A new algorithm for forecasting stock index based on heuristic fuzzy time series model.

Abstract

In this report a new improved method is proposed for time series forecasting based on results of fuzzy time series. Since Song sand Chissom (1993) proposed the concept of fuzzy time series and first order fuzzy time series model, a  number of reseachers are developped this idea for time series forecasting. However the accuracy of these algorithms is necessary to improve. Therefore finding the effective algorithms for time series forecasting is priority. In this article, we develop an algorithm for heuristic time series model in order to improve the forecast accuracy of time series. Our modification in this algorithm is exposed in two aspects: Partition the Universe of discourse into intervals so that distribution of value observations in these intervals is regular and using the information of first and second orders differences of time series for definition the law in defuzzification process. An example for forecasting Tapei index stock is used for demontration of effectiveness of the algorithm.