The Forecasting of Istanbul Stock Market with a High Order Multivariate Fuzzy Time Series Forecasting Model

The fuzzy time series approaches, which recently are intensively considered by the researchers, consist of three stages of fuzzification, determination of fuzzy relations and defuzzification. Several studies using different approaches in these steps have been conducted in literature. In most of the studies related fuzzy time series, the membership degrees of belonging to every fuzzy set of each observation are ignored in the stages. This conflicts the fuzzy sets theory and causes the loss of information. The fuzzy set theory is required to regard these membership degrees. In that case, how the membership degrees are defined is considered as an important issue. On the other hand, there is another problem that many fuzzy time series are necessarily modeled by a multivariate fuzzy time series forecasting method to discover the fuzzy relations between that time series and the others. In accordance with this purpose, Yolcu (2011) present a new multivariate fuzzy time series procedure in which the method of fuzzy Cmeans is used to define the membership degrees and a neural network with the input and output, which are composed of the membership degrees, is used to define fuzzy relations. In this study, we intend to introduce this high-order multivariate fuzzy time series forecasting model and it was applied to 100 in stocks and bonds exchange market of Istanbul which is frequently used in the literature.