If the overlay framework defines the governance architecture for management adjustments, then post-model adjustments and emerging risk capture explain how that architecture is used in practice. This is where the Expected Credit Loss framework confronts one of its most difficult realities: even a well-designed model can be directionally sound and still incomplete at a particular moment in time. New risks often emerge before they become visible in historical default rates, stage migration patterns or calibrated model variables. Concentrations can worsen faster than segmentation frameworks adjust. Regulatory or legal changes can alter recoveries before LGD assumptions are refreshed. Portfolio behaviour can drift. Macroeconomic stress can enter through channels not yet reflected in the model. When this happens, the institution needs more than a general philosophy of prudence. It needs an operating discipline for noticing the gap, assessing its significance and translating it into a controlled post-model adjustment.
This is the role of post-model adjustments.
A post-model adjustment is not merely an extra reserve placed on top of the allowance because management feels uneasy. Nor is it simply another name for every overlay. It is more specific than that. It is the process by which identified model blind spots, emerging risks or transmission gaps are converted into a measured, reviewable and temporary addition or refinement to the ECL estimate. If overlays provide the governance shell, post-model adjustments provide the practical method of execution.
This distinction matters because many institutions have overlay policies but weak emerging-risk discipline. They know, in principle, that management adjustments should be used carefully, but they do not have a repeatable process for identifying what exactly is missing, how to evidence it, how to quantify it and how to prevent it from becoming a permanent, unexamined layer in the allowance. Others go to the opposite extreme: they identify emerging risk constantly, but apply adjustments in a fragmented or reactive way without a coherent link to model architecture, scenario design or portfolio analysis.
A mature ECL framework does something more demanding. It develops a structured process to capture risks that the model has not yet fully absorbed, while maintaining discipline, traceability, challenge and eventual model remediation.
This article explores that process in depth: what post-model adjustments are, how emerging risks should be identified, what types of model blind spots commonly require them, how to distinguish true residual risk from already captured risk, how quantification can be approached, how concentration and thematic risks should be handled, how adjustments should age and unwind, and what common failures most often undermine this sensitive but necessary part of the ECL framework.
It may seem that post-model adjustments are already covered by the overlay discussion. But they deserve a separate article because the challenge here is not only governance. It is also detection and translation.
An institution may have a strong overlay approval process and still fail to capture emerging risk well. Why? Because the real difficulty often lies upstream:
- How does the institution know the model is missing something?
- What evidence indicates that the missing risk is real rather than anecdotal?
- How can management be confident it is not already reflected through scenario weighting, segmentation or parameter movement?
- How should the missing element be quantified without inventing false precision?
- How should the institution avoid turning a temporary blind spot into a permanent manual reserve?
These are practical and methodological questions, not just governance questions. Post-model adjustments exist to answer them. They are the bridge between observed emerging conditions and a model that has not yet fully absorbed those conditions.
A post-model adjustment is a targeted change to the model-derived allowance intended to reflect a specific, identified risk or limitation that remains insufficiently captured after the core ECL model, staging logic and macroeconomic framework have been applied.
This definition has several important features.
It is targeted, meaning it should relate to a defined issue, not general discomfort.
It is post-model, meaning it is applied after understanding what the core engine already captures.
It is residual, meaning it addresses what remains missing, not the totality of the risk.
It is temporary or transitional, unless and until the relevant logic is embedded in the model itself.
This makes post-model adjustments conceptually narrower and more disciplined than a broad management reserve. A good post-model adjustment has a clear reason for existing and a clear path to being reviewed, changed or removed.
One of the strongest justifications for post-model adjustments is that credit deterioration often becomes visible in forms that precede formal model capture.
For example:
- Relationship managers may report consistent customer stress in a specific segment before defaults rise.
- Collections teams may observe unusual payment behaviour that has not yet affected historical roll rates materially.
- Industry intelligence may show demand collapse or regulatory disruption affecting a subset of borrowers.
- Treasury or funding teams may see refinancing pressure building in leveraged sectors before it becomes visible in stage migration.
- Recovery teams may report worsening disposal conditions or legal delays ahead of formal LGD recalibration.
In each case, the institution has information. But the information may still be too new, too fragmented or too non-standardised to flow immediately into the base model. This is precisely the environment in which post-model adjustments become valuable. They allow the institution to respond to credible early warning without pretending that the model already contains the full signal.
Not every concern justifies a post-model adjustment. This is one of the key disciplines in this pillar.
Markets are always changing. Individual business teams are often more cautious or more optimistic than the model. Anecdotes arise constantly. If the institution turns every concern into a reserve, the allowance quickly becomes unstable and ungovernable.
The real task is to distinguish emerging risk from ordinary noise.
A strong framework asks:
- Is the issue supported by more than isolated observations?
- Does it affect a defined portfolio, segment or concentration?
- Is the risk economically meaningful to expected loss?
- Is it likely to persist long enough to matter for the reporting date?
- Is the model actually missing it, or is it already reflected indirectly?
- Can the issue be linked to plausible effects on PD, LGD, EAD or staging?
These questions act as a filter. They help ensure that post-model adjustments are used to address real emerging credit concerns, not every fluctuation in sentiment.
Emerging risks often arise in a few recurring forms.
Sector-specific weakness
A defined industry segment shows stress not yet fully represented in portfolio-level model parameters.
Concentration deterioration
A particular customer group, geography, channel or collateral class becomes materially more vulnerable.
Regulatory or policy disruption
New legal frameworks, borrower relief schemes, tax measures or regulatory constraints alter credit behaviour or recovery conditions.
Model data lag
Relevant information exists but is not yet fully reflected in the source data or model refresh cycle.
Recovery environment shift
Collateral markets, disposal conditions or enforcement processes weaken in a way that the LGD framework has not yet absorbed.
Portfolio drift
The characteristics of new origination or recent book composition differ materially from the historical portfolio on which the model was trained.
Emerging contagion or correlated stress
The institution identifies linked exposures or interconnected counterparties whose vulnerability is not adequately represented by current segmentation.
These categories matter because they give the institution a vocabulary for emerging risk. Without such a vocabulary, discussions remain abstract and difficult to convert into structured adjustments.
A post-model adjustment is not always a response to a model that is clearly "wrong." Often the model is directionally reasonable but has a blind spot.
This is an important distinction. A blind spot may arise when:
- The model is calibrated on historical relationships that have not yet seen the new stress regime.
- The model is portfolio-wide, but the risk is concentrated in a small segment.
- The model uses macro variables that do not fully capture a sector-specific disruption.
- The model incorporates point-in-time adjustment, but only at a broader level than the risk now requires.
- The staging framework is accurate on average, but under-sensitive in one concentrated part of the book.
These are not model failures in the sense of incompetence. They are limitations in scope, granularity or timing. Post-model adjustments are often at their best when used to bridge these limitations rather than as a substitute for the entire model.
Concentration risk deserves special attention because it is often where emerging risk first becomes material while remaining partly invisible in broad model outputs.
A portfolio can look acceptable at overall level while containing vulnerability concentrated in:
- One industry
- One region
- One borrower ecosystem
- One counterparty chain
- One product subtype
- One collateral class
- One origination cohort
- One sponsor network
Model averages may not fully capture this. The losses of the broader pool can dilute the stress in the vulnerable segment. This is why many institutions use post-model adjustments to address concentration risks when the segmentation framework has not yet been refined enough to isolate them.
However, concentration adjustments must be done carefully. They should be based on identified exposure groups, not broad unease. They should also prompt a strategic question: is this concentration material and persistent enough that the model itself should be redesigned to capture it more directly?
Some emerging risks cut across multiple portfolios. These are sometimes called thematic risks.
Examples may include:
- Stress in commercial real estate
- A broad refinancing squeeze
- Energy-transition-related borrower vulnerability
- Supply-chain disruption affecting several industries
- Large-scale fraud patterns in a product channel
- Persistent recovery delays in a jurisdiction
- A sharp drop in used-asset resale markets affecting several secured products
These risks do not fit neatly into one model parameter. They may affect loans, receivables, guarantees and lease receivables simultaneously, but in different ways. A mature institution therefore sometimes needs cross-portfolio post-model adjustments or, at minimum, a coordinated thematic review to determine whether the issue is sufficiently captured across the framework.
The challenge here is balance. A thematic adjustment should be coherent, but it should not become so broad that it loses portfolio specificity.
Before applying any post-model adjustment, the institution should perform one of the most important tests in this entire area:
How much of this risk is already captured in the model?
This question is essential because it protects against double counting and forces a more disciplined articulation of the residual gap.
The institution should examine:
- Whether scenario design already includes the relevant macro or sector effect
- Whether stage migration has already responded in the affected segment
- Whether recent PD or LGD recalibration has partly absorbed the change
- Whether concentration is already reflected in segmentation
- Whether recent overlays or other adjustments already address the issue
Only after that review can the institution determine what remains missing. The best post-model adjustments are not based on total risk. They are based on residual uncaptured risk.
Quantification is often the hardest part. Emerging risk is, by definition, not yet fully visible in historical loss experience. This makes exact measurement difficult. But lack of precision does not justify arbitrariness.
A disciplined quantification process may draw on:
- Exposure-weighted sensitivity analysis
- Alternative scenario outcomes
- Leading risk indicators
- Observed deterioration in comparable segments
- Judgment ranges linked to specific balance subsets
- Temporary stress multipliers on defined model components
- Targeted stage uplift or severity uplift analysis
- Historical analogues from similar episodes
- Expert credit review translated into bounded quantitative effect
The goal is not to pretend certainty. It is to convert evidence and informed judgment into a range narrow enough to support a defendable adjustment and clear enough to be reviewed later against realised outcomes.
Even when a post-model adjustment is booked as an aggregate amount, it should ideally be linked conceptually to the underlying mechanism of risk.
For example:
- An emerging distress adjustment may represent higher PD in a particular sector.
- A property-market adjustment may represent higher LGD in a defined secured portfolio.
- A liquidity-stress adjustment may represent higher EAD due to expected drawdowns in revolving facilities.
- A staging sensitivity adjustment may represent under-recognised SICR in a specific borrower population.
This linkage matters because it improves conceptual clarity, supports governance and helps determine whether model redevelopment is required. It also makes later validation easier, because the institution knows what kind of risk it thought it was capturing.
A recurring danger is that repeated post-model adjustments in the same area begin to function as an alternative modelling layer.
For example, an institution may repeatedly apply a sector uplift every quarter for the same vulnerability. Or it may maintain a recurring recovery haircut on a collateral class because the base LGD model never catches up. Or it may use ongoing concentration reserves without redesigning segmentation.
At that point, the institution is no longer using post-model adjustment as a temporary bridge. It is effectively running a shadow model outside the controlled modelling framework.
A mature ECL process recognises this pattern and responds by escalating the issue into model redevelopment or methodological change. Temporary adjustments should remain temporary unless there is a strong, documented reason they cannot yet be embedded.
The more precisely the institution can identify the affected population, the more credible the post-model adjustment becomes.
A broad top-line reserve across all portfolios is harder to justify than a targeted adjustment to:
- a specific sector within SME lending,
- a defined office-property secured sub-book,
- a group of export receivables tied to one market,
- a set of guarantees linked to a stressed counterparty class,
- or a cohort of restructured borrowers not yet performing long enough for normal model capture.
Granularity creates accountability. It allows the institution to explain why the adjustment exists, what it affects and how it should unwind. Broad adjustments may sometimes be unavoidable in early crises or data-poor settings, but they should be treated as exceptions, not the norm.
Post-model adjustments are strongest when supported by management information that is leading, segment-specific and independently observable.
Useful indicators may include:
- watchlist growth,
- covenant waiver frequency,
- requests for relief or restructuring,
- utilisation spikes,
- sector arrears-leading signals,
- early delinquency drift,
- cancellation rates,
- dispute build-up,
- valuation slippage,
- legal recovery delays,
- counterparty downgrades,
- or concentration migration.
These indicators matter because they often reveal stress before realised loss metrics do. A mature institution treats them as inputs into emerging risk capture, not merely as side reports disconnected from the impairment process.
Because post-model adjustments are often judgment-heavy, governance must do more than rubber-stamp them. It should actively challenge them.
Useful challenge questions include:
- Why is this not already in the model?
- What evidence shows the issue is material?
- Why is the proposed amount proportionate?
- Which portfolios are truly affected?
- What alternative quantifications were considered?
- What could cause this adjustment to be released next period?
- Does the same logic appear elsewhere in the allowance already?
- Is this an emerging issue or a structural weakness?
Challenge is especially important because post-model adjustments often arise in areas where management concern is high and model visibility is low. Without challenge, they can become comfort reserves rather than disciplined estimates.
Once an adjustment is introduced, it should be tracked actively. The institution should ask each period:
- Has the risk intensified, stabilised or faded?
- Has the base model now begun to capture some of the issue?
- Has the affected exposure group changed in size or quality?
- Have actual defaults, stage movements or recoveries validated the concern?
- Should the adjustment be increased, reduced, segmented differently or released?
This ongoing review is crucial because emerging-risk adjustments are by nature transitional. They sit between observation and formal model incorporation. If they are not monitored dynamically, they quickly become stale or overbearing.
One of the most useful side effects of a strong post-model adjustment framework is that it can reveal where the model needs to evolve.
Repeated adjustments in the same area may indicate:
- segmentation gaps,
- weak macro transmission,
- under-sensitive staging logic,
- poor concentration capture,
- stale recovery assumptions,
- or insufficient data architecture.
A mature institution therefore uses its adjustment inventory as a signal for model remediation priorities. This turns judgment into learning. The adjustment is no longer just a reserve change; it becomes evidence that the modelling framework should deepen in a particular place.
Several failures recur repeatedly.
One is confusing vague concern with identified residual risk.
Another is failing to test whether the issue is already reflected in the model, leading to double counting.
A third is applying large broad-based reserves without sufficient scope definition.
A fourth is using recurring adjustments without escalating them into model redevelopment.
A fifth is quantifying adjustments as round-number cushions without traceable reasoning.
A sixth is keeping adjustments alive through inertia because release criteria were never defined.
A seventh is allowing business or reporting pressure to shape the adjustment more than risk evidence does.
These failures are particularly damaging because post-model adjustments sit close to the boundary between prudent judgment and uncontrolled discretion. Weakness here quickly affects credibility.
Consider a lender with a diversified SME book. The core ECL model is functioning well overall, but one segment of export-linked textile borrowers begins to weaken sharply due to abrupt trade restrictions and order cancellations. Defaults have not yet risen significantly, and the PD model still reflects the prior period's broader sector averages. Relationship managers report distress, line usage is rising and several borrowers have requested informal relief. The concentration is material but not yet fully segmented in the model.
A mature institution would identify this as a classic candidate for a post-model adjustment. The risk is specific, evidence-based, likely under-captured and concentrated in a defined group. A targeted reserve can then be applied to that segment using exposure-level review, sensitivity analysis and documented rationale, while the institution evaluates whether the sector should be separately modelled going forward.
This is exactly how post-model adjustments are supposed to work: not as general fear, but as disciplined early recognition of a defined blind spot.
A strong institutional framework for post-model adjustments and emerging risk capture usually includes:
- a defined process for identifying emerging risks,
- clear criteria for distinguishing noise from material residual risk,
- formal assessment of what the base model already captures,
- targeted scope definition,
- reasoned quantitative translation,
- conceptual linkage to PD, LGD, EAD or stage effects,
- governance with real challenge,
- tracking and ageing of adjustments,
- release criteria,
- and escalation into model redevelopment where issues persist.
The strength of this framework lies in its honesty. It acknowledges that models are incomplete in real time, but refuses to let that incompleteness become an excuse for imprecise management reserves.
Post-model adjustments and emerging risk capture are among the most important disciplines in a mature Expected Credit Loss framework because they address the time lag between reality and model absorption. Credit risk often changes before historical evidence is strong enough to recalibrate the model. Concentrations can deteriorate before segmentation catches up. Recovery conditions can weaken before LGD assumptions are refreshed. In those moments, the institution needs more than model output and more than vague caution. It needs a structured way to recognise what is missing, assess how material the gap is and translate that gap into a controlled adjustment.
A strong institution does this with precision. It identifies residual risk rather than total uncertainty. It scopes adjustments narrowly where possible. It links them conceptually to the components of loss. It monitors them, challenges them and removes them when their purpose has been served. And, crucially, it learns from them, using recurring adjustments to improve the base model rather than letting manual reserves quietly take over the impairment framework.
In that sense, this pillar teaches an essential lesson about ECL maturity: the best judgment is not the judgment that replaces the model. It is the judgment that notices what the model has not yet seen, and then helps the framework evolve.
