What are the traditional risk control modes of banks?

Recently, the supervision of cash loans is imminent. On the one hand, the actual borrowing interest rate of many small and medium-sized cash lending platforms is too high; On the other hand, the risk control of some platforms on cash loan business is more like "the emperor's new clothes", so that the bad debt rate of the whole industry remains high.

From the point of view of supervision, if the cash loan business wants to continue to operate, it will have to significantly reduce the loan interest rate and reduce various handling fees. Therefore, it is the only way to ensure the operating profit of cash loan business by improving the level of risk control and reducing the loss of bad debts.

First, a brief analysis of cash loan risk control system: the process of point-face-point.

The first "point" is the starting point. The design of cash loan risk control system needs to take the product itself as the starting point. Cash loan products are nothing more than four elements: interest rate (including various rates), term, quota and target population. For each target group, they have certain regularity and * * * in the dimensions of liquidity demand, expected future cash flow, consumption concept, income level and credit status, which further affects their application quota, loan interest acceptance level, repayment ability and repayment willingness. Therefore, the reasonable design of products can effectively reduce the difficulty of risk control and maximize the benefits. For example, for white-collar workers, the monthly salary is more suitable for the loan period of less than one month.

In addition, in addition to the characteristics of cash loan products, its promotion channels are also very important. If a large number of non-target people are introduced through a promotion channel, it will not only reduce the use efficiency of promotion cost and the accuracy of judgment in the later risk control process, but also produce a large number of biased data, which is not conducive to the iterative upgrade of risk control model and product redesign.

"Face" refers to the specific risk control process. Differentiated by time period: the risk control process includes three stages: before lending, during lending and after lending. The pre-lending stage is the core stage of the whole risk control process. This stage includes three steps: application, review and credit granting. Figuratively speaking, the pre-lending stage is a stage of filtering impurities. Third-party credit data, blacklists, anti-fraud rules and risk control models are filters with different apertures. The intermediate stage of the loan is mainly to track and monitor the personal information of the borrower. Once abnormal information is generated, the risk control personnel can find and contact the borrower in time to ensure the safety of the loan as much as possible. The focus of the post-loan stage is collection. In addition, if the borrower applies for loan extension or renewal, it needs to re-examine it with historical data and behavior scorecard at this stage, and make corresponding quota adjustment and risk pool management. In the whole risk control process, we should properly manage the concentration of loans to prevent the problem of insufficient liquidity caused by centralized loans and overdue centralized loans.

The second "point" refers to the key point. There are two key points in the whole cash loan risk control system.

First, anti-fraud. Compared with the risk control under the traditional lending mode, the cash loan risk control is a mild risk control. Because of its small amount and short-term characteristics, cash loan risk control pays more attention to the borrower's willingness to repay rather than ability to repay. Moderate overdue will not only affect the normal operation of the platform, but also increase its income through overdue fees.

Therefore, anti-fraud is the primary topic of cash loan risk control. At present, online loan fraud includes agency, gang crime, machine behavior, account theft, identity fraudulent use, serial transactions and so on. In view of these fraudulent behaviors, common anti-fraud rules include cross-checking, cross-checking, strong feature screening, risk relationship and user behavior data analysis.

Second, the identification of multi-head lending behavior. Multi-head lending refers to the lending behavior of the same borrower in multiple lending institutions. At present, the identification of multi-head lending behavior includes two aspects: (1) obtaining multi-head lending data. Because most of the target groups of cash loans are long-tailed people not covered by traditional lending institutions and lack complete central bank credit data, some platforms engaged in cash loans will cooperate with each other to realize the enjoyment of loan application data. In addition, the cash loan platform will inevitably leave a lot of identity information of the loan applicant when the third-party credit reporting agency inquires about each loan application record. This information will form a reliable multi-head lending database after being filtered by query anomaly detection algorithm. (2) Identification of vicious long-term lending behavior. Vicious long-term lending behavior means that the borrower borrows the new and returns the old or owns a large number of long-term loans at the same time. The determination of the behavior of borrowing new and returning old can be combined with the interval of loan application and the term of loan. If the loan application interval is significantly shorter than the loan term, it means that there is a greater risk of borrowing new and returning old loans.

Second, challenges: contradictions and changes

1. fraudulent means are diversified, technical and internet-based.

Fraud and anti-fraud have always been one of the main contradictions in the lending industry. With the rapid development of online loan business, online loan fraud based on information technology has also intensified. When scammers also started to play big data and machine learning, it is conceivable that many risk control personnel collapsed.

For example, mobile phone verification is one of the most commonly used online auditing methods. Including SMS verification code and filling in operator service password. But this method is also an opportunity for fraud gangs. Because they have a technical equipment-cat pool. Simply understood, it is a simple mobile phone with the function of sending and receiving short messages. A computer can connect multiple cat pools, and a cat pool can insert 8-64 SIM cards. Along with it, there are so-called "card receiving" and "card raising" businesses. When the number time reaches a certain standard, it is possible to verify this anti-fraud means through mobile phones.

In addition, the use of some simulators can help fraudsters to easily modify the IMEI, MAC, IP, GPS and other equipment and environmental information of mobile phones. Under this layer of camouflage and packaging, the anti-fraud measures using equipment and environmental information seem a bit pale and powerless. Moreover, some personal information, such as ID card information, social account number, bank card account number and even U shield, can be bought online by fraud gangs or searched by search engines. Many times, some anti-fraud measures are effective, not because they cannot be cracked, but because of the high cost of cracking, which leads fraud gangs to give up this method.

2. Cold start of wind control model

"Cold start" is the primary problem to be faced in the construction of big data risk control model. Especially for some start-up cash lending platforms, data accumulation is a process from scratch. In the early stage of data accumulation, it is bound to pay a great price. On the one hand, in addition to ensuring the normal risk control process, the platform also needs to invest a lot of labor costs to collect data, build models and back test data; On the other hand, the platform has to invest a high capital cost to purchase third-party data. Compared with the total number of long-tail users not covered by the central bank's nearly 654.38 billion credit data, the customer base of cash loans is still limited, and most platforms are facing the problem of "cold start".

At present, the common method to solve the problem of data cold start is to start with external data. Due to the lack of historical credit records and personal credit data of borrowers, the risk control model loses the basis of directly considering the default risk of borrowers. Therefore, if external data such as user behavior can be combined with collaborative filtering algorithms such as Eigentaste, people with high fraud risk can be identified and filtered to the maximum extent. However, the current situation is that most platforms lack the motivation and ability to process external data, and often use manual audit supplemented by some simple anti-fraud rules.

3. The contradiction between user experience and anti-fraud

In the eyes of users of cash loans, the user experience is reflected in the speed of borrowing and the difficulty of applying. But anti-fraud requires users to provide all kinds of personal information, which greatly reduces the quality of user experience. In the past, many platforms adhered to the principle of "high returns cover high risks" and paid too much attention to traffic. And there are all kinds of cash loan products on the market. In order to ensure traffic, many platforms have slogans such as "just ID card and mobile phone number" and "XX minutes after application". However, with the tightening of supervision, the operation mode of "high income and high bad debts" will be phased out. In order to control bad debts, the cash lending platform has to face the opposite problem of user experience and anti-fraud again. On the one hand, the platform needs to optimize the anti-fraud model, reduce the dimension of input data as much as possible, and shorten the review time of the risk control model; On the other hand, optimizing the user experience from other angles such as customer service and repayment convenience is also one of the feasible methods to alleviate the contradiction between user experience and anti-fraud.

Third, the direction of development

1. Use of unstructured data

The sparsity of structured data such as personal credit data will exist in the cash loan industry for a long time to come. Accordingly, unstructured data is flooding. Due to the serious disclosure, theft and trafficking of personal basic information, the anti-fraud efficiency of conventional structured data is greatly reduced. Compared with structured data, human behavior data is more difficult to be simulated, and it can describe the loan applicant more comprehensively, which has obvious effect on reducing the error rate of anti-fraud model.

From the application of unstructured data, it is difficult to unify the logic between them, and the problems of abnormal, redundant and missing data are serious and difficult to deal with. Therefore, seeking cooperation from third parties such as big data companies and traditional Internet giants will be the first choice for small and medium-sized cash lending platforms. At present, there have been some products in the market that serve finance by refining unstructured data, such as semantic analysis service of a business short message. In addition, Tencent, as one of BAT, also cooperated with Qian Niuniu to launch a pure model cloud risk control system-"Fiona Fang". The biggest feature of this system is the introduction of massive social data from Tencent.

2. Differentiated pricing

Differentiated pricing can also be understood as precise pricing. The essence of differential pricing of cash loans is to accurately price the credit and fraud risks of each loan applicant. At present, the pricing standards of cash lending platforms are too simple, and interest rates and miscellaneous fees are basically adopted. Some platforms will adjust the rate for users who renew their loans. There are also a few platforms that refer to the dimension of personal information provided by lenders when applying. But generally speaking, the current pricing standard is not suitable for the "low rate" characteristics of the future cash loan industry. The so-called price advantage between platforms will be minimal. Customized microfinance demand under precise pricing may become the highlight of the platform.

The construction of big data risk control model provides technical support for the realization of differentiated pricing. Based on a large number of network behavior data, user transaction data, third-party data, partner data, etc. Through natural language processing, machine learning, clustering algorithm, etc. The model can establish a multi-dimensional data portrait for each loan applicant, including personal basic information, behavioral characteristics, psychological characteristics, economic situation, hobbies and so on. With these dimensional characteristics and a large number of historical loan records, differentiated pricing strategies for different lenders, different quotas and different maturities will become a reality.

abstract

Behind the reshuffle of the industry is the efforts made by the cash loan platform to survive. How to ensure compliance, how to obtain low-cost funds, how to replace manpower with technology, and how to find a balance between risk control cost and bad debt rate are all issues that the platform needs to consider and solve in the future. I believe that under the searchlight of industry policy, gold will always shine to the end.