過去銀行授信風險之評估多採5C、5P與財務比率分析，其缺點為過於主觀。因此本研究採用較為客觀的量化模型(風險中立評價模型)衡量國內某家銀行放款部位的信用風險。旨在探討1.風險中立評價模型是否比傳統授信方法更為精確。2.信用價差對於不同的資料特性是否存在顯著的關係。利用市場利率、台灣經濟新報社資料庫（TEJ）及銀行放款資料來建立模型，求出個別的信用價差、違約機率及信用差福（加減碼與理論信用價差之差異）。 實證結果發現：1. 加減碼與資料特性(企業型態、貸放狀況、貸款期別、擔保內容與資金用途)之卡方檢定皆呈現顯著關係，表示該銀行的經驗判斷雖主觀，但方向並未錯誤。2.信用差幅的平均值均為負數，推論該銀行之加碼數不足。為檢視本研究之推論是否正確，經實證顯示違約機率(或損失率)與信用差幅存在顯著關係。表示違約機率高（或損失率高）時信用差幅皆為負值，亦即實際值小於理論值，證實該銀行加碼不足，因此建議銀行應採量化模型，算出精確的加減碼以加強信用風險的控管和放款決策的釐定。 For the past decades, we usually use 5C, 5P methods and financial ratio analysis to measure credit risk for bank loans; however, they are too subjective. Therefore this paper uses an objective and quantitative model, Risk Neutral Valuation, to measure credit risk of loan positions. The model is used to study: 1) If Risk Neutral Valuation is better than traditional methods; and 2) If credit spread is significant correlated with different types of data. Those data are from market interest rates, TEJ, and bank loans. We estimate credit spread, the probability of default and the differences between actual spread (AS) and credit spread (CS). The empirical results: 1) By Chi–Square Test, actual spread is significant correlated with different types of lending companies, conditions of loans, terms of loans, collateral and purposes. It means the direction of the subjective analysis is not wrong.2) the average of the differences between AS and CS are negatives, so we infer actual spread is too low. In order to know if our inference is right, the probability of default (or loss given default) is significant correlated with the differences between AS and CS. In other words, actual spread is lower than credit spread, and it proves our inference. Hence, we suggest the bank should use quantitative model to calculate the accurate spread to enhance credit risk management and policies.
