自組織特徵映射圖類神經網路應用於動態公司體質檢定之選股研究

Abstract

金融風暴後，台灣許多體質不健全的公司陸續爆發財務危機引起股市重挫，造成了投資者的重大損失。本研究著重在公司體質上之變化，將靜態的財務指標及股價資料以自組織映射圖神經網路進行時間序列的路徑分析。進一步本研究結合灰關聯分析與自組織映射圖神經網路建構出半監督式學習的模型以分析各公司體質變化之情形。預測出未來最有可能轉變的方向，並建構出股票投資組合予以檢驗。研究成果顯示，在市場多頭時期透過自組織映射圖神經網路之動態軌跡式分析可獲得顯著超額報酬；並且應用灰關聯分析之方法可改善自組織映射圖神經網路於空頭時期預測錯誤之情形，顯著提高獲利能力及預測訊號之準確率。

After Asia Crisis, many unhealthy companies busted out emerging financial crisis and led to the crash of the stock market. Therefore numerous investors came across huge loss. Having this tragedy in mind, this research focuses on the financial constitution of the corporate. Financial ratios and stock price data are analyzed by the self-organizing map according to the trajectory of a firm. Further, grey relation analysis method and self-organizing map are integrated to construct a semi-supervised model on this research for anticipating the future trend of a company. A stock portfolio is also established to check the result of these models. It is revealed that not only excess profit is earned by the self-organizing map based dynamic analysis significantly but also the profit and accuracy of self-organizing map can be improved by combining grey relation analysis method.