The Modeling of Cash Card’s Churning Rate

• Main Authors: Ching-Fang Hsu, 許瀞方
• Other Authors: Min-Hua Kuo
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/55192202328611948960

碩士 === 世新大學 === 財務金融學研究所(含碩專班) === 95 === Keeping good relationships with good customer is one of the essential factors of enterprise success. If we prevent customers from churn as soon as possible, we may increase the earnings of an enterprise. This paper is about customer retention of cash card business, and we construct cash card’s churning rate prediction model. The sampling frame is one of the commercial banks in Taiwan. To make sure that our samples in the prediction model are good credit customers, those who used to pay behind time or those who are uncollectible accounts are eliminated. After sieving out the customers, we have about 60 thousand cash card holders in research. The research period is from 2001 to 2005. The main prediction variables are basic information of customer, the dealings between the customers and the bank, and the dealings between the customers and the other banks. Because of the difference between male and female, we construct not only the entirety customer retention prediction model, but also male and female customer retention prediction models. We want to find out which are the main factors to predict customer attrition, and compare these three models to see whether the prediction is better when we separate our sample by gender. We find that the significant variables are different among these three models. The basic population statistic variables which significantly influence the male churning rate are ‘age’, ‘job’, ‘salary per year’ and ‘education’. As for female prediction model, only education variable is significant among the basic population statistic variables. The significant variables which customers dealing with the sample bank and the significant variables which customers dealing with other bank of both male and female prediction model are the same, which include ‘the number of owning other bank’s cash card’, ‘the number of owning other bank’s credit card’, ‘the total credit amount’, ‘the increase cash card debt in the last three months’, ‘the average least pay amount in the last three months’, ‘the average withdrawal amount per month’, ‘the average withdrawal times per month’, ‘the average revenue increase per year’, ‘dealings period’, ‘the first time the cash card use rate up to 90%’and ‘the margin increase of the cash card use rate in the last three months’. The whole prediction rate of the entirety model is higher than the other gender prediction models, but the attrition prediction rate of the entirety model is lower than the both gender prediction models. The type I error of these two different gender prediction models are both lower than the entirety prediction model. In addition, when we use these different gender prediction models to predict customer attrition, and combine these two potential effects together, we find that the bank minimize more loss than the entirety prediction model. In our different gender model, if the bank can retain all its customers who are predicted to lose, it can save loss up to $40,166,000(TWD) more than the entirety prediction model. And if it can only retain 30% its customers who are predicted to lose, it can save $12,054,000(TWD) as well. These findings reveal that the cash card’s churning rate prediction model has good predictability which may increase the revenue of the enterprise. Besides, constructing different gender prediction model is also valuable and is worthy of consideration.