By implementing a scoring system for customers, we can judge whether the customers are of high quality or not. A card (Application Score Card), application score card
B card (Behavior Score Card), behavior score card
Since the retail credit business has a large number of transactions and a small single amount, The rich characteristics of data determine the need for intelligent and probabilistic management models. The credit scoring model uses modern mathematical statistical model technology, and through in-depth data mining, analysis and refinement of borrowers' credit history and business activity records, it discovers knowledge contained in complex data that reflects consumer risk characteristics and expected credit performance. and rules, and summarize them through scoring as a scientific basis for management decision-making.
Differences:
1. The time of use is different, focusing on before the loan, during the loan, and after the loan.
2. The data requirements are different, A card generally does The data of 0 to 1 year before the loan, B card is carried out after the customer has certain behavior and has larger data, usually 3 to 5 years,
3. All models of each score card Differently, A card mostly uses logistic regression, while the latter two commonly use multi-element logistic regression, which has better accuracy.
B Card
1. Definition: Predict the future overdue probability based on the lender’s behavior after lending (observation behavior)
2. Usage scenario: Loan issuance The time period before maturity
3. Pay attention to the observation period, performance period, and time slicing issues
Division
1. Based on repayment willingness and repayment Depending on the ability, different risk levels are divided
Mild: Good repayment willingness and repayment ability, overdue for special reasons
Medium-light: Good repayment willingness and repayment ability appear Problem
Moderate: Willingness to repay has deteriorated, but ability to repay is available
Severe: No willingness to repay, ability to repay has deteriorated or been lost
2. Collection Process
SMS collection, phone collection, on-site collection, legal proceedings, third-party collection (overdue assets are packaged and sold)
3. Model composition
Repayment Rate model: predicts the rate of debt collection after collection
Aging rolling model: predicts the probability of the number of overdue people converting from mildly overdue to severely overdue
Loss of contact model: In the overdue stage, predict the probability of loss of contact for those who can still be contacted
4. Common indicators
Number of overdue days
Historical repayment rate information
Proportion of overdue amount
Proportion of debt burden
Personal information (gender, age, income, job, education, etc.)