Prediction Model for Creating the Default Risk of Cash Cards

• Main Authors: Cheng Ya-Chen, 鄭雅真
• Other Authors: 謝昆霖
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/h43j9s

碩士 === 國立臺東大學 === 資訊管理學系碩士班 === 97 === Changes in industrial structure and consumer consciousness made cash cards the most popular financial product during 2002. However, numerous issues of cash cards rapidly increased defaults on cash cards, with issuing banks attracting high-risk customers by offering easy application and no collateral and guarantor requirements. Moreover, to increase their share of the cash card market and thus their profits, most issuing banks offered low and attractive prices, as well as simplified credit evaluation and approval procedures. All of these factors contributed to loose credit evaluation standards and approval processes. The cash card market gradually shrank after experiencing the crisis of credit card and cash card in 2005. Given the deterioration in the consumer financial environment, learning from past experience of bad debt and identifying risk factors for default on cash card debt has become a key research focus. This investigation is based focuses on random samples from a single cash card issuing bank. The sample data were obtained between 2004 and 2005, and the target population comprised cash card holders who had used their cash cards after they were activated. Three-hundred of the 600 random samples comprise customers whose payments are over 2 months overdue. Meanwhile, the remaining 300 samples comprise customers who with a good payment record. This study applies analyses including Discriminant Analysis from SPSS, Logistic Regression and Back-Propagation Network which is based on the financial crisis prediction mode. These analysis methods are used to identify factors most closely related to the likelihood of cash card holders defaulting on their cash card. The main factors include the following. Cash card holder annual income, available credit, number of cash cards and credit cards already held, the frequency of checks by other banks, the approved loan limit in the second month following initial cash card use, actual credit drawn in the second month following cash initial card use, amount of loan repaid one year after initial card use. Among three analysis methods, Discriminant Analysis achieves accuracy of 69%, Logistic Regression achieves 69.5% and the Back Propagation Network Model achieves 90%. Regarding the accuracy of prediction of overdue customers, the accuracy ratio of Discriminant Analysis is 69.3%, while that of Logistic Regression is 69.7% and that of the Back-Propagation Network Model is 90%.