建構台灣中小企業兩階段風險評估模型

Abstract

由於全球性的經濟不景氣，導致銀行與金融機構承受相當大的財務風險，因此國際清算銀行於2004年公佈之新巴塞爾資本協定(Basel II)，允許銀行及金融機構可以自行利用內部評等方法建立風險評估模型來衡量借款客戶之風險。目前中、外文獻雖已發展出許多風險評估模型，但大多是針對上市、上櫃公司，少有以中小企業為研究對象，由於中小企業佔台灣企業數九成以上，為國內金融機構主要放款對象，現有文獻所建議之風險評估模型若直接應用到台灣中小企業上，預測可能不準確。現有的風險評估模型多是建構一個分類判別模型（如區別分析模型、邏輯斯迴歸模型等），將借款客戶分成違約(default)及不違約(non-default)兩類，然而利用這些判別模型評估借款企業之風險時，雖然有不錯的整體準確率，但常會出現某類借款客戶（如違約客戶）準確率高，而另一類客戶(如不違約客戶)之準確率卻偏低的情況，此種判定模型即使整體判別準確率不錯，但對於金融機構而言，其實用性不高。因此，本研究針對中小企業之特性，發展出一套兩階段的風險評估模型，以改善這種準確率偏向某一類客戶的問題，並提升傳統風險判別模型之準確率。本研究利用邏輯斯迴歸(logistic regression)與支持向量機(support vector machine, SVM)建構此兩階段風險評估模型，然後依照判定模式給予中小企業一個信用風險等級，以供銀行或金融機構能夠制訂出最佳之放款策略。最後，本研究利用國內某金融機構所提供之中小企業借款歷史資料，驗證了本研究之兩階段風險評估模型確實有效可行。

Due to the global economic recession, enterprises are facing strong financial stress. For this reason, banks or financial institutions are suffered from serious financial risk. In order to reduce the global financial risk, banks or financial institutions need to develop to their own internal measures for assessing the borrower’s credit according to the New Basel Capital Accord (BASEL II). Most risk assessment models found in literature were constructed for the publicly traded companies. However, 90% of enterprises in Taiwan are small and medium enterprises. It is not quite appropriate to apply the risk assessment models for publicly traded companies directly to those banks or financial institutions whose borrowers are mainly small and medium enterprises. Furthermore, most available risk assessment models use classification methods (such as the discriminant analysis model and logistic regression model) to construct the models and classify the loan borrowers into default and non-default groups. Although the total accuracy rate of classification may be good, but the accuracy rate for a particular group (such as the default group) is significantly higher than the other. It is often found that the accuracy rates for both groups are not balanced. It causes serious problem in practice use for the financial institutions. Therefore, the objective of this study is to develop a two-stage risk assessment model to improve the unbalanced accuracy rate for different group, furthermore, to increase the total accuracy rate. This study utilizes logistic regression at the first stage and support vector machine (SVM) at the second stage to construct this two-stage risk assessment model. Finally, a real case from a Taiwanese financial institution is utilized to demonstrate the effectiveness of the proposed procedure.