運用資料採礦技術及企業評價模式建立投資策略之研究-以台灣上市櫃非金融類公司為例

Abstract

本研究以台灣公開上市(櫃)非金融類股公司為研究對象，用股價報酬率以及財務報表中所揭露各項數據作為研究變數。首先進行相關性分析，藉以瞭解兩者間之關係並選取相關性較高之變數，建構資料採礦分析模型。研究發現；前後期股價報酬率之間的相關性最強、股價報酬率與財務數據間之相關強度有隨時間遞減之現象，顯示兩者之間的確有因果關係。 資料採礦分析顯示，分類預測的準確性約為75%，以羅吉斯迴歸之表現最佳，而推估預測能力比較之結果，各演算法間之表現很接近，均只能對報酬率表現突出之族群做出較精確的預測（誤差約為40%），較具有參考之價值，因此，本研究所建構之資料採礦選股模型，雖然在實務上有其價值，可作為投資人選股之參考，但使用時仍需配合其它財務數據或評價模式作研判較為適當。 使用自由現金流量折現法作企業評價有許多的限制以及假設，在本研究中僅用財務數據進行假設與計算，但估算結果仍可適用於許多股票之上，且適用之股票多數能產生高的股價報酬率，因此，可搭配使用來判斷股票超漲或超跌作為買賣時機選擇之依據。

This study subject is non-banking companies in Taiwan stock market,base on investment return rate and financial statement.Using correlation analytic to find out the relationship between the two at first,choose variable with higher correlation coefficient,to construct data mining analysis model. Close period of investment return rate the correlation is strong most, phenomenon decreased progressively at time in between the two,show that really causality. Data mining show out,in classify prediction the accuracy about 75%,the logistic regression have best performance,each perform among algorithms is almost same in estimate prediction,can well doing prediction in the outstanding clustered by the investment return rate (error is about 40%),so,the data mining model can help investor to choose investment target ,but still need enterprise evaluation model to doing the judge comparatively. Use DCF model have many constrain and assumption,in this study base on financial data doing the enterprises evaluate, still be suitable for a lot of stocks and those target can be produce high investment returns,so,use it to make judge the stock prices is over rise or fall, doing the buy or sell choose.