_________________________________________________

For:

 National Conference of IPSJ (Information Processing Society
 of Japan),  3/12-15 1997, Chiba.

_________________________________________________


Title: 
----


Neural Networks for Time-Series prediction; Stock Exchange Forcasting.  

Author: 
------
Alexandra Ioana Cristea 
Toshio Okamoto 

Affilliation:
---------

The Graduated School of Information Systems
University of Electro-Communication, Tokyo

Keywords: 
--------
3105

Abstract:
-------

Stock Exchange Events are examined as Time-Series, and solutions are proposed for the different
components: Trend, Cyclus, Season and Irregular events. Neural Networks (NN) are used for their 
parallel processing  power, and for their ability to learn three of the four components mentioned above.
For the forth, a economy based calculus is proposed. 
As market occurencies do not follow a given function or equation, they can be most probably approached 
with a better solving chance by approximative methods, such as NN methods (This is debateable, stochastic 
methods seem to give good results,too).
For testing the above mentioned theory, a program has been developed. Data input has a three level 
hierarchy: it is either stock market prices for a given time period (training), or the weights matrix 
computed at the previous step,for new data testing on the trained system (testing) and at last, current 
stock market prices,for active forcasting (prediction).
Some results are presented and comparision with former developed systems is also performed. The main 
difference from previous systems consists in the mixt solution of mathematical and economical rules. 


Adress:
------
Alexandra Ioana Cristea
Prof. Toshio Okamoto's Laboratory (AI)
The Graduated School of Information Systems (IS)
University of Electro-Communication,
1-5-1 Chofugaoka, Chofu, Tokyo 182

Tel.: 0424-896070
Fax.: 0424-815934
e-mail: alex@ai.is.uec.ac.jp
http:  http//www.ai.is.uec.ac.jp/u/alex/