Expected Credit Loss (ECL) measurement under Ind AS 109 - Financial Instruments is inherently forward-looking.
Unlike an approach that waits for an observable loss event before recognising impairment, the ECL framework requires entities to estimate credit losses using information about past events, current conditions and forecasts of future economic conditions.
That makes ECL measurement heavily dependent on data, methodology, assumptions and management judgement.
For entities with material lending, receivable or other credit exposures, these estimates may be generated through models ranging from relatively simple provision matrices to more sophisticated frameworks using Probability of Default (PD), Loss Given Default (LGD) and Exposure at Default (EAD).
But a model producing a number does not automatically mean that the number is reliable.
The model may contain weak assumptions. Historical data may be incomplete. Segmentation may no longer represent the portfolio. Forward-looking adjustments may be poorly supported. Staging rules may not identify increases in credit risk appropriately.
This is where ECL model validation becomes important.
Independent validation provides structured challenge to the methodology, data, assumptions, implementation and performance of an ECL framework before management relies on its output for financial reporting.
There is, however, an important distinction:
Ind AS 109 establishes the principles an ECL estimate must satisfy. It does not, by itself, prescribe a universal requirement that every ECL model must be independently validated by a separate external validator.
Independent validation is therefore best understood as a model-risk and governance mechanism that helps management demonstrate that its ECL methodology is reasonable, supportable and appropriately controlled. Additional regulatory requirements may apply to particular regulated entities.
Ind AS 109 sets clear principles for measuring expected credit losses.
Paragraph 5.5.17 requires ECL measurement to reflect:
These principles make ECL fundamentally different from a mechanical provisioning percentage.
The entity must develop an approach capable of translating credit risk information into a reasonable estimate of expected cash shortfalls.
The Standard does not prescribe one mandatory model architecture for every entity. Its application guidance recognises that different approaches may be appropriate for different financial instruments. An approach does not even have to use an explicit probability of default input, provided that it meets the relevant requirements.
This is an important point for model validation.
The purpose of validation is not to ask:
"Does this model look like everyone else's model?"
The more appropriate question is:
"Does this methodology appropriately satisfy the requirements of Ind AS 109 for this portfolio and these circumstances?"
ECL model validation is a structured assessment of whether the framework used to calculate expected credit losses is:
A proper validation exercise goes beyond checking formulas.
It should consider the entire path from source data to the final impairment allowance.
For a sophisticated lending portfolio, this might include:
Data → segmentation → default definition → credit-risk assessment → PD/LGD/EAD or other loss methodology → forward-looking adjustments → scenario weighting → staging → ECL calculation → management adjustments → financial reporting.
A weakness anywhere along that chain can affect the final provision.
Model developers naturally understand their models better than anyone else.
But that creates a potential weakness.
The same team that designed a methodology may also have made the assumptions being challenged.
This does not imply that the development team has done anything wrong.
It means that independent challenge can identify issues that are difficult to see when reviewing one's own methodology.
A validator can question matters such as:
Independent validation therefore acts as a control over model risk—the risk that incorrect design, assumptions, data or implementation lead to unreliable results.
This distinction is also important.
The purpose of model validation is generally to examine the design and performance of the ECL framework in depth.
An external financial statement audit has a different objective and operates under auditing standards.
A model validation report therefore does not replace the work of the statutory auditor.
Likewise, external audit does not remove management's responsibility for having an appropriate ECL methodology.
The responsibility for the financial statements and the ECL estimate remains with management and, where applicable, those charged with governance.
Validation provides additional evidence and challenge to support that responsibility.
A meaningful validation should consider more than the mathematical model.
The following areas are particularly important.
The first question is whether the entity has correctly identified the financial instruments to which the impairment requirements apply.
Depending on the circumstances, the ECL requirements of Ind AS 109 can apply to financial assets measured at amortised cost, relevant debt instruments measured at FVOCI, loan commitments, financial guarantee contracts and certain receivables.
Different approaches may also apply to particular instruments.
For example, Ind AS 109 contains a simplified approach for specified trade receivables, contract assets and lease receivables under which lifetime ECL may be recognised without applying the general significant-increase-in-credit-risk assessment in the same manner.
A validation should therefore first establish whether the methodology matches the nature of the exposures.
An ECL model is only as reliable as the data feeding it.
A technically sophisticated model built on poor-quality data can produce misleading results.
Validation should therefore assess whether relevant data is:
Examples of data issues include:
Where historical data is limited, the validator should assess how that limitation has been addressed and whether any additional assumptions are reasonable.
Validation should not simply state that "data is limited."
It should consider what that limitation means for the reliability of the ECL estimate.
Credit exposures are often grouped for modelling purposes.
But grouping should make economic and credit-risk sense.
Combining exposures with substantially different risk characteristics can weaken the model.
For example, depending on the business, relevant differences might exist between:
The appropriate segmentation depends on the portfolio.
There is no universal segmentation structure required by Ind AS 109.
The validator should therefore assess whether the groups used are sufficiently meaningful for measuring credit risk and whether changes in the portfolio require the segmentation to be revisited.
The definition of default is fundamental to many ECL methodologies.
It affects:
An inconsistent default definition can distort the entire framework.
The validator should understand:
Ind AS 109 does not prescribe one single numerical definition of default for every entity. The framework therefore requires judgement, making documentation and consistent application particularly important.
For RBI-regulated NBFCs implementing Ind AS, RBI guidance places emphasis on Board-approved documentation of matters including the ECL model and definition of default.
For financial instruments subject to the general ECL approach, one of the most important judgements is whether credit risk has increased significantly since initial recognition.
This determines whether the loss allowance is generally measured using:
12-month expected credit losses, or
lifetime expected credit losses.
Ind AS 109 requires an entity to compare the risk of default at the reporting date with the risk at initial recognition and to consider reasonable and supportable information.
A validation should therefore examine the framework for identifying Significant Increase in Credit Risk (SICR).
This may include reviewing:
The validator should also examine whether exposures are moving between risk categories at appropriate points.
A framework that recognises deterioration too late can understate lifetime ECL.
A framework that moves exposures too aggressively may produce an unnecessarily volatile allowance.
The objective is not to minimise or maximise the provision. It is to apply the Standard appropriately.
Many sophisticated ECL frameworks calculate losses using components commonly described as:
An estimate of the likelihood of default over a relevant time horizon.
An estimate of the loss if default occurs, considering relevant expected recoveries.
An estimate of the exposure when default occurs.
These components can provide a useful structure for ECL modelling.
However, Ind AS 109 does not require every entity to use a PD × LGD × EAD model.
Other methodologies can comply with the Standard if they appropriately measure expected credit losses.
This distinction matters particularly for:
A validator should therefore assess whether the chosen methodology is appropriate rather than treating PD/LGD/EAD as a mandatory template.
Where PD, LGD and EAD models are used, each component should be evaluated separately as well as collectively.
One of the defining characteristics of the ECL framework is its use of forward-looking information.
Ind AS 109 does not permit ECL to be based solely on historical experience when relevant reasonable and supportable forward-looking information is available.
Historical data may therefore need to be adjusted to reflect:
The Standard expressly requires ECL measurement to incorporate reasonable and supportable forecasts of future economic conditions where available without undue cost or effort.
Validation should therefore examine:
Selecting an economic variable simply because data is available is not enough.
There should be a reasonable relationship between the variable and the credit behaviour being modelled.
ECL should not simply reflect one "most likely" future.
Ind AS 109 requires an unbiased, probability-weighted amount determined by evaluating a range of possible outcomes. It does not require entities to model every theoretically possible scenario, but the possibility of both a credit loss and no credit loss needs to be reflected appropriately.
Depending on the portfolio and methodology, an entity may use scenarios such as:
But the number of scenarios itself is not the key test.
Validation should examine whether:
A three-scenario model is not automatically better than a different approach.
The question is whether the methodology produces an unbiased probability-weighted estimate.
Expected credit losses are not simply undiscounted future loss amounts.
Ind AS 109 requires the estimate to reflect the time value of money.
Validation should therefore examine how expected cash shortfalls are discounted and whether the treatment is consistent with the applicable requirements.
This becomes particularly relevant for:
Ignoring the timing of recoveries can distort loss estimates even where the eventual recovery amount itself is reasonable.
For secured lending, expected recoveries can materially influence credit losses.
The validator should therefore examine whether assumptions relating to:
are supported by evidence.
A common risk is relying on current collateral value without adequately considering:
An asset worth ₹100 today does not necessarily produce a ₹100 recovery after default.
The timing and cost of recovering that asset matter.
A model may be conceptually reasonable but poorly calibrated.
Calibration determines whether model parameters appropriately reflect the risk characteristics of the portfolio.
A validator may examine whether:
A model should not continue indefinitely with parameters developed for a portfolio that no longer resembles the business.
Backtesting compares previous model expectations with subsequent actual outcomes.
For example:
Backtesting does not mean a model is wrong whenever actual results differ from estimates.
ECL is an estimate of uncertain future outcomes. Actual experience will naturally differ.
The purpose is to identify systematic or persistent differences that may indicate:
This is one of the clearest ways to assess whether a model continues to perform as intended.
A validator should understand how the ECL result responds to important assumptions.
For example:
What happens if:
Sensitivity analysis helps identify which assumptions have the greatest impact on the reported allowance.
This is particularly useful where ECL is highly dependent on judgement.
It also helps management understand the uncertainty around the model output rather than treating the final provision as an exact prediction.
Models cannot always capture every emerging risk immediately.
An organisation may therefore consider a management adjustment or overlay when there is a material risk not adequately reflected in the underlying model.
Examples could include:
But overlays require careful governance.
A validation should examine:
Without such discipline, overlays can become unsupported additions to the provision.
They can also create double counting if the same risk is already reflected through macroeconomic scenarios, model parameters or staging.
Even a sound methodology can fail if it is implemented incorrectly.
Validation should therefore consider the actual calculation process.
Possible issues include:
For material models, reproducing selected calculations or independently testing model outputs can be valuable.
A methodology document and the production model should tell the same story.
A strong ECL framework should be understandable to someone other than its original developer.
Documentation should ordinarily explain matters such as:
This is not merely good administrative practice.
Ind AS 107 – Financial Instruments: Disclosures requires entities to explain the inputs, assumptions and estimation techniques used to apply the ECL requirements, including how forward-looking information is incorporated and significant changes in estimation techniques or assumptions.
Weak documentation can therefore create difficulties not only for model governance but also for financial reporting and audit support.
Independence does not necessarily mean that validation must always be outsourced.
The appropriate structure depends on:
The underlying principle is that the validator should be able to challenge the model objectively.
Ideally, the person or team validating the model should not have primary responsibility for developing the methodology being assessed.
For some organisations, this may involve an independent internal model-validation or risk team.
For others, particularly where specialist expertise or organisational independence is difficult to achieve internally, an external validation may be appropriate.
The exact model-governance requirements should also be considered alongside any regulations applicable to the specific entity.
Validation should not become a checklist exercise.
A report that merely states:
provides limited value if it does not explain what was tested and what was found.
A meaningful validation should identify:
Findings should also be prioritised according to their potential impact.
Not every issue is equally material.
This point needs careful treatment.
Ind AS 109 itself does not prescribe a blanket rule stating that every entity must obtain an independent ECL model validation annually.
Its requirements focus on how credit risk is assessed and how expected credit losses are measured.
However, validation is a strong governance mechanism for determining whether a model continues to satisfy those requirements.
The appropriate validation frequency may depend on factors such as:
A material model may warrant periodic validation and additional review after significant changes rather than being treated as a model that is approved once and never reconsidered.
Entities should separately assess any specific RBI or other regulatory requirements that apply to them.
Even where an organisation follows a regular validation cycle, certain events can justify an earlier review.
These include:
A lender enters a new borrower segment or launches a materially different credit product.
PD, LGD, EAD or other methodologies are materially changed.
Actual credit outcomes repeatedly differ from model predictions.
Existing relationships between macroeconomic variables and credit losses may no longer behave as expected.
Better historical data becomes available and could materially improve calibration.
Large recurring overlays may indicate that the core model no longer captures important risks adequately.
New requirements may affect model design, governance or reporting.
A useful validation exercise can be organised around seven broad questions:
1. Scope:
Are the correct exposures covered by the ECL framework?
2. Data:
Is the information complete, accurate and appropriate for the model?
3. Methodology:
Is the model conceptually suitable for the portfolio and consistent with Ind AS 109?
4. Assumptions:
Are default, recovery, staging and forward-looking assumptions reasonable and supportable?
5. Performance:
Does backtesting show that the model behaves reasonably?
6. Implementation:
Has the approved methodology been translated correctly into systems and calculations?
7. Governance:
Are changes, limitations, overrides and validation findings properly documented and monitored?
This approach helps move validation away from checking one final ECL number and towards understanding the complete estimation process.
ECL models are often seen as the responsibility of credit risk or modelling teams.
But their results directly affect financial reporting.
Changes in ECL can influence:
Finance teams therefore need sufficient understanding of the model to challenge unexpected movements.
If ECL increases materially, they should be able to ask:
Independent validation can make those drivers more transparent.
For material ECL estimates, the issue extends beyond the modelling team.
ECL frequently involves significant judgement.
Board and Audit Committee oversight can therefore benefit from understanding:
For RBI-regulated NBFCs implementing Ind AS, RBI guidance specifically places emphasis on governance and Board-approved documentation of critical matters such as the ECL model and related judgements.
Independent validation can support this oversight by providing challenge that is separate from the team responsible for producing the model output.
There is no single model that can perfectly predict future credit losses.
Credit risk is uncertain.
Economic forecasts are uncertain.
Borrower behaviour is uncertain.
The objective of validation is therefore not to prove that the ECL figure is exactly what future losses will ultimately be.
The objective is to determine whether the process used to estimate those losses is conceptually sound, evidence-based, consistently implemented and appropriately governed.
That is a much more realistic - and useful - standard.
Ind AS 109 gives entities flexibility in how they estimate expected credit losses, but that flexibility comes with responsibility.
ECL measurement must appropriately reflect:
The more modelling and judgement involved in producing that estimate, the more important robust governance becomes.
Independent model validation helps organisations assess whether:
Importantly, independent validation should not be presented as a universal standalone requirement imposed by Ind AS 109 itself. Rather, it is an important governance tool that can help entities demonstrate that their ECL methodologies continue to meet the principles required by the Standard, subject also to any entity-specific regulatory requirements.
For organisations using ECL models, the real question is not simply whether the model produces a provision.
It is whether management can understand, support, challenge and defend the methodology behind that provision.
With experience across actuarial, financial and risk modelling assignments, KA Pandit can support organisations in reviewing ECL methodologies, assumptions, data and model performance, helping strengthen model governance and the reliability of financial reporting.
Get in touch with our experts for personalized solutions.
