The Use of Neural Network and Portfolio Analysis in Forecasting Share Prices at the Stock Exchange

• Main Author: Przemyslaw Stochel
• Format: Article
• Language: English
• Published: AGH University of Science and Technology Press 2000-01-01
• Series: Computer Science
• Subjects: Neural networks
• Online Access: http://www.csci.agh.edu.pl/22/1/cs2000%2D02.pdf

The article presents the use of neural networks in decision making process on the capital market. The author tried to show the efficiency of established solution in Polish reality which features different conditions in comparison with the markets of higher developed countries. The aim of the paper was to prove that neural networks are flexible tools which on one hand might be adjusted to investor's requirements and on the other, can reduce equirements to his experience. The article is based on the author's own research carried out by modelling neural network operation with a simulation program. The established solutions are input which employs stocks portfolio computed on the basis of Markowitz portfolio theory and Sharpe's model. According to the established propositions, the portfolio created in such a way is modified by neutral network in order to optimise a criterion which maximises the income of such a modified portfolio. A detailed genesis of the established input vector and network structure are presented. It allows the reader to carry out his own research and create his own attitude towards applied values. The research results based on a real stock market database with the use of one-output networks predicting thc price of a single company - Agros as well as networks predicting the desirable structure of the whole portfolio are presented. The effect of the network structure leaming parameters, input vector (not only as to the input quantity but also as to period of time they were collected) was examined. The dependence between the factors mentioned above such as input vector and network structure were discussed. lt seems that the presented paper has proved that some not widely spread methods with neural networks can become at competitive tool to optimisation methods.