應用自組性演算法建構多階段信用風險評估模型

Abstract

近年來美國次級房屋信貸危機引發全球性的經濟蕭條，導致眾多企業因無法支付龐大的債務而紛紛倒閉，金融機構逾期放款比率劇增，使金融機構承受莫大的財務壓力。為了有效地提升金融機構對信用風險之應變能力，國際清算銀行在2004年公佈新巴塞爾資本協定(Basel II)，明定金融機構可自行建構內部評等模型，來衡量欲借款企業放款之信用風險，提早採取應變措施，擬定授信放款之策略。目前中外文獻已發展出許多風險評估模型，然而在應用這些方法建構風險評估模型時，其過程相當複雜且耗時，導致金融機構不易將這些模型應用於授信業務上。由於應用自主性演算法(Group Method of Data Handling, GMDH)建構預測模式可獲得一個數學方程式，實務應用上簡便且模型不需要任何統計假設。因此，本研究應用自主性演算法發展出一套多階段風險評估模型，然後根據各階段的判別結果，判定借款企業之風險等級與制定放款策略。本研究最後以台灣中小企業借款客戶資料及UCI Repository of Machine Learning 資料庫中澳洲與德國信用資料，驗證本研究所提出之多階段風險評估模型確實能有效的判別借款客戶之風險，所制定之風險等級與放款策略可做為金融機構放款之參考依據。

In recent years, subprime mortgage crisis in U.S.A. has resulted in global economical recession. Due to the huge debts, thousands of enterprises went bankrupt and financial institutions suffered serious loan risk. In order to enhance the reactive capability of financial institutions when facing credit risk, the New Basel Capital Accord (Basel II) issued by the Bank for International Settlements allows financial institutions to construct their own internal measures for credit risk. Therefore, financial institutions can adopt some emergency measures and strategies to deal with loan risk. Many risk assessment models have been developed in many studies. However, It is usually too complicated and time-consuming to employ the developed risk assessment models for banks or financial institutions. The Group Method of Data Handling (GMDH) method can be easily applied in practice since it can develop an effective discriminant function. Therefore, this paper proposes a multiple-stage risk assessment model using GMDH method to evaluate the credit rank and loan strategies based on the outcomes from each stage of the proposed model. Finally, real cases from Taiwanese small-and-median sized enterprises and UCI Repository of Machine Learning database are utilized to demonstrate the effectiveness of the proposed multiple-stage risk assessment model.