Trend weighted fuzzy time series model for TAIEX forecasting

• Main Authors: Chen-Han Chiang, 江承翰
• Other Authors: none
• Format: Others
• Language: en_US
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/52122715375400620133

碩士 === 國立雲林科技大學 === 資訊管理系碩士班 === 94 === Trend forecasting plays an important role in decision making, and time series models are usually used to forecast future trend. Throughout the years, there have been various fuzzy methods proposed to solve traditional time series problems. Song and Chissom (1993a) defined fuzzy time series and proposed their methods to model fuzzy relationships among observations. On the other hand, different fuzzy time-series models have been proposed for various domain problems, such as enrollment forecasting, temperature forecasting the stock index forecasting, etc. It is obvious that trend of fuzzy relationships, which were currently ignored in previous studies, should be considered in forecasting. Further more; it is clear to see that, the major drawback of traditional methods is the lack of consideration in determining reasonable universe of discourse and the length of intervals. Moreover, we find that the neglect of information, which tells patterns of trend changes in the past history, should be considered in forecasting. Besides that, we should apply to smoothen our forecast value in the case of stock index prediction so that each day’s forecast value is clearly different, thus we can set some criteria in order to make trading possible. In order to handle these kinds of problems, an objective and reasonable approach is then proposed. Traditionally, fuzzy relationships weights are either determined based on domain know-how, which could be elicited from domain experts or determined based on their chronological order. However each fuzzy relationship is likely to occur again once in a while, so it is critical to classify them into different trends in order to make more precise prediction. In this paper we propose a trend weighted fuzzy time series model to improve the forecast accuracy, and apply parameter to smoothen our forecast value, thus we can set some criteria in order to make trading possible. Using the Taiwan stock index and enrollment of Alabama as the forecasting data, the results show that our trend-based weighted fuzzy time series model outperforms other fuzzy time series models.