1.   Background to the launch of trim

1.1.   Unjustified variability of RWAS of banks using internal models highlighted by the 2007-2009 crisis

Following the financial crisis of 2007-2009, concerns were raised about unwarranted (i.e. non-risk based) variability in the results of models used to calculate regulatory capital requirements. This coincided with concerns from banking supervisors and external stakeholders about the complexity of these models and the resulting opacity of the modelling approaches. This opacity also led to increasing difficulties for supervisors. It was difficult for supervisors to assess whether risks were being captured correctly and consistently by the models developed by the banks.

1.2.   The launch of the trim audit to correct the banks’ internal models and restore their credibility

The targeted review of internal models TRIM was a multi-year project launched by the ECB in early 2016 in close cooperation with the ACN authorities within the MSU. TRIM aimed to assess whether the Pillar I internal models used by the MSU’s major institutions are appropriate in light of the applicable regulatory requirements and whether their results are reliable and comparable. The TRIM was an important step in improving the comparability of the results of the internal models used by the major banks.

In addition, TRIM aimed to harmonise supervisory practices regarding internal models within the MSU.  From a supervisory perspective, TRIM thus complemented the regulatory initiatives of the Basel Committee and the EBA to address the perceived drawbacks of internal models following the financial crisis.

These shortcomings included the high non-risk based variability of model output. The exercise helped to harmonise supervisory practices by developing a common assessment methodology and by strengthening collaboration within European banking supervision. In particular, the richness of the results obtained through TRIM and the subsequent detailed follow-up by supervisors with the institutions involved have played and will continue to play a key role in promoting a level playing field and high quality standards for institutions.

1.3.   TRIM: the largest project to date within the MSU

TRIM reviewed 200 internal models (IMIs) on-site at 65 major institutions. The project covered internal models for credit risk, market risk and counterparty credit.

Given the large number of approved internal models at major institutions and the time required for on-site investigations, TRIM adopted a targeted approach to examine the topics that have the greatest contribution to unwarranted variability in risk-weighted assets (RWAs). The review of credit risk models focused only on those considered to be the most important and critical. In this way, TRIM was able to cover virtually all major institutions with internal models and all credit, market and counterparty risk models.

The ECB published a review of this exercise in April 2021. This is the largest project conducted within the MSU to date, both in terms of :

  • Duration of the exercise.
  • ECB and ACN core teams involved.
  • The scope covered (banks, model types).
  • Depth of analysis.
  • A number of findings were made.
  • Lessons learned.

2.   serious gaps in IRB models identified

2.1.   A large proportion of credit models still valid

A very significant and representative proportion of credit risk models of major institutions can still be used for the calculation of capital requirements. This is the most encouraging part of the audit, although it is mitigated by the shortcomings identified.

2.2.   Gaps in credit models with multiple causes

2.2.1.    more than 5,800 findings identified for credit risk, 30% of which were at severity level 3 or 4

For a number of credit models, add-ons were needed to ensure an appropriate level of capital to cover the underlying risk. This was particularly the case for a number of loss given default (LGD) and credit conversion factor (CCF) models related to low default portfolios (LDPs), for which supervisory measures were imposed following TRIM investigations.

A number of shortcomings were identified and require considerable effort on the part of the institutions to remedy them. More than 5,800 findings were identified for all types of risks, of which at least 30% were rated at severity level 3 or 4.

2.2.2.    Deviations due to misinterpretation of EU requirements, ECB guidelines or ABB guidelines by institutions

In some cases, these gaps are due to historical differences in the way the requirements have been understood at national level or to a previous lack of clarity or guidance on the implementation of certain requirements.

There have also been deviations from draft provisions arising from EBA-led regulatory initiatives, such as the regulatory review of the Internal Ratings Based (IRB) approach to credit risk. In the future, large institutions should benefit from the detailed guidance provided by the ECB (in its internal models guide) as well as the relevant regulatory technical standards and EBA guidance to clarify the interpretations of the different requirements.

2.2.3.    Serious weaknesses identified in the PD and LGD models related to retail and SME portfolios

Institutions generally have the capacity to build adequate IRB models. More specifically, for the probability of default (PD) parameter, more than 70% of the surveys ended with no serious findings on the calculation of one-year default rates and the long-term average default rate.

However, a significant number of findings have been raised concerning the low risk differentiation of these models, due to the low discriminatory power of the scoring functions:

  • Further improvements in calibration methods are still needed, in particular the need for adequate data to ensure that PD estimates reflect long-term average default rates and are sufficiently conservative. At least one high severity finding was identified in 67% of the surveys on the PD parameter.
  • The calculation of the achieved LGD was a frequent cause of compliance problems. Findings on this calculation were raised in every survey. In addition, 42% of the surveys contained severe findings on risk differentiation. In 95% of the surveys where the LGD parameter was examined, at least one high severity finding (i.e. F3 or F4) was raised in relation to this parameter.

2.2.4.    Serious weaknesses also identified in the pd and LGD models related to LDPS leading to the implementation of conservative supervisory measures on capital levels

A large number of findings were made in relation to the rating process and risk quantification. These mainly concerned the calibration methodology and the calculation of long-term average default rates.

One reason for these shortcomings is that there are considerably fewer internal observations available for this type of portfolio (compared to retail and SME portfolios). This means that institutions have to make greater use of other observations (e.g. external default data) in order to calculate default rates and, subsequently, PDs.

For the LGD parameter, most of the findings concerned the calculation of realised LGD and long-term average LGD. It was observed that some institutions had difficulties in finding representative data for these portfolios, which led to cases where the LGD estimate was not based on realised LGD nor representative data.

As a result, supervisory measures (add-ons) were increasingly used to avoid underestimating capital requirements. Overall, in 96% of the surveys, at least one level 3 or 4 finding was made in relation to the PD and LGD parameters.

2.2.5.    Efforts still need to be made on the quality of IRB data

All on-site credit risk surveys included dedicated reviews of IRB data quality. While institutions have made efforts to ensure that robust data management and quality frameworks are in place, in line with BCBS 239 on the ability to effectively aggregate risk data and risk reporting, there are still some important areas that need to be modified or adapted to ensure compliance with requirements. These areas include data quality monitoring and internal control, as well as the assignment of roles and responsibilities for data management.

3.   Gaps also noted in governance, validation of credit models, roll out and SSP

Other topics related to aspects not specific to the technical part of the models were also the subject of findings. These include findings on :

  • The organisation and activities of the internal validation function.
  • Deployment of internal models (Roll out).
  • Permanent partial use (PPU).
  • Managing model changes.

All institutions covered by the TRIM on this subject received follow-up letters containing recommendations because some of their internal practices were not in line with the ECB’s understanding of the applicable regulatory requirements. A subset of institutions received a supervisory decision containing obligations to remedy deviations from the applicable regulatory requirements without delay.

4.   significant gaps in VAR and SVAR methodology, regulatory back-testing and the scope of the internal market risk model approach

Around 60% of TRIM market risk surveys resulted in at least one high severity finding on VaR and sVaR methodology. In addition, just over 80% of large institutions that used incremental default and migration risk charge (IRC) models had at least one high severity finding against these models.

5.   significant weaknesses in the governance and validation process for counterparty risk models

All counterparty risk surveys had at least one finding related to relating to model governance and the model validation process.  These topics made up the bulk of the findings and in 60% of the cases, these findings were issued with a high severity rating.  Observations were also made on specific modelling topics such as

  • Hedging of derivative transactions.
  • The margin at risk period.
  • The consideration of collateral.
  • Taking into account the initial margin.
  • Risk factors.
  • Calibration.

6.   Actions to be taken within the banks following the TRIM

The findings reported in TRIM were followed by binding supervisory decisions requiring the institutions to remedy their shortcomings within the specified timeframe.

Institutions should work intensively to address the findings raised and ensure that these remedies also take due account of the new requirements arising from the EBA’s regulatory review of the IRB approach.

The general structure of the bonds can be understood as follows:

  • In areas where institutions have the freedom to design their own approaches, banks need to better document, analyse and justify aspects of their approaches.
  • In areas where regulation is more prescriptive, for example in the calculation of the default rate, institutions have to modify this specific aspect, which can have a direct impact on risk parameters and, consequently, on capital requirements.
  • Where there are specific concerns about the lack of evidence of compliance, institutions should amend or explain this.
  • In the event of clear non-compliance with the regulatory requirements of any part of the approach, the institution is required to return to compliance.
  • If an institution’s approach does not take account of future requirements, a non-binding recommendation is issued.

More than 40% of the obligations imposed have an implementation period of between 12 and 18 months after the publication of the respective decision, while for a quarter of the obligations the period is less than 12 months.

7.   A substantial impact in RWA to be expected following the TRIM

The ECB estimates that the aggregate impact of the conservative supervisory measures following TRIM and the model changes approved in the TRIM surveys will result in a 12% increase in the aggregate RWAs covered by the models assessed.  The results of the TRIM correspond to an absolute in RWA increase of €275 billion and a very significant impact on the CET1 ratios for the institutions concerned.

8. The ECB’s long-term expectations for improving the framework for the use of internal models

8.1.   Strong expectations expressed by the ECB on the framework for the use of internal models in the future

Institutions must continue to invest in the development and maintenance of their internal models.

Institutions should continue to work on their internal models to maintain the high quality of models obtained through TRIM. This includes defining internal model strategies for the development and maintenance of internal models. In particular, the independent internal validation function needs to be strengthened in line with TRIM requirements to ensure a continuous internal challenge of the performance of internal models and an appropriate follow-up of corrective actions.

8.2.   More ECB guidance to regulate the use of internal models

In pursuit of the main objectives of reducing the variability of non-risk based RWAs and harmonising supervisory practices, TRIM has helped to develop comprehensive approaches that will promote consistency and quality in the supervision of internal models used by major institutions, thereby contributing to a level playing field.

These include:

  • The development of the ECB’s guidance on internal models.
  • The improvement of the in-depth on-site audit approach was implemented with the allocation of sufficient resources to carry out a large number of on-site IMIs.
  • A systematic overview of the main features and weaknesses of internal models used in major institutions.

9.   References


Abbreviations and glossary

TRIM: Target Review of Internal Model

NCA: National Competent Authority

MSU: Single Supervisory Mechanism

SI: Important institution

IMI: internal Model Inspection

EBA: European Banking Authority

LGD: Loss Given Default

PD: Probability of Default

CCF: credit conversion factor

HDP: High Default Portfolio

LDP: Low Default Portfolio

IRB: Internal Rating Based

VaR: Value at Risk

SVaR: Stress VaR