Risk Management for Loan Portfolio

Context

SHBFinance operates in unsecured lending, which carries inherently high credit risk.

Rising late payments and delinquencies increase portfolio exposure and required provisions.

As part of the risk management data team, I prepared risk reports and presented findings to leadership.

The request was to monitor repayment behavior, identify high-risk loans, and ensure compliance through data-driven analysis.

Request

Monitor and manage customer payment behavior to prevent defaults.

Use predictive modeling to identify high-risk loans and classify them by risk level.

Analyze data to identify risk patterns and implement risk mitigation strategies.

1. Monitor and manage customer payment behavior:

Evaluation Criteria 1: Delinquency Matrix

Delinquency Bucket:

Delinquency matrix reflects that SHBFinance will have difficulty in collecting debts if customers do not pay for more than 30 days

• 94% of loans in bucket B0 are maintained on time
• Once they have moved to B1 (1<=DPD<=9) -> tend to be more late (B2, B3) and mostly late 1 month
• From B3, when customers have owed >30 days, they tend to move to the next higher bucket and the “recovery” rate (return to B0) is close to 0

→ SHBFinance can hardly recover the capital

=> Encourage customers to pay their debts when there are signs of late

Flow rate matrix shows that the stay and back rate from B3 onwards is below 10%

→ Their will to pay is decreasing.

Customer psychology: if they can default for 1 month without any problem, they will default for the remaining months.

Besides, there is still a small number of customers who are late on their first month on book (MOB=1)

• %FPD5: 3.2% Customers who are 5 days late in their first MOB
• %FPD10: 2.9% Customers who are 10 days late in their first MOB

It is worth mentioning that in this group of overdue customers, up to 97% of loans will be rolled to group B7→B14, meaning that 97% of loans that are 10 days late will default

=> Pay attention to your customers' first month on book to avoid fraud and control late payments

94% loans in bucket B0 maintain timely payments

97% Loan paid in full for the first 3 terms of the loan (MOB = 1,2,3)

-> We can consider the group of customers who are not late in the first 3 payments as “good customers”

But only 80% 4th full payment loan, down more than 15% in quantity

2. Build predictive modeling:

Build a model to predict the credit roll rate

Goal: Capture as many loans that are likely to default, even if the customer has paid very well in the first three months.

Features:

• PII Info: Age, Num. Dependants, Working Years, Education, Marital Status, Labour Contr. Type
• Loan Info: Loan Purpose, Loan Term, Loan Amount, Product Category, Month Interest
• Financial: DTI Ratio, Income Amount, Income Resource, Credibility
• History: Max DPD of previous contracts

Machine Learning Model: Random Forest

Critical for targeting risky customers while preserving profit

High Precision for Default: Minimizes false positives for defaulters

After using Random Forest, apply for the data to classify users into 9 groups: AAA, AA, A, BBB,..., which shows the risk score of customers.

3. Analyze data to identify risk patterns

With score classification, we can analyze some characteristics of the data and find some points:

• The roll rate exceeds > 90% for score bands B to C, making it clear that we should reject customers with scores from B and below.
• The company faces a high risk of losing both principal and profit with C-group customers.

4. Propose strategy and risk management:

Retain high-quality customers

• Target: Maintain Good Customer Group (Low Bucket - B0, B1, B2)

• SMS/Voicebot reminders 5/3/1 days before the due date, 3 days after if payment is not made.

• Offer reduced interest rates when old customers sign up for new loan products

• Provide cashback or loyalty points for consistent on-time payments throughout a cycle

Monitor and early intervene with at-risk customers:

• Target: Early intervention for Medium-risk Group (B3, B4, B5)

• Dedicated customer monitoring team

• Flexible products

Manage risk of lower score band groups:

• Target: Risk management for score band group B and C

• Tightening loan approval criteria

• Flexible Loan Packages

• Post-Covid financial status monitoring

Enhance value for premium segment customers

• Target: Customer engagement for Score band A customer group

• Enhanced financial products offerings

• Channel Optimization

Expand the offering of digital financial products on the platforms of other partners: Leverage Online Channels (Momo, Zalopay, ShopeePay,...) and Partner with Local Businesses (Grab, Thegioididong,...)

Slide proposal: P1

Slide proposal: P2