Executive summary
Forward-looking information is not made more defensible by adding scenarios, variables or overlays without a clear transmission mechanism. A sound process explains how economic change affects a portfolio segment, what is already captured by the model, where judgement enters and how the resulting provision movement is reviewed.
The practical objective is traceability. Credit, Finance, model-risk and governance stakeholders should be able to move from economic evidence to model input, scenario result, management adjustment and final provision without losing the logic between them.
The business problem
IFRS 9 requires expected credit losses to reflect reasonable and supportable forward-looking information. In practice, organisations must make decisions under uncertainty while avoiding two opposite failures: a mechanistic process that ignores emerging risk, and an unconstrained process in which judgement produces material movements without sufficient evidence.
The issue becomes particularly visible when macroeconomic conditions move faster than model history, portfolio composition changes, or a known limitation is not captured in the core ECL model. A management adjustment may be appropriate, but it must not duplicate risk that already sits in staging, model calibration or scenario weighting.
Start with the transmission logic
Before selecting an economic variable, document the expected causal path. For example, a rise in unemployment may affect unsecured consumer lending through household income pressure, but the timing and magnitude can differ by employment segment, geography and existing arrears position.
A useful assessment asks:
- Which portfolio segment is exposed?
- Through which borrower or business mechanism should the factor affect default or recovery?
- Over what time horizon should the effect emerge?
- Is the relationship stable enough to model, or should it remain a governed judgement?
- Where is the risk already represented in the ECL framework?
This prevents statistical correlation from being treated as sufficient business evidence.
Scenario design should be proportionate
More scenarios do not automatically create a better estimate. The scenario set should represent materially different and plausible paths, with weights that can be explained and reviewed. The organisation should understand whether provision movement comes from the scenario values, their weights, non-linear model response or a separate overlay.
Sensitivity analysis is often more informative than a single final number. It shows which assumptions materially affect the result and helps governance forums focus on the uncertainty that matters.
Treat overlays as controlled model components
An overlay should have a defined purpose, owner, calculation method, review date and exit condition. Without these elements, temporary judgement can become a permanent, opaque component of the provision.
For each overlay, record:
- the model limitation or emerging risk being addressed;
- the affected exposures and evidence supporting the scope;
- the method used to quantify the adjustment;
- checks for overlap with modelled risk;
- sensitivity and materiality;
- approval, monitoring and release criteria.
Subsequent outcomes should be compared with the original rationale. This does not mean every overlay must predict the future perfectly. It means the organisation should learn whether the risk emerged as expected and whether the method remains proportionate.
A practical governance view
Governance reporting should reconcile the current provision to the previous period and separate the main drivers: portfolio volume and mix, staging movement, model changes, economic scenarios, parameter updates, write-offs and post-model adjustments. Narrative should explain the decision consequence, not only the calculation.
Questions for a review forum include:
- Can we trace the movement from source data to reported provision?
- Which assumptions create the greatest sensitivity?
- Are adjustments correcting a known limitation or duplicating risk?
- What evidence would cause us to increase, reduce or release an overlay?
- Are model and judgement uncertainty communicated clearly to decision-makers?
Implementation recommendation
Build a controlled evidence pack around each reporting cycle. It should combine data-quality checks, scenario inputs, model outputs, overlay registers, provision reconciliations, sensitivities and documented approvals. Automate repeatable calculations where practical, but preserve accountable review of assumptions and exceptions.
The strongest process is not the one with the most complex economics. It is the one in which material choices are visible, evidence is proportionate and the path from uncertainty to management decision can be explained.