SME Credit Rating Model
Rates SME borrowers on the Bank's 1 to 10 master scale.
AUC (out-of-time)
0.804
Gini
0.608
KS
0.461
Rating accuracy within one notch
0.87
1 more metric
Binomial test (grades passing)
10 / 10
About this model
Internal rating model for SME borrowers: combines financial statement ratios, account conduct, bureau data and a qualitative management assessment into a 12-month probability of default and a rating grade on the Bank's 1 to 10 master scale. Explains each rating through factor contributions so credit officers and committees can see what drives it. Tier 1 model; v3.0 is in independent validation by Model Risk Management, replacing the v2.4 champion.
Intended use
Credit origination, annual reviews, pricing and capital calculation for Business Banking SME exposures. Ratings can be overridden by credit officers within the override policy; overrides are logged and reviewed.
Training data lineage
Rebuilt from the SAS Enterprise Miner rating model used since 2017. Developed on sme-portfolio-mart and loan-applications-history (2016 to 2025 obligor history with default flags) in BigQuery, with through-the-cycle calibration using GDP and OPR series from macro-indicators. Logistic model with monotonic binning, trained in Vertex AI Pipelines. Independent validation by Model Risk Management covers discrimination, calibration, stability, fairness and explainability; annual revalidation thereafter.
Limitations & bias notes
Newly incorporated businesses with under two years of financials rely heavily on the qualitative module. Low-default segments (large SMEs) have few observed defaults, so calibration there leans on expert judgement. Sector shocks move ratings with a lag of one financial year.
Ownership and sensitivity
Who approves access
- Owner, SME Lending, Lim Wei Jie
- Data Governance Office, Ahmad Faizal
- Model Risk Management, Dr. Kavitha Subramaniam
Customer-level data under banking secrecy and PDPA, and models that decide on customers.
Named-user entitlement with the data owner's consent and a stated purpose; reviewed on expiry.
- Tier 1 model
- Material model: independent validation before use and annual revalidation.
- Credit decision
- Outputs feed a decision on a customer's credit; fairness and explainability tests apply.
- Banking secrecy
- Customer account information protected under the Financial Services Act 2013.
- Customer PII
- Contains or processes personal data about customers (PDPA 2010).
Try it
Live sandboxobligor=B-TKN-1180&years=12
fixtureEvaluation only: restricted models run against synthetic or masked sample data until your business unit's access request is approved.
Feedback
Details
- Updated
- 2026-07-29
- Latest version
- v3.0
- Licence
- BU Restricted
- Access
- Restricted
- Framework
- scikit-learn
- Language
- N/A