標題: 利用風險評估與存活期預測模式構建台灣中小企業信用評等流程
Developing a Credit Rating Process for Small and Medium Enterprise in Taiwan Integrating Risk Assessment Model and Prediction Model of Survival
作者: 黃啟峰
Chi-Feng Huang
唐麗英
Lee-Ing Tong
工業工程與管理學系
關鍵字: 中小企業;風險評估模式;存活期預測模型;信用評等;支向機;判別分析;自組性演算法;small and medium enterprises;risk assessment model;prediction model of survival;credit rating;Support Vector Machine;Discriminant analysis;Group Method of Data Handling
公開日期: 2005
摘要: 近年來由於全球化以及國內政經形勢的不穩定,使得各金融機構皆承受相當大的放款風險。有鑑於此,巴塞爾監理委員會在2001年提出新的巴塞爾資本協定,明文規定銀行應有內部風險管理機制,以有效管控自身的風險及降低呆帳。過去有許多研究使用倒傳遞類神經網路(Back-Propagation Neural Network, BPN)和核心法(Kernel method)構建風險評估模式,但是由於這些方法皆難以得到分類之方程式,使的業界在實際應用上遭遇很大的困難。過去曾有文獻利用風險評估模式與存活期預估模式構建信用評等模式,但模式中並只考慮存活期,未考慮到債務期限長短,導致無法準確衡量違約後之風險程度。本研究利用支向機(Support Vector Machine、SVM)、判別分析(Discriminant analysis)和自組性演算法(Group Method of Data Handling、GMDH)等可以得到分類方程式的方法構建風險評估模式,再整合存活期預測模式與一個可以同時考慮存活期與債務期限之指標,另提出一套中小企業信用評等模式,使金融機構能有效評估借款企業之風險程度和信用等級,快速的作出適當之放款決策。本研究利用國內某金融機構所提供近五年向其借款之中小企業歷史資料,以驗證本研究所發展之信用評等模式確實能有效地幫助金融機構作出正確之放款決策。
Many financial institutions have faced serious loan risk due to globalization and instability of political and economic situations in Taiwan. For this reason, the internal risk management for financial institutions becomes an important issue in recent years.Many research about risk assessment model use methods like Back-Propagation Neural Network and Kernel method. But these methods can not provide mathematic function.The proposed procedure uses Support Vector Machine, Multivariate Discriminant analysis and Group Method of Data Handling, respectively, to construct the risk assessment model.Than, this study propose a procedure to assess credit rating, utilizing the risk assessment model and prediction model of survival. A case study is provided by a financial institution in Taiwan to demonstrate the effectiveness of the proposed procedure.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT009333502
http://hdl.handle.net/11536/79462
Appears in Collections:Thesis


Files in This Item:

  1. 350202.pdf