金融機構住宅房屋貸款信用評分系統之建構研究The Building of Credit Scoring System on the Residential Mortgage Finance

本研究以金融機構分析住宅房屋貸款授信風險的方向，評估最能衡量借款戶信用、償債能力的預測變數，由發掘房貸授信風險評估因素，並依照各個因素對於信用狀況的影響程度給予不同的權重，利用勝算比的估計，決定迴歸估計係數的權重，建立較完整評估系統與事前的風險量化研究模型，藉以評定授信條件策略，以期提高信用良好比例，減少銀行呆帳的發生。本研究顯示分別以月付比例、過去信用狀況、貸款成數、借保關係為住宅房屋貸款信用好壞的顯著相關因素。在線性條件預測模型下，建立邏輯斯迴歸分析模型，並依此結果製成業界所能直接應用的評分卡，揭露依評分模型所計算出評分分數等級與所屬百分位點，協助資訊使用者瞭解評分所代表的實際風險意義，並評估風險模型適用性。因每一借款者違約所造成金融機構的呆帳損失並不相同，應用決策樹方法偵測不同風險程度的信用組合，作為建構以住宅房屋貸款信用評分分級為基礎的授信審核政策之參考。
關鍵詞：住宅房屋貸款，授信風險，評分系統，決策樹，邏輯斯迴歸模型

The volume of credit business in the residential mortgage finance has greatly expanded and the use of credit scoring through the evaluation of large credit portfolio becomes crucial to guard against any management risk. The objective of this study is to devise a credit scoring system for finance granting decisions. We describe statistical method to create scorecards and show how the result of the model is applied to calculate score point weights. Scorecards are built using the logistic regression method which estimates the relationship between the individual characteristics and the log of the odds (risk) so that the score point weights can be calculated directly from the regression coefficients.The model performance is usually monitored by the model validation and classification error. We propose an alternative measure for power of model discriminations and credit-granting decisions.
Key words: residential mortgage, credit risk, credit scoring system, decision tree, logistic
regression model