|
PureBytes Links
Trading Reference Links
|
Thanks for your post, sorry I didn't post the site
http://ourworld.compuserve.com/homepages/ftpub/jcif.htm
Journal of Computational Intelligence in Finance
We are pleased to present the ASCII text versions of the award-winning
papers for 1995 and 1996. Abstracts from each paper are described below.
These papers (ASCII text versions) may be downloaded as described below.
----------------------------------------------------------------------------
----
Winning Paper 1995
Supervised Evolution of the Neural Trader Component of a Stock Portfolio
Trading System
David L. March
Abstract
This paper describes a method that can be used to adjust neural network
weights in situations where there is no advance knowledge about the
correspondence between the network input and output. The method, which is
partially based on genetic algorithm concepts, will work in any situation in
which the target objective or profit function is stepwise instead of
continuous.
The validity of the method is illustrated using a trading system that has a
neural network as one of its components. The system is designed to maximize
the profits from trading a portfolio of diverse stocks using input derived
from weekly price data. A number of tables and equity graphs are used to
show how the system performed when it traded the training portfolio and two
other portfolios that contained stocks that were not part of the training
set.
Download - download an ASCII text version of this paper.
----------------------------------------------------------------------------
----
Winning Paper 1996
Forecasting the 30-year U.S. Treasury Bond with a System of Neural Networks
Wei Cheng, Lorry Wagner, and Chien-Hua Lin
Abstract
A forecasting model based on a system of artificial neural networks (ANN) is
used to predict the direction of the 30-Year U.S. Treasury Bond on a weekly
basis. At the close of Friday's market, the 23-variable database is updated
with the latest information and the data pre-processed for input to the
prediction system. Thirty-two feed-forward neural networks are trained on
the new data and then individually recalled to predict the following
Friday's market direction. These results are then used as input to a
decision model that ultimately determines the final prediction.
Forecasting began in 1989, with live trades commencing in 1990. The average
accuracy of the buy prediction was 67% over a five-year period, with an
average annualized return on investment (ROI) of 17.3%. This compares with
an ROI of 13.9% for the Lehman Brothers T-Bond Index. 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.
Download - download an ASCII text version of this paper.
==================
Best regards
Walter
|