programming Approach 26
Christian Keber
In this paper we derive analytical approximations for the valuation of
American put options on non-dividend paying stocks using the genetic
programming approach. Using experimental data sets we can show that the
genetically determined formulas outperform other formulas presented in the
literature. Furthermore, we derive a pure analytical approximation for
determining the killing price used in several classical option valuation
models. We can show that the results obtained by our formula are very close
to the numerically calculated killing prices.

A Basic Neural Network-based Trading System
Project Revisited (Parts 5 and 6) 37
Mark R. Thomason
Due to reader interest, we revisit The Neurophyte, one of the most popular
series ever published on the application of neural networks in finance for
the novice. This included a neural network-based trading system project
published in 6 parts. In this issue, we present an updated version of
parts 5 and 6 of that project, first published in July and September 1995,
respectively.

GUEST EDITORIAL advancements in option pricing using
computational intelligence (part 2) 4
FROM THE EDITOR 5
REVIEWS IN BRIEF 47
derivatives: a powerplus picture book
MUSINGS OF NOTE reflections on the literature 47