vintage analysis is widely used in the financial and credit industry. The analysis method is to track the credit accounts generated in different periods and compare them synchronously according to the age, so as to understand the asset quality of approved accounts in different periods. It is a so-called vertical concept. The following examples illustrate the Vintage analysis of delinquent second-cycle accounts based on aging (see Table 1)

In Table 1, it is listed as card issuing time and operating time. The data of 2.12% is the amount of the credit card issued in April 26 that was in arrears for the second period in July 26 divided by the overdraft balance of this batch of credit cards in July 26, and so on, to get the data of the whole table. On this basis, according to the aging as the operating time minus the card issuing time, the data between tables is converted to get Table 2, and a line chart is made (see Figure 1).

before talking about mobility, let's define the concept of overdue stage. Overdue means that you don't repay the loan when it is due, then you are overdue. According to the number of overdue days, it is divided into eight stages, such as M-M7+. M is not overdue, M1 is overdue for 1 29 days, M2 is overdue for 3 59, and so on, and M7+ is overdue for more than 18 days. With the concept of overdue stage, the mobility is easy to understand. Simply put, it is the change of customers in one overdue stage to other overdue stages. Mobility can usually be used to predict future bad debt losses at different overdue stages. For example, M2-M3 refers to the ratio from overdue stage M2 to overdue stage M3. The following chart 2 is a breakdown of the aging of online loans in various periods, where WO stands for Write Off. The cells in the same color in Table 3 indicate the migration path of non-performing loans. It can be seen that out of the normal overdraft of RMB 1,4,844 in July, RMB 237,327 became overdue in M1 after August, and RMB 55,372 became overdue in M2 in September.

in risk control, our fundamental purpose is to identify bad customers, grasp the characteristics of bad customers that are significantly different from normal customers through historical data, and use this as a standard to predict future bad customers. In fact, it is difficult to define the quality of users. It cannot be said that users who have expired are bad users. Maybe people actually want to pay back, but they just accidentally forget to pay back. Moreover, sometimes, "appropriate" overdue can also increase the company's overdue interest income. The bad customers we are concerned about are those who are bad to a certain extent, that is, customers with high overdue grades and do not pay back.

? The vintage mentioned above is to judge the quality of customer groups from the time dimension, and the rolling rate mentioned below is to judge the quality of customers from the degree of behavior, which can help us judge whether some overdue customers can be rescued again and recover some costs.

? The rolling rate simply means taking a certain time point as the observation node, observing the worst overdue stage of the customer in a period of time before that point (such as half a year), and tracking the development of the customer to other overdue stages in a period of time after the observation point, especially to a worse degree. Take a chestnut as an example. Today is May 25th, 218. Take today's 1, customers and count their biggest overdue stage in the past six months. Then track their performance in the second half of the year. The following figures are purely fictitious, just to illustrate the problem. Each company has its own observation data and tracking data.

? M customers in the next six months, 98% of customers will still maintain the normal M state < P >? In the future, 8% of customers at the maximum overdue stage M1 will become M, but 2% will continue, and even 5% will develop to a worse degree < P >? At the maximum overdue stage, 4% of customers in M2 will continue to deteriorate in the future, and about 22% will become M (completely good);

? At the maximum overdue stage, 6% of customers in M3 will continue to deteriorate in the future, and about 15% will become M (completely good);

? 8% of customers in M3+ with the maximum overdue period will continue this state (hopeless) in the future.

? According to the above data, we may come to the conclusion that customers who are overdue for more than 3 periods are all bad customers who can't be rescued. If we want to tighten the conditions a little, then we may choose more than 3 or 2 issues. If I want to give a bad rating, then I can set more than three periods as extremely bad, more than three or two periods may be set as moderate bad, and people who exceed one period may be inadvertently bad. These features can be put into the sample features of risk control modeling in the future.

with the previous preparation, the rush rate is relatively simple. It refers to the proportion of customers changing from M to M1 on a certain repayment date. For example, today, there are n M customers who have reached the repayment date, and there are m customers who have repaid on time, so today's rush rate is (n-m)/n. It is different from the FBD below.

(If you have different opinions, please let me know! ! ! )