------- Forwarded Message Follows -------
From RBCALDWELL(a)DELPHI.COM Sun Mar 19 17:27:57 2000
From: RBCALDWELL(a)DELPHI.COM
To: "SNDE Mailing List"
<SNDE(a)fas-econ.rutgers.edu>
Subject: JCIFinance - Final CFP - Special issue on "Improving Generalization
Date: Thu, 22 May 1997 20:36:11 -0400 (EDT)
Status: RO
X-Status:
X-Keywords:
X-UID: 265
*******************************************************************
F I N A L C A L L F O R P A P E R S
*******************************************************************
Journal of Computational Intelligence in Finance
Final Call for Papers
Special Issue and Competition on
"Improving Generalization for Nonlinear Financial Forecasting Models"
The Journal of Computational Intelligence in Finance, a peer-reviewed
technical journal, published by Finance & Technology Publishing, is
seeking papers for review and publication in 1997 on "Improving
Generalization for Nonlinear Financial Forecasting Models". For
comparison of methods submitted, the target variable series and
performance metrics are specified (though not required).
The Journal of Computational Intelligence in Finance publishes applied
research and practical applications of high quality that are based on
sound theoretical, empirical or quantitative analysis. It provides the
international forum for the convergence of the new multi-disciplined
field of computational intelligence in finance.
Papers published in the Journal are eligible for entry in an Annual
Essay Award Contest. The Editorial Advisory Board of the Journal
selects the best paper for which a cash award is presented each year.
EDITORIAL ADVISORY BOARD
Emilio Barucci, University of Florence - Italy
Richard J. Bauer, Jr., St. Mary's University, Texas - USA
Neil Burgess, London Business School - UK
Oscar Castillo, UABC University - USA
Jerry Connor, London Business School - UK
Eric de Bodt, Universite Catholique de Louvain - France
James F. Derry, Mgmt. Engineering Productivity Systems - USA
Athanasios Episcopos, National Bank of Greece
Andrew Flitman, Monash University - Australia
Susan Garavaglia, Dun and Bradstreet - USA
Ramo Gencay, University of Windor - Canada
Sabyasachi Ghoshray, Florida International University - USA
Lee Giles, NEC Research Institute - USA
Christian Haefke, University of California at San Diego - USA
Ypke Hiemstra, Vrije Universiteit - The Netherlands
Yuval Lirov, Lehman Brothers - USA
Ralph Neuneier, Siemens AG Corporate Research Center - Germany
Zoran Obradovic, Washington State University - USA
Marimuthu Palaniswami, University of Melbourne - Australia
Carlos E. Pedreira, Catholic University, Rio - Brazil
David B. Skalak, University of Massachusetts - USA
Stephen Slade, Stern Business School, New York University - USA
Leon Sterling, University of Melbourne - Australia
Manoel F. Tenorio, University of Purdue - USA
Halbert White, University of California at San Diego - USA
Lei Xu, The Chinese University of Hong Kong
SPECIAL TOPIC
Improving Generalization for Nonlinear Financial Forecasting Models
PUBLICATION DATE
November 1997
PAPER SUBMISSION DEADLINE
June 30, 1997
MOTIVATION
The critical issue in applying neural networks and other data-driven
forecasting systems is generalization, the performance on data not used
for training. The key to generalization behavior is model complexity.
Too simple a model cannot approximate the true relationship, and overly
complex models adjust to the noise in the data. Nearly all financial
applications of nonparametric models (such as neural networks and genetic
algorithms) vary model complexity by adjusting the number of parameters.
This special issue intends to highlight other methods to improve
generalization, in particular regularization (e.g., neural network
weight decay and smoothing) and techniques for combining models. Of
particular interest are nonlinear methods including neural networks,
genetic algorithms, nearest neighbor networks, polynomial networks,
fuzzy logic, and hybrids.
Nearly all studies apply cross-validation to select the best model.
Alternatives to cross-validation include 'analytical' selection rules
such as Akaike's Information Criterion, Schwartz's Information Criterion,
and a number of others. Of particular interest are the statistical
properties (i.e., bias and variance) of model selection methods in
estimating out-of-sample performance.
DATA, TARGET VARIABLES and PERFORMANCE METRICS
Data: daily prices of a financial time series (see below)
Target Variable: the relative difference in percent (RDP) between
today's closing price and the price five (5) days ahead
Performance Metrics: MSE (target). nRMSE and DS (to be used in the
analysis).
Participants are encouraged to use the forecast data, target variable and
performance metrics specified for this special issue, which are available
on the Web to those who submit a satisfactory abstract (including brief
biography) as outlined below. Participants are not be restricted regarding
the data used as inputs to their predictors. Especially interesting
original methods using other forecast data, target variables and
performance metrics will also be considered.
The forecast series is derived from daily closing prices for a financial
time series. The target variable is the relative difference in
percent (RDP) between today's closing price and the closing price
five (5) days ahead. The date, the underlying price series and the
target variable series are all provided in the downloadable data file.
The target metric is the MSE. Also, authors' analysis should include
the normalized RMSE (RMSE normalized using the standard deviation of
actual RDP values), and Directional Symmetry (percentage of correctly
predicted directions with respect to the target variable).
The forecast data provided is separated into in-sample (10 years of
daily data) and out-of-sample (2 years of daily data) sets. Participants
are not restricted regarding the data used as input to their predictors.
However, all data used should be disclosed in the paper presentaton,
including the details of all techniques and formulas used to pre-process
the data. Details on the predictor and the methods used for improving
generalization should be presented in the paper.
FORECAST HORIZON AND RE-TRAINING
Participants should test performance of their predictors over the entire
two-year out-of-sample dataset. Of interest are results of analyses and
performance of predictors over the entire two-year prediction period:
(1) without re-training and
(2) with re-training (optional).
The results from (1) and (2) can be useful for estimating the limits
of the forecasting horizon for the prediction methods presented.
For additional details on the forecast data, target variable and
performance metrics, see:
http://ourworld.compuserve.com/homepages/ftpub/call.htm
Suggested references for the topic include:
1) Abu-Mostafa, J.S. [1990] "Learning from hints in neural networks",
Journal of Complexity, 6, June, pp. 192.
2) Bishop C.M. [1995] Neural Networks for Pattern Recognition, Oxford
University Press.
3) Caldwell, R.B. (editor) [1997] Nonlinear Financial Forecasting:
Proceedings of the First INFFC, Finance & Technology Publishing.
4) Elder, John F. and Mark T. Finn [1991] "Creating `Optimally Complex`
Models for Forecasting," Financial Analysts Journal, Jan/Feb, pp. 73-79.
5) Haykin, Simon [1994] Neural Networks: A Comprehensive Foundation,
IEEE Press.
6) Ripley, Brian D. [1996] Pattern Recognition and Neural Networks,
Cambridge University Press.
7) Swanson, N.R. and H. White [1995] "A Model Selection Approach to
Assessing the Information in the Term Structure Using Linear Models
and Artificial Neural Networks", Journal of Business and Economic
Statistics 13.
ABSTRACTS
Submit 150 to 300 word abstract including full name(s) and
affiliation(s) of the author(s), complete mailing address,
email address and telephone numbers of all authors. Authors
should provide a brief biographic sketch of themselves. Send
to either the postal or email addresses below:
Post:
Editors
JCIF
P.O. Box 764
Haymarket, VA 20168
USA
E-mail:
72672.261(a)compuserve.com
PAPERS
Submit three copies of each paper. Papers should be double-
spaced, single-sided. Authors should provide a brief
biographic sketch of themselves. Each copy submitted should
include a page that contains the title of the paper, the full
name(s) and affiliation(s) of the author(s), complete mailing
address, email address and telephone numbers of all authors,
and a 150 to 300 word abstract. The Journal reserves the right
to edit all material to meet space requirements and to make
grammatical and typographical corrections.
The final text should be 4000 to 5000 words in length,
containing no more than about 10 references, and be provided
as follows:
(1) Hardcopy: printed and double-spaced, with notations
for the location of graphics, mathematical equations, given
thereon, as necessary,
(2) Softcopy: The preferred media format is IBM PC 3.5", 1.44MB.
The preferred file format is Word 6/95/97 for Windows 3.1/95.
Other acceptable software files (in the IBM PC format) are the following:
Word/DOS 3.0 or later
Word/Mac 4.0 or later
Word/Win 2.0 through 7
WordPerfect 5.1 or later (for DOS or Windows 3.1/95).
Any standard ASCII text file format using the preferred
media format, including bracketed notations for
the locations of symbols, equations or other
non-ASCII characters.
Tex and LaTex may be used for the development and
generation of the hardcopy version of the
paper, provided that a softcopy version is also
submitted in any standard ASCII text file
format using the preferred media format,
including bracketed notations for citations and
for the locations of symbols, equations or
other non-ASCII characters.
GRAPHICS
The preferred graphics format is a Windows compatible format
(.pcx, .bmp, .wmf). For other graphics formats, submit high-quality,
camera-ready hardcopy.
TEXT CITATIONS AND REFERENCES
Papers should be limited to about 10 references. Encouraged are
references to peer-reviewed journals as well as to books.
Conference proceedings/compendiums are discouraged.
Text citations must use the following format: last name(s) of
author(s), publication date and suffix (as necessary) in
brackets. Example:
Watkins and McCoy [1993a]
References must be listed alphabetically by the last name of
the first author according to the following formats:
Journal Article: authors' names, publication date and
suffix (as necessary) in brackets, article title (in double
quotations), periodical title (in italics), volume and number,
pages cited.
Book: authors' names, publication date and suffix (as
necessary) in brackets, book title (in italics), publisher,
publisher location, pages cited.
Chapter in Book: authors' names, publication date and
suffix (as necessary) in brackets, chapter title (in double
quotations), editors' names, book title (in italics),
publisher, location, pages cited.
Send all manuscripts to the following postal address:
Editors
JCIF
P.O. Box 764
Haymarket, VA 20168
USA
***********************************************************************
F I N A L C A L L F O R P A P E R S
***********************************************************************
------------------------------------------------------------------
Dr. Andrea Gaunersdorfer
Department of Business Administration
University of Vienna Tel.: +43-1-29 1 28-466
Bruenner Strasse 72 FAX: +43-1-29 1 28-464
A - 1210 Wien e-mail: gauner(a)finance2.bwl.univie.ac.at
http://www.bwl.univie.ac.at/bwl/fiwi1/members/gauner/gauner.htm
=========================================================================