Application of fuzzy cognitive maps for stock market modeling and forecasting

In this paper, we investigated the possibility of discovering complex concepts for modeling and forecasting the stock market. We started from a short overview of existing domain knowledge and discussed the usefulness of well-known stock market indicators for predicting share prices. As opposed to other statistical or artificial intelligence approaches, we decided to concentrate on the modeling of cause and effect relationships among concepts within stock market behavior. After preliminary analysis, we made the case for the application of the resulting model for the forecasting stock market performance, using complex concepts that involve a mixture of diverse simple indicators and causal relationships. For the construction and evaluation of such complex concepts, we applied FCMs (fuzzy cognitive maps), a relatively easy approach that allows human interpretation of the results from the scheme. On the basis of the proposed formalism and the adapted evolutionary learning method, we have developed an FCM with the ability to support decisions relative to the stock exchange. We have verified the usefulness of the proposed approach using historical transactions of the Warsaw Stock Exchange.