Fe of mL: introduction to vintage curve /vintage analysis, calculation logic and detailed introduction to case application

ML And FE：Vintage curve /Vintage Introduction to analysis 、 Calculation logic 、 Detailed introduction of case application

Catalog

Vintage Introduction to curves — Understand through the wine industry Vintage curve

(1)、Vintage source

(2)、Vintage The meaning of the curve

Vintage Curve application case extension — Credit field

(1)、 Wine VS credit →Vintage The meaning of the curve

(2)、 Indicators to be determined in the credit field —— Account performance period — Credit products Vintage Curve mining

(3)、Vintage Curve drawing logic

---

Vintage curve brief introduction — Understand through the wine industry Vintage curve

(1)、Vintage source

         first ,Vintage The original meaning is , A wine made in a particular year . Because the grapes picked every year will be exposed to sunshine 、 The temperature 、 Precipitation and other factors , There will be differences in the quality of the final wine . After a certain year of hoarding , The quality of wine will be stable , That is, the quality is mature , This period of years is called maturity （maturity）.
        Vintage The analysis was originally used to evaluate the quality of wines from different years , Changes with the passage of cellar time . For simplicity , We use alcohol concentration as a measure of wine quality , Appointment ： The higher the concentration , The better the quality . First , Record the year of cellar entry as the label of the batch of wine , This is also called Vintage perhaps Cohort. after , We will regularly sample and measure the alcohol concentration every year , Save record data , As shown in the figure .

         After years of data accumulation , We can plot the change of alcohol concentration over time Vintage curve . It is said that , The more wine is brewed, the more mellow it is ,Vintage The curve is usually monotonically increasing .

(2)、Vintage The meaning of the curve

         Of wine Vintage The flow chart of curve analysis is as follows ：

Vintage What can a curve do ？

1)、 Analyze the law of change ： To evaluate the change of wine quality in different years with the passage of cellaring time . such as ,** The wine concentration of the year is 1 It can reach a higher level in , But the rise is slow ;** Vintage concentration , Although the starting point is low , But the rise is fast .
2)、 Determine final quality ：Vintage The final stable value of the curve , It indicates the final alcohol concentration of this batch of wine .
3)、 Determine maturity ： according to Vintage The graph shows , Generally, when entering the cellar 6 After year , The alcohol concentration is stable , It can be determined that the maturity period is 6 year , We started at the 6 You can open a barrel to taste in .
4)、 Analyze the influencing factors ： according to Vintage diagram , We can analyze the environmental impact factors of grapes in a certain year , So as to improve the production process . such as , Due to insufficient light in a certain year , Less sugar accumulation , The alcohol concentration may eventually be lower . We can artificially increase the light intensity .

Vintage curve Application case extension — Credit field

​​​​​​​         In the field of financial risk control , It is usually necessary to observe the full cycle risk situation of users , Define the maturity of user risk ,Vintage The curve just does this .
​​​​​​​         Now? ,Vintage Analysis It is usually translated as Aging analysis , Because in the field of credit ,Vintage It is an important tool for asset quality analysis , because Vintage The loss rate is static , Therefore, it is convenient to analyze the quality changes of loans of the same product at different times .
​​​​​​​         Different products often have different Vintage shape , Behind this is the difference in migration rates , Behind the migration rate may be the difference between the overdue situation or the recovery situation , Disassembly and analysis are required .
         actually , Of different products or institutions Vintage It is impossible to compare directly , The premise of comparison is to know two necessary conditions ： Number of product periods and product repayment method , And then Vintage The loss rate is converted to the annualized loss rate for comparison . Special attention is paid to ： Without knowing the number of installments and the method of repayment , Unable to accurately compare different products Vintage.

Reference article

Vintage The loss rate is converted to the annualized loss rate ( Nominal interest rate / Internal rate of return / Concepts such as annualized interest rate ),Vintage The loss rate is converted to the annualized loss rate ( Nominal interest rate / Internal rate of return / Concepts such as annualized interest rate )

In financial risk control Vintage And rolling rate

In financial risk control Vintage And rolling rate - You know

(1)、 Wine VS credit →Vintage The meaning of the curve

​​​​​​​         In the field of credit , We can also use Vintage Curve analysis assets （portfolio） The change law of quality maturity process . To make it easier to understand , Here is a list of Vintage The correspondence between the two fields in the analysis process .

industry	Wine	credit	remarks
batch	Year of cellar entry	Mortgage month	Batch label
The quality of	Alcohol concentration	Overdue risk	Ordinate
Time	Cellar age	Account age MOB	Abscissa
mature period	Quality Maturity	Maturity of the account	Vintage Curve features
Influencing factors	The temperature 、 sunshine 、 Precipitation, etc	Risk control strategy 、 Customer group 、 Market environment, etc	Cause analysis

​​​​​​​        Vintage A data table usually consists of three parts , The horizontal and vertical coordinates are usually in the account month （MOB） And billing month , The content can be various statistics , The overdue rate is 、 Overdue amounts are common （ chart 1 The statistics are M1+ The overdue rate of ）, But other proportions can also be used 、 ratio 、 Values and quantities , For example, the total amount 、 Average outstanding balance, etc .Vintage Analysis has three dimensions ：
1)、 Looking horizontally , Indicates the user life cycle effect , It reflects the changes that occur in the process of user maturity , It reflects the change of default rate of users with different account ages .
2)、 Look lengthwise , It shows the new account effect , It reflects the changes of users with the same aging over time , It reflects the change of default rate in different months .
3)、 Diagonals Look sideways , It shows the combination effect , It reflects the combination change of users with different account ages in the same period .

 ​​​​​​​         Follow the same analytical thinking , By aging （MOB） Compare after long and short alignment , We can Know the same product , stay At different times Lending Asset quality .
1)、 Analyze the law of change ： Asset quality （ For example, the overdue rate index ） The change of , If the overdue rate of previous periods rises rapidly , Then it means that the short-term risk is not captured , The risk of fraud is high ; conversely , If the curve keeps rising , It shows that the ability of credit risk identification is poor .
from MOB9 Start , The overdue rate tends to stabilize in most months , This explanation 9 A performance period of months , The risks of users are basically completely exposed , This has a guiding role in determining the sample performance period in our modeling process .
2)、 Determine asset quality ： Asset quality is generally defined by the overdue rate , That is, the corresponding overdue rate after the curve is flat .
3)、 Determine the maturity of the account ： The time factor used to judge the customer's performance , To help define the performance period .
4)、 Analyze the influencing factors ： Tighten or relax the risk control strategy 、 Changes in customer base 、 market environment 、 Policies and regulations will affect asset quality . Analyze the influencing factors , It can be used to guide the adjustment of risk control strategy .

(2)、 Credit sector to be determine Indicators of —— Account performance period — Credit products Vintage Curve mining

​​​​​​​         In the field of financial risk control , Especially in risk control modeling , It is usually necessary to observe the full cycle risk situation of users , Define the maturity of user risk ,Vintage The curve just does this .

• 1)、 For risk control modeling , It is often necessary to determine Target, For fraud , We can simply use the first timeout indicator to define Target. But for PD In terms of models , It is usually necessary to determine the performance period through analysis . If the definition of performance period is not appropriate , Then a good user may be defined as Bad, A bad user can also be defined as Good. We usually use Vintage Curve to Determine the overdue rate When tends to Stable , thus Choose the right performance period .
• 2)、Vintage The analysis can reflect the risk situation of customers with different aging periods , By observing the change of risk situation in different cycles , We can analyze customer risk changes in different cycles （ For example, changes in the macro situation will lead to fluctuations in customer risk ）、 Change of risk control strategy （ For example, the risk control model has been updated or the risk has been reduced due to strategy adjustment ）. in addition , Through customers who have completed the whole life cycle , Sure See the overall overdue situation of users , To use To adjust the risk control strategy .

​​​​​​​         Further reflection , If the asset quality is defined by the overdue rate , Then the overdue risk （ Target variable Y） How is it defined ？ How to determine the ？M3, still M6？ This indicator , It needs to be determined in combination with the rolling rate .
         First , Let's analyze why we need to determine the performance period of the account ？ as everyone knows , The longer the performance period , The more thorough the credit risk exposure will be , But it means that the observation period is farther away from the present , The historical data used to extract sample features will be older , The greater the difference between the modeling sample and the future sample ; conversely , The shorter the performance period , The risk has not been fully exposed , But the advantage is that you can use more recent samples .

​​​​​​​         for example , For one 12 Credit products with installment payments , In theory, when the user is 12 The end of the period , And pay off all the money , We can be defined as absolutely good customers ; conversely , We can only say that we have been a good customer so far , However, it is not known whether the users in the next few installments will not pay back the money overdue . therefore , We need to identify a Appropriate performance period Enough bad customers can be covered .

​​​​​​​         According to the credit product shown in the figure Vintage curve , What information can we get ？
1)、 The longest aging is 12 Months , The term of representative products is 12 period . With 12 The end of the period , The life cycle of the account has come to an end .
2)、 Account age MOB1、MOB2、MOB3 The overdue rate of is 0, Explain that the overdue indicator is M4+（ Overdue more than 90 God ） risk .
3)、 From the month of lending to 2018 year 1 month ～12 The final overdue rate of accounts is decreasing in all months , It shows that the quality of assets is constantly improving , It may be because the level of risk control is constantly improving .
4)、2018 year 5 The moon is opposite 2018 year 1～4 The monthly overdue rate has decreased significantly , It shows that the risk control strategy at this stage has been significantly improved .
5)、 Loans made in different months M4+ after 9 individual MOB Then it began to stabilize , Explain that the maturity period of the account is 9 Months .

(3)、Vintage curve Draw logic

​​​​​​​         draw Vintage Curve time , I have to mention the definition of overdue rate in the ordinate . There are usually two kinds of calculation caliber ：
T1、 Order caliber ： Overdue rate = Number of overdue orders / Total lending orders
T2、 Amount caliber ： Overdue rate = Overdue remaining principal / Total loan principal
         At present, the caliber definitions of Internet financial institutions are different , So just according to the Vintage curve , Sometimes it is not possible to objectively analyze the asset quality and risk control level .

Recommended articles ：
Vintage Detailed explanation of calculation process of analysis table ,Vintage Detailed explanation of calculation process of analysis table - You know

AI Intelligent risk control ( 3、 ... and ）—— Challenges in actual risk control products ​​​​​​​,https：//zhuanlan.zhihu.com/p/158588024