CLUSTERING OF THE STOCKS (SECURITIES) USING SELF-ORGANIZING NETWORKS AND NON-LINEAR RANGING TECHNIQUES

摘要

The problem of comparison of different companies is facing, when analyzing company's performance in stock exchange market. Screening of the companies, using "well-known" trading strategy parameters, is one of the ways to solve this problem. Actually, much more companies appear on the list, than the trader is willing to buy. To define the best companies on the list, the non-linear ranging technique could be used. The certain meanings of weights could be given to the factors, which characterize the companies. Then, by estimation of all weights company could be assigned to its place in the general listing. Ranging technique with the parameters proposed by T. Allrich and with the parameters proposed by the authors themselves was applied and researched. Self-organizing (Kohonen's) neural network was applied as an alternative to ranging technique. Using fundamental financial parameters as inputs, the output of neural network was the different groups of companies located into a number of disjoint clusters. Three different portfolios were formed as a result of the research. The performance of these portfolios showed which of the researched techniques gave better result. The real data from USA stock markets was used for the realization of the whole idea.