Developing a Procedure to Determining the Timing of Updating a Credit Risk Model

• Main Authors: Chang, Che-Yuan, 張哲源
• Other Authors: Chang, Yung-Chia
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/26999089338255878490

碩士 === 國立交通大學 === 工業工程與管理學系 === 99 === The principal activities of banks or financial institutions is the management of customer deposits, depositors of funds through the lessons, and then lending those funds to the need for profits. In lending, it is necessary to consider the lending risk and return. If banks or financial institutions could not control the credit risk of the customer, the credit default or bad loans would lend to heavy losses. Literature of credit risk model, the most part are studying the forecast accuracy about the probability of default. There are few about the timing point for the revision of model studies. The modified model too often lead to practical difficulties in operation. If it is too long to modified the model, would result in model prediction decreased for the future accuracy. First, this study used partial least squares path model to identify variables with the interaction, and then using stepwise logistic regression to constructed model. The input use original variables and the interaction of factors. Find the significant variables of the model, and then compare the inter-annual forecast sample. Count apparent variation of the number of significant variables. If the apparent variation of significant variables ratio larger than 1/3, then the model forecast for the future accuracy rate will be error. It should add new information to correct the model. If the total significant variables ratio less than 1/3, then the model predict for different years of the accuracy of error is small. This model can predict the future sample. After the construction of this process, banks and financial institutions can use this mechanism to determine the timing to modified the model. In order to prevent too often or too long to modified model.