Definition of Default – Modelling Considerations

Backscoring Definition of Default (DoD), quantification of DoD-related Appropriate Adjustments (AAs) and Margin of Conservatism (MoC) might be the most important factors of an IRB-modelling process in terms of model outcome and capital requirements. However, these elements do not always get the attention they deserve. To great dissatisfaction of the ECB, Dutch banks are still not fully compliant with regulatory DoD requirements. During informal conversations the ECB indicated that continued incompliance may threaten the IRB status of a bank which would then lead to a return to the standardized approach with accompanying increase in capital requirements. Some Dutch IRB banks have already received IMI (internal model investigation) findings related to DoD and others can expect them soon. To address the findings a new iteration of the DoD will be necessary, triggering new redevelopment of (mortgage) IRB models. This paper highlights some key considerations while developing a model with a new DoD, in particular in light of the IMI findings on mortgage models.

Introduction

In 2016, the EBA released the EBA guidelines on the Definition of Default (DoD) with the aim to harmonize the criteria across EU institutions for classifying borrowers in default. In particular, the guidelines clarified the days past due criterion, the materiality threshold, unlikeliness to pay (UTP) triggers and rules for returning from defaulted status.

Furthermore in 2018 EBA released the EBA guidelines on management of non-performing (NPE) and forborne exposures clarifying how banks should manage the non-performing exposures and forborne exposures (borrowers in financial difficulties).

Since the release of the EBA guidelines, several key developments have occurred as banks and regulators fine-tune their practices to comply with the standards.

The ECB has conducted the Targeted Review of Internal Models (TRIM) process to assess the consistency and adequacy of banks’ internal models, including assessment of the DoD.

To align with the guidelines and IMI and TRIM findings, many banks have gone through multiple iterations of their DoD definition and frameworks. These iterations often include calibration of the materiality threshold and improving the Unlikeliness to Pay (UTP) criteria and processes. Note that a change in DoD is classified as a material change of the IRB models.

Furthermore, forbearance treatment under the DoD has been closely scrutinized. Many banks have refined their forbearance processes, especially to avoid misclassification of defaults when forbearance is applied. This has led to closer monitoring of restructured loans to ensure that loans in forbearance are properly flagged for potential default risk.

While a lot of improvements have been made, (most) banks are still not fully compliant with DoD. As of 2024, UTP triggers are by far the most important source of incompliance with DoD. Some UTP triggers involve subjective judgment which can result in inconsistencies and significantly reduce comparability across banks, contradicting the regulator’s ambition.

During informal conversations the ECB indicated that continued incompliance may threaten the IRB status of a bank which would then lead to a return to the standardized approach with accompanying increase in capital requirements.

New IMI findings

Some Dutch IRB banks have already received IMI (internal model investigation) findings on their mortgage models related to DoD and others can expect them soon. UTP assessments for interest-only mortgages receive specific supervisory attention.

For instance, ECB observed that assessment for financial difficulties was not always or incorrectly performed, e.g. when providing extension of interest-only loans. Furthermore, the performed affordability and viability tests were determined to be not prudent enough in some cases. Additionally, an LTV-based UTP trigger is required for interest-only loans that depend on the collateral as primary source of repayment.

The deficiencies potentially lead to an underestimation of the default risk of mortgages, and interest-only mortgages in particular. In order to address the IMI findings, banks will need another iteration of the implementation of the DoD in the IRB models which results in a material change. As a consequence, also the mortgages IRB models need to be redeveloped. In the next section some key considerations while developing a model with a new DoD will be highlighted.

Modelling new DoD

Some banks are already at their second or third iteration of their DoD implementation in the IRB models. For example, some UTP triggers were introduced at different moments in time, some were phased out, and for some triggers the scope or implementation changed. This leads to historical data that is not representative for the application portfolio. During model development, representativeness of data for risk differentiation and calibration of risk parameters needs to be ensured.

In order to create a modelling dataset (reference data set, RDS) that is representative for the new DoD, the dataset is ‘backscored’. This is done by retrospectively applying the new DoD to the historical data.

In some cases, e.g. due to data quality issues, this backscoring can’t be performed for the complete historical period or there are limitations to the backscoring. These can be addressed by appropriate adjustments (AA). In cases where performing AAs is impossible or  there is uncertainty relating to the quantification of AAs, a DoD-related Margin of Conservatism (MoC) needs to be applied.

Reducing the impact of backscoring limitations on capital requirements

To isolate  uncertainties regarding the approximations of the current DoD on the historical dataset and the consequential increase in capital requirements (e.g. through MoC), banks can consider modelling the impacted parts of the portfolio as a separate sub-model. For instance, if data quality in a particular source system is limited, modelling teams can examine to separately model the labels that use this source system.  The other labels can be treated in another submodel. Likewise, if there is uncertainty in backscoring a certain UTP trigger, modelling teams can assess if modelling segments with high prevalence of this UTP trigger separately might lead to an overall reduction of capital requirements. In this way, the uncertainty for the other segments will decrease. In terms of model outcome and capital requirements, improving DoD Backscoring and quantification of DoD-related AAs might be the most important factors of a modelling process.

Backtesting

Incomplete backscoring negatively affects the trustworthiness of backtesting results because the historical data is not fully reflective of actual credit risk behaviour. It hides weaknesses, biases model calibration and leads to incorrect conclusions about the model’s predictive power. Ensuring a complete and accurate backscoring of defaults is essential for meaningful model validation and reliable risk assessments.

For IFRS 9 risk parameter backtesting the impact of having an incomplete or biased historical dataset is more severe than for capital models. Backtests of risk parameters for capital models simply have to ensure that the capital models are on the conservative side, whereas backtests of IFRS 9 models and risk parameters have to prove that the predictions are accurate. Proving this with incomplete or biased historical data is more difficult.

Representativeness of default distribution

Historically (pre-DoD), most defaults were arrears-based. Currently, in some portfolios most defaults result from UTP triggers, which are often idiosyncratic. These idiosyncratic defaults are sometimes more difficult to predict compared to arrears-based defaults.

Furthermore, specific kind of defaults can be expected to result in different loss profiles.

Therefore, large shifts in the default distribution can greatly affect the model performance. During model development, model validation and model monitoring the distribution of UTP triggers needs to be carefully assessed to gain insight in the impact on model performance.

Significantly changed risk perception with new DoD

A new DoD can greatly affect the risk differentiation and risk calibration of models or model segments. Given that a significant portion of the DoD-related IMI findings are focused on the interest-only segment, banks can expect the risk parameters of this segment to be affected significantly. Furthermore, the relationships between defaults and risk drivers needs to be re-assessed completely. Potentially, the new DoD might also lead to a higher perception of increased riskiness of the interest-only segment.

Conclusion

In January 2025, a few Dutch IRB banks can expect IMI findings on their mortgage models. UTP assessments for interest-only mortgages receive specific supervisory attention. A new iteration of the DoD will be necessary, also warranting redevelopment of the mortgage IRB models. During the redevelopment, backscoring DoD on the historical dataset and quantifying the DoD-related appropriate adjustments and MoC will be crucial. Furthermore, key decisions, e.g. on segmentation, can greatly affect the material impact on model output and capital requirements. Additionally, the impact of backscoring limitations should be assessed on the backtesting. Moreover, the distribution of UTP triggers over time needs to be studied to gain insight in the impact on model performance. The new DoD might significantly impact the risk parameters and thus affect the perception of the inherent risk in the interest-only segment.

How Triple A – Risk Finance can help you

Triple A – Risk Finance has assisted multiple Dutch banks with both model development and validation of mortgage models. In particular, we are experienced in:

• Quantification of appropriate adjustments and Margin of Conservatism for DoD
• Implementation of new DoD in models (both IRB, IFRS 9 models and model overlays) and backscoring model development datasets
• Interpretation of regulations and following up on ECB findings

Dit artikel is geschreven door Robert Jan Sopers, Bas Jordans en Stephan Runderkamp