Application of a Neural Network Approach to Implement a Bivariate Fuzzy Time Series Model

• Main Authors: Yu-Fang Lin, 林鈺芳
• Other Authors: Dat-Bue Lock
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/85843829438332117042

碩士 === 逢甲大學 === 企業管理所 === 93 === Time series studies have been so popular for a long time. To solve nonlinear problems, fuzzy time series models are believed to be more suitable. Different fuzzy time series studies have been proposed for various applications, such as enrollment (Song & Chissom, 1993), temperature (Chen & Hwang, 2000), stock index (Yu, 2005), etc. However, most of these studies were for one variable only. Recently, some models have been proposed for two-variables (Huarng, 2001; Hsu, Tse & Wu, 2003), which rendered better forecasting results. Hence, this study also proposes a model for two-variable problems. We use the neural network to establish fuzzy relationships. TAIEX (Taiwan Stock Exchange Capitalization Weighted Stock Index) and its corresponding index futures TAIFEX are used as inputs to forecast the TAIEX of the next day. The data used are the daily closing prices of the stock market and futures market in Taiwan. The data are selected from TEJ Database (Jan. 1999 – Dec. 2004). We apply an neural network to implement this Bivariate Fuzzy Time Series model. The inputs are the closing prices of a stock index and those of a futures index, and the output is the closing price of the stock index of the next day. The results are compared with those from Chen’s Model (Model 1) and Standard Neural Networks (Model 3). In most of years, the proposed model (Model 2) outperforms the previous fuzzy time series model. However, the forecasting results from Model 2 and 3 are not so good when the volatility of the stock index is high. To improve the forecasting results, we apply some forecasts in Model 1 to those from Model 2 in Model 4. The results show that Model 4 could solve this problem.