Investigations of cascade neo-fuzzy neural networks in the problem of forecasting at the stock exchange

The problem of the forecasting stock prices and indexes in the stock market is considered. The application of new class of neural networks - cascade neo-fuzzy neural networks for the forecasting is investigated. For the forecasting data of the company´s stock quote NYSE and the RTS index and the Dow Jones over the last year have been used. The comparison of results for the cascade neo-fuzzy neural networks with varying types of membership functions with the classical fuzzy neural networks, Group Method of Data Handling (GMDH) and fuzzy GMDH has been performed. The best results among the fuzzy neural networks showed cascaded NF network with Gaussian membership functions, their error does not exceed 3%.