- 學位論文
住宅貸款授信風險評估─多層函數連結網路(MFLN)之應用
The Risk Evaluation for House Loan Credit─Application of Multilayer Functional-Link Network
摘要
本研究旨在分析住宅貸款之授信風險評估。授信工作係金融機構最主要的業務之一，故授信品質的良窳攸關金融機構的生存發展。近年來，本國銀行整體的逾放比從90年底7.48%、91年底6.12%至92年底4.33%，三年來下降3.15%，更凸顯平日應做好授信風險管理的重要性。 研究變數包括:性別、年齡、雙薪家庭、教育程度、服務機關、職業、年資、所得、擔保品地區、擔保品現況、屋齡、貸款額度、貸放成數、貸款方式、共14項變數。比較正常戶及逾期戶兩群體樣本之差異，並運用多層函數連結網路(MFLN)的模式建立住宅貸款授信風險評估模式，分析模式實用上之可行性。 本研究實證結果可歸納成下列幾點結論： 一、在14個影響住宅貸款授信風險之變數，樣本特性交叉分析得知，性別、年齡、雙薪家庭、教育程度、服務機關、職業、年資、所得、擔保品地區、擔保品現況與逾期還款有顯著相關。 二、本研究以多層函數連結網路(MFLN)，以雙薪家庭、職業、學歷、年資、所得做為輸入變數，繳息正常與否做為輸出變數，收斂情形理想，然敏感度分析上對雙薪家庭、職業有明顯的正相關，而年資呈微弱負相關。另轉換成C程式，此模式的正確率為81%。
並列摘要
The main purpose of this Research is to analyze the risk evaluation of house loan credit. Granting credit is one of the essential practices for financial institutions, hence, the existence and development of a financial institution depend on whether its quality of granting credit is bad or good. In recent years, the national banks’ total over loan rates, from 7.48% of the end of 2001 to 4.33% of the end of 2003, has reduced by 3.15% for these three years, which apparently indicates the importance of having good control for granting credit risk in ordinary days. The variables of this Research totally include 14 variables such as sex, age, family with double salary, education level, working place, occupation, working accumulating year, income, security area, security condition, house age, loans-to-value ratios, loan types, etc. The ways for the Research is to compare their differences between normal households and overdue households and construct a risk evaluation mode for house loan credit by using MFLN and analyze the feasibility of the modes in practice. The results demonstrated by this Research can be concluded into several points as follows: 1.In the 14 variables which affect credit risks for house loans, known from the analysis of sample characteristics, overdue reimbursement is remarkably related to sex, family with double salary, education level, working place, occupation, working accumulating year, income, security area, and security condition. 2.By means of MFLN, the variables such as family with double salary, education level, working places, occupation, working accumulating year, and income are considered as input variables in this Research, and the condition of interests paying as an output variable. Convergence is good, but in sensitivity analysis there is an apparent positive relevance to family with double salary and occupation, and a weak negative relevance to working accumulating year. Further, it changes into C formula, which of correct rate is 81%.