Discussing the Return of Investment under Different Classification by Combining Statistical Methods and Artifical Neural Networks

• Main Authors: Bao-SyuanFu, 傅寶萱
• Other Authors: Chung-Cheng Wu
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/87128558624491559111

碩士 === 國立成功大學 === 統計學系 === 102 === This study from the Taiwan Top 50 ETF screened which from 2008 to 2013 had never been kicked out of the non-financial constituent stocks. After screened, a total of 24 constituent stocks meet this condition, and object of this study is the 24 companies. Financial ratios were used as a measure of operation performance of the 24 companies over the years, by using Statistical Methods and Artificial Neural Networks. This study is based on 15 financial ratios as variables to classify the operation performance of 24 companies from 2008 to 2013. The former five years data as training data, 120 data altogether, and data in 2013 as testing data, 24 data altogether. After analyzing the performance of 24 companies, a total of four classification modes were constructed to compare the accuracy of each mode clustering results, and then constructing the portfolio. In the classification part, judging distinguished effect by the accurate classification rate of testing data, and matching up Cluster Analysis and Discriminant Analysis as result get better effect than other modes; terms of testing data classification results further construct the portfolio, and then compare the investment return of four modes, matching up Cluster Analysis and Discriminant Analysis as result also get better effect than other modes, the highest rate of return on the testing results.