------- Forwarded Message Follows ------- 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 =========================================================================