Modelling stock market via fuzzy rule based systems

Abstract

A rule based fuzzy logic model is implemented to forecast the monthly return of the Istanbul Securities Exchange 100 (ISE100) Index by combining common stock market data analysis techniques. These are technical analysis, financial analysis and macroeconomic analysis.Starting with the technical analysis, an index level observation by using classifier systems is used as a long term input for the rule based modelling of the fuzzy logic. The negative correlation of the level of the ISE100 Index and the daily returns is obtained. The observation is statistically tested and found to be significant by using bootstrap method. A basic technical analysis rule using moving average is also modified as a short period input for the model.In the financial analysis that borrows from the methodology of the Altman's bankruptcy prediction analysis, a model for predicting next period's stock return is developed with the logistic regression by using 81 financial ratios before the factor analysis. It is shown in the ANOVA analysis that the output of the analysis is statistically significant as good companies perform better than bad companies. Thus, the weighted average of the logit score of the stocks is used to forecast the next period return of the index in the fuzzy logic model.After the calculation of the technical and financial inputs for the model, macroeconomic data is gathered in three main groups: real economy, FX market and TL market. The data is ruled and modeled within the period from 1996 to 2002 and optimized with steepest descent learning algorithm. The R-Square of the model is obtained to be better than those of other stock market return estimation models in the literature. The reasons are: a) the ISE is not a developed and an efficient market, thus, predicting future prizes are easier than the other developed markets; b) rule based fuzzy logic modelling with large enough data set improves the explanation of the future movements. Furthermore, the model is also tested for optimal investment decision in 2003. The algorithm is as follows: If the next month's predicted return of the index is positive, the model suppests to invest all of the existing money in the stock market. Otherwise, it is optimal to invest in repo only. The performance of the investor is compared according to the returns of the ISE100 Index, USD, repo and TL in the same period. Consequently the statistical comparison of the results by using the bootstrap method is promising and the model's suggestion performs better than the return of the repo and the ISE100 Index in 2003.