雙卡因素與房貸風險評估關係之研究

A Study of the Influence of Credit Card And Cash Card in Home Loan Risk Evaluation

楊庭芳(Ting-Fang Yang)

指導教授 ： 劉立倫

中原大學/商學院/會計研究所/碩士(2007年)

https://doi.org/10.6840/cycu200700379

摘要

國內歷經雙卡風暴，房屋貸款是否為下一波金融風暴，已成為關注之焦點。傳統上銀行對房貸授信審核大部分依靠的是授信人員個人的經驗與主觀的判斷，尤其是著重擔保品價值做為決定授信的條件。但隨著新巴塞爾協定預定實施行程，「個人房貸評分」已是目前金融業正在著手推行的重點。

過去研究者對有關金融授信風險評量的文獻，所研究的變數皆以表內變數為主，而忽略借款戶是否具有運用所得以攤還借款之意願，故本次除了延續過去的研究變數，並增列了聯合徵信中心查詢之信用卡及現金卡使用情況為表外變數做增額分析，探討雙卡因素與房貸逾期之關係；並以近年來深受各領域學者所重視，並可模擬專家系統的倒傳遞類神經網路模型為研究方法，期望能準確及客觀的評量申請人信用風險程度，從而能降低房屋貸款違約的風險及降低呆帳率，增進經營績效。

本次研究樣本以國內某大行庫2004 ∼ 2005年提供不動產為擔保品之房貸戶為研究之對象，抽取正常還款的正常戶138件、逾期三個月以上的逾期戶44件，合計182件，做為建立授信評量模式的樣本。並就授信申請書表選出表內變數，包含：年齡、婚姻狀況、職業、戶籍、服務年資、家庭年收入、貸款金額、貸款成數、保證人、過去往來情形等十種。再由財團法人金融聯合徵信中心查詢資料選出表外變數，包含：信用卡張數、現金卡張數、是否使用循還利息、是否動用現金卡等四項，共計十四項做變數分析。

在進行倒傳遞類神經網路模型測試前，為了讓樣本有良好的收斂效果，首先將樣本資料進行交叉分析，藉此選出表內及表外顯著變數，再以顯著表內變數測試模型Ⅰ；其次再加入表外變數做增額測試模型II。研究結果顯示：本次建立的倒傳遞類神經網路模型Ⅱ比模型I對授信風險判別評估上，辨識能力更能夠接近100%判決精確度，顯示加入雙卡因素後確實能增進房貸授信風險之評估，故此網路模式Ⅱ已具備預測房貸風險評估之能力，可作為房貸事前授信決策與事後預警管理之參考。

關鍵字

授信評估 ； 房屋貸款 ； 倒傳遞類神經網路 ； 雙卡風暴

並列摘要

In the wake of the dual-card storm (credit cards and cash cards), house loans have turned out to be the focal point of the next financial storm in Taiwan. Conventionally, banks rely on their credit personnel to screen the house loans according to the personnel’s personal experience and subjective judgment. It is especially so in determining the value of collateral in order to set the credit terms. However, as the New Basel Accord is on the way to being implemented, “personal house loan rating” has become the key task requiring promotion by the financial industry.

The variables shown in earlier research papers, related to the risk valuation of financial credit, are mainly based on endogenous variables while ignoring the loan borrowers’ intention concerning using their income to repay their loans. Thus, in addition to continuing the variables used in past research, this study has added the utilization status of credit cards and cash cards, enquired by Joint Credit Information Center, as exogenous variables with which to carry out an increment analysis and investigate the correlation between the dual-card factor and overdue house loans. The study uses back propagation neural network (BPN), which is increasingly emphasized by academics in various fields and which can simulate the expert system, as the research methodology in an attempt to accurately and objectively evaluate the degree of an applicant’s credit risk, so that the default risk of house loans and bad debt rate can be effectively reduced while enhancing operating performance.

This study uses the house loan borrowers who provided real estate to a major local financial institution as collateral in the period between 2004 and 2005, as its sample target, and 182 accounts, including 138 normal accounts and another 44 accounts overdue for more than 3 months, are randomly selected as the samples to establish a credit valuation model. At the same time, the 10 endogenous variables, including: age, marital status, occupation, household register, working seniority, annual family income, loan amount, loan to value percentage, guarantor, and past credit records are selected from the items in the credit application form. Moreover, four exogenous variables: the number of credit cards, the number of cash cards, whether revolving interest is used, and whether cash card is used, are selected from the data enquired by Joint Credit Information Center. There are a total of 14 variables used for analysis.

In order to obtain a good convergence effect on the samples before proceeding with the test of back propagation neural network model, the sample data is first cross analyzed to identify exogenous and endogenous significant variables, followed by using significant endogenous variables to test model I, before adding exogenous variables to do an increment test on model II. The study results show that in terms of credit risk judgment valuation, the model II of back propagation neural network established this time can approach almost 100% accuracy of judgment. It illustrates that dual-card issuance did enhance the valuation of house loan credit risks, so the network model II has been equipped with the ability to predict house loan risk. The prediction can be used as reference for the credit decision-making prior to giving house loans and the pre-and post-warning management.

並列關鍵字

Credit Valuation ； Back-Propagation Neural Network ； Dual-Card Storm ； House Loans

參考文獻

1.江百信、張金鶚（1995），「我國購屋貸款放款條件之研究」，住宅學報，第三期， 頁1-20。

1.吳萬益（2005），「企業研究方法」，二版，華泰文化事業股份有限公司。

1.王思評（2005），「房屋抵押貸款授信風險評估研究 ─ 以X銀行為例」，大同大學事業經營學院碩士論文。

10.林國順（2003），「房屋貸款逾期還款預警模式之研究」，大同大學事業經營研究所碩士論文。

1.Balakrishnan, P.V.,Cooper, M.C., Jacob, V.S., & Lewis, P.A. 1994, A study of the Classification Capabilities of Neural Networks Using Unsupervised Learning-A Comparison with K-Means Clustering. PSYCHOMETRIKA, Vol.59(4), pp.509-525.