| 摘要: | 企業財務危機分類模型一直是熱門研究議題，從早期傳統統計方法來建立模型，時至今日已經有大量的人工智慧演算法所建立的模型出現。而「支援向量機」與「最小平方法支援向量機」是現今較新穎的人工智演算法，究竟兩種方法在企業財務危機分類的診斷上有何差異？本研究嘗試建立此兩種方法之企業財務危機分類模型並比較差異，另外再以區別分析、t檢定與因素分析來相結合，以建立一個二階段的財務危機分類模型，並探討結合上述變數篩選方法所建立的模型之分類績效差異。
此研究以2001年至2011年底台灣非金融業之上市及上櫃公司為樣本，取財務比率、智慧資本和公司治理指標資料來建立企業財務危機分類模型。
本研究發現，以t檢定先進行變數篩選後，將有助於提升模型的整體分類績效。另外，在「支援向量機」和「最小平方法支援向量機」所分別建立的財務危機分類模型中，各組模型的分類績效並無明顯差異。此研究結果希望能提供日後學者在建立財務危機分類模型時能當參考之用。
Many previous studies have examined classification models for enterprise financial distress. While earlier models were built using traditional statistical methods, other machine learning algorithms are being used for building models nowadays. Support Vector Machine (SVM) and Least Squares Support Vector Machine (LS-SVM) are relatively new machine learning algorithms. How are they different? This study builds two classification models for financial distress using these two methods and compares the differences between them. This study also combines discriminant analysis, t-test, and factor analysis with SVM and LS-SVM to build a two-step classification model of enterprise financial distress and discusses the predictive performance of the models built using the abovementioned feature selection methods.
The sample includes listed and OTC companies in Taiwan, which were observed during the period 2001–2011. The study uses financial ratios, index of intellectual capital, and corporate governance for building the models.
It was found that models based on feature selection by t-test could forecast enterprise financial distress more accurately, and both SVM and LS-SVM had similar classified ability for building prediction models of financial distress. These findings could be useful for future studies. |
