本國銀行將於2007年起，採取巴塞爾新資本協定計提資本，須依據資產風險性的高低給予不同的風險權重以計算需要之資本額，用以確實反應銀行資產之品質。鑑於新資本協定對於自行建置風險評等以供資本計提使用，有相當嚴謹的要求，本國銀行對於相關量化方法尚屬陌生，多花費龐大的顧問費建置，但往往得不到完整的技術移轉。本研究嘗試以特定銀行資料進行違約機率模型的評估，採用羅吉斯迴歸方法、離散時間危險模型與類神經網路模型並加以比較，發現羅吉斯迴歸方法具有較佳的解釋能力。本研究進一步對此模型進行新資本協定要求之穩定性、同質性檢測，期盼可供本國銀行在巴賽爾資本協定下建置違約機率模型與進行資本計提時參考使用。

By the end of 2007, banks in Taiwan must follow the BASEL Ⅱ Accord to calculate required capital based on risk weights of assets which will reflect their credit qualities. The BASEL Ⅱ Accord provides extensive rules for risk modeling and validation. However, most banks in Taiwan spend a substantial amount for consultation and know-how transfer as they do not have enough experience to construct the model. This study tries to use data from a specific local bank for model development, adopting logistic regression, discrete-time hazard model and neural networks analysis. We find that logistic regression fits better than others. Based on the BASEL Ⅱ requirements, we further conduct the stability test and homogeneity test. Our research will be useful for local banks that want to develop internal default models to comply with BASEL Ⅱ Accord.