應用比率風險模型探討信用貸款之違約時間

Discussing the Default Risk Time of Credit Loan by the Proportional Hazards Model

蘇容嫻(Jung-Hsien Su)

指導教授 ： 李孟峰

國立臺北大學/商學院/統計學系/碩士(2014年)

摘要

依據中央銀行所發佈之消費者貸款餘額統計資料，觀察近10年貸款餘額及年增率之變動，發現「其他個人消費性貸款」自2009年開始明顯呈上升趨勢，顯示個人消費性貸款依舊是近年來銀行招攬的目標之一。而根據金融監督管理委員會所公布之資料，過去幾年本國銀行在逾放比方面，呈現不穩定且表現不一的狀態。近三年逾放比甚至有銀行曾高達7%，可說明整體上依然存在放款體質較差的銀行。因此本研究期望在銀行信用貸款之風險控管上，使用存活方法所建置之Cox比率風險模型加強其違約預警能力。目的在於信貸案件進行審核階段時，除了將原銀行內部違約預警模型之風險等級視為第一道門檻外，以Cox模型產出的存活率預測值為第二道門檻。
根據本研究實證結果，有4個變數能顯著解釋信用貸款戶對應違約的存活期數變動：「擔保現欠訂約金額比率」、「貸款利率」、「近三個月銀行查詢家數(2家以上)」、「信用卡近12個月全額繳清次數比率」，而其模型預測能力之敏感度為19.15%，特異度為94.27%，而精確度為86.49%。顯示原銀行已通過審核的案件，在Cox模型預估的第二道門檻下，能再排除近20%之違約戶，而婉拒良好客戶的誤差僅5%左右。顯示此模型確實能增強預警能力，進而達到更嚴謹的風險控管。

關鍵字

信用貸款 ； 違約時間 ； 存活分析 ； Cox比率風險模型

並列摘要

According to the statistics of the balance of consumer loans published by the Central Bank, we may find that there is a clearly increasing trend on "other personal consumer loans" since the beginning of 2009 by observing the balance and the changes of annual growth rate of consumer loans in the past 10 years. The statistics also show that personal consumer loan is still one of the major attracted goals to the banks in recent years. The data released by the Financial Supervisory Commission shows the NPL ratio was as high as 7% for some banks in past three years. This also indicates the considerable risk of consumer loans still exist overall. Therefore, this study not only considers the probability of default risk but also expects to predict the default time of a consumer loan. A default time prediction model built by Cox model to predict the possible default time for those customers already granted consumer loans. This model will pay a role as a second threshold after the original evaluate held by the bank.
The empirical study of this research shows that there are four variables can explain credit default survival time significantly: "the ratio of the secured debt balance to the contracted amount," "loan interest rate," "nearly three months banks query number (2 or more)," and "Credit card in last 12 months full pay frequency ratio." The Sensitivity of predict ability of the model is 19.15%, a specificity of 94.27%, and an accuracy of 86.49%. Applying this model as the second threshold can rule out almost 20% of default loans. And the loss of good customers is only 5%. It shows that this model can really enhance the early warning capabilities, thus achieving the more rigorous risk management.

並列關鍵字

Credit Loan ； Default Risk Time ； Survival Analysis ； Cox Proportional Hazards Model

參考文獻

梁德馨,卓聖恆,陳冠宇,張博彥(2008),「消費者小額信用貸款違約風險評估區別分析模型建構」,創新與管理,第五卷,第一期,pp.101-134。

石雨宸(2010),「應用存活分析模型探討借款人提前清償之行為」,朝陽科技大學財務金融系碩士論文。

呂孟育(2008),「消費信用貸款授信風險評估之實證研究－以○銀行為例」,中興大學高階經理人碩士論文。

張素燕(2009),「消費者信用貸款授信評量模式之研究」,中興大學應用經濟學系碩士論文。

莊昊翰(2011),「應用邏輯斯迴歸與存活分析建構台灣中小企業貸款提前清償預測模型」,國立交通大學工業工程與管理學系碩士論文。