Forecasting Stock Market based on Trend and Variation Patterns

• Main Authors: Hung-Hsiang Chiang, 江宏翔
• Other Authors: none
• Format: Others
• Language: en_US
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/63508096332817318634

碩士 === 雲林科技大學 === 資訊管理系碩士班 === 96 === Since human cannot handle the changes of unknown tomorrow completely, forecasting what will occur mostly has been one challenging issues in many areas such as weather and stock market forecasting. As the emergence of information technology has arisen for a long time, it played an important role to help people make forecasts for the future. A lots of forecast model based on computer system has applied to assistant stock analyst and investors for forecast stock market. When forecasting models produce reasonable predictions with less bias, great profit will be obtained. In stock markets, many models were proposed by researchers to forecast stock price, such as time series, technical analysis, and artificial intelligence. However, there are some problems in the former models: (1) strict mathematical assumptions are required; (2) objective human judgments are involved in forecasting; and (3) a proper threshold is not easy to be found. For this reasons above, a new forecasting model based on variation and trend patterns is proposed in this dissertation. After choosing patterns to forecast, a proper threshold is adopted to choose better patterns. Besides, the adaptive expectation model is built up with an adaptive parameter to refine forecast result. To verify the performance of proposed model, the TAIEX (Taiwan Stock Exchange) and TSMC (Taiwan Semiconductor Manufacturing Company) stock price from 1997 to 2005 are applied as experiment dataset. Due to the stockholders in Taiwan stock market prefer short-term investments based on recent stock information, the time series models such as AR(1) (Autoregressive), AR(2), and ARMA(1, 1) (Autoregressive-Moving Average) are taken as comparison models. Mean absolute percentage error (MAPE) and root mean square error (RMSE) are used as performance indicators to evaluate the forecast models. The comparison results have shown the proposed model outperforms the listing methods when there are lager fluctuations.