影響銀行授信戶協商違約成敗因素之探討 ─以消金無擔保客戶為例

A study of the factors that affect the success of negotiation on trouble debt restructuring of customer business loans.

葉宣萱(Hsuan-Hsuan Yeh)

指導教授 ： 劉立倫

中原大學/商學院/企業管理研究所/碩士(2010年)

https://doi.org/10.6840/cycu201000359

摘要

消費性放款中無擔放款相對於企業授信及擔保放款，其戶數十分龐大，但每筆金額小，如未獨立觀察其借戶違約情形，很可能因企業授信及擔保放款之餘額而稀釋個人信用貸款之逾放比，從而忽略個人信用貸款授信政策。
尤其銀行業受卡債風暴衝擊以來，信用卡、現金卡及個人信用貸款總體業績呈直線下滑，不再有先前這麼大的誘因，面對日益惡化的消費性放款，若未作好風險控管，未來銀行必須增提備抵呆帳，影響銀行的收益。
具體而言「協商」之本意是具正面積極的，當客戶信用發生警訊時，銀行必須確認風險來源，究係因客戶信用不當擴張或其收入結構改變，亦或是發生道德危險。
但實質整個與客戶之「協商」程序對銀行人員是極花成本及時間的，而依照以往的實證顯示從客戶申請協商到銀行審核通過，到日後有持續正常履約者，所佔比率很低，本研究則想探討；1.對於違約協商之申請客戶可更準確知道何者係真正具能力及誠意者，2.無誠意及能力之協商者也事先判別而不予受理，並將有限的成本和人力運用在對的協商客戶。
本研究以資料探勘(Data Mining)之「決策樹」、「類神經」二種模式，與傳統統計技術之「區別分析」來進行實證結果之分析比較，建構可辨識具實效之潛在協商授信戶之預測模式。
研究結果顯示，於27個研究變數中,任二種模型共有之變數共有7項，分別為「餘額/額度」、「預定繳息日」、「利率」、「期付金」、「失業率(%)」、「已正常繳款6個月以上」、「貸款種類」，而各模型單獨有之變數則有5項「出生日期」、「協商註記」、「職業」、「學歷」、「通訊地址」，而各模式準確率(%)比較上，在訓練值上以「類神經」96.07 %最優，其次為「決策樹」92.94%，在測試值上以「決策樹」最優92.53%，其次為「類神經」92.47%，準確率均達90%以上，但「區別分析」在原始的及交叉驗證的準確度分別為75.15%、75.14%，兩者技術在準確率上亦差距甚大。
由研究結果得知，在進行本研究預測授信戶協商違約之風險上，是較適宜以資料探勘人工智慧改良技術之方法來建構判別模式的。

關鍵字

協商違約 ； 資料探勘 ； 類神經 ； 決策數 ； 區別分析

並列摘要

The customer business loans relative to the enterprise loans and secured loans, its number is very huge, but each amount is small, if has not observed its borrower violation situation independently, possible only pay attention to sum of the enterprise loans and secured loans，dilutes the DPD ratio of the personal unsecured, thus neglects the policy of personal unsecured loan .
Especially,since the card debt storm impact the banks, the total loan balance of credit card, cash card and personal unsecured loans has gone down , no significant performance. Facing the personal unsecured which day by day worsens loans, if has not control the risk , in the future the bank must add raises the allowance for uncollectible account, affects the bank the income.
Specifically speaking "the negotiation" the original intention has frontage to be positive, when the customer credit has the danger signal, the bank must confirm and investigate the risk source ,maybe is the customer credit expands improperly or income structural change, also perhaps has the moral hazard.
Actually for the banks, the negotiation program is spends the cost and the time .According to the former real results, the number of continued normal payment very low percentage of total that the customers apply to negotiation, this research wants to discuss; 1. The application that trouble debt customers may know accurately which one is the true ability and the sincerity, 2. The application that trouble debt customers may know accurately which one is not the true ability and the sincerity in advance.Bank will not give accepts, and restricted cost and manpower utilization to correct negotiation customer.
This research by the Data Mining method of “Decision Tree” &“ Artificial Neural Network”two kind of models, Traditional Statistics method of “Discriminated analysis” ，comparison of test results and analyze the reasons . Building forecast of model may recognize has the latent success of negotiation customer .
The findings showed that in 27 research variables, any two kind of models altogether has the variables total of 7 , respectively was "the Principal balance /amount", "predetermined the payment date", "the interest rate", " Installments amount ", "the unemployment rate (%)", "to pay for above normally 6 months", "the loan type", The model has separate variables to have 5 items "the date of birth", "the restructuring remark", "the occupation", "the school record", "the mailing address", bvarious Rate of accuracy (%) compared on, in training value by “ Artificial Neural Network” 96.07% most superior, next is "Decision Tree" 92.94%, in test value by “Decision Tree” most superior 92.53%, next is “ Artificial Neural Network”92.47%, the rate of accuracy reaches above 90%, but “Discriminated analysis” in primitive and the overlapping confirmation's accuracy respectively is 75.15%, 75.14%.The rate of accuracy has a big difference of both technology methods .
Knew by the findings, this research forecast to default risk of negotiation on trouble debt restructuring of customer bussiness loans, the Data Mining method is suited.

並列關鍵字

Decision Tree ； Data Mining ； negotiation on trouble debt restructuring of customer bussiness loans ； Artificial Neural Network ； Discriminated analysis

參考文獻

郭一聰，2006年，應用決策樹與類神經網路於應收帳款之呆帳預警模式研究，中原大學

吳長洲，2004年，運用資料探勘技術在產險行銷之個案研究，中原大學/企業管理研究

邱挺晏，2005年，消費分期市場評等評分模型建置的方向，淡江大學/財務金融學系碩

蔡介文，2007年，2006年台灣卡債風暴之紛爭解決機制，臺灣大學/法律學研究所碩士論文。

陳龍泉，2006年，卡債問題及消費金融不良債權處理之研究，臺灣大學/財務金融組研