房貸放款業務為銀行業資金運用最主要方式，因此積極建立客觀、效率化且準確度高的授信評估模型，藉以降低逾期放款，提升銀行經營績效與獲利率為銀行首要目的。有鑑於時下一般研究皆以可被觀察到還款行為的房貸戶為主，進行首次授信模型研究，如此將無法反應出於銀行初審即已被婉拒的申請人對模型之影響，本研究提出因應建模樣本限制之二階段模型，以期能有效運用更多的訊息加以修正。
本研究以國內某家金融機構針對台南市於2004年6月至2006年4月期間之個人房屋貸款申請案件中4,274筆房貸申請人資料及其中的2,890筆房貸戶資料為樣本，分別建立第一階段與第二階段模型。研究中採用邏輯斯迴歸分析、Cox迴歸分析及區別分析建立模型，以分析銀行業在房貸授信模型中，必須考慮之風險評估因素，納入研究中有個人基本屬性、擔保品屬性、審核條件、還款能力及聯徵查詢資料等變數。其研究結果如下：
(1) 由二階段模型顯示，申請金額、年齡、性別、學歷、房屋型態、屋齡、房屋鑑價及近三個月被聯徵查詢總次數、保証人之有無、償還方式、貸放成數及客層等12個變數皆是重要變數，且它們與逾期違約之影響關係和文獻結果皆一致。
(2) 邏輯斯迴歸分析模型在切點 表現較其它切點為佳，就訓練組樣本而言，邏輯斯迴歸分析在逾期戶下判別之正確率達52.4%，較區別分析模型的24.83%高出近一倍，本研究評斷邏輯斯迴歸分析模型之表現較區別分析佳。而在Cox等比例風險模型上，就其測試組預測結果而言，好壞參半，可能是因為設限資料過多的原故。
(3) 本研究為因應模型於實務上應用之便利，將模型以 Nomogram形式展現，藉以提供銀行業者在模型使用上之推廣，提高授信作業效率。

Housing mortgage plays a major role in the banking system, building a effective and correct leading evaluation models is very important. We can use it to reduce Non-Performing loan, provment performance and enhance profits. Typical research involves gathering information on the background of the applicant and how the applicant will make the loan payments. This model does not reflect how the bank rejects the applicant. This new research uses a different method which involves a two-stage model, which will gather more information to correct the errors found in the current method.

This research is based on the the information gathered from 2,890 out of 4,274 cases of mortgage loans from an unanimous bank inTainan, Taiwan between June 2004 and April 2006. There are two stages in this research, which are named Stage 1 and Stage 2. The research utilizes Logistic Regression Analysis, Cox Regression Analysis, and Discriminant Analysis. In the model using these analyses, we must take into account the risk as a factor, especially when the bank is involved. The credibility of the individual, the home, and ability to make payments were also accounted for in the research process. The results are as follow:

(1) The information gathered in stage 2 includes: amount applied for, applicant’s age, sex, and education, the condition of the house, the age of the house, and how many times within 3 months the house has been appraised. In addition, the bank will ask about the guarantor, the source of income, and the loan term. Additionally, how these factors will affect the non-performing loan.
(2) The Logistic Regression Analysis has a figure of , which shows it is better than the other analyses. From the trial samples, Logistic Regression Analysis has 52.4% of correctly evaluating the applicant, which doubles the percentage when using Discriminant Analysis, which is 24.83%. This research results in finding that Logistic Regression Analysis is the best method to be used. However, when compared to the Cox Regression Analysis, the samples’ results fluctuates between good and bad, mainly due to too many detailed information provided.
(3) This study use the type of graphics to show the models, we hope tha can provid a extension way to help the people who do not toch models before.

一、中文文獻
1. 王濟川，Logistic迴歸模型-方法及應用，第二版，台北：五南圖書出版股份有限公司，2004。
2. 李桐豪和呂美慧，「金融機構房貸客戶授信評量模式分析-Logistic迴歸之應用」，台灣金融財務季刋，頁1-20，2000。
3. 李海麟，「銀行消費者房屋貸款授信評量之實証分析」，國立中正大學國際經濟研究所，碩士論文，2003。
4. 周建新、于鴻褔和陳進財，「銀行業房貸授信風險評估因素之選擇」，《中華管理評論國際學報》，第7期，頁77 -103，2004。
5. 林建州，「銀行個人消費信用貸款授信風險評估模式之研究」，中山大學財務管理研究所，碩士論文，2001。
6. 金融出版社編，銀行授信教戰準則，台北：金融出版社，1995。
7. 施孟隆，「Logit 模式應用於信用卡信用風險審核系統之研究-以國內某銀行信用卡中心為例」，《金融財務季刊》，第4期，頁85-104，1999。
8. 財團法人台灣金融研訓院編，銀行授信實務概要，台北：財團法人台灣金融研訓院，2006。
9. 張文生，「銀行建構信用卡信用風險即時預警系統之研究」，中原大學企業管理研究所，碩士論文，2000。
10. 張文智，「應用Logistic Regression於個人房貸戶信用評估之研究」，中山大學國際經濟研究所，碩士論文，2003。
11. 陳清河，「銀行房貸客戶授信風險評估模式之研究」，南華大學財務管理研究所，碩士論文，2004。
12. 鄒香蘭，「我國股票上市公財務危機預警模型之比較」，國立彰化師範大學商業教育研究所，碩士論文，2001。
13. 劉應興譯，類別資料分析導論，台北：華泰文化事業公司，2003。
14. 蔡明憲，「金融機構消費信用貸款授信評量模式」，國立中山大學財務管理學系碩士在職專班，碩士論文，2002。
15. 鄭婷月，「汽車貸款客戶之風險研究」，國立成功大學統研所，碩士論文，2003。
16. 鍾岳昌，「以比例危險模型估計房貸借款人提前清償及違約風險」，國立政治大學財務管理研究所，碩士論文，2004。
二、英文文獻
1. Collins, R.A. and Green, R.D., “Statistical Method for Bankruptcy Forecast,” Journal of Economics and Bussiness, Vol. 34, pp.348-354, 1982.
2. Clive, L., “Identifying failing companies: A Revaluation of the Logit, Probit and DA approaches, ” Journal of Economics and Business, Vol. 51, pp. 347-364, 1999.
3. Collett, D., Modelling Survival data in Medical Research. London: Chapman & Hall, 1994.
4. Espahibodi, P., “Identification of Problem Bank and Binary Choice Models,” Journal of Banking and Finance, Vol. 15, pp. 53-71, 1991.
5. Lee, S. H. and Urrutia, J. L., “Analysis and Prediction of Insolvency in the Property-Liability Insurance Industry：A Comparison of Logit and Hazard Models,” The Journal of Risk and Insurance, Vol. 63, pp. 121-130, 1996.
6. Petetta, M., Survival Analysis Using the Proportional Hazards Model:course notes. N.C.: Cary, 2001.
7. Steenackers, A. and Goovaerts, M.J., “A Credit Scoring Model for Personal Loans,” Insurance Mathermatics Economics, vol. 8, pp.31-34, 1989.