High-Order Fuzzy Time Series Based on Rough Set for Forecasting Taiex

Time series data consists of consecutive observations with continuous value changing in time. The data itself implies a fluctuation characteristic which can be forecasted. But different periods of stock prices results from non-equivalent data generating models. Therefore, this study presents a high-order fuzzy time-series method based on rough set for forecasting a ten-year period of the TAIEX (Taiwan stock exchange capitalization weighted stock index). In empirical analysis, this study takes 4 different lag periods (t-1, t-2, t-3, and t-4), which mean different orders, from 1-order to 4-order, as the input attributes to evaluate the proposed method and compares the forecasting results with those derived from Chen´s (1996) and Yu´s (2004) methods. The experimental results show that the best performance for the proposed method is fall on 1-order and the proposed method outperforms Yu´s and Chen´s in the all testing periods except 1997.