Credit risk modeling has been a critical and established procedure in the financial services industry. In the past 10 years, China's consumer lending market has experienced rapid growth and many alternative (non-bank) lending firms expanded into sub-urban and rural areas, against the backdrop of policies of financial inclusion. But traditional credit bureau data covers limited proportion of China's population, especially outside tier one cities. Many individuals lack credit bureau data or even banking histories, making it extremely challenging for banks and lending firms to assess individual’s credit quality, and for potential consumers to receive credit at reasonable cost and speed. As of Jan 2018, the PBOC (People’s Bank of China) credit bureau covered less than 400 Million individuals, which is lower than 30% of the total population. Bureau coverage has not been able to keep up with the pace of industry growth. However, most of individual consumers do have personal mobile phones, with rich behavioral data being generated continuously and with minimal cost.

This paper is to discuss a credit modeling work based on mobile behavioral data, instead of traditional structured credit bureau data. Research data are collected from a handset financing product rolled out since Aug 2017 in some provinces by one major telecom company in China. This paper shows that behaviors captured in mobile data can be used to predict consumer credit quality, using call/sms/data usage records matched to installment repayment. On a sample of individuals with no historical credit bureau data available, our analysis shows good prediction power, on both validation data set and test data set. Customers in the highest decile of risk by Our measure are 8.8x – 17.6x times more likely to default on the installment payment than those in the last decile. The method discussed in this paper forms a new way to measure and quantify the credit risk for telecom related lending product.