In February 2024, ECB Banking Supervision released its revised ECB guide to internal models. We will focus on selected aspects of use of data in a credit risk context, as e.g. challenges resulting from intergration of external data, external ratings or pooled data.
Prolog
„Articles 143, 283 and 363 of Regulation (EU) No 575/2013 (CRR)1 require the European Central Bank (ECB) to grant permission to use internal models for credit risk, counterparty credit risk and market risk where the requirements set out in the corresponding chapters of the CRR are met by the institution(s) concerned. Based on the current applicable European Union (EU) and national law, the ECB guide to internal models provides transparency on how the ECB understands those rules and how it intends to apply them when assessing whether institutions meet these requirements.“ (see ECB 2024, p.4, clause 1, para. 1).
The ECB also states: „This guide should not be construed as going beyond the current existing applicable EU law including, among others, adopted regulatory technical standards (RTS), and national law and therefore is not intended to replace or overrule applicable EU and national law.“ (see ECB 2024, p.4, clause 3, para. 1).
Furthermore, the ECB notes that it provides in the guide its understanding of the CRR provisions which apply currently. (see ECB 2024, p.5, clause 3, para. 3).
And finally, still in the foreword, it is set out that: „The concept of „best practice“ as used in this guide can be described as actions or measures which – in the view of the ECB – ensure compliance with certain prudential requirements in a prudentially sound manner. In this context, it should be noted that (i) there can also be other actions or measures that ensure compliance, and (ii) institutions are not required to apply that practice where compliance is ensured in another way.“ (see ECB 2024, p.5, clause 4).
Credit risk: Use of data
At the beginning, the credit risk chapter of the ECB guide defines its scope and states that it does not intend to cover all topics of the CRR for the IRB approach in full, which could be subject to review in the course of a model investigation. Rather, that the goal is to offer transparency on how the ECB understands a series of topics related to internal models used for the IRB approach. And that on the selected topics the chapter is aligned with the EBA Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures (EBA 2017) (see ECB 2024, p.61).
The topical areas addressed in the guide with respect to credit risk are overall · data maintenance for the IRB approach · use of data · definition of default · probability of default · loss given default · conversion factors · model-related MoC (Margin of Conservatism) · review of estimates and · calculation of maturity for non-retail exposures. From which we will focus now on specific aspects for the use of data.
A prerequisite and often a challenge in empirical model building and risk quantification is the availability, quality and appropriate use of data.
The ECB refers in its elaborations on the use of data to Article 144(1)(d) of the CRR to set the floor – and reminds that institutions must collect and store all relevant data to effectively support their credit risk measurement and management processes. Furthermore, the guide references paragraphs 15-34 (Data requirements: · quality of data · governance for data representativeness · representativeness of data for model development · representativeness of data for calibration of risk parameters) of the EBA Guidelines on PD estimation (EBA 2017), before turning to the data topics addressed in detail.
Of those, let us specifically look at the following issues raised:
Use of external data: Data-related requirements under the CRR do not only apply to internal data, but also to external or pooled data. The principles on collection and storage of data hold for own data as well as for data received from a pool. With respect to risk differentiation, risk quantification and review of estimates institutions should assess whether external data can be used to complement internal data (if felt to be needed).
Problem of common obligors: When banks use different data sources for risk quantification, there is a risk of common obligors in the data sets, and therefore for double-counting and biases in the calculation of 1Y-default rates. Banks should identify common obligors and care for only single-counting. If they can´t do so, they should evaluate the potential bias in the 1Y-default rates and reflect them in the calculation of the 1Y-default rates and the long-run average (LRA) default rates.
Assigning exposures to obligors / facilities grades / pools on the basis of (statistical) models: The data used for building the model must be representative for the actual population of bank obligors / facilities.
Representativeness in case of internal data scarcity: Proving representativeness of external data when internal data is scarce is more difficult. In case the bank cannot prove the representativesness of the external data, it may still use the external data if the informational benefits of doing so outweigh the drawbacks. Therefore the institution should provide evidence that the model´s performance does not deteriorate, the parameter estimates are not biased, and that an approriate margin of conservatism (MoC) is applied, and it should do so by conducting quantitative and qualitative analyses specifically designed for this purpose.
Use of external bureau scores or external ratings as input variables in the rating process: To mitigate the risk that not all relevant informations are considered in the internal rating model when a bank uses external credit scores or external ratings as input variables for their rating, especially when external input plays a dominant role, the ECB sees the following principles to comply with: The external information is regularly updated. Institutions understand the structure, nature and key drivers of the external scores / ratings and regularly verify the continued appropriateness as input variables. With respect to validation requirements, they are seen to be similar to the requirements for other internal and external input variables. Furthermore, the bank ensures that all relevant internal information on an obligor´s creditworthiness is sufficently accounted for and sufficiently weighed in the internal rating. The institution shows that the internal contribution is sufficient in the sense that the internal rating does not only take on the external scores or external ratings. The institution remains responsible for the performance of the model. And the institution takes care that potential correlations between the risk drivers do not cause bias or double-counting effects in the risk parameter estimates (might be particularly relevant here due to potential duplicated information consideration).
Use of pooled data: First, the treatment for the use of pooled data is similar to the use of external data. Second, when an institution uses data pooled across institutions the rating systems and criteria of the other institutions must be similar to the own. Among other things, this implies that an institution should ensure that there is a common defintion of key drivers and processes, and that policies / procedures for human judgement (incl. overrides) can be applied in a comparable and similar manner across the institutions.
Use of purchased rating systems or models (pool models): Where institutions work with rating systems / models purchased from third-party vendors, Art 144 (1), last sentence, of the CRR becomes relevant. It says: „The requirements to use an IRB Approach, including own estimates of LGD and conversion factors, apply also where an institution has implemented a rating system, or model used within a rating system, that it has purchased from a third- party vendor.“ The ECB guide continues with stating that in order to comply with this requirement instititutions should make sure that for model building and parameter calibration all relevant internal information is taken into consideration. Especially, long-run average default rates, LGD and CCFs based only on internal data should always be calculated and considered for calibration. And: The institution remains responsible for the performance of the rating system / model.
When institutions use pool models the following principles should be pursued, states the ECB guide:
- If PD estimates are computed using pooled data, banks should verify that the data requirements for default rate calculation (as elaborated in para 121 of the ECB guide) are met by the data used for risk quantification – or that the data are adjusted accordingly.
- When using a joint pool model, each participating institution should make sure that its rating process is aligned, to the extent that all input risk drivers are defined in the same way. Plus, all assessments of the qualitative rating components should be done in a comparable way.
- If the model-relevant parts of the processes for managing distressed obligors are not aligned for the participating institutions, and if the model is used for the estimation of risk parameters, these differences should be taken into account within the model or by an adjustment in accordance with the EBA Guidelines on PD estimation (para 37(a)(viii), EBA 2017). If a pool model is used for the estimation of LGD parameters, differences in the model-relevant parts of the workout processes should also be taken into account within the model or through an appropriate adjustment.
- Banks should provide for, that all relevant internal information on an obligor´s creditworthiness is reflected and the rating is updated with new information in a timely manner.
- Again, each institution should remain responsible for the performance of the rating model on its portfolio.
And, finally, if an institution starts doing systematic adjustments to the outputs of the pool model, this institution should carry out internal procedures to evaluate whether significant weaknesses in the model exist and a model change has to be induced.
Let us conclude with essential information:
The relevant regulatory references for the „Use of data“ chapter of the ECB guide to internal models are2
- CRR: Art 144, para (1)(d); Art 171, para (1)(a), (b); Art 172, para (3); Art 174, para (b), (c), (e); Art 176; Art 178, para (4); Art 179, para (1)(a), (c), (d), (2)(a), (b).
- Commission Delegated Regulation (EU) No. 2022/439: Art 42, 45, 47, 53.
- EBA Guidelines on PD estimation: Para 15-35.
2 Source: Table 10, p. 71 (ECB 2024).
References
(1) Capital Requirements Regulation (CRR): Regulation (EU) No 575/2013 of the European Parliament and of the Council of 26 June 2013 on prudential requirements for credit institutions and investment firms and amending Regulation (EU) No 648/2012 (OJ L 176, 27.6.2013, p. 1).
European Banking Authority (EBA 2017), EBA Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures, EBA/GL/2017/16 (referred to here as EBA Guidelines on PD estimation – if abbreviated).
European Central Bank (ECB 2024), ECB guide to internal models, February 2024.
European Commission (EC 2022), Commission Delegated Regulation (EU) No. 2022/439 of 20 October 2021.
Disclaimer
Ed.: LBMS Advisory Services GmbH, 1070 Vienna, Austria, FN 417881g, www.lbms.at
The contents of this document are for general information and marketing purposes only and do not contain any recommendations whatsoever. In particular, the information does not constitute an offer or legal, tax, investment or financial advice and is not suitable as a substitute for individual business advice.
The content has been compiled with the greatest possible care. However, LBMS Advisory Services GmbH accepts no liability whatsoever for the accuracy, completeness and up-to-dateness of the information and data contained herein, nor for any loss or damage of any kind (whether direct or indirect) arising from or in connection with this document. The information does not constitute a decision-making aid.
The content created and published by the publisher is protected by copyright. All rights are reserved, also regarding any disposal, exploitation, reproduction, editing, distribution, as well as in the event of applications for industrial property rights. Unauthorized use of such content is not permitted.