Models / Credit Risk

Collections Roll-Rate Forecaster

Forecasts delinquency roll rates under base and stress scenarios.

ConfidentialBNM reportableAggregated onlyLimited riskForecasting · 1.2.0 · collections-roll-rate-forecasterOwner Credit Risk
44 1.6K/30d 27

MAPE (roll rate, 1 month)

5.2%

MAPE (roll rate, 6 months)

11.9%

90+ DPD balance error (6 months)

4.1%

Parity vs SAS champion (MAPE delta)

-0.7 pp

About this model

Forecasts monthly roll rates between delinquency buckets (current, 1 to 29, 30 to 59, 60 to 89 and 90+ days past due) for cards, personal loans, mortgages and hire purchase, six months ahead, under base and stressed macro scenarios. Used to plan collections capacity and as an input to the expected credit loss overlay discussion.

Intended use

Collections capacity planning, ECL management overlay discussions and portfolio reporting to Credit Risk committee. Works on portfolio aggregates, not individual accounts.

Training data lineage

Rebuilt from the SAS roll-rate model maintained by Credit Risk since 2019. Trained on monthly bucket transitions aggregated from collections-accounts, with unemployment, OPR and household-debt series from macro-indicators as scenario drivers. Built in BigQuery and Vertex AI Pipelines; Model Risk Management signed off parity with SAS.

Limitations & bias notes

Repayment assistance programmes and moratoria distort history (2020 to 2021 is down-weighted). Stressed scenarios assume historical relationships between macro drivers and roll rates hold. Small portfolios (below 5,000 accounts) show wide bands.

#collections#roll-rates#credit-risk#ecl#stress-testing#rebuilt-from-sas

Ownership and sensitivity

Owned by

Group Risk ManagementCredit Risk

Accountable owner: Rajesh Kumar

Who approves access

  1. Owner, Credit Risk, Rajesh Kumar
Confidential

Business-sensitive. Models and data products scoped to named business units.

Entitlement per business unit, approved by the owner; conditions attach.

BNM reportable
Feeds a regulatory return or a figure reported to Bank Negara Malaysia.
Aggregated only
No row-level customer data; aggregates with small-cell suppression.

Try it

Live sandbox

portfolio=personal-loan&horizon=12

fixture

Evaluation only: restricted models run against synthetic or masked sample data until your business unit's access request is approved.

Feedback

Details

Updated
2026-07-09
Latest version
1.2.0
Licence
BU Restricted
Access
Restricted
Framework
statsforecast
Language
N/A