M銀行非政策性購置型房貸信評建構之研究

Building Credit Scoring Model for Non-Policy House Purchasing Mortgages -A Case Study of M Bank

向大龍(Ta-Lung Hsiang)

指導教授 ： 吳榮杰

國立臺灣大學/生物資源暨農學院/農業經濟學研究所/碩士(2010年)

https://doi.org/10.6342/NTU.2010.10138

摘要

在全球化發展之推波效應下，次貸金融危機由美國境內蔓延至境外，由工業先進國家波及新興經濟體及開發中國家，由房屋市場波及金融機構與金融市場，由金融面波及實質經濟，甚至出現經濟衰退影響金融，金融再影響經濟之惡性循環，顯著抵銷各國紓困措施之效果。因此，良好之風險控管已成為新巴塞爾資本協定規範中最主要的要求，而在使用信用風險管理工具時，其最重要的就是信用評分，也就是運用各種主、客觀認定與交易對手償債能力相關的資訊，對於其整體信用能力加以衡量，並依照評量結果給予適當的評分。房貸評分卡模型會根據每一個房貸客戶之相關變數進行分析，然後將這些資訊歸納成一個授信違約機率。本研究採用邏輯斯迴歸方法，並以M銀行非政策性一般型房貸產品為樣本資料，從西元1997年至2003年共7年樣本中，共收集37,231筆好樣本及違約樣本348筆進行模型建置,並隨機抽樣30%做為樣本驗證；從多個變數中，透過AUC值及單變量迴歸原則，先篩選出數個預測力高的變數為候選變數；並經過分群方法最後選出5個最終變數模型，有效區分好壞客戶，AUC值為84%，屬於預測力強之模型，將建置樣本之最終顯著變數及係數代入驗證樣本中，AUC值也有80%之區隔好壞樣本之效度。並透過KS及PSI母體穩定性測試發現，KS值達0.62，PSI值也幾乎在0.001以下，實證結果顯示模型具一定之預測能力及穩定度，最後以預測之違約機率進行切等，共區分10個風險等級。透過房貸評分卡建立，提升銀行授信風險之控管。面對房貸景氣熱絡之際，本研究建議金融機構應及早建立相關之早期預警指標，針對各資產組合之風險做評估分析，適時搭配評分工具，執行更全面之風險管理。

關鍵字

信用評分 ； 邏輯斯迴歸 ； 分群 ； 效度 ； PSI值

並列摘要

Under the effect of globalization, the Financial Tsunami started from USA and rapidly spread to other parts of the world. This crisis first affected the highly industrial developed countries before extending its influence to the newly emerging markets and developing countries. The crisis of mortgage market resulted in catastrophes of financial institutions and the financial market and caused a full scale meltdown in economy. Like a vicious circle, the slowdown of economy backfired. It not only influenced the financial markets but also offset the relief measures from the government. Therefore, good risk control has become a main requirement in BASEL II regulation. On the other hand, when using tools of credit risk management, the most important thing is credit scoring. Credit scoring is to use subjective as well as objective means to identify the solvency ability of the counterparty in the transaction. At the same time, we use this information to evaluate and to give a suitable score according to overall credit ability of the object of study. A scoring model will be analyzed by variables of mortgages customers, aggregating this information to a credit default probability. On a conservative basis, we use logistic regression methodology to evaluate the non -policy general mortgages with the case of M bank as samples, and exercise 37,231 good samples as well as 348 defaulted samples. The samples were took from the year of 1997 to 2003. Also, we use random sampling way about 30% to validate model. First of all, we filter several variables of higher prediction as candidate variables through AUC values and one variable regression rules. Then we filter the final 5 variables by group methodology. From The AUC value of discriminatory development model is 84%, and validation model was used by the final coefficient and variables from development model. Also, AUC is 80% to separate the good or bad customers. We find KS value to be 0.62 and PSI value to be under 0.001 by KS and PSI test, and prove that there is very credible and stable with our model. Finally, we use predicted PD to separate ratings to 10 rankings of risk. Facing the business cycle from recession to boom, we suggest banks build warning index in the early period, and focus on credit scoring card to evaluate assets and portfolio and to execute the overall risk management.

並列關鍵字

Credit Scoring ； Logistic Regression ； Binning ； validity ； PSI

參考文獻

3.江百信、張金鶚，1995。「我國購屋貸款放款條件之研究《住宅學報》3」。

7.鍾經樊、黃嘉龍、黃博怡、謝有隆，2006。「台灣地區企業信用評分系統的建置、驗證和比較」。經濟論文。

4.沈大白、張大成、楊佳寧、陳漢沖編著，2003.4。「金融業風險管理實證論文集」。台灣經濟新報與數位財經合作出版。

5.張大成，2003。「違約機率與信用評分模型」。台灣金融研訓院財務季刊第四輯。

1.Calhoun, C. A., and Y. Deng. , 2002.“A Dynamic Analysis of Fixed and “Curtailment as a Mortgage Performance Indicator,” Journal of Housing Economics, 14. 294–314.