In developing markets, the credit evaluation of lenders is usually subjective, inefficient and expensive. Generally speaking, the loan officer will visit the lender at home and interview him and his neighbors to judge whether they can borrow money. The data coverage of credit reporting agencies to lenders may be incomplete or even missing. This reflects the fact that in these markets, financial institutions have little or no historical information about many consumers. In such an environment, lending institutions can only focus more on cross-selling existing customers and cater to those customers with better credit data. (Of course, these people tend to be richer. Therefore, customers without credit records are often excluded from the formal credit business, or have to accept loans far higher than the market interest rate. It is conceivable that for most of these people, because they can't afford the high borrowing cost, they will eventually seek loans from family and friends.
The problem of being rejected by credit institutions is also common in some high-income economies, and the reason is nothing more than the lack of credit data. Even in some of the most developed economic entities (see figure 1), the information of credit reporting agencies is only a part of the population, while the information of the rest of the population is very poor, including people without bank accounts, new entrants to the labor market (such as graduates), housewives returning to the labor market, new immigrants or foreigners.
The fundamental problem is that, whether in mature markets or developing markets, lenders often rely heavily on their financial data for risk assessment. However, for many people in the world, these traditional financial data are seriously missing. Therefore, all this is bound to become a stumbling block to consumer credit and inclusive finance.
However, some measures to replace traditional credit risk management are constantly emerging. For example, information about prepaid mobile phones, psychological test data, social media activity information, and e-commerce behavior data. The introduction of these data has injected new vitality into risk assessment. At the same time, this paper will introduce four different evaluation methods based on these data to illustrate.
1) Prepaid mobile phone information:
It is estimated that among the 2.5 billion people who cannot enjoy financial services, 654.38+06 billion people have mobile phones, most of whom are prepaid users. Some risk management service providers (such as ci grami, FirstAccess, MasterCardAdvisors, etc.). ) have developed corresponding risk control models for this group of people. According to these models, the repayment willingness and repayment ability of the lender can be predicted to some extent by displaying the payment, telephone and internet usage of prepaid users (see Figure 2). A person's mobile phone usage data for several months can provide enough sample size for modeling and calculation.
For example, there is a positive correlation between the number of calls initiated (rather than the number of calls received) and the call duration. On the contrary, in some modes, it is a feature of low-credit customers if they answer more calls or have a relatively small circle of friends during working hours.
Therefore, risk control modeling based on prepaid mobile phone data can greatly help some developing markets that lack credit data to realize the healthy growth of inclusive finance.
2) Psychological test:
Psychological test refers to measuring the knowledge, ability, attitude towards life and personality characteristics of the subjects by questionnaire. Companies such as VisualDNA and EFL can predict risk factors such as the willingness to repay through the feedback of the subjects on a series of questions or tests. At the same time, these results can also generate personal credit risk scores for lenders to refer to when lending (see Figure 3).
For example, some problems can be deeply understood by measuring the impression of the loan applicant and the relationship between the lender and others. Similarly, measuring the logical thinking ability of loan applicants in adverse circumstances can help lenders better evaluate how to react once the financial situation of loan applicants deteriorates.
In some countries, psychological tests have been widely used in credit scoring (see Figure 4). This technology is most suitable for those markets that have a large number of personal credit data, but their data are seriously incomplete and the Internet has become highly popular.
3) Social data:
Generally speaking, the use of social media and other online behaviors can be collectively referred to as "social data". For companies like Lenddo, DemystData and Kreditech, social media has a huge data mine (see Figure 5).
These solution providers use social data provided by more than 100 information sources (public or private channels) to build risk control models. They found that the analysis of online data helps to evaluate a person's identity, income level and employment status. Solution providers usually cross-match the data obtained from multiple channels and compare their results with the information registered by lenders at the time of application, thus giving the evaluation results.
In some models, what habits loan applicants use to apply for loans online can often show their credit indicators. For those users who carefully read the loan application instructions and complete the online loan application at home, it will be considered that the credit risk is low. In addition, users who fill in all uppercase or lowercase fonts when applying for loans will be considered to have higher credit risk. Similarly, if the loan applicant is a network "celebrity", such as the size of Weibo, the forum moderator can improve his credit score.
Risk modeling based on "social data" is suitable for evaluating individuals with incomplete credit data in high-income markets or evaluating the middle class in relatively underdeveloped markets.
4) E-commerce behavior:
Payment-related data provided by wholesalers and e-commerce companies are conducive to evaluating the credit status of small and micro enterprises and business owners. Ali Finance is a company specialized in providing loan services for small and medium-sized enterprises in China. It evaluates the credit status of small and micro enterprises that conduct e-commerce transactions on Alibaba's website. Ali Finance can accept the loan applicant to lend with the unexpired notes receivable as collateral, and its published bad debt rate is lower than 1%.
Another different example is Kabbage, a company that provides loans to small and medium-sized enterprises in the United States and Britain. It uses credit scoring model data from Amazon, UPS and Intuit. The data used for scoring may include the sales and shipments of loan applicants, and even the feedback rating of customers on their products.
After introducing the above cases, the question is, how to establish an effective strategy to better play the application of non-financial data in the field of risk control?
Any lender who wants to use non-financial data for risk scoring needs to find a solution provider who has accumulated some experience in using non-financial data for risk control modeling. An effective strategy at least includes the following steps:
1) Determine the data source needed by the market:
As mentioned in the above case, psychological tests and social data are more suitable for markets with high Internet penetration.
2) Establish and improve the risk control model:
In the case of small samples, it is necessary to construct a rule-based risk control strategy. By cross-checking the risk modeling results of people with financial data samples with the risk modeling results of mobile phone data, hydropower data, online behavior data and other data, the validity of non-financial data is relatively ideal. The models trained from these samples can be applied to a wider range of people.
3) Ensure data stability:
In some markets, there is a certain threshold for obtaining effective third-party data sources. For some mobile operators, there is a lack of motivation to transform the existing network to meet the requirements of risk control modeling such as DPI convergence, security control and data cleaning.