IFRS 9 has been in force long enough that the standard itself is no longer the news. The news is what is happening inside the implementations — particularly in smaller markets where the volume of historical default data is genuinely thin.
For Maldivian banks and finance companies, two questions keep coming up in technical review: how do you estimate Probability of Default when your portfolio does not generate enough events, and how do you defend Loss Given Default assumptions to an auditor who has seen the international playbook?
PD when the data is thin
The temptation is to anchor on through-the-cycle PDs from a regional peer or a published benchmark. This is defensible as a starting point, but it is not sufficient. The standard requires entity-specific calibration — and "we used a published table" is not calibration.
What works in practice
Three approaches we see succeed in Maldivian engagements: scorecard-based segmentation with judgmental overlays, vintage analysis on portfolios old enough to support it, and Markov transition matrices for portfolios with sufficient stage-movement data.
Auditor scepticism about ECL models is not unreasonable. It is what the standard requires.
LGD in a small market
LGD is where the work gets interesting. Recovery patterns in the Maldives reflect local realities — collateral types, enforcement timelines, recovery costs — that do not map cleanly to international defaults.
What good looks like
- Segmentation that reflects how the portfolio actually behaves
- PD, LGD and EAD models that are independently testable
- Forward-looking information that affects the numbers, not just the documentation
- Stage transfer criteria that are operational, not theoretical
- Governance — board-level oversight, model risk management, validation