To Build the Prediction Model by Using Data Mining Approach - A Case Study in Taiwan ETF-50 Stock Price

• Main Authors: Yen-Chou Lai, 賴彥舟
• Other Authors: Jau-Shin Hon
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/a6nhx2

碩士 === 東海大學 === 工業工程與經營資訊學系 === 102 === Taiwan 50 Exchange Traded Funds (Taiwan 50 ETF) fluctuate similarly to Taiwan’s stock market, therefore it’s chosen as a representative for Taiwan’s stock market for purposes of our study. This research is based on the data generated from Taiwan 50 Index, applying "rule induction" and "back-propagation neural network" to predict short-term future Taiwan 50’s volatility. First, index historical data and common indicators are selected as variables. Then "rule induction" is conducted to extract seven principles, and highly related variables are filtered out as a tool to recognize the upcoming direction of price variation. Data is subjected to the seven principles, and seven prediction models are constructed applying “propagation neural network". Finally, the study performs "Mean Absolute Percentage Error (MAPE)" method to assess the predictability of the model. The results show that, highly related variables selected through "rule induction" are able to quickly determine whether the future price rise or not with at least 70% accuracy, and five of seven prediction models established based on the "back-propagation neural network" proved reasonable. Therefore, the predicted volatility is able to be converted into closing price of stock as an indicator for investors and/or traders.