應用資料包絡法與邏輯斯迴歸建構中小企業之信用評等與放款風險整合模型

Abstract

信用評等(credit rating)及風險評估(risk assessment)是金融機構用以評量借款企業償債能力的兩項重要工具。有鑑於全球性的經濟不景氣及國內政策不穩定，且「新巴塞爾資本協定」( New Basel Capital Accord )已明文規定自2006年底金融機構可使用自行創立之內部評等方法來衡量借款企業信用風險，因此銀行或各金融機構構建一個適用於其特殊放款特性之有效信用評等及風險評估模型是非常重要的課題。然而至今中外文獻所提出之信用評等與風險評估模型大多是以上市、上櫃公司為研究對象，故對在我國扮演經濟主體之中小企業而言，可能並不適用。因此本研究應用資料包絡判別分析法(Data Envelopment Analysis - Discriminant Analysis, DEA-DA)與邏輯斯迴歸建構一個中小企業之信用評等與放款風險之整合模型。本研究之模型建構流程是針對所蒐集之中小企業之財務變數、非財務變數以及總體經濟因素等應用資料包絡判別分析法建立信用評等分數並訂定各信用等級之標準；再應用邏輯斯迴歸建立違約率(probability of default, PD)預測模型。本研究成果可寫成軟體程式，銀行業者只需輸入中小企業之財務、非財務及總體經濟因素資訊，即可得到其信用等級及違約率，因此可快速準確做出適當之放款決策。本研究之貢獻為提供金融機構一個準確且應用簡易之信用評等與風險評估之整合模型，以有效降低逾放款比率。本研究最後以台灣某金融機構所提供之中小企業借款客戶的實際歷史資料，證實了本研究所建構之中小企業信用評等流程確實有效。

Many banks and financial institutions have suffered serious loan risk due to the globalization and the instability of political and economic situations. To meet the requirements from the New Basel Capital Accord (Basel II) and to minimize the credit risks due to incorrect loan decisions, it is very important for banks and financial institutions to develop a reliable procedure to evaluate the credit risks resulting from defaults. Two important methods - credit rating and risk assessment are commonly utilized to evaluate solvency of enterprises. Many studies on credit rating model are based on the financial data drawn from the publicly traded companies. However, it is not appropriate to apply the credit rating model for publicly traded companies directly to those banks or financial institutions whose customers are mainly small and medium-sized enterprises. Therefore, this study employs Data Envelopment Analysis - Discriminant Analysis (DEA-DA) and logistic regression to construct a integrating credit rating and risk assessment model for small and medium-sized enterprises. The proposed procedure consists of three stages: 1.selecting appropriate financial variables and collecting the data; 2. utilizing DEA-DA to evaluate the credit rating score for each loan business and establish a standard for rating the credit of the loan business; 3.constructing a prediction model of default probability using logistic regression. Finally, a real case from a Taiwanese loan company is utilized to demonstrate the effectiveness of the proposed procedure