Building Stock Selection Models Using Classification and Association Approaches - An Empirical Research on Taiwan Stock Market

• Main Authors: Hung-Ju Hou, 侯宏孺
• Other Authors: I-Cheng Yeh
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/19092527124593475274

碩士 === 中華大學 === 資訊管理學系(所) === 97 === This study employed two kinds of data mining method and thirteen fundamental and market quarter indicators of stock to build the stock selection model. The data mining methods included classification analysis and association analysis; indicators included the current quarter return rate, ß risk factor, debt to equity ratio (D/E), return on equity (ROE), trade volume, turnover ratio, market value, stock price, book value, book value to market value ratio (B/M), earnings per share to price ratio (E/P), growth-value rate (GVR), and modified GVR. The dependent variable is the t+2 quarter return rate of individual stock. Using the t+2 quarter return rate as the dependent variable is to reflect the one quarter time lag of the quarter finance report of company. Data were collected from the first quarter of 1996 to the third quarter of 2007 (total 47 quarters) for the companies listed in Taiwan stock market. The first 24 quarters (total 7638 data) were used as the training examples, and the latter 23 quarters (total 9838 data) were used as the testing examples. The classification analysis methods included logistic regression, back-propagation network (BPN), and classification tree; the association analysis used the Aporiori algorithm associated with two-segmentation variables. The results showed that the mean (± standard deviation) of quarter return rate of testing period: (1) The best classification algorithm is BPN, obtaining 6.85% ± 20.64% quarter return rate; (2) The association analysis obtained 6.84% ± 26.57% quarter return rate; (3) These two approaches were much better than the market quarter return rate 4.74% ± 22.35%.