Forecasting the Taiwan Stock Market with a Novel Momentum-based Fuzzy Time-series

Fuzzy time-series models have been utilized in making reasonably accurate predictions in many areas, such as academic enrollments, weather forecasting and stock markets. To refine past fuzzy time-series models, this paper proposes a new model, which employs the concepts of ¡°momentum¡± along with Chebyshev¡¯s theorem in the forecasting process. The proposed model applies a ¡°momentum¡± index to generate forecasting rules (fuzzy logical relationships) to reduce the probability of rules not being found in cases where no rules are available to forecast a testing dataset. Chebyshev¡¯s theorem is adopted to define a ¡°reasonable¡± universe of discourse for the observations in a training dataset. From the refined process, two types of universe, symmetrical and asymmetrical, are given. To verify the proposed model, this paper employs experimental datasets, derived from a seven-year period of the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX). Model comparison results show that the proposed model surpasses in accuracy one traditional fuzzy time-series model and two advanced models, based on neural networks and rough set algorithms.