ECB guide to internal

models

October 2019
Document releases
Release Date of issue Release number Sections/Chapters Rationale underlying the
modified release

Guide for the Targeted Review of 17/02/2017 1.0 -

Internal Models (TRIM) – First
version

Public Consultation – ECB guide 28/03/2018 2.0a General topics Incorporation of institutions’
to internal models – general topics feedback on version 1.0,
chapter outcomes of the supervisory
review on general topics,
and related horizontal
analyses

Public Consultation – ECB guide 07/09/2018 2.0b Credit risk, market Incorporation of institutions’
to internal models – risk-type- risk and counterparty feedback on version 1.0 and
specific chapters credit risk experience gained with TRIM
investigations; addition of
dedicated principles for low-
default portfolios

Revised ECB guide to internal 15/11/2018 2.1a General topics Inclusion of industry’s
models – general topics chapter feedback received during the
public consultation

Revised ECB guide to internal 08/07/2019 2.1b Credit risk, market Inclusion of industry’s
models – risk-type-specific risk and counterparty feedback received during the
chapters credit risk public consultation

Consolidated version 01/10/2019 2.2 Consolidated Compilation of the general

document topics and risk-type-specific
chapters in one consolidated
document

ECB guide to internal models – Contents 1

Contents
Foreword 4

General topics 6

1 Overarching principles for internal models 6

2 Roll-out and permanent partial use 13

3 Internal governance 17

4 Internal validation 23

5 Internal audit 32

6 Model use 35

7 Management of changes to the IRB approach 43

8 Third-party involvement 49

Credit risk 56

1 Scope of the credit risk chapter 56

2 Data maintenance for the IRB approach 57

3 Use of data 65

4 Probability of default 73

5 Loss given default 88

6 Conversion factors 104

7 Model-related MoC 112

8 Review of estimates 113

9 Calculation of maturity for non-retail exposures 116

Market risk 117

1 Scope of the market risk chapter 117

2 Scope of the internal model approach 118

3 Regulatory back-testing of VaR models 133

4 Aspects of internal validation of market risk models 144

5 Methodology for VaR and stressed VaR 150

ECB guide to internal models – Contents 2

 6 Methodology for IRC models focusing on default risk 162

7 Risks not in the model engines 174

Counterparty credit risk 186

1 Scope of the counterparty credit risk chapter 186

2 Trade coverage 187

3 Margin period of risk and cash flows 192

4 Collateral modelling 196

5 Modelling of initial margin 201

6 Maturity 202

7 Granularity, number of time steps and scenarios 205

8 Calibration frequency and stress calibration 207

9 Validation 208

10 Effective expected positive exposure 215

11 Alpha parameter 216

Annex 218

Acronyms 224

Glossary 225

ECB guide to internal models – Contents 3

Foreword

1. 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 institutions
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.

2. The guide is also intended as a document for the internal use of the different
supervisory teams, with the aim of ensuring a common and consistent approach
to matters related to internal models. When applying the relevant regulatory
framework in specific cases, the ECB will take into due consideration the
particular circumstances of the institution concerned.

3. This guide should not be construed as going beyond the current existing
applicable EU and national law and therefore is not intended to replace,
overrule or affect applicable EU and national law.

In accordance with the requirements set out in the CRR, the European Banking
Authority (EBA) has drafted regulatory technical standards (RTS). These
include the Final Draft RTS on assessment methodology for the Internal
Ratings-based (IRB) Approach, the Final Draft RTS on the specification of the
nature, severity and duration of an economic downturn and the Final Draft RTS
on the specification of the assessment methodology for competent authorities
regarding compliance of an institution with the requirement to use internal
models for market risk and assessment of significant share. 2 These specify how
competent authorities should assess compliance with the regulatory framework

1
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), referred to in this guide as the “CRR”. For the purposes of this
document the reader’s attention is also drawn to the corrigendum published on 30 November 2013 (OJ
L 321, 30.11.2013, p. 6).
2
Final Draft Regulatory Technical Standards on the specification of the assessment methodology for
competent authorities regarding compliance of an institution with the requirements to use the IRB
Approach in accordance with Articles 144(2), 173(3) and 180(3)(b) of Regulation (EU) No 575/2013
(EBA/RTS/2016/03), referred to in this guide as the “Final Draft RTS on assessment methodology for
IRB”.
See also: Final Draft Regulatory Technical Standards on the specification of the nature, severity and
duration of an economic downturn in accordance with Articles 181(3)(a) and 182(4)(a) of Regulation
(EU) No 575/2013 (EBA/RTS/2018/04), referred to in this guide as the “Final Draft RTS on the
specification of the nature, severity and duration of an economic downturn”.
See also: Final Draft Regulatory Technical Standards on the specification of the assessment
methodology for competent authorities regarding compliance of an institution with the requirements to
use internal models for market risk and assessment of significant share under points (b) and (c) of
Article 363(4) of Regulation (EU) No 575/2013 (EBA/RTS/2016/07), referred to in this guide as the
“Final Draft RTS on assessment methodology for IMA and significant share”.
Note that there are no RTS on assessment methodology mandated for the assessment of the Internal
Model Method (IMM) for calculating counterparty credit risk (CCR) exposures.

ECB guide to internal models – Foreword 4

 defined in the CRR. The Final Draft RTS have not yet been adopted by the
European Commission, but those Final Draft RTS are referred to in the guide as
good practice for interpretative purposes. Some parts of this guide may require
revision once the European Commission has adopted the RTS by means of a
Delegated Regulation. The ECB will amend or delete those parts of the guide
when the RTS enter into force.

The ECB provides in this guide its understanding of the CRR provisions which
apply currently. The legal references in this guide to the CRR provisions
(including the number of the Articles) are references to the CRR before the
amendments introduced by Regulation (EU) 2019/876 3. The amendments
introduced by Regulation (EU) 2019/876 which have an impact on the guide will
only apply from July 2021.

3
Regulation (EU) 2019/876 of the European Parliament and of the Council of 20 May 2019 amending
Regulation (EU) No 575/2013 as regards the leverage ratio, the net stable funding ratio, requirements
for own funds and eligible liabilities, counterparty credit risk, market risk, exposures to central
counterparties, exposures to collective investment undertakings, large exposures, reporting and
disclosure requirements, and Regulation (EU) No 648/2012 (OJ L 150, 7.6.2019, p. 1).

ECB guide to internal models – Foreword 5

 General topics

1 Overarching principles for internal models

1.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRD IV, 1 as implemented in the relevant 26/06/2013 3, 76, 85 Paragraph 1, sub-paragraphs

national law 7, 9 and 11

CRR 26/06/2013 175, 179, 185, 189, 190, 191,

287, 288, 292, 293, 368

Other references

Final Draft RTS on assessment methodology 21/07/2016 3, 9, 10, 12, 13, 14, 16, 17,
for IRB 30, 32, 33

Final Draft RTS on assessment methodology 22/11/2016 7-34

for IMA and significant share

EBA Guidelines on SREP 2 19/12/2014 235

EBA Guidelines on internal governance 3 21/03/2018 Section 22

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB and the Final Draft RTS on assessment
methodology for IMA and significant share will become additional relevant legal
references. Currently these RTSs only exist in a final draft version.

1. The principles listed in this section relate to internal models that are subject to
supervisory approval for the calculation of own funds requirements for credit,
market and counterparty credit risk (Pillar 1 models), unless stated otherwise.

1.2 Guidelines at consolidated and subsidiary levels

2. An inconsistent implementation of internal model-related tasks within a banking

group bears the risk of an inappropriate coverage of the risks measured by
internal models at group level. Therefore, institutions should either develop
binding group-wide (i.e. consolidated) principles and guidelines relating to the
life cycle 4 of internal models 5, or ensure that each relevant entity has

1
Directive 2013/36/EU of the European Parliament and the Council of 26 June 2013 on access to the
activity of credit institutions and the prudential supervision of credit institutions and investment firms,
amending Directive 2002/87/EC and repealing Directives 2006/48/EC and 2006/49/EC (OJ L 176,
27.6.2013, p. 338), referred to in this guide as “CRD IV”.
2
EBA Guidelines on common procedures and methodologies for the supervisory review and evaluation
process (SREP) (EBA/GL/2014/13), referred to in this guide as the “EBA Guidelines on SREP”.
3
EBA Guidelines on internal governance (EBA/GL/2017/11), referred to in this guide as the “EBA
Guidelines on internal governance”.
4
The model life cycle generally includes the following steps: development (including data preparation),
calibration (including data preparation), validation, supervisory approval (if necessary), implementation
in internal processes, application and review of estimates.

ECB guide to internal models – General topics 6

 appropriate and independently audited principles and guidelines in place with a
high degree of consistency between one another.

3. A consistent and integrated level of application of the group principles and

guidelines in the relevant entities is expected. This could be ensured by
applying controls that verify that these principles are implemented correctly in
all relevant entities. Examples of such controls include periodic monitoring
procedures by the parent entity on implementation at local level or a gap
analysis between group-wide principles and local application, including local
guidelines. These controls can be used to identify those gaps and to mitigate
any associated risk. The group-level policies should clearly define under which
circumstances deviations from the group-wide principles would be accepted.
The parent entity should be informed about such deviations. The effective
implementation of the policies should be periodically monitored and assessed.

1.3 Documentation of internal models

4. All internal models should be documented to allow a qualified third party to

independently understand the methodology, assumptions, limitations and use of
the model and to replicate its development and implementation.

5. The institution should therefore define principles and guidelines for model
documentation. These should encompass guidance for the governance of the
model documentation itself. The scope of the model documentation should be
defined by type of model. In-scope areas should include at least the technical
aspects of the model (methodology and assumptions), data (processes),
instructions for model users and performance/validation (including the results of
implementation testing).

6. The institution should be able to demonstrate how its documentation and the
register of its internal models facilitate the internal and external understanding
of the models. The institution can choose to have one or more registers for
models in use, according to the different risk types (credit, market or
counterparty credit risk). The register(s) should contain at least the following
information for each model: 6 the model owner(s), range of application,
materiality, approval date, potential restrictions on the model’s use (e.g.
conditions), as well as the key model weaknesses and change versioning. The
register should also include models purchased from third-party vendors.

7. Adequate controls of the register of the institution’s internal models, together

with an inventory of the documentation, including an annual review, should be in
place. This includes a policy for document management that clearly states the
roles and responsibilities involved in approving documents, as well as how
changes in documentation are applied and communicated internally. In addition,
5
In the case of credit risk, “internal models” should be read hereinafter as “IRB rating systems”.
6
Regarding credit risk, Article 33 of the Final Draft RTS on assessment methodology for IRB specifies
the contents of a register of rating systems.

ECB guide to internal models – General topics 7

 the institution should have a policy regarding the adequate archiving and
maintenance of information, access permissions and the assessment of the
completeness and consistency of the information held.

8. Documentation should be kept up-to-date and the institution should keep

documents for an appropriate period of time, taking into account legal or
regulatory retention periods.

1.4 Implementation of a model risk management framework

9. Effective model risk management allows institutions to reduce the risk of

potential losses and underestimation of own funds requirements as a result of
flaws in the development, implementation or use of the models. To mitigate
these risks, institutions should have a model risk management framework 7 in
place that allows them to identify, understand and manage their model risk for
internal models across the group. This framework should comprise, at least, the
following.

(a) A written model risk management policy. This policy should, as a minimum,
include a concept or a definition of what constitutes a model, provide the
institution’s interpretation of model risk 8 and describe the model risk
framework with reference to its different components (e.g. model
governance, risk control function, validation function, internal audit) and
their related documented policies.

(b) A register of the institution’s internal models, as described in paragraph 6.

This register should facilitate a holistic understanding of the application
and use of the models and provide the institution’s management body and
senior management with a comprehensive overview of the models in
place.

(c) Guidelines on identifying and mitigating any areas where measurement

uncertainty and model deficiencies are known to exist, according to their
materiality. In particular, those elements that relate to qualitative aspects of
model risk (such as data deficiencies, model misuse or implementation
errors) should be considered. This methodology should be applied
consistently across the group (e.g. in subsidiaries or regions).

7
Institutions are expected to implement an effective model risk management framework for all models in
use. However, Section 1 focuses only on internal models approved for use for the calculation of own
funds requirements for credit, market and counterparty credit risk and the respective requirements for a
model risk management framework for these models.
8
“Model risk” as defined in Article 3(1) sub-paragraph (11) of CRD IV.

ECB guide to internal models – General topics 8

 (d) Guidelines and methodologies for the qualitative and/or quantitative
assessment and measurement 9 of the institution’s model risk.

(e) Guidelines with respect to the model life cycle as set out in paragraph 2.

(f) Procedures for model risk communication and reporting (internal and
external).

(g) Definition of roles and responsibilities within the model risk management
framework (e.g. define which unit(s) is/are in charge of or involved in
independent reviews of risk estimates).

1.5 Identification of management body and senior management

10. Institutions should clearly define the roles and responsibilities of their
management body and senior management as defined in Article 3(1) sub-
paragraphs (7) and (9) of CRD IV, as implemented in the relevant national law
with regard to internal models and in relation to each risk type. The institution’s
internal documentation should also clearly describe the composition of the
management body and the senior management.

11. The term “management body” could refer to the single board, in a one-tier
system, or to the function of the management and supervisory boards in a two-
tier corporate governance system. The institution should document the roles
and responsibilities of each individual in the management body. For the
purposes of this guide, the management body refers to the members of the
management body in its management function (executive members).

12. The institution should assess the appropriateness of designated committees of

the management body (if applicable) in order to ensure that they provide an
adequate support function for effective decision-making procedures. This holds,
in particular, for decisions concerning material aspects of the institution’s
internal models. The institution should clearly document the composition,
mandate and reporting lines of committees responsible for internal model
governance and oversight, as well as the decisions taken. These committees
should be given a mandate by the management body; the mandate should
clearly define their tasks and authority. In order for the designated committee to
be set at an appropriate level, it should be chaired by a member of the
management body.

13. The institution should also be able to identify which individuals constitute its
senior management with respect to the credit, market and counterparty credit

9
Article 85(1) of CRD IV refers to model risk in operational risk. However, and specifically with regard to
Pillar 1 models, the EBA Guidelines on SREP (paragraph 235) state that competent authorities should
consider model risk as part of the assessment of specific risks to capital (e.g. IRB model deficiency is
considered as part of the credit risk assessment) and for the capital adequacy assessment. In
particular for IRB models, the expected range of estimation errors should be reflected in the margin of
conservatism of the model, in accordance with Article 179(1)(f) of the CRR. The guidance for the
calculation of the margin of conservatism for IRB models is defined in the credit risk chapter.

ECB guide to internal models – General topics 9

 risk Pillar I model frameworks. In addition to the specifications of Article 3(1)
sub-paragraphs (7) and (9) of CRD IV, senior management can be deemed to
constitute the highest hierarchical level(s) below the management body with a
clearly defined responsibility for internal models. 10 The senior management
should either report directly to the management body or be responsible for
providing it with the necessary information to carry out its duties, especially with
regard to its oversight role. The senior management’s decision-making
procedures relating to all aspects of internal models should be clearly
documented.

1.6 General principles for internal validation

14. All internal models and internal estimates should be subject to an initial, and
subsequently to an annual, internal validation. For the avoidance of doubt, the
term “initial validation” in the guide refers to the validation of new models as well
as the validation of material changes and extensions to approved models.

15. To ensure the effective independence of the internal validation function from the
model development process (i.e. model design, development, implementation
and monitoring), institutions should have appropriate organisational
arrangements in place. The ECB understands that the possible organisational
arrangements are as follows: 11

(a) separation into two different units reporting to different members of the
senior management;

(b) separation into two different units reporting to the same member of the
senior management;

(c) separate staff within the same unit.

The decision on which organisational arrangement to adopt should take into

account the nature, size and scale of the institution and the complexity of the
risks inherent in its business model.

16. Consequently, the ECB understands that large and complex institutions should
implement the most robust independence option.

17. With option (b) of paragraph 15, two different units report directly 12 to the same
member of senior management, but act separately at any level below that.
When using this second option, the ECB considers that it is good practice if the
institution fulfils the additional requirements specified in Article 10(3) of the Final

10
This includes clarity on the role, authority and responsibilities of the various positions within senior
management.
11
The principles set out in paragraphs 15 to 18 do not apply to the organisational structures for the
management of counterparty credit risk due to the specific requirements of Article 287(1) and (2) and
Article 293(1)(c) of the CRR. Further guidance is given in Section 9 of the counterparty credit risk
chapter.
12
Crossing the units’ reporting lines on a lower level would impede fulfilling the requirement.

ECB guide to internal models – General topics 10

 Draft RTS on assessment methodology for IRB and Article 22(1)(e) of the Final
Draft RTS on assessment methodology for IMA and significant share. The
internal audit should regularly assess whether these additional requirements
are being met.

18. The ECB considers that option (c) of paragraph 15 could be suitable for small
legal entities which are not classified as global systemically important
institutions (G-SIIs) or other systemically important institutions (O-SIIs). 13 When
using this option, the ECB considers that it is good practice if the institution
fulfils the additional requirements specified in Article 10(4) of the Final Draft
RTS on assessment methodology for IRB and Article 22(2) of the Final Draft
RTS on assessment methodology for IMA and significant share, and in
particular that the internal audit regularly assesses that these additional
requirements are met.

19. A proper separation of the staff of the development function from the staff of the
validation function enables institutions to limit the risk of conflicts of interest
resulting in an ineffective challenge from the validation. To mitigate this risk, the
institution should ensure that the staff of the validation function is separate from
the staff involved in the model development process. 14

20. The validation function 15 should be adequately staffed following the

proportionality principle. It should have suitable resources and experienced, 16
qualified personnel (who have appropriate quantitative and qualitative
knowledge) to enable it to conduct an effective independent challenge of the
internal models and internal estimates and their performance.

1.7 General principles for internal audit

21. The CRR requires internal models to be subject to regular review by the internal
audit or another comparable independent auditing unit (hereinafter internal
audit). 17 In the understanding of the ECB, also taking into account the EBA
Guidelines on internal governance 18, this regular review needs to be efficient
and effective to meet that objective.

22. To enable an objective assessment, the internal audit should be granted an

adequate level of independence from the processes and units reviewed to
ensure that:

13
SIs not considered as O-SIIs are those not included in the list available on the EBA banking website.
14
In particular regarding option (c) of paragraph 15, this means different sub-teams with different tasks.
15
Regardless of whether the related validation tasks are allocated internally or delegated to a third party.
16
The use of external resources has to comply with the institution’s internal validation guidelines. See
also Section 8.3.1 of this chapter.
17
See also Articles 191, 288 and 368(h) of the CRR.
18
See Section 22 “Internal audit function” of the EBA Guidelines on internal governance.

ECB guide to internal models – General topics 11

 (a) there is an effective separation of the internal audit from the staff involved
in the operations of the internal models, e.g. the validation function, the
risk control unit and all other relevant business areas;

(b) it reports directly to the management body 19;

(c) no undue influence is exerted on the staff responsible for the audit
conclusions.

23. To enable a sufficient number and adequate scope of internal model reviews
the internal audit should:

(a) have adequate resources 20 and experienced, qualified personnel (with the
appropriate quantitative and qualitative knowledge) to undertake all
relevant activities;

(b) be adequately equipped and managed in proportion to the nature, size and
degree of complexity of the institution’s business and organisational
structure.

24. In the follow-up process to the internal audit conclusions and findings and to
ensure that the internal audit reviews have a timely and effective impact, the
following are considered to be good practice.

(a) Conclusions, findings and recommendations should be reported to the

audit committee 21 and/or the appropriate 22 management level of the
audited areas.

(b) Where weaknesses are identified, action plans and related measures
should be approved by the audit committee and/or the appropriate24
management level of the audited areas. The internal audit function should
monitor whether the audited areas implement the corrective measures in a
timely manner. The institution should use appropriate information
technology (IT) tools in order to ensure the effectiveness of the monitoring
procedures.

(c) Regular (at least annual) status reports should be prepared and the results
discussed in the appropriate committees to ensure the timely and proper
implementation of follow-up actions. Institutions should submit a summary
of the outcomes of the relevant audit reports, action plans and the status
of findings to the competent authority.

19
Definition provided in paragraph 11.
20
The use of external resources should comply with the institution’s internal audit guidelines. See also
Section 8.3.1 of this chapter.
21
As defined in Article 76(3), sub-paragraph 4, of CRD IV.
22
The appropriate level of management (the management body or senior management) depends on the
corporate governance model and the severity of the results.

ECB guide to internal models – General topics 12

2 Roll-out and permanent partial use

2.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 148, 149, 150, 189

23
Commission Delegated Regulation (EU) No 529/2014 12/03/2014

SSM Regulation 24 15/10/2013 10

Other references

Final Draft RTS on assessment methodology for IRB 21/07/2016 1, 6, 7, 8

25
EBA Consultation Paper 2014/10 26/06/2014 Entire paper and responses received and
published were considered

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become additional relevant legal
reference. Currently the RTS only exists in a final draft version.

2.2 Application of the IRB approach

25. In accordance with Article 148 of the CRR, institutions must implement the IRB
approach for all exposures, unless they have received the permission of the
competent authority to permanently use the standardised approach on some
exposure classes or some types of exposures.

26. The criteria used to define the scope of application and sequential
implementation of the IRB approach should be clearly documented and agreed
with the competent authority. The ECB understands that these criteria include
the following.

(a) Quantitative aspects: as a minimum, the materiality and risk profile of the
exposures and internal thresholds/ratios (for both exposure at default
(EAD) value and risk-weighted exposure amounts (RWEAs)) for the initial
and targeted IRB coverage. For institutions for which the ECB is the
competent authority of the consolidated entity, the initial IRB coverage ratio
(rating systems included in the initial IRB application) is expected to be

23
Commission Delegated Regulation (EU) No 529/2014 of 12 March 2014 supplementing Regulation
(EU) No 575/2013 of the European Parliament and of the Council with regard to regulatory technical
standards for assessing the materiality of extensions and changes of the Internal Ratings Based
Approach and the Advanced Measurement Approach (OJ L 148, 20.5.2014, p.36), referred to in this
guide as “Commission Delegated Regulation (EU) No 529/2014”.
24
Council Regulation (EU) No 1024/2013 of 15 October 2013 conferring specific tasks on the European
Central Bank concerning policies relating to the prudential supervision of credit institutions (OJ L 287,
29.10.2013, p. 63), referred to in this guide as the “SSM Regulation”.
25
Consultation Paper “Draft regulatory technical standards on the sequential implementation of the IRB
Approach and permanent partial use under the Standardised Approach under Articles 148(6), 150(3)
and 152(5) of Regulation (EU) No 575/2013 (CRR)” (EBA/CP/2014/10), referred to in this guide as
“EBA Consultation Paper 2014/10”.

ECB guide to internal models – General topics 13

 above 50% (in terms of both EAD and RWEAs 26) at consolidated level.
This takes into account that institutions must implement the IRB approach
for all exposures included in the roll-out plan in a reasonable time period. 27

(b) Qualitative aspects: importance of the exposure classes or types of

exposures for the institution’s business model, data availability, operational
capacity, staffing, length of experience and homogeneity of treatment of
similar exposures.

27. Institutions which have already received permission to use the IRB approach
are also expected to reach a 50% minimum IRB coverage ratio (in terms of both
EAD and RWEAs) in the implementation of the approved roll-out plan.

28. The IRB approach can be implemented sequentially across the different
exposure classes or types of exposures, across different business units or for
the use of own estimates of losses given default (LGDs) and conversion factors
(CCFs) (roll-out). The ECB considers that the time frame for the initially
approved roll-out plan should generally not exceed five years. 28

29. In the light of the ECB’s need to know the regulatory treatment of all exposures
for its ongoing supervision, institutions are expected to provide the competent
authority with full transparency and regular communications regarding this
treatment. 29 These communications should include information on all
subsidiaries and all portfolios (together with clear exposure assignment criteria).
Institutions should receive explicit permission from the competent authority to
use the standardised approach for exposure classes or types of exposures that
are not in the scope of the IRB approach, following the requirements
established under Article 150 of the CRR.

30. Decisions of the institutions on the application and sequential implementation of

the IRB approach should be triggered by internal criteria (as defined in
paragraph 26) with the main purpose of enhancing risk management and risk
sensitivity. In particular, sequential implementation should not be selectively
used for the purpose of achieving reduced own funds requirements. 30

26
The EAD and RWEAs should be calculated by the institution in accordance with the regulatory
approach envisaged for the calculation of own funds requirements in the initial approval request.
27
For the computation of the IRB ratio, institutions should exclude the following exposures:
• equity exposures as referred to in Articles 133(1) and 147(6) of the CRR;
• exposures covered under Article 150(1) paragraphs (d), (e), (f), (i) and (j) of the CRR;
• exposures classified as “other non-credit obligation assets” and “items representing securitisation
positions” under Article 147(2) paragraphs (f) and (g) of the CRR;
• exposures classified as “other items” under Article 112(q) of the CRR.
The rationale for these exclusions is to limit the scope to exposures for which the CRR envisages the
implementation of a rating system.
28
See Article 148(2) of the CRR.
29
Under Article 10 of the SSM Regulation, the ECB can request all relevant information.
30
See Article 148(3) of the CRR.

ECB guide to internal models – General topics 14

2.3 Governance of the roll-out plan for the IRB approach

31. “All material aspects of the rating and estimation processes shall be approved
by the institution’s management body or a designated committee thereof and
senior management.” 31 As the roll-out plan determines the intended application
of the IRB approach and its sequential implementation, it too should be
approved by the institution's senior management and management body (or a
designated committee thereof).

32. As a corollary and in accordance with Article 189(2)(a) of the CRR, the ECB
understands that the reporting by senior management on the status and
progress of the sequential implementation of the IRB approach should be a
regular agenda item for the management body or designated committee. The
ECB would consider the following to be good practice: that the status of the roll-
out plan is reported frequently (at least annually) and that such reports include
the exact scope of application (exposure class), the planned dates of approval
and/or use, and the “initial, current and planned” 32 exposure amounts and
RWEAs.

33. To ensure compliance with Article 189(1) of the CRR with respect to the roll-out
plan, institutions should have a framework or policy for the governance of their
roll-out plan that includes, as mentioned above, the following:

(a) indication of which internal bodies, including the management body or

other committees and persons, are responsible for approving the roll-out
plan and any changes to it;

(b) the frequency of reporting on the implementation of the roll-out plan to the
management body (or designated committee) and to the competent
authority;

(c) the criteria used for introducing changes to the roll-out plan (see also
Section 2.4, below);

(d) controls to assess compliance with the roll-out plan, for example second
line of defence attestation or internal audit review (see also paragraph 79).

2.4 Changes to the roll-out plan for the IRB approach

34. Under Article 148(2) of the CRR, institutions are required to follow the roll-out
plan approved by the competent authorities. In the event that a change in the
approved roll-out plan is necessary, this change needs to be subject to a
supervisory decision.

31
See Article 189(1) of the CRR.
32
Where “initial” refers to the initial application, “current” to the moment the roll-out plan is updated and
“planned” to the target level.

ECB guide to internal models – General topics 15

 35. Article 7(3) of the Final Draft RTS on assessment methodology for IRB 33
provides a good understanding of the conditions under which competent
authorities may approve any changes to the sequence and time period of the
plan. The ECB intends to assess any application for a change to a roll-out plan
against these conditions, on the basis of the documentation provided by the
institution regarding the rationale for the change, the materiality of the portfolios
affected, and governance arrangements for the change (e.g. which body will
approve it). In addition, when assessing an application for a change to the roll-
out plan, the competent authority will determine, if necessary, whether the
proposed amendment to the time frame for the implementation of the roll-out
plan is acceptable. If institutions have already completed the implementation of
their roll-out plan but would like to extend the use of the IRB approach (for
instance following a merger or acquisition), they should also seek approval 34
from the competent authority.

36. Furthermore, institutions should pay attention to the following.

(a) Resource constraints and re-prioritisation may affect their operational

capability to develop and maintain rating systems. Institutions should seek
to minimise disruptions to the implementation of the roll-out plan as a
result of such factors, taking appropriate mitigation or contingency actions
to demonstrate compliance with the CRR requirements.

(b) General uncertainty caused by potential changes to the IRB regulatory

requirements should not be considered a valid reason for changing the
roll-out plan (or for delaying its implementation). If such regulatory
changes take place and become binding, the institution can then reflect
the impact on its plan by submitting a revised roll-out plan for approval.

2.5 Monitoring of compliance with permanent partial use provisions

37. Since permanent partial use (PPU) requires compliance with certain conditions,
institutions need to ensure on an ongoing basis that exposures under PPU fall
within the categories listed in Article 150(1) of the CRR. 35 To avoid a risk of non-

33
“In the course of their regular assessment of the institution’s compliance with the plan for sequential
implementation of the IRB Approach, which has been subject to permission of the competent
authorities in accordance with Article 148 of Regulation (EU) No 575/2013, competent authorities may
approve any changes to the sequence and time period only where any of the following conditions is
met:
a) there are significant changes in the business environment and in particular changes in strategy,
mergers and acquisitions;
b) there are significant changes in the relevant regulatory requirements;
c) material weaknesses in rating systems have been identified by the competent authority, internal
audit or validation function;
d) there are significant changes to the aspects referred to in paragraph 2, or any of the aspects
referred to in paragraph 2 were not taken into account adequately in the plan for sequential
implementation of the IRB Approach which was approved.”
34
In line with the requirements of Commission Delegated Regulation (EU) No 529/2014 for model
changes and extensions in the scope of that regulation.
35
See also Article 8(4) of the Final Draft RTS on assessment methodology for IRB.

ECB guide to internal models – General topics 16

 compliance with these requirements, the ECB is of the view that institutions
should implement the following.

(a) Measures and triggers for a re-assessment of the suitability for PPU of
PPU-authorised exposure classes or types of exposures. Examples of
measures that could be used include the number of obligors, EAD,
proportion of group EAD, and average risk weight.

(b) A reporting process monitoring the materiality (in terms of both EAD and
RWEAs) of the exposure classes or types of exposures in PPU over time.
The PPU reporting should identify any changes in PPU exposures and
RWEAs over time as well as the exposure classes or types of exposures
that are at risk of no longer fulfilling the PPU conditions.

(c) Processes and guidelines to assess whether further exposure classes or

types of exposure – currently treated under the IRB approach – may
become eligible for PPU (e.g. business in run-off or planned to be
discontinued, excessive maintenance costs induced by the regulatory
requirements vis-à-vis number of obligors).

3 Internal governance

3.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 175, 179, 187, 189, 190

Other references

Final Draft RTS on assessment methodology for IRB 21/07/2016 2, 14, 15, 16, 33

Basel Committee on Banking Supervision (BCBS) 08/07/2015 Principles 1, 2, 3, 4, 5, 6

328 36

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become additional relevant legal
reference. Currently the RTS only exists in a final draft version.

38. The principles on internal governance have been organised along the following
lines:

(a) the materiality of rating systems (Section 3.2);

(b) the management body and senior management:

(i) decision-making responsibilities (Section 3.3);

36
Basel Committee on Banking Supervision “Guidelines: Corporate governance principles for banks”,
referred to in this guide as “Basel Committee on Banking Supervision (BCBS) 328”.

ECB guide to internal models – General topics 17

 (ii) internal reporting (Section 3.4);

(iii) understanding of the rating systems (Section 3.5);

(c) responsibilities of the credit risk control unit (CRCU) (Section 3.6).

3.2 Materiality of rating systems

39. Whether a rating system is material depends on quantitative criteria (such as

the share of total EAD and RWEA covered by the material rating systems) and
qualitative criteria (such as the type, riskiness and strategic importance of the
exposures, the complexity of the rating systems and risk parameters, and the
model risk – in alignment with the model risk management framework). The
more material a rating system is, the higher are the risks resulting from any
potential shortcomings in it. Consequently, materiality should be an essential
consideration in model risk management and the rating systems classified as
material by the institution should be subject to particular scrutiny.

40. In line with the above, institutions should assess and determine the materiality
of their rating systems and communicate it to the competent authority. Material
rating systems should at least include the rating systems covering material
types of exposure. As the classification of the rating systems according to their
materiality may be subject to changes resulting from internal or external factors
(e.g. changes in economic factors, changes in business strategy), institutions
should review their classification on a regular basis.

41. Although the same requirements apply to all rating systems throughout the
model life cycle, additional requirements may apply to material rating systems,
in particular with regard to internal reporting (see Section 3.4 of this chapter)
and internal validation (see Section 4.3 of this chapter). This concept, that
additional requirements may apply for material rating systems, is supported by
the higher potential risk resulting from material rating systems and by Article
189 of the CRR, which also embeds the concept of materiality. 37

3.3 Decision-making responsibilities

42. In accordance with Article 189(1) of the CRR, material aspects of all rating and
estimation processes must be approved by the institution's management body
or a committee designated by it, as well as by senior management. 38 The ECB
considers that Article 14(b) of the Final Draft RTS on assessment methodology
for IRB provides a good understanding of the aspects that should at least be
considered as material, and should therefore be approved at both levels.

37
See Article 189(1) and (3) of the CRR.
38
If the decision-making process takes place in the same forum where management body and senior
management members meet, institutions should ensure that the information is approved beforehand by
senior management and that senior management is responsible for presenting it to the forum.

ECB guide to internal models – General topics 18

 43. The ECB understands that the approval process envisaged by Article 189(1) of
the CRR should include the documentation of the approvals, so that they can
be made available for review at the request of the institution’s internal audit or
the competent authority.

44. The institution should define which policies should be approved at both levels
(management body and senior management) and this should be clarified in
their respective mandates. At a minimum, the following should be considered as
requiring approval at both levels: (i) risk management policies that could have a
material impact on the institution’s rating systems and risk estimates, and (ii)
policies covering the risk of a third-party provider for model-related tasks
ceasing to operate (in relation to IT infrastructure and contingency planning). 39

45. Senior management must provide notice to the management body or a

designated committee thereof of material changes or exceptions from
established policies. 40 Consequently, institutions should be able to demonstrate
which material changes or exceptions from established policies are
communicated to the management body or a committee designated by it, and
how this is done. To ensure compliance with this provision each institution
should, in the ECB’s view, have a policy in place which defines material
changes or exceptions. They may use quantitative and/or qualitative criteria to
do so. The use of expert judgement within the classification process should be
clearly explained and documented.

3.4 Internal reporting

46. To ensure consistent oversight of the functioning of the rating systems, the CRR
requires internal reporting on their performance. 41

47. To comply with this reporting requirement, institutions should determine the
level of detail of the information and data to be presented to senior
management and the management body (or designated committee thereof),
and the frequency of the reporting. In view of proportionality, the level of detail
of the information and data and the frequency of reporting may differ depending
on the recipient and the materiality of the rating systems concerned (see
Section 3.2 of this chapter on the materiality of rating systems). Reports
regarding non-material rating systems may be provided in a more aggregated
form. The procedures encompassing reporting for non-material rating systems
should be clearly defined in the institution’s policies and differences with respect
to the reporting of material rating systems should be clearly identified.

48. The ECB considers that Article 15 of the Final Draft RTS on assessment
methodology for IRB provides a good understanding of the elements to be

39
This view of the ECB coincides with Article 14(b) of the Final Draft RTS on assessment methodology
for IRB.
40
See Article 189(2)(a) of the CRR.
41
See Article 189(2), sub-paragraph 2, and Article 189(3) of the CRR.

ECB guide to internal models – General topics 19

 included as part of the institution’s internal reporting. 42 In particular, these
reports should include information regarding the materiality of each rating
system, its perceived strengths and limitations, and its current status in the light
of validation and/or audit actions.

49. As regards the level of detail of content, reports to the management body are
expected to be more concise than reports to senior management. They should,
however, include the necessary information for sound and appropriate decision-
making. As regards frequency, institutions should provide reports to senior
management as often as, or more frequently than, to the management body (or
designated committee). Risk profiles and the comparison of estimated
probability of default (PD) with realised default rates should be reported to
senior management more frequently than annually (at least for material rating
systems – see also paragraph 41) to enable senior management to ensure, on
an ongoing basis, that the rating systems are operating properly in accordance
with Article 189(2)(c) of the CRR and to avoid risks that could justify supervisory
measures.

50. At least annually, both senior management and the management body should
receive an aggregated overview of the validation results for each rating system
(see also paragraph 69).

3.5 Understanding of the rating systems

51. The management body must possess a general understanding of the rating
systems and senior management must have a good understanding of the rating
systems designs and operations. 43 Institutions should therefore be able to
provide evidence of the processes they use to improve and maintain the
management body and senior management’s understanding of the rating

42
“In assessing the adequacy of the internal reporting as referred to in Article 9(1)(b)(ii), competent
authorities shall, in particular, verify that:
(a) the internal reporting includes all of the following:
(i) a risk profile of the obligors or exposures, by grade;
(ii) the migration across grades;
(iii) an estimation of the relevant parameters per grade;
(iv) a comparison of realised default rates, and, to the extent that own estimates are used, of
realised LGDs and realised conversion factors against expectations;
(v) stress test assumptions and results;
(vi) information about the performance of the rating process, areas needing improvement and the
status of efforts to improve previously identified deficiencies of the rating systems;
(vii) validation reports;
(b) the form and the frequency of internal reporting correspond to the significance and type of the
information and to the level of recipient, taking into account the institution’s organizational structure;
(c) the internal reporting facilitates the senior management’s monitoring of the credit risk in the overall
portfolio of exposures covered by the IRB Approach;
(d) the internal reporting is proportionate to the nature, size, and degree of complexity of the
institution´s business and organizational structure.”
43
See Article 189(1) and (2)(b) of the CRR.

ECB guide to internal models – General topics 20

 systems, including those implemented after receiving permission to use the IRB
approach.

52. The format and content of these processes (for example workshops, seminars
or dedicated training on IRB models) should match the roles and responsibilities
of the management body and senior management, in particular those related to
the model approval process. Especially for the management body, an adequate
balance between collective and individual knowledge should be ensured. In the
case of third-party involvement (see Section 8 of this chapter), the institution
should maintain adequate internal knowledge of the outsourced tasks.

53. While reporting or monitoring can be considered as part of the management

body and senior management’s knowledge-building process, it is expected that
these will not be the only means of ensuring that they have an adequate
understanding of the rating systems.

54. One outcome of an effective internal understanding of the rating system is that
the management body or the designated committee should be able to hold an
objective debate on, and challenge, the rating systems. This applies in
particular to the approval of material changes or the escalation process
contemplated in paragraph 71. In the ECB’s view, it is good practice if the
evidence of such debates is visible in the minutes of management body or
designated committee meetings in which such a challenge is raised, as the
management body or the designated committee should be able to discuss the
outcomes, use, strengths and limitations of the IRB models.

3.6 The credit risk control unit (CRCU)

55. To ensure that the CRCU is independent from the personnel and management
functions responsible for originating and renewing exposures, 44 institutions
should clearly determine which individuals and/or teams make up the credit risk
control function and which personnel and/or units are responsible for originating
and renewing exposures, 45 and why they are independent from one another. In
addition, institutions need to ensure a direct line of reporting of the CRCU to
senior management in accordance with Article 190(1) of the CRR.

56. Institutions should have a clear written mandate for their CRCU which clarifies
its roles and responsibilities in order to ensure that they are aligned to the
responsibilities mentioned in Article 190(2) of the CRR. Institutions should also
determine which units are responsible for performing which subset of the tasks
allocated to the CRCU(s), 46 especially those tasks related to model design and
development and the ongoing monitoring of the rating systems.

44
See Article 190(1) of the CRR.
45
This refers in particular to those persons with authority or direct responsibility for decisions to originate
or renew facility or obligor-level credit lines (for example by underwriting).
46
See Article 190(2) of the CRR.

ECB guide to internal models – General topics 21

 57. The CRCU is responsible for ensuring the satisfactory performance of the rating
systems and their ongoing maintenance. The CRCU may provide the validation
function, when required and in line with paragraph 65, with the necessary input
for the validation of internal estimates. In addition, the CRCU should address
any deficiencies identified by the validation function and conduct the approved
remediation activities as described in paragraph 70.

3.7 Review of estimates

58. In accordance with Article 179(1)(c) of the CRR, risk estimates should be
reviewed 47 when new information comes to light but at least on an annual basis.
The ECB considers it good practice to do this on the basis of:

a) the ongoing monitoring performed by the CRCU;

b) the annual validation of internal estimates performed by the validation

function (as described in Section 4 of this chapter).

47
See the credit risk chapter.

ECB guide to internal models – General topics 22

4 Internal validation

4.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 144 (1)(f)

143, 170, 172,

174, 185, 189

Commission Delegated Regulation (EU) No 529/2014 Annex 1 Part 2 Section 1 and 2

Other references

Final Draft RTS on assessment methodology for IRB 21/07/2016 10 (1)(a) and (b), (2)(a)

11 (1)(b), (2)(c), (4)

12 (a), (c), (d), (e)

13 (b)

14 (b), (d)

15, 34, 38

Basel Committee on Banking Supervision (BCBS) – Newsletter 01/2005

No. 4 48

Basel Committee on Banking Supervision (BCBS) Regulatory 04/2016

49
Consistency Assessment Programme (RCAP)

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become additional relevant legal
reference. Currently the RTS only exists in a final draft version.

59. In the context of rating systems, the term “validation” encompasses a range of
processes and activities that contribute to an assessment of whether ratings
adequately differentiate risk, and whether estimates of risk parameters (such as
PD, LGD and CCF) appropriately characterise the relevant aspects of risk.

60. The main role of the validation function is to perform a consistent and
meaningful assessment of the performance of internal rating and risk estimation
systems. The term “validation function” encompasses the personnel responsible
for performing the validation.

4.2 Validation level and responsibilities

61. In general, internal validation should be performed at all relevant levels. In

particular, institutions should pay attention to the following.

48
Basel Committee Newsletter No. 4 (January 2005), “Update on work of the Accord Implementation
Group related to validation under the Basel II Framework”, referred to in this guide as “Basel
Committee on Banking Supervision (BCBS) - Newsletter No. 4”.
49
Basel Committee on Banking Supervision “Regulatory Consistency Assessment Programme (RCAP) –
Analysis of risk-weighted assets for credit risk in the banking book”, referred to in this guide as “Basel
Committee on Banking Supervision (BCBS) - Regulatory consistency assessment programme
(RCAP)”.

ECB guide to internal models – General topics 23

 (a) If the institution has approval for a rating system on a consolidated basis
only, the validation of that rating system should be performed at least at
consolidated level.

(b) If the institution has approval for a rating system on a consolidated basis
as well as on a sub-consolidated and/or individual basis, the validation of
that rating system should be performed at the consolidated as well as on
the sub-consolidated and/or individual levels. The results of the validation
at the sub-consolidated and/or individual levels should be taken into
account for consistency reasons in the validation performed at
consolidated level.

62. In order to ensure consistency in validation activities across the different levels,
the group validation function can provide support to validation functions at lower
levels (sub-consolidated and/or individual level). However, responsibility for the
validation tasks should be retained at the level at which the rating system is
approved (sub-consolidated and/or individual level). See also Section 8.3.1 on
internal and external outsourcing of internal validation and internal audit tasks
and Section 8.3.5 on in-house knowledge of this chapter.

4.3 Content and frequency of tasks of the validation function

63. To meet the requirements of Article 185 of the CRR with regard to the
assessment of the performance of the internal rating and risk estimates by the
validation function, the ECB understands that institutions should implement the
following.

64. Validation policy

Institutions should have internal validation policies involving proven procedures
and methods which adequately validate the accuracy, robustness and stability
of their estimation of all relevant risk parameters. 50 The procedures and
methods stipulated in the validation policy should be in line with the institution’s
classification of material and non-material rating systems as defined in
paragraphs 39 and 41. Institutions should follow their internal validation policy
when carrying out validations.

65. Validation process and content:

(a) The validation process should assess the performance of the rating
systems by means of qualitative and quantitative methods, in particular
with regard to the ranking of borrowers by creditworthiness (ranking
power) and risk parameter estimation (calibration appropriateness).

(b) To be able to meaningfully and consistently assess the performance of the

rating systems, the content of the validation process should be consistent

50
Validation policies are assumed by Article 185(d) of the CRR, which stipulates that the methods and
data used for quantitative validation must be documented and consistent through time.

ECB guide to internal models – General topics 24

 across rating systems and through time. 51 The analyses and tests
described in this paragraph should be considered as the minimum
activities required to ensure that the assessment is meaningful. However,
the implementation of all of these analyses and tests does not necessarily
mean that the validation requirements according to CRR have been
fulfilled, nor should it prevent the institution from developing additional
tests when deemed relevant.

(c) The content of the validation process should include quantitative analyses,
which in turn should include thresholds. If such thresholds are breached,
further investigation should be initiated and, if necessary, adequate
measures or actions should be triggered.

(d) All analyses and tests should be performed in such a way that the
validation function is in a position to propose an effective and independent
challenge to model development and use. To that end, the institution
should ensure that the validation function has its own access to the
relevant databases. 52 The results of the validation analyses and tests
should be documented (validation report) and verifiable by third-party
experts (e.g. the internal audit and the competent authority). This also
includes the preparation of the validation data.

(e) It is not expected that institutions develop a uniform validation process, as

the relevant tests and their frequency may differ from one rating system to
another (e.g. corporate vs. retail rating systems or material vs. non-
material rating systems).

(f) To ensure that the systems in place to validate the accuracy and
consistency of rating systems are sufficiently robust, 53 institutions should
ensure that any statistical tests or confidence intervals that they use are
appropriate from a methodological point of view and sufficiently
conservative.

(g) A meaningful validation of the rating systems requires (as also anticipated
by Article 185(b) of the CRR with respect to certain elements) not only an
initial validation but also assessment on a regular basis. In line with Article
185(b) of the CRR, these assessments should be carried out annually. 54
Therefore, the validation analyses listed hereinafter should be carried out
annually, unless otherwise advised below. As mentioned in paragraph 14
the term “initial validation” in the guide refers to the validation of new
models as well as the validation of material changes and extensions to

51
See Article 185(a) of the CRR for “consistently and meaningfully” and Article 185(d) for “consistent
through time”.
52
When an institution is using rating systems developed from pooled data, the validation function should
have access to all relevant internal data of that institution. In addition, any third party involved may
assist the institution in its validation activities by performing those validation tasks which require access
to pooled data (see also paragraphs 129 and 132(e)).
53
See Article 185(a) of the CRR.
54
See Article 174(d) of the CRR.

ECB guide to internal models – General topics 25

 approved models. In the case of material changes and extensions the
validation should be conducted on those aspects directly or indirectly
affected by the change. 55

(h) In addition, and to ensure a robust validation, the analyses should be

performed by the validation function without considering input from the
CRCU 56 unless otherwise specified below.

(i) Back-testing 57

For the annual validation of non-material rating systems, the

validation function can also take into account the back-testing
analyses performed by the CRCU.

(ii) Discriminatory power

(ii.a) Analyses of discriminatory power for PD models should be

designed to ensure that the ranking of obligors/facilities
resulting from the rating methodology appropriately separates
riskier and less risky obligors/facilities. 58 Similarly, analyses of
discriminatory power for LGD (respective CCF) models should
be designed to ensure that the LGD (respective CCF) model is
able to discriminate between facilities with high values of LGD
(respective CCF) and those with low values of LGD (respective
CCF).

(ii.b) These analyses should be performed at least at the level of the

overall model and, when relevant, at the level of individual risk
factors and other possible segments including, for example,
scorecards and modules.

(ii.c) For the annual validation of non-material rating systems, the

validation function can also take into account the analyses of
discriminatory power performed by the CRCU.

(iii) Analyses of representativeness

(iii.a) The data used to build the model must be representative of the
actual obligors or exposures (Article 174(c) of the CRR). This
should include the following checks.

• Ensuring that the range of application of the model is in line

with the approved one, in accordance with Article 143(3) of
the CRR. Obligor characteristics should be compared for

55
See also Article 11(4) of the Final Draft RTS on assessment methodology for IRB.
56
Or credit risk control function for institutions using option (c) of paragraph 15.
57
Comparison of risk estimates with realised default rates, in accordance with Article 185(b) of the CRR.
58
Further elaborated in Article 170(1), sub-paragraphs (d), (e) and (f), and 170(3)(c) of the CRR.

ECB guide to internal models – General topics 26

 PD models, and facility types and characteristics should be
compared for LGD and CCF models.

• Monitoring the modifications undertaken in the definition of

default, with the aim of identifying any changes that would
affect the representativeness of the dataset with respect to
the obligors or facilities within the range of application of
the model.

• Analysing lending standards or workout procedures,

external market and economic conditions, and other
relevant characteristics surrounding the model
development process.

(iii.b) Where an institution uses data that are pooled across

institutions, the analyses should also cover the requirements of
Article 179(2) paragraphs (a) and (b) of the CRR.

(iii.c) For the annual validation of rating systems, the validation

function can also take into account the analyses of
representativeness performed by the CRCU.

(iv) Analyses of overrides

(iv.a) Overrides should not only be monitored but also assessed as

part of the validation process (Article 172(3) of the CRR). See
also Section 6.6.2 of this chapter.

(iv.b) Analyses of overrides should be conducted on an annual basis

(and not at initial validation). The validation function can also
take into account the analyses of overrides performed by the
CRCU.

(v) Stability analyses of the internal ratings and risk parameters over
time 59

(v.a) Examples for analysing the stability of internal ratings and risk
parameters over a specific observation period for PD
estimates can be the following:

• obligor/facility migrations;

• stability of the migration matrix;

• concentration in rating grades.

(v.b) Institutions should justify excessive or unexpected variability.

59
This is required by Article 174(d) of the CRR.

ECB guide to internal models – General topics 27

 (v.c) For the annual validation of non-material rating systems, the
validation function can also take into account the stability
analyses performed by the CRCU.

(vi) Analyses of model design stability 60

(vi.a) The institution should analyse:

• the differences between the original weights of the risk

drivers (derived from the development sample) and the
weights estimated from a different sample (longer or more
recent historical sample);

• the continued homogeneity61 of rating grades or pools

used as a basis for the estimation of risk parameters.

(vi.b) Institutions should justify excessive or unexpected variability.

(vi.c) These analyses should be performed at initial validation and at

an adequate frequency thereafter, but at least every three
years. The validation function can also take into account the
analyses performed by the CRCU for non-material rating
systems, except at initial validation.

(vii) Evaluation of input data 62

(vii.a) This should ensure all of the following:

• that the data treatment process is reliable and well-

founded;

• that the necessary information is available and up-to-date

for the majority of the application portfolio’s 63 obligors and
facilities by tracking the age of model input data, especially
in the case of financial statements;

• that all defaults that occurred in the institution within the

scope of application of the model are correctly identified
and fully documented and registered in the appropriate and
intended IT systems;

• that the number and reasons for technical past-due

situations are tracked.

60
This is required by Article 174(d) of the CRR on “monitoring of model performance and stability” and
“review of model specifications”.
61
Homogeneity as defined in Article 38 of the Final Draft RTS on assessment methodology for IRB. This
understanding is also supported by Article 34(d) of the Final Draft RTS on assessment methodology for
IRB.
62
See Article 174(b) of the CRR.
63
“Application portfolio” means the actual portfolio of exposures within the range of application of the PD
or LGD model at the time of estimation of a risk parameter.

ECB guide to internal models – General topics 28

 (vii.b) For the annual validation of the rating systems, the validation
function can also take into account the evaluation of the input
data performed by the CRCU.

(viii) Benchmarking analyses

(viii.a) The institution should carry out comparisons with up-to-date

data from representative and comparable external data
sources, in particular with regard to low-default portfolios
(Article 185(c) of the CRR). The institution should provide
sufficient evidence in the event that no usable external data
are available.

(viii.b) Benchmarking analyses should be performed at initial

validation and at an adequate frequency thereafter, but at least
every three years. For benchmarking analyses at initial
validation the validation function can take into account the
benchmarking analyses performed by the CRCU.

(ix) Data cleansing analyses 64

(ix.a) These analyses refer to the exclusion of observations (and the

reasons behind this) from the risk database for the
construction of the reference dataset (RDS) for the modelling.

(ix.b) These analyses should be performed at least at initial

validation.

(x) Review of the model specification

(x.a) This analysis should include challenging the model design,

assumptions and methodology. A stepwise initial validation
process involving interaction with the model development at
each step of the development phase may not be sufficient to
perform this challenge effectively. Regardless of the validation
approach followed, the validation unit should provide an overall
conclusion on the model to ensure that individual model
strengths and weaknesses are evaluated on an overall basis.

(x.b) This analysis should be performed at least at initial validation.

(xi) Quality assurance of the computer codes used

(xi.a) This should include at least the following:

• that the implementation of the rating system in the relevant

IT system is compliant with and reproduces exactly the
documented model under review; 65

64
The rationale for these analyses is set out in Article 174(c) of the CRR.

ECB guide to internal models – General topics 29

 • that the descriptions of the data sources, variables and risk
drivers used for development purposes are properly
documented.

(xi.b) These analyses should be performed at least at initial

validation.

(xii) Additional qualitative analyses 66

These analyses should include, among other things, the following:

• qualitative assessments of assumptions and expert-based

estimates and of the integrity of the rating assignment
process;

• assessment of the use of the models and their correct

application in practice;

• assessment of legal or macroeconomic changes that may

impact the risk parameters;

• assessment of downturn phases and the correct application

of margins of conservatism.

Not all of the above-mentioned qualitative analyses need to be

performed annually, but institutions should have a clear policy in
place defining the appropriate frequency of each assessment.

66. The duration from the start (reference date of data) to the end (approval of the
validation results) of the yearly validation should not be more than one year.
Deviations from this requirement should be clearly justified and documented by
the institution and also reported to senior management.

67. Quantitative thresholds (see paragraph 65(c)) should be set up for at least the
following tests:

(a) back-testing;

(b) discriminatory power;

(c) analyses of overrides;

(d) stability analyses of the internal ratings and risk parameters over time.

68. In particular for tests where no thresholds are applied, a consistent qualitative
assessment of the results should be performed and documented. In the event
of a negative qualitative assessment, adequate measures or actions should be
triggered.

65
See also Article 144(1) of the CRR, which requires that the institution’s systems for the management
and rating of credit risk exposures be sound and implemented with integrity.
66
This is contemplated in Article 174(e) of the CRR.

ECB guide to internal models – General topics 30

4.4 Reporting and follow-up

69. Institutions should ensure that senior management and the management body
(or the committee designated by it) are informed about the conclusions and
recommendations of the validation results as set out in the rating systems’
validation reports, and in particular about any exceeded thresholds and
deficiencies identified, as this is required by Article 189(2)2 of the CRR. Such
information should be provided, where appropriate, 67 in a summary document
(or documents) to ensure that a sufficient level of information is provided to
senior management and to the management body and to enable them to
assess the performance of the rating systems. This summary document (or
documents) should present an aggregated view and comparison of the results
for all the rating systems.

70. Institutions should be able to demonstrate that, on the basis of the validation
results and recommendations, measures are initiated to remedy the identified
deficiencies of the rating systems (e.g. model change, recalibration) as
contemplated by Article 189(2), sub-paragraph 2, of the CRR.

71. Article 189(2)(c) of the CRR requires senior management to ensure, on an

ongoing basis, that the rating systems are operating properly. Article 189(2)2
requires senior management to be informed of the status of the measures to
remedy any previously identified deficiencies. From this, it can be inferred that
institutions should have a process in place (e.g. through a committee or another
comparable body) to decide on such measures, who will be responsible for
them, and the timelines for their implementation, on the basis of the validation
results and recommendations. To ensure sufficient senior management
engagement as contemplated by Article 189(2) of the CRR, the ECB is of the
view that at least for material rating systems the direct involvement of senior
management should be envisaged (e.g. by senior management chairing the
committee). The process should involve persons with the appropriate level of
seniority and responsibility from both the CRCU and the validation function. If
affected, business units should also be involved. An escalation process up to
management body level should be in place in the event of conflicts between the
validation function, the CRCU and/or business units.

72. Institutions should have adequate processes in place for tracking the status of
the measures adopted to remedy deficiencies. 68

73. Institutions should always apply to/notify the competent authority in the event of
changes to their validation methodology and/or processes in accordance with
Annex 1 Part 2 Section 1 (material model change) or Section 2 (ex ante
notification) of Commission Delegated Regulation (EU) No 529/2014.

67
Such a summary document (or documents) is (are) expected to be appropriate, for example in the case
of institutions with a significant number of rating systems.
68
See Article 189(2), sub-paragraph 2, of the CRR.

ECB guide to internal models – General topics 31

5 Internal audit

5.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 145, 175, 191

Commission Delegated Regulation (EU) No 529/2014 8

Other references

Final Draft RTS on assessment methodology for IRB 21/07/2016 17

Basel Committee on Banking Supervision (BCBS) 328 08/07/2015 43, 139, 140, 141, 142, 143

EBA Guidelines on internal governance 26/09/2017 V. 22. 201, 202, 204, 205, 206,
207

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become additional relevant legal
reference. Currently the RTS only exists in a final draft version.

5.2 Scope and frequency of the review of the rating systems

74. In accordance with the existing regulatory requirements under Article 191 of the
CRR, the internal audit or another comparable independent auditing unit
(hereinafter internal audit) must review the institution's rating systems and its
operations at least annually. The areas for review must include adherence to all
applicable requirements.

75. The ECB considers that an institution fulfils the above-mentioned requirements
of Article 191 of the CRR if the internal audit carries out, annually and on the
basis of up-to-date information, a general risk assessment of all aspects of the
rating systems for the purpose of drawing up the appropriate internal audit work
plan, as elaborated in paragraphs 76 to 78, and executes this plan.

76. Depending on the outcome of the general risk assessment, the intensity and
frequency of the audit assignments 69 may differ between specific areas.

(a) An area showing signs of increased/high risk in the general risk

assessment is subject to a thorough audit assignment (“deep dive”).
Reasons for the increased risk might include, but are not limited to, new
processes, new regulatory requirements, new types of exposures in the
range of application of a rating system, material model changes, findings
by the competent authority or by any other function, open issues and
areas with high risk identified in previous risk assessments. The audit
assignments should be included in the annual audit work plan established
on the basis of the general risk assessment.

69
An audit assignment can refer to a separate audit, a range of audits or themes of audits.

ECB guide to internal models – General topics 32

 (b) Areas not showing any sign of an increased/high risk are subject to a deep
dive mission at least every three years in order to provide a thorough
update of the internal audit opinion on them. 70

77. For the purpose of the general risk assessment, the internal audit should
develop its own opinion on the areas of rating systems to be reviewed but can
take into consideration the analysis performed by the internal validation function
where appropriate. This general risk assessment should include at least the
opinion of the internal audit unit on the following aspects.

(a) The development 71 and performance 72 of the rating systems.

(b) The use of the models. The assessment of model use shows that the
rating systems play an essential role in the most significant areas of the
institution’s risk management, decision-making, credit approval, internal
capital allocation and corporate governance functions.

(c) The process for the materiality classification, 73 the impact assessment and
the compliance with regulatory requirements of all changes to the rating
systems as well as their consequent implementation. The internal audit is
informed of all changes to the rating systems.

(d) The quality of the data used for the quantification of risk parameters.

(e) The integrity of the rating assignment process.

(f) The validation function, in particular with regard to its independence from
the CRCU as described in paragraphs 17 and 18 of this chapter, as well as
the scope and suitability of the tasks performed and outputs obtained.

(g) The process for calculating own funds requirements.

78. The procedures and results of the general risk assessment and prioritisation,
the annual work plan, the guidelines, and the subsequent production of the
internal audit reports should be properly documented and approved by the
management body. The auditing techniques used and applied by the institution
should be documented to ensure that assessments are consistent. 74

79. To avoid delays in the procedures related to an initial IRB application or an

application to extend the IRB approach to an exposure class or a type of
exposure that is currently treated using the standardised approach, it is
beneficial if the internal audit provides the competent authority with an

70
These deep dives may be either thematic, covering one area for several rating systems (e.g. rating
assignment process, data quality management, definition of default), or targeted on a specific rating
system.
71
The scope should include the initial validation tasks described in the internal validation principles.
72
The scope should include the annual/regular basic tasks described in the internal validation principles.
73
The internal audit function may be involved in the classification as an independent function confirming
the assessment of materiality, as set out in Section 7.4 of this chapter.
74
See Article 175(1) and (2) of the CRR.

ECB guide to internal models – General topics 33

 independent assessment of the compliance of the initial IRB application or
extension package with all applicable requirements. This applies in particular to
compliance with the experience test requirements of Article 145 of the CRR.
The benefit of carrying out an internal review of this nature is that shortcomings
can be addressed by the institution before submitting the application, and the
completeness of the initial IRB application or extension package can be
ensured.

80. For extensions and changes to the IRB approach, 75 institutions must submit,
among other things, reports of their independent review or validation. 76 In the
case of material changes in the validation methodology and/or validation
processes or process-related aspects of material changes, including changes in
the definition of default, the review of those aspects should be subject to an
independent assessment by the internal audit function before they are
submitted to the competent authority for approval.

75
See Section 7.3 “Notification” of this chapter on suggested templates to be used.
76
Article 8(1)(e) of Commission Delegated Regulation (EU) No 529/2014.

ECB guide to internal models – General topics 34

6 Model use

6.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 143 (3)

144 (1)(b) and (h)

145

148 (1)

150

171 (2)

172 (3)

173 (1)(b), 2

174 (e)

179 (1)

189 (3)

Other references

Final Draft RTS on assessment methodology for IRB 21/07/2016 18, 19, 20, 21

22 (a) and (b)

23 (1)(a)

24 2(a), (b), (c), (d) and (e)

25 2(a) and (b)

39 (c)

42

Basel Committee on Banking Supervision (BCBS) - 18/09/2006 Principles: 1, 2, 3, 4

77
Newsletter No. 9

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become additional relevant legal
reference. Currently the RTS only exists in a final draft version.

6.2 Use test requirement

81. The ECB acknowledges that the degree of use of internal ratings and default
and loss estimates in the institution’s risk management and decision-making
process, and in its credit approval, internal capital allocation and corporate
governance functions, is more extensive for PD/internal ratings than for
LGD/loss estimates and CCFs.

82. Moreover, the IRB risk parameters can be used in an adjusted form or indirectly
through relevant risk measures/indicators stemming from the rating systems,

77
Basel Committee Newsletter No. 9 (September 2006), “The IRB Use Test: Background and
Implementation”, referred to in this guide as “Basel Committee on Banking Supervision (BCBS) -
Newsletter No. 9”.

ECB guide to internal models – General topics 35

 provided that differences from the regulatory parameters are fully justified and
properly documented. 78 For example, institutions may use adjusted or
transformed IRB parameters by removing certain constraints (e.g. downturn
effect, conservative add-on, floor) or adjusting the time horizon.

83. The use of IRB risk parameters and their inclusion in internal policies and
procedures enables institutions to continuously improve their accuracy and
reliability by receiving feedback from model users. The conditions for an
effective and beneficial feedback loop include a good understanding of the
model, its assumptions and constraints and an adequate level of interaction
between users, the CRCU and the validation function.

6.3 Risk management, credit approval and decision-making process

84. As set out in Article 144(1)(b) of the CRR, internal ratings and default and loss
estimates produced by the rating systems must play an essential role in the risk
management and decision-making process and in credit approval. To ensure
that they are able to play this essential role, institutions should use internal
ratings and default and loss estimates in the approval, restructuring and
renewal of credit facilities, and in lending policies and the monitoring process for
obligors and exposures. This should be formally included in the institution’s
internal policies.

(a) Credit approval

The ECB considers that institutions comply with the requirement to use the
internal ratings and default and loss estimates in the approval,
restructuring and renewal of credit facilities if they establish the following.

(i) Institutions incorporate the internal rating systems in the overall credit
granting, restructuring and renewal process. Related policies are
calibrated on the basis of rating classes or groups of rating classes or
risk parameters.

(ii) For the incorporation of internal rating systems to be beneficial, staff

involved in the credit granting, restructuring and renewal process
need to have sufficient knowledge of the rating systems, including
their strengths and limitations. This encompasses the inclusion of
rating system users’ feedback in model development and
maintenance as well as having appropriate training programmes in
place.

(iii) The assignment or updating of ratings is a prerequisite for the

assessment underlying the granting and reviewing of credit lines.

The ECB considers that institutions comply with the requirement to use the
internal ratings and default and loss estimates in their lending policies,

78
See Article 179(1), sub-paragraph 2, of the CRR.

ECB guide to internal models – General topics 36

 including exposure limits and mitigation techniques, if those policies have
the following features.

(iv) Lending policies include specific references to the use of internal

rating systems and related parameters (for instance, use of a grid of
parameters in the decision-making process). These parameters serve
as an indicator of riskiness (e.g. in terms of expected loss (EL)). They
may be differentiated by institutions’ portfolios (e.g. retail/non-retail)
and by facility type.

(b) Risk management – monitoring process for obligors and exposures

The ECB considers that institutions comply with the requirement to use
their internal ratings and default and loss estimates in respect of the
monitoring process for obligors and exposures if the following is
established.

Individuals in charge of the monitoring process are promptly provided with

adequate information on the development of counterparties’ credit risk as
expressed by ratings, so that the relevant information can be easily
incorporated in the process and trigger appropriate actions.

85. In addition, institutions should consider taking into account the internal ratings
and default and loss estimates produced by the rating systems in the five areas
shown below. If an institution decides to take into account the internal ratings
and default and loss estimates in any of these five areas, this should be
formally included in its internal policies. 79 If an institution is not using internal
ratings or risk parameters in one or several of these areas, it should properly
document and justify the rationale for that to ensure that discrepancies are
explained in a sound and understandable manner.

(a) Pricing of transactions

(i) Internal ratings and risk estimates can be considered in the pricing of
transactions, in particular for non-retail exposures.

(ii) The methodology underlying pricing can be documented and the use
of risk-adjusted performance indicators (e.g. return on risk-adjusted
capital – RORAC) or adjusted IRB parameters is considered as good
practice for pricing estimation.

(b) Early warning systems

(i) Early warning systems can be applied to all the institution’s

exposures and can be tailored to its specific sub-portfolios (with at
least a distinction between retail and non-retail exposures).

79
See Article 19(2) of Final Draft RTS on assessment methodology for IRB.

ECB guide to internal models – General topics 37

 (ii) PD/ratings dynamics (i.e. downgrades) and other indicators linked to
other risk measures (e.g. EL, loan-to-value, overdraft) can be taken
into account in the institution’s early warning system – at least for the
most relevant portfolios. Whenever an anomaly is detected,
appropriate risk management decisions can be triggered. This
process can be adjusted depending on the persistency and intensity
of the warning. It can also be designed according to other variables
such as exposure size or facility type.

(c) Collection and recovery policies and processes

(i) Regarding the collection process, institutions can have risk

management procedures in place which are triggered in advance of
the exposure’s default (e.g. early collection calls) and are based –
among other indicators – on their internal ratings or risk drivers.

(ii) Regarding the recovery process, institutions can have in place rules,
strategies or procedures that take into consideration, inter alia, their
LGD/expected loss best estimate (ELBE) values, as well as their set-
aside provisions.

(d) Credit risk adjustments

(i) The credit risk adjustments methodology for both performing

exposures and exposures in default (or share of exposures in default)
can be aligned to the calculation of own funds requirements, although
some adjustments might be needed to comply with accounting
standards (for example, International Financial Reporting Standards 9
(IFRS 9)). The institution should justify and document any significant
deviations to ensure that the rationale for discrepancies is sound and
understandable.

(e) Allocation or delegation of competence for the approval process

(i) Along with the materiality of credit lines, the delegation of

competences for credit approval can take risk estimates into account
through one or several IRB parameters or through EL (for example,
an increase of some EL-driven measures above a pre-defined
threshold should typically trigger an escalation process).

(ii) The allocation and delegation process can include the criterion of
proportionality, taking into account portfolio risk and facility types.

6.4 Internal capital assessment and allocation

86. Internal ratings and the default and loss estimates produced by the rating
systems play an important role in the assessment, calculation and allocation of

ECB guide to internal models – General topics 38

 institutions’ internal capital 80 under the framework of the internal capital
adequacy assessment process (RWEAs can also be used as an additional
driver). This role should be reflected within the institutions’ internal policies and
procedures on internal capital assessment and allocation.

6.5 Corporate governance functions

87. Internal ratings and the default and loss estimates produced by the rating
systems play an important role in the institution’s corporate governance
functions. 81 To ensure that they are able to play this essential role, institutions
should use internal ratings and default and loss estimates, in particular in their
internal reporting and portfolio credit risk monitoring procedures. This role
should be reflected within the institutions’ internal policies.

88. The ECB considers that institutions comply with the requirement to use the
internal ratings and default and loss estimates in their corporate governance
functions if they establish the following.

(a) Institutions’ internal reporting 82

(i) Institutions have a structured reporting system on risk measured by

their IRB risk parameters. This reporting framework contains
information about frequency, recipients and contents (if possible
broken down by segment, portfolio and product).

(ii) The reports are accompanied by comments and explanations on the

numbers provided as well as by qualitative assessments, to enable
recipients to fully understand the potential underlying risks.

(b) Portfolio credit risk monitoring

(i) In its ongoing model monitoring function, the CRCU performs

descriptive analyses of portfolio riskiness (distribution of exposures
among rating classes, average risk estimates and their realisations,
ELs). These analyses are progressively refined to include the
analytical insights derived from the information on ratings. 83

80
See Article 144(1)(b) of the CRR.
81
See Article 144(1)(b) of the CRR.
82
See also Section 3.4 of this chapter for further details.
83
For example: highlighting, in aggregate terms, the volume of credits whose rating has worsened by
more than one class (“double downgrade”); rating stability; the speed and frequency of rating
modifications; the incidence of defaults; the relationship between “upgrade” and “downgrade” at
portfolio level in a given period of time; and changes in rating by line of business, market segment or
type of credit line.

ECB guide to internal models – General topics 39

 (ii) Reporting to senior management provides a concise but complete
overview of the relevant variables 84 so that the evolution of credit risk
can be monitored at portfolio level.

6.6 Assignment of exposures to grades or pools

6.6.1 Non-rated exposures and outdated ratings

89. Article 144(1)(h) of the CRR requires institutions to assign and continue with
assigning each exposure in the range of application of a rating system to a
rating grade or pool of this rating system. Furthermore, and in accordance with
Article 173(1)(b) of the CRR, institutions must review those assignments at
least annually. Nevertheless, the ECB observes that institutions’ portfolios
occasionally show a certain proportion of non-rated exposures and/or outdated
ratings. The ECB considers that this should be properly investigated, justified,
documented and monitored.

(a) Root causes

(i) Non-rated exposures are temporary exceptions to the “ordinary”

rating assignment process and should therefore be investigated,
documented and justified in detail.

(ii) Outdated ratings include both ratings that have not been updated
within the 12-month period following the last rating date 85 and ratings
based on outdated information. 86

(b) Materiality

(i) Non-rated exposures and outdated ratings present a risk for

institutions of not being compliant with the CRR requirements related
to rating systems. To mitigate that risk, institutions should implement
specific policies and procedures to identify these non-rated
exposures and outdated ratings and monitor their materiality (in terms
of number, EAD and RWEAs). A formal check should be carried out
on these issues, at least annually, and reported to senior
management. These items should also be reviewed periodically by
the internal validation function.

90. With regard to ratings not updated within the 12-month period following the
most recent rating date, a transition period during which the current rating is
carried forward can be considered provided all the following conditions are met:

84
At least those defined in Article 189(3) of the CRR.
85
See related requirements in Article 173(1)(b) and 173(2) of the CRR.
86
For instance, when financial information is used in the rating assignment process, it should be taken
from financial statements dating back no more than two years.

ECB guide to internal models – General topics 40

 (a) the transition period begins 12 months after the date the current rating was
assigned or when new material information on the obligor or exposure
became available (if earlier);

(b) the transition period does not exceed three months;

(c) the transition period applies:

(i) to wholesale exposures whose rating assignment relies on external

information (e.g. financial statements), which may be
available/published not exactly 12 months apart;

(ii) in the presence of exceptional internal impediments that affect the

timely review of the rating assignment;

(d) adequate monitoring and reporting policies, together with escalation

procedures, are implemented to ensure a rapid return to compliance;
restrictions to credit granting or to delegation of powers for credit approval
are also envisaged for those counterparties whose rating falls within the
transition period;

(e) conservative treatments apply after three months, as detailed in paragraph

91.

91. All exposures within the range of application 87 of an IRB rating system must
eventually be rated 88 and are not expected to be treated under the standardised
approach, unless they have received the permission of the competent authority
to be permanently treated under that approach in accordance with Article 150 of
the CRR. 89 Non-rated exposures and outdated ratings present a risk of potential
underestimation of own funds requirements. To mitigate that risk, institutions
should adhere to the following guidance.

(a) Institutions should have internal policies in place establishing a process to

monitor and manage non-rated exposures and outdated ratings
prudentially. In particular, they should be able to prove that their
procedures allow for a conservative measure of risk, such as time-
dependent downgrading for outdated ratings and at least the application of
the worst-performing rating grade for unrated exposures. 90

(b) The calibration of the prudential treatment should be validated at least

annually (evidence of conservativeness).

87
The range of application refers to Article 143(3) of the CRR and thus to all exposures of the relevant
type of exposure for which that rating system has been developed and approved.
88
See Article 144(1)(h) of the CRR.
89
See Article 148(1) of the CRR.
90
To avoid any distortion of risk estimates, institutions should ensure that these ex-post conservative
adjustments are not included in the calibration dataset.

ECB guide to internal models – General topics 41

6.6.2 Analysis of overrides

92. For grade and pool assignments institutions must document the situations in
which human judgement may override the inputs or outputs of the rating system
and the personnel responsible for approving these overrides.

93. For the purposes of this subsection, instances of overruling of internal policies
and/or procedures are not considered as overrides. If institutions overrule
internal policies and/or procedures, they should, as in the case of overrides,
document these instances and report them to the relevant functions (e.g.
CRCU, internal validation function). The functions concerned should assess the
performance of the exposures affected by the overruling and its potential impact
on the rating systems.

94. As a general principle, the rating of retail exposures is less likely to be affected
by an override process, given the high degree of standardisation of information
processing – including in qualitative terms – and the small margins of discretion
in the evaluation.

95. “For grade and pool assignments institutions shall document those situations in
which human judgement may override the inputs or outputs of the assignment
process.” 91 Accordingly, institutions should have documented policies that
incorporate the following principles:

(a) the policies include clear and exhaustive justifications for triggering the
override process on the basis of pertinent and significant information for an
accurate assessment of the counterparty’s creditworthiness;

(b) the policies define the maximum extent of overrides (in terms of, for
example, maximum number of notches up 92 and maximum share of
overridden exposures), also considering model/portfolio specificities;
institutions should aim to be more restrictive with positive overrides than
with negative ones.

96. Institutions must document each override. 93 To this end, they should retain the
quantitative and qualitative information concerning each phase of the rating
process. In particular, all decisions taken throughout the process – including
interim ratings – should be recorded, as should the reasons for any override.
The information should be proportionate to the severity and extent of the
override.

97. “Institutions shall analyse the performance of those exposures whose

assignments have been overridden.” 94 To comply with this requirement, the
institutions should carry out the following procedures.

91
See Article 172(3) of the CRR.
92
Where “up” refers to the direction of non-conservative overwrites.
93
See Article 172(3) of the CRR.
94
See Article 172(3) of the CRR.

ECB guide to internal models – General topics 42

 (a) Performance analysis of the rating systems, to assess whether the
judgemental adjustments improve their discriminatory power. The analysis
may be extended to all of the underlying components (modules) of the
rating system. It is deemed particularly useful to measure the difference in
terms of performance and impact of the “pre-override” and “post-override”
stages.

(b) Other analyses, including the assessment of the distribution of overrides

by override root cause, i.e. if there is a situation that systematically triggers
an adjustment.

98. To mitigate the risks identified through the aforementioned analyses (paragraph
97), institutions should identify specific criteria for assessing whether or not the
number of and justifications for overrides indicate significant weaknesses in the
rating system and whether this is a reason to take ad hoc actions (e.g. a model
change). In general, situations where there are too many overrides could be a
strong indicator of weaknesses in the model (i.e. systematic and material
adjustments can be the consequence of a misspecification of the model).

7 Management of changes to the IRB approach

7.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 143 (3) and (4)

Commission Delegated Regulation (EU) No 529/2014 12/03/2014 2, 3, 4, 5, 8, Annex 1

Other references

Final Draft RTS on assessment methodology for IRB 21/07/2016 86, 87

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become additional relevant legal
reference. Currently the RTS only exists in a final draft version.

99. Changes to a rating system’s range of application or to a rating system itself are
subject to approval by the competent authorities if assessed as material, or to
ex ante or ex post notification if non-material.

100. Commission Delegated Regulation (EU) No 529/2014 requires institutions to

assess and classify the materiality of changes to rating systems. To comply with
these requirements in a comprehensive and consistent way, institutions should
establish a policy related to changes to the IRB approach (“change policy”). In
line with the above, this policy should include, in particular, detailed criteria to
ensure that the classification of changes is consistent and that any arbitrage in
that regard is avoided. Institutions are encouraged to share their policy with the
competent authority and inform the latter about any implemented modifications

ECB guide to internal models – General topics 43

 to it, in order for both sides to have a common understanding of the
classification process.

7.2 Content of the change policy

101. The change policy should include provisions relating to the operationalisation of
the requirements of Commission Delegated Regulation (EU) No 529/2014 with
respect to the materiality assessment, classification, impact assessment,
notification and documentation of changes and extensions. 95 To that end, it
should include, in particular, the following. 96

(a) Responsibilities, reporting lines and procedures for the internal approval of
changes, taking into account the institution’s organisational characteristics.
This policy should define at least the unit(s) responsible for the
assessment and classification of changes or extensions, as well as the
function/committee responsible for confirming and countersigning the
classification.

(b) Definitions, methods and, where applicable, metrics and significance

levels for the impact assessment, threshold calculation and classification
of changes; in particular, the quantitative/qualitative criteria referred to in
Commission Delegated Regulation (EU) No 529/2014. In addition, as
required by Annex 1, Part 2, Section 1 of the Regulation and to ensure
consistency, the following should be observed:

(i) the institution should specify metrics and significance levels that
define the significance/materiality of changes in the distribution
across rating grades produced by changes to the rating methodology
(paragraph 2(d)(ii) of Annex 1, Part 2, Section 1); these metrics and
significance levels should be complementary to those of Articles 4(2)
and (3) and Article 5(2) of the same Regulation;

(ii) the institution should specify metrics and significance levels that
define the significance/materiality of rating migrations produced by
changes in the rating system's assumptions on the impact of
economic conditions (paragraph 2(c));

(iii) the institution should specify metrics and significance levels that
define the significance/materiality of changes in the rank ordering of
clients/exposures (paragraph 2(d)(i));

(iv) in its change policy, the institution should define which changes
constitute a change in the fundamental methodology for estimating
PDs, LGDs (including best estimate of EL) and CCFs and are

95
Articles 2, 3, 4, 5 and 8 of Commission Delegated Regulation (EU) No 529/2014.
96
Article 87 of the Final Draft RTS on assessment methodology for IRB provides for a comparable set of
policy elements.

ECB guide to internal models – General topics 44

 considered as material in the sense of paragraph 2(f) (as opposed to
the changes referred to in paragraph 2(h) of Annex 1, Part 2, Section
2);

(v) the institution’s change policy should include a definition of changes

in the validation methodology and/or validation processes which lead
to changes in its judgement of the accuracy and consistency of the
estimation of the relevant risk parameters, the rating processes or the
performance of the rating systems (paragraph 4 of Section 1).

(c) Procedures to identify and monitor changes, and to notify and apply to the
competent authorities for permission to make such changes. In particular,
institutions should establish an end-to-end process from identification to
notification/application and describe how they perform the activities at
each step.

(d) Procedures for the implementation of changes, including their

documentation; in particular, the re-rating process should be defined (if no
other document is already in place).

7.3 Notification

102. To facilitate the process for submitting the documentation package defined
under Article 8 of Commission Delegated Regulation (EU) No 529/2014, on the
banking supervision website the ECB has made available to institutions
standardised templates for notifying ex ante and ex post non-material
changes/extensions and for submitting applications for material model
changes/extensions. Institutions are invited to use these templates to facilitate
the process and to ensure consistency and completeness. 97

7.4 Classification

103. For the reasons mentioned in paragraph 100, institutions should have
processes in place which specify, in detail, that the classification of a
change/extension is adequate and consistent with the classification of other
changes/extensions. The institution should ensure that the classification
process is not subject to any arbitrage. In line with Article 3(3) of Commission
Delegated Regulation (EU) No 529/2014, “One material extension or change
shall not be split into several changes or extensions of lower materiality”.
Similarly, an extension or change that requires notification before its
implementation (ex ante) should not be split into several changes or extensions
to produce one that is notified after implementation (ex post). Nor should
several unrelated changes/extensions be combined to produce one change of

97
In addition, further forms and guidelines have been made available on the Single Supervisory
Mechanism (SSM) website to support institutions in the pre-application process.

ECB guide to internal models – General topics 45

 lower materiality (e.g. two different model changes that affect RWEAs in
opposite ways).

104. In accordance with Article 3(4) of Commission Delegated Regulation (EU) No

529/2014, “in case of doubt institutions shall assign extensions and changes to
the category of the highest potential materiality”.

105. To ensure the accuracy of the impact assessment and the correctness and
consistency of the resulting classification, the institution should establish a four-
eye principle. This means that the assessment and classification should be
confirmed by a unit independent of the one responsible for the assessment and
classification of the change/extension.

7.5 Impact assessment

106. The impact assessment process must fulfil the requirements of Commission
Delegated Regulation (EU) No 529/2014. It should consist of a quantitative and
a qualitative assessment.

107. The quantitative assessment focuses on the impact of the change or extension
on RWEAs. Before and after the change or extension the institution should
calculate the difference in these amounts for credit and dilution risk associated
with the range of application of the internal rating system.

108. This quantitative impact assessment is based on the specifications of Article

4(2) and (3) and Article 5(2) of Commission Delegated Regulation (EU) No
529/2014. The institution should use transparent definitions and internal
procedures.

(a) The institution should document the relevant reference date on which the
calculations are based. In accordance with Article 3(2)(a) of Commission
Delegated Regulation (EU) No 529/2014, the institution should use the
most recent data available. In the case of material change or ex ante
notification the time between the reference date and the date of notification
should not exceed nine months. 98

(b) The institution should give a precise definition of the range of application of
the rating system applied in the calculations as referred to in Article 4(1)(c)
and Article 5(1)(a)(iii) of Commission Delegated Regulation (EU) No
529/2014. It should ensure that the change to the IRB approach is directly
related to exposures within the range of application of the rating system.

(c) The institution should generally perform a precise impact assessment 99 (all
exposures of the relevant range of application) when the rating system is

98
For impact assessments provided during an on-site inspection, the most recent data should be used.
99
As derived from Article 3(2) of Commission Delegated Regulation (EU) No 529/2014.

ECB guide to internal models – General topics 46

 automatic and does not require any human judgement of the qualitative
variables (e.g. behavioural scoring).

(d) If the institution applies Article 3(2)(b) of Commission Delegated

Regulation (EU) No 529/2014 and performs an impact assessment based
on the re-rating of a representative sample of the population (only possible
for rating systems that require qualitative assessment to define the final
rating/score), this sample and its relation to the population should be
described in detail (number of observations/exposures,
minimum/maximum exposure amount, mean/median exposure amount,
first/third quartile). The representativeness of the sample should be
documented.

(e) If the institution applies Article 3(2)(b) of Commission Delegated

Regulation (EU) No 529/2014 and performs an impact assessment based
on other reliable inference methodologies, these methods should be
described in detail and their reliability corroborated by qualitative and
quantitative means.

109. The qualitative assessment is based on the specifications of Article 4(1)(a) and
(b) and Article 5(1)(a) points (i) and (ii) of Commission Delegated Regulation
(EU) No 529/2014, which refer to the Annex I of that same Regulation. The
institution should thoroughly examine each of these criteria. In addition, as
reported in the change policy and to ensure consistency, institutions should
examine the metrics and significance levels for the impact assessment and
threshold calculation (see also Section 7.2, content of the change policy).

110. If a criterion specified in Annex 1 of Commission Delegated Regulation (EU) No

529/2014 may be applicable a priori (and within reason) to a change or an
extension and the institution’s assessment concludes that this particular
criterion is not fulfilled, the institution should document this conclusion in the
notification.

7.5.1 User acceptance test

111. To avoid the risk of having unexpected consequences in the use of the changed
rating system (e.g. altered role of the changed rating system in the risk
management of credit exposures), institutions should assess and document the
impact of a material change/extension on the use of the parameters and ensure
that the related internal policies and procedures for the areas described in
Section 6.3 of this chapter remain relevant.

112. In the context of rating systems which contain qualitative inputs and/or any
expert judgement component, the exposures of the representative sample
referred to in point (d) of paragraph 108 should be fully re-rated under the
amended rating system (including the material change or extension); adherence
to the entire rating assignment process should be ensured. The feedback

ECB guide to internal models – General topics 47

 received from users on the application of the amended rating system and on the
rating results is expected to be analysed and documented.

113. Changes or extensions that are classified as non-material do not generally

require the preparation of a use test sample, unless there is evidence of a
potential impact on the use of the parameters.

7.6 Re-rating process

114. Where competent authorities have provided their permission in relation to a

material extension or change, Article 3(5) of Commission Delegated Regulation
(EU) No 529/2014 requires institutions to calculate their own funds
requirements on the basis of this approved extension or change from the date
specified in the new permission. The ECB understands that this process (i.e.
the re-rating process) should be covered in the institution’s change policy.

115. Re-rating refers to the computation of a rating using the changed or extended
rating system and the assignment of this new rating to an obligor previously
rated using the rating system as it was before the change or extension.

116. In the context of changes or extensions that are classified as material, the
change policy should ensure that the re-rating process is immediate. All former
ratings and estimates should therefore be replaced by ratings and estimates
calculated using the changed or extended model from the date specified in the
approval decision – if the conceptual design allows this. This applies, for
example, to rating systems that are exclusively based on behavioural scoring or
in the case of recalibration not affecting the rating process.

117. If an immediate re-rating is not possible (for example if the rating assignment
requires new manual input and human judgement), and only for non-retail rating
systems, the policy should ensure the following.

(a) The obligors/exposures/facilities are rated using the amended rating

system within the time frame of the yearly re-rating process, i.e. within a
maximum of 12 months.

(b) To mitigate the risk of underestimation of own funds requirements, in the

event that a material change would lead to a material increase in the
RWEA (i.e. more than 10% on the range of application of the rating system
subject to change), the institution should apply the RWEA impact,
simulated on the basis of the representative sample. The impact thus
produced is the positive difference between the simulated RWEA after the
material change and the RWEA before the material change approved by
the competent authority. This should be done at the first Common
Reporting date after the date of implementation.

For the purpose of point (b) above, the ECB would consider the following
approach as the most appropriate:

ECB guide to internal models – General topics 48

 • apply the simulated RWEA impact until all exposures within the range of
application are rated using the changed model;

• remove the simulated RWEA impact linearly, i.e. 25% every quarter.

118. The re-rating process for changes/extensions that are classified as non-material
may take up to one year from the date of implementation.

8 Third-party involvement

8.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 144 (1)

171

179 (2)

190 (3) and (4)

Other references

Final Draft RTS on assessment methodology for IRB 21/07/2016 4 (1) to (5)

CEBS Guidelines on Outsourcing 100 14/12/2006

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become additional relevant legal
reference. Currently the RTS only exists in a final draft version.

8.2 Preliminary principles

119. Outsourcing, as defined in the CEBS Guidelines on Outsourcing, refers to an

authorised entity’s use of a third party (the “service provider”) to perform
activities that would normally be undertaken by the authorised entity itself, now
or in the future. For the purposes of this section, outsourcing in the context of
IRB models refers to the involvement of third parties in any IRB-related tasks,
including data provisioning and the use of external data (e.g. development data,
calibration data, external ratings as input for internal models, pooled data). The
specific case of delegation of IRB-related tasks to different legal entities within
the same group (internal outsourcing) is also considered as outsourcing and
hence is subject to the expectations set out below. 101

100
Committee of European Banking Supervisors, “Guidelines on Outsourcing”, referred to in this guide as
the “CEBS Guidelines on Outsourcing”.
101
For generic requirements on outsourcing, such as the existence of an outsourcing policy, contract
requirements, monitoring of third-party performance and a contingency plan for interruption of service,
institutions must take into consideration the generic guidelines set out in the CEBS Guidelines on
Outsourcing, as well as the review of those guidelines by the EBA once published. The ECB’s thematic
review on outsourcing should also be considered once published.

ECB guide to internal models – General topics 49

8.2.1 Contract requirements

120. All outsourcing arrangements for IRB-related tasks should be subject to a

formal and comprehensive contract or similar documented agreement in
accordance with the proportionality principle (in the case of internal outsourcing
between different entities within the same group, provisions such as service
level agreements (SLAs) or other written agreements may be considered as
sufficient, subject to the criticality or importance of the tasks outsourced).
Outsourcing institutions should take into account the aspects set out in
Guideline 8 of the CEBS Guidelines on Outsourcing when preparing written
outsourcing contracts. To avoid operational risks which could hinder the
performance and operations of the rating systems, outsourcing agreements
should provide for the following.

(a) The agreed terms do not impede the institution in performing its validation
activities.

(b) The agreed terms do not impede the necessary communication between
the institution and the competent authorities in performing their supervisory
duties. In this sense they should include:

(i) full and timely access for competent authorities to all information
required (e.g. all of the models’ development details, where an
externally developed rating system is used);

(ii) a requirement for the third party to provide support to the institution in
the event of a request for information by the competent authority.

(c) The agreed terms should ensure that the provider gives the institution
access to relevant information in order to maintain sufficient in-house
knowledge. The delivery of training and workshops is considered good
practice.

8.3 Third-party involvement in internal functions and tasks

8.3.1 Internal validation and internal audit tasks

121. Although institutions are allowed to delegate some of their tasks, activities and
functions to a third party, this should be done in accordance with all existing
legal requirements and after due consideration of this guide, in particular the
principles relating to internal validation and the internal audit included in this
general topics chapter. If an institution plans to delegate certain internal
validation or internal audit tasks to a third party that would perform them outside
the EU, it should discuss this plan with the competent authority in advance.

ECB guide to internal models – General topics 50

 122. The ECB considers that responsibility for delegated tasks should be retained by
the outsourcing institution. 102 This understanding is also expressed in Guideline
2 of the CEBS Guidelines on Outsourcing.

123. To ensure consistency in the content, quality and governance of the activities
performed internally and externally, the ECB understands that the following
practices in particular should be observed:

(a) Reports should carry the logo and name of the institution and of the third
party performing the tasks.

(b) Reports should be approved by the senior management and the members
of the management body (or the designated committee thereof)
responsible for the function within the institution.

(c) The institution should assess the quality/performance of the outsourced

tasks.

(d) The independence requirements set out in Section 1.6 for internal
validation and Section 1.7 for internal audit of this chapter also apply to
external parties. Institutions must ensure that model reviews are
independent and free from any undue influence, also when performed by
third parties. In this respect institutions should establish independence
guidelines/policies with regard to third parties and those participating in
internal model tasks equivalent to the internal guidelines and rules set
internally. In particular, third parties and individuals that have performed or
are currently performing model development or CRCU tasks should not
perform model validation tasks within the same institution until a prudent
cool-off period has elapsed. 103, 104

8.3.2 Use of external credit risk parameters/ratings

124. Although institutions are allowed to use external credit risk parameters as a
component of their rating systems, the following practices should be observed:

(a) Internal ratings and estimates methodology should also take internal
information into account. 105 When institutions use external ratings or
parameters, they should ensure that these are incorporated in their
estimation process in an appropriate manner and adjusted in accordance
with the specificities of the institution. 106

102
This also applies in the case of internal outsourcing.
103
In the case of internal outsourcing or delegation of tasks within different legal entities of the same
group, “third party” refers to the unit/function that would perform the delegated tasks.
104
In the light of Article 4(3) of the Final Draft RTS on assessment methodology for IRB, the third party
may provide the institution with the information necessary to conduct the validation activities.
105
See Article 171(2) of the CRR.
106
See also the credit risk chapter on the use of external data.

ECB guide to internal models – General topics 51

 (b) Analogously to the requirements for data pooled across institutions, 107 the
institution should demonstrate good knowledge of the work performed by
the third party in producing the estimates. In particular, the institution
should demonstrate a good understanding of the data cleansing process,
assumptions used, methodological choices and resulting limitations. The
institution should also monitor the performance of the rating systems
involved and be able to audit them, and have clear triggers for requesting
a model review.

8.3.3 Model development and maintenance

125. When institutions delegate the development and/or maintenance of internal

models, this should be done in accordance with all existing regulatory
requirements and the institutions’ internal guidelines and policies, also taking
this guide into account. If an institution plans to delegate such tasks to a third
party that would perform them outside the EU, it is encouraged to discuss this
with the competent authority in advance.

126. As the ultimate model owners and users, institutions should do the following:

(a) maintain an appropriate level of in-house knowledge (see Section 8.3.5 of

the this chapter);

(b) have a robust contingency plan in place to ensure that they are prepared
for the risk that could derive from insufficient maintenance of the rating
systems.

8.3.4 Use of pool models

127. In accordance with Article 179(2)(d) of the CRR, an institution using pool
models must remain responsible for the integrity of its rating systems. In
particular, the institution is ultimately responsible for the performance of the
internal validation activities as elaborated in Section 4 of this chapter.

128. To comply with the requirement to ensure integrity of the rating system specified
in Article 179(2)(d) of the CRR, if deficiencies are identified in the pooled rating
system at institution level, the institution should be able to independently trigger
a procedure designed to amend the system, if necessary, at individual or pool
level. This applies regardless of the performance of the rating system at the
pool level or at the level of the other participating institutions.

129. Where a third party is involved in the tasks of developing a rating system and
risk estimation for an institution, the institution should verify that the validation
activities with regard to those rating system and those risk estimates are not

107
See Article 179(2) of the CRR and in particular sub-paragraph (e).

ECB guide to internal models – General topics 52

 performed by that third party. Where, for the purpose of developing a rating
system and risk estimation, the institution uses data that is pooled across
institutions and a third party is developing the rating system, the third party may
assist the institution in its validation activities by performing those tasks of
validation which require access to the pooled data. 108

8.3.5 In-house knowledge

130. To ensure that institutions are able to identify, manage and monitor the risks
connected with internal models, they should maintain adequate in-house
knowledge and core competence, as they are ultimately responsible for
outsourced tasks and functions. Institutions should have access to all relevant
information; this will enable them to take direct control of an outsourced activity
in extremis. Best practice to ensure that this in-house knowledge is maintained
includes ad hoc training at all levels (not only at management level) and proper
oversight of the outsourced activities.

131. In the relationship with the third party, and for the purpose of maintaining
appropriate in-house knowledge and responsibilities, the ECB would consider
the following practices as being advisable in the event of third-party involvement
in IRB-related tasks:

(a) the terms of the contract include transparency requirements;

(b) the institution has full access to all relevant information regarding internal
model-related topics;

(c) the institution receives regular reports;

(d) on request, the institution is provided with specific reports;

(e) on request, the third party provides support and attends interviews with the
competent authorities.

132. In cases where third parties are involved in model (re-)development and/or
parameter (re-)calibration, to ensure that the institution maintains sufficient in-
house knowledge and an adequate understanding of the rating system or that
part of the rating system obtained from the third party 109, the ECB would
consider the following practices as being advisable with regard to both the
methodology and the data used for (re-)development and (re-)calibration:

(a) The institution has access to all relevant information that enables it to
understand the main model assumptions and risk estimation processes.

(b) The institution has access to its own obligors’ information.

108
See Article 4(3) of the Final Draft RTS on assessment methodology for IRB.
109
See Article 4(2)(b) of the Final Draft RTS on assessment methodology for IRB and paragraph 128 of
this chapter.

ECB guide to internal models – General topics 53

 (c) In cases where pooled or external data are used for model development or
calibration, the institution is able to assess to what degree the portfolio on
which the model is based/developed is representative of its own portfolio,
for the purposes of risk differentiation and risk quantification.

(d) The institution has sufficient knowledge of the definition of default applied
for the purposes of risk differentiation and risk quantification.

(e) The institution has access to all necessary information to enable it to

perform independent validation, including the validation of the model
assumptions and performance of its own portfolio.

(f) The institution has a specific change policy in place for models developed
by third parties.

(g) The institution is able to assess the need for a model change. The criteria
that trigger a model change should be reflected in both the institution’s
model change policy and the contract with the third party, to ensure that
the institution is able to make or request changes to the models.

8.3.6 Independent monitoring of third-party performance

133. To ensure that it is able to identify and manage the risks connected with internal
model-related outsourced tasks, the institution should also independently
monitor the performance of third parties and have appropriate processes in
place in this regard. This practice reinforces the fact that the institutions are the
ultimate users of the rating systems and thus have the ultimate responsibility for
their operations.

134. The following are considered by the ECB as good practices with regard to
monitoring third-party provisioning of external data.

(a) Similar data vetting should be performed as would be the case if the data
or service were provided in-house. Data quality checks should be
automated (IT/batch processes) when possible, and technical issues as
well as reasonableness and consistency should be considered.

(b) Historical differences in the data provided should trigger inquiries if

justified, or if there has been an error.

(c) Where external data are used, their representativeness, appropriateness

and consistency with regard to the institution should be assessed.

(d) Cross-checks should be carried out between different databases (when

available) or between different providers. This is a sign of consistency and
robustness.

(e) SLAs/contract agreements should include the required specific key

performance indicators (KPIs) and performance metrics.

ECB guide to internal models – General topics 54

 (f) It should be recognised that the data quality of the information provided
cannot be determined by its predictive power or by the performance of the
model itself.

135. The following are considered by the ECB as good practices with regard to
monitoring third-party IRB-related tasks:

(a) the same standards of monitoring and audit should be applied to external
tasks as to those performed in-house;

(b) SLAs/contract agreements should include the specific KPIs and

performance metrics that the service should include;

(c) specific bodies should be designated, with clear responsibilities regarding

the monitoring of external IRB activities (such as a monitoring committee).

ECB guide to internal models – General topics 55

 Credit risk

1 Scope of the credit risk chapter

1. The purpose of this chapter is to provide transparency on how the ECB
understands a number of topics related to internal models used for the IRB
approach, including an initial section covering data maintenance for this
approach. It is important to note that this chapter does not aim to cover
exhaustively all topics of the CRR for the IRB approach that could be subject to
review during internal model investigations. On these selected topics, the
chapter is aligned with the EBA GL on PD and LGD 1.

1
EBA Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures
(EBA/GL/2017/16), referred to in this guide as the “EBA GL on PD and LGD”.

ECB guide to internal models – Credit risk 56

2 Data maintenance for the IRB approach

2.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 142 (1)(1)

144

174 (b)

175 (1)

176

189 (1), (2)(c)

190 (4)

Other references

Final Draft RTS on assessment methodology for IRB 2 21/07/2016 32, 33, 75, 76, 77, 78
3
Basel Committee on Banking Supervision (BCBS) 239 09/01/2013 Principles 1-11

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently the RTS only exist in a final draft version.

2. In accordance with Article 144(1) of the CRR, an institution’s systems for the
management and rating of credit risk exposures must be sound and
implemented with integrity. In particular, the institution must collect and store all
relevant data to provide effective support to its credit risk measurement and
management processes. The ECB understands that, in order to comply with
these requirements, institutions should deploy robust, well-documented and
adequately tested IT systems, together with sound data management practices.

3. Consequently, this section of the guide sets out the principles regarding the
following elements for the management of IRB data: 4

(a) IT systems: infrastructure and implementation testing;

(b) policies, roles and responsibilities in data processing and data quality
management;

(c) components of the data quality management framework.

2
Final Draft Regulatory Technical Standards on the specification of the assessment methodology for
competent authorities regarding compliance of an institution with the requirements to use the IRB
Approach in accordance with Articles 144(2), 173(3) and 180(3)(b) of Regulation (EU) No 575/2013,
referred to in this guide as the “Final Draft RTS on assessment methodology for IRB”.
3
Basel Committee on Banking Supervision “Principles for effective risk data aggregation and risk
reporting”, referred to in this guide as “Basel Committee on Banking Supervision (BCBS) 239”.
4
The ECB acknowledges that there are other relevant elements of data management not covered in this
guide which institutions should take into account.

ECB guide to internal models – Credit risk 57

2.2 IT systems: infrastructure and implementation testing

2.2.1 Infrastructure

4. Sound and robust IT infrastructures play an essential role in supporting the

institution’s rating systems. In addition, and in accordance with Article 175(1) of
the CRR, institutions must document the design and operational details of their
rating systems.

5. With regard to the soundness and robustness of institutions’ IT infrastructure,

the ECB considers that Article 78(2) and (3) of the Final Draft RTS on
assessment methodology for IRB provides a good understanding of the
elements that institutions should take into account in order to comply with the
data-related requirements of the CRR. 5

6. Furthermore, to comply with the documentation requirements for the rating

systems as established under Article 144(1)(e) and Article 175(1) of the CRR, it
is the ECB’s view that institutions should document and keep an updated
register of all current and past versions of the following elements of a rating
system:

(a) the model’s data 6 flow (from data entry 7 to reporting and for both historical
data and current exposure data), identifying the relevant workflows and
procedures relating to data extraction, data collection, data storage and
data transformations;

(b) the relevant sources of data and the global map of IT systems and
databases involved in the calculation systems used for the purposes of the
IRB approach;

(c) the relevant functional specification of IT systems and databases, including

their size, date of construction and data dictionaries, specifying the content
of the fields and of the different valid data values that could be inserted in
them, with clear definitions of data items;

(d) the relevant technical specification of IT systems and databases, including

the type of database, tables, database management system, database
architecture, and data models given in any standard data modelling
notation;

(e) the audit trail for critical IT systems and databases.

5
See Articles 144(1)(d) and 176 of the CRR.
6
This refers to the model’s internal data, external data or pooled data.
7
This refers to the first entry or registration of data in the institutions’ systems and applications or in the
core systems of the institutions where the raw data first originated.

ECB guide to internal models – Credit risk 58

 To allow an independent knowledgeable third party to obtain a detailed outline
of the different IT elements of the rating systems, the documentation produced
by the institution should be clear and understandable.

To comply with the requirement to document its rating system and the rationale
for its design, 8 the institution should keep the register of all rating systems,
including all current and past versions of rating systems, updated for a period of
at least three years and, whenever necessary, an extended period beyond that.
This is also reflected in Article 33(1) of the Final Draft RTS on assessment
methodology for IRB. In any case, the institution should ensure that the
elements mentioned in paragraph 6(a) to (e) for the current rating system are
adequately recorded in the register and enable a clear understanding of all
relevant data of the current rating system that must be stored by the institution. 9

2.2.2 Implementation testing

7. In order to ensure the integrity and robustness of IT systems 10 and in particular

that, in terms of IT, the implementation of the models is successful and error-
free, institutions should have in place a consistent process for testing the
relevant IRB systems and applications upon first implementation and on an
ongoing basis. This IT-testing process should be clearly defined and
documented in an organisation-wide policy and procedure.

8. To achieve its objective the policy should consider all potential events that
should trigger a testing procedure and their impact on the tests to be
conducted. The trigger events that should be considered include: software
releases or material IT-related changes, regulatory changes, model
methodology changes and the extension of the range of application of a rating
system.

9. IT implementation tests to be considered include the following:

(a) unit/component/module tests;

(b) integration tests (of units and between systems);

(c) system tests (this includes functionality, performance – in normal and

stress scenarios – and security and portability tests);

(d) user acceptance testing (functional testing);

(e) regression testing.

10. In principle, the unit(s) responsible for performing the implementation tests
should be clearly identified and the results of the tests should be documented. It

8
See Article 144(1)(e) of the CRR.
9
See Articles 144(1)(d) and 176 of the CRR.
10
See Article 144(1) of the CRR.

ECB guide to internal models – Credit risk 59

 is the ECB’s view that as a general rule institutions should develop a
standardised format for the documentation of test results.

2.3 Policies, roles and responsibilities in data processing and data

quality management

11. For institutions to be able to comply with the requirement to collect and store all
relevant data established under Article 144(1)(d) of the CRR, it is the ECB’s
understanding that policies and rules on data management should be defined at
group level 11 for both of the following aspects: data processing (i.e. data
collection, storage, validation, migration, actualisation and use), and data
quality management (see Section 2.4 of this chapter).

12. As for data processing, and in particular with regard to manual interventions
and data transfers, the following principles should be considered:

(a) to ensure that all data transformations are traceable and controlled,
general guidelines and rules should be clearly formalised with regard to
manual interventions within the data processing;

(b) to ensure timeliness and accountability, all data transfers should be

formally agreed upon (for example by means of service level agreements)
by data providers and data users (for both outsourced and in-house
processes).

13. To ensure the integrity of the data processes, the policies and rules on data
management should clearly set out the relevant data governance
arrangements. It is also expected that these policies and rules will specify the
different roles and responsibilities assigned to data management. These include
data quality roles and responsibilities for both business owners and IT functions
and data and systems ownership throughout the entire credit risk modelling life
cycle (including all IT systems used). These policies should take into account
the following principles.

(a) The responsibilities of business owners include:

(i) ensuring data are correctly entered, kept up-to-date and aligned with
the institution’s data definitions;

(ii) ensuring that data aggregation capabilities and reporting practices

are consistent with the institution’s policies.

(b) IT functions are responsible for supporting the operation of the systems for
data collection, processing, transformation and storage during the entire
life cycle of the data.

11
See Section1.2 of the general topics chapter for the definition and implementation of group-wide
principles and guidelines.

ECB guide to internal models – Credit risk 60

 (c) Different business owners and IT systems owners may be appointed
throughout the data life cycle. However, each data source, IT system and
process step should have an assigned business owner and/or IT systems
owner that can be formally identified.

2.4 Data quality management framework

14. Institutions must have in place a process for vetting data inputs into the model.
This must include an assessment of the accuracy, completeness and
appropriateness of the data. 12 To comply with this requirement and to ensure
the quality of the data used for credit risk measurement and management
processes, it is the ECB’s view that institutions should establish and implement
an effective data quality management framework that is formalised in a set of
policies and procedures. This framework should be applicable to all data used
in IRB-related processes, i.e. internal data, external data and pooled data, if
any. In addition, it should ensure that reliable risk information is available to
enable an institution’s risk profile to be assessed accurately and drive sound
decision-making within the institution and by external stakeholders, including
competent authorities.

15. The ECB considers that the data quality management framework is effective
when it encompasses the following components:

(a) sound underlying governance principles (see Section 2.4.1 of this

chapter), particularly those relating to the allocation and fulfilment of roles
and responsibilities for the management of data quality in a manner that
ensures that data quality management activities are independent of data
processing activities;

(b) a description of the scope in terms of risk data coverage (see

Section 2.4.2);

(c) data quality standards covering all relevant data quality dimensions,
i.e. completeness, accuracy, consistency, timeliness, uniqueness, validity,
availability and traceability (see Section 2.4.3);

(d) consistent criteria and a systematic metrics approach to assess

compliance with data quality standards; this should be supported by
sufficient data quality controls along the entire IRB data chain (see
Section 2.4.4);

(e) procedures for constantly assessing and improving the quality of data (see
Section 2.4.5);

(f) reporting procedures on data quality allowing for sufficient understanding

of the quality of the data supporting the IRB models (see Section 2.4.6).

12
See Article 174(b) of the CRR.

ECB guide to internal models – Credit risk 61

 The following sections further develop the above-mentioned elements.

2.4.1 Governance principles for the data quality management framework

16. The data quality management framework:

(a) should be approved by the institution's management body or a designated

committee thereof and senior management as part of their responsibilities;

(b) should be distributed throughout the organisation to the relevant staff;

(c) should be regularly assessed in order to verify its adequacy, and be

updated and improved whenever necessary;

(d) should be subject to regular review by the internal audit function or another
comparable independent auditing unit. 13

17. The roles of the different units, internal bodies and staff involved in the data
quality management process should be defined in such a way as to ensure that
the data handling process is sufficiently independent of the data quality
management process.

18. The ECB considers it good practice for institutions to have a dedicated
independent unit with an overall view of and responsibility for the management
of data quality. Where an independent unit is established, the size of this unit
should be proportionate to the nature, size and degree of complexity of the
institution’s business and organisational structure.

2.4.2 Scope of the data quality management framework

19. The data quality management framework:

(a) should cover all relevant data quality dimensions: completeness, accuracy,
consistency, timeliness, uniqueness, validity, availability and traceability
(see paragraph 21);

(b) should cover the whole data life cycle, from data entry to reporting, and
encompass both historical data and current application databases.

20. If institutions use data provided by third parties, the ECB considers it good
practice for them to ensure that the third party has data quality processes in
place to ensure the accuracy, completeness and appropriateness of the data
provided. 14

13
For further details on the review of the rating systems by internal audit, see Section 5 of the general
topics chapter of this guide.
14
See Article 174(b) of the CRR.

ECB guide to internal models – Credit risk 62

2.4.3 Data quality standards in the data quality management framework

21. In accordance with Article 174(b) of the CRR, institutions must implement a
process for vetting data inputs into the model which must include an
assessment of the accuracy, completeness and appropriateness of data. The
ECB understands that, in order to comply with this requirement, institutions
should establish data quality standards that set the objectives and overall scope
of the data quality management process. To this end, these standards should
be defined for the following data quality dimensions 15 for all data inputs into the
model and at each stage of the data life cycle:

(a) completeness (values are present in any attributes that require them);

(b) accuracy (data are substantively error-free);

(c) consistency (a given set of data can be matched across the institution’s
different data sources);

(d) timeliness (data values are up-to-date);

(e) uniqueness (aggregate data are free from any duplication arising from
filters or other transformations of source data);

(f) validity (data are founded on an adequate and rigorous classification

system that ensures their acceptance);

(g) availability/accessibility (data are made available to the relevant

stakeholders);

(h) traceability (the history, processing and location of the data under
consideration can be easily traced).

2.4.4 Data quality controls

22. Data quality should be measured in an integrated and systematic way. The
measurement system and the frequency of its application should be formalised.

23. Indicators and their corresponding tolerance levels and thresholds should be
set in order to monitor compliance with the standards established and should be
combined with visual systems (e.g. red/amber/green traffic-light system) and
dashboards for monitoring and reporting purposes.

24. Indicators should be supported by effective and sufficient data quality checks
and controls throughout the data life cycle, from data entry to reporting, and for
both historical data and current application data. Data quality checks and
controls should include reconciliation across and within systems, including

15
It is the ECB’s view that the CRR reference to appropriateness of data inputs encompasses the
following additional data quality dimensions: consistency, timeliness, uniqueness, validity,
availability/accessibility and traceability.

ECB guide to internal models – Credit risk 63

 between accounting and IRB data. An effective control framework should
therefore be in place to ensure that sound controls and related procedures are
implemented, especially for manual processes.

2.4.5 Remediation of data quality issues

25. A process for the identification and remediation of data quality deficiencies
should be in place in order to constantly improve data quality and promote
compliance with the data quality standards.

26. Data quality assessments should be carried out independently (see

paragraphs 17 and 18) and recommendations should be issued with an
indication of their priority, based on the materiality of the incidents identified. All
such data quality incidents should be recorded and monitored. For each of the
data quality incidents, an owner responsible for resolving the incident should be
appointed and an action plan for dealing with the incident drawn up on the basis
of the priority assigned. Remediation timelines should depend on the severity
and impact of the incident and the implementation timelines required to resolve
it. Data quality incidents should be resolved at source level 16 or, if this is not
possible, mitigated by taking a prudent approach.

2.4.6 Data quality reporting

27. In accordance with Article 189(2)(c) of the CRR, the institution’s senior
management must ensure, on an ongoing basis, that the ratings systems are
working properly. To accomplish this, the ECB understands that a formal
reporting process on the quality of risk data should be in place with the
objective of improving the quality of data and enabling an assessment of the
potential impact of data quality in own fund requirements calculations. In
general, this reporting should be presented in a standardised format with clear
and concise content, including the following:

(a) comprehensive overview of the performance of the model in terms of data

quality, including external data and pooled data, if any, at all stages of the
IRB life cycle, from data entry to reporting, for both historical data and
current exposure data;

(b) findings and, where applicable, recommendations to address detected

weaknesses or shortfalls;

(c) sufficient and appropriate evidence that the recommendations have been
adequately addressed and properly implemented (e.g. by means of a
status report).

16
From the source system in which the incidents are present down to the IRB datasets or systems.

ECB guide to internal models – Credit risk 64

 28. In accordance with Article 189(1) of the CRR, the management body or a
designated committee thereof and senior management must possess a general
understanding of the rating systems of the institution and a detailed
comprehension of its associated management reports. To comply with this
requirement, the ECB understands that reports on the quality of risk data should
be submitted to these parties. In addition, the ECB considers it good practice for
these reports to also be submitted to all other relevant staff, including
modellers, internal validation, internal audit, data quality managers, data owners
and other business units involved.

29. Data quality reports should be produced and submitted to senior management
more frequently than annually to enable senior management to ensure, on an
ongoing basis, that the rating systems are operating properly in accordance
with Article 189(2)(c) of the CRR.

3 Use of data

3.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 144 (1)(d)

171 (1)(a), (b)

172 (3)

174 (b), (c), (e)

176

178 (4)

179 (1)(a), (c), (d), (2)(a), (b)

EBA GL on PD and LGD 20/11/2017 15-35

Other references

Final Draft RTS on assessment methodology for IRB 21/07/2016 45, 48, 50, 56

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

30. In accordance with Article 144(1)(d) of the CRR, institutions must collect and
store all relevant data to provide effective support to their credit risk
measurement and management processes. Furthermore, good data quality is a
fundamental condition for developing a robust rating system. The ECB
considers that, to comply with these requirements and ensure the quality of
data, institutions should have sound policies, processes and methods in place,
under paragraphs 15 to 34 of the EBA GL on PD and LGD for assessing and
improving the quality and representativeness of the data used in the modelling
and risk quantification process.

ECB guide to internal models – Credit risk 65

 31. Since the data-related requirements of the CRR also apply in cases where an
institution estimates CCFs, paragraph 30 is also relevant for such institutions.

3.2 Use of external data

32. Data-related requirements established under the CRR apply to all data: internal,
external or pooled. In the ECB’s understanding, therefore, paragraph 30 is also
relevant in the event that an institution uses external or pooled data. The
principles on the collection and storage of data are relevant to the institutions’
own data and to the data received from the pool.

33. To ensure that credit risk management and measurement processes are built
on appropriate data, for the purposes of risk differentiation, risk quantification
and review of estimates institutions should assess whether external data can be
used to complement internal data when they consider they do not have
sufficient available internal data.

34. In addition, when institutions use different data sources (including different
external databases, whether or not combined with internal data) for the purpose
of risk quantification, there could be a risk that the sources include common
obligors among the data sources used for risk quantification. This could lead to
bias or double-counting effects in the calculation of one-year default rates. To
avoid this risk, institutions should develop the necessary processes in order to
identify common obligors within these databases and ensure that each common
obligor is only taken into account once in the calculation of one-year default
rates. 17 When institutions are not able to identify common obligors, they should
analyse potential bias or double-counting effects in the calculation of one-year
default rates. These bias or double-counting effects should be reflected
appropriately in the computation of one-year default rates and long-run average
(LRA) default rates.

35. If an institution uses statistical models and other mechanical methods to assign
exposures to obligors or facilities grades or pools, the data used to build the
model must be representative of the population of the institution’s actual
obligors or facilities. 18 If external data are used, the same requirements with
regard to representativeness 19 must be applicable vis-à-vis the bank’s portfolio
or portfolio subset for which the external data are used.

36. Proving representativeness in cases where an institution uses external data is

generally more difficult, as internal data are scarce. If an institution cannot
provide sufficient proof that the external data are representative, in the ECB’s
view it may still use external data if it shows (by quantitative analysis and/or
qualitative argumentation) that the information gained from the use of the

17
If the default identification is at the level of an individual credit facility rather than at obligor level, this
principle will not be relevant.
18
See Article 174(c) of the CRR.
19
As established under Articles 174(c) and 179(1)(d) of the CRR.

ECB guide to internal models – Credit risk 66

 external data outweighs any drawbacks stemming from the deficiencies
identified and an appropriate margin of conservatism (MoC) is applied. In
particular, institutions should provide evidence that the model’s performance
does not deteriorate when including information derived from the external data,
and that the parameter estimates are not biased. To assess these issues, the
institution should conduct quantitative and qualitative analyses specifically
designed for this purpose.

37. In accordance with Article 174(b) of the CRR, if an institution uses statistical
models and other mechanical methods to assign exposures to obligors or
facilities grades or pools, it must have in place a process for vetting data inputs
to the model, which should include an assessment of the accuracy,
completeness and appropriateness of the data. In addition, and in accordance
with Article 179(1)(a), in quantifying the risk parameters to be associated with
rating grades or pools institutions must incorporate all relevant data, information
and methods. To comply with these requirements, institutions should ensure
that, when external data are used for risk differentiation, risk quantification or
review of estimates, they know the data sources and the most relevant data
processing operations of the variables acting as direct model inputs performed
by the data provider. Institutions should be able to differentiate between internal
and external data and to document which information is internal and which
information is received from external data sources. To ensure that the data
remain appropriate, institutions should provide an adequate rationale in the
event that, for the purpose of risk differentiation, risk quantification or review of
estimates, they modify the external data acquired, select only part of a wider
external database or use different external providers.

3.3 Use of external bureau scores or external ratings as input variables

in the rating process

38. Where an institution uses external credit bureau scores or external ratings as
input variables in the rating process, and in particular when externally sourced
scores are the main (or one of the main) input variable(s) of the overall internal
rating, there is a risk that an internal model may not consider all relevant
information. In the ECB’s understanding, institutions mitigate this risk when they
comply with the following principles.

(a) The external scores or ratings and/or data are regularly updated or
refreshed, especially where credit bureau information is dynamic and is
used not only for the application rating but also for the ongoing behavioural
rating.

(b) Institutions understand the structure and nature of external scores or

ratings and their key drivers. They also regularly verify that the results of
the credit bureau score continue to be appropriate input variables in their
credit rating process, for example by reviewing any changes in the credit
bureau score methodology. The greater the importance of the external

ECB guide to internal models – Credit risk 67

 scores or ratings, the greater understanding institutions should
demonstrate of their structure, nature and key drivers and the higher the
frequency that should be considered in monitoring the appropriateness of
these external scores or ratings.

(c) Validation requirements are similar to those applied to other internal and
external input variables.

(d) Even when the external score or rating is the main (or one of the main)
driver(s) of the internal rating, the institution ensures that all relevant
internal information regarding the creditworthiness of the obligor is taken
into account with sufficient weighting in the internal rating. In addition, the
institution demonstrates that the additional relevant internal information
considered in the model and its weighting are sufficient to ensure that the
internal rating does not merely take on the results of the external bureau
scores or the external ratings used.

(e) When institutions make use of external scores or ratings or any other
judgement-based assessment provided by a third party as input variables
in the rating process, they should ensure that any potential correlation
between the relevant risk drivers does not lead to bias or a double-
counting effect in the risk parameter estimates. This can be especially
relevant in these cases, due to the potential use of duplicated information.

(f) The institution remains responsible for the performance of the model.

3.4 Use of pooled data 20

39. The use of pooled data is treated similarly to the situation where internal data
are combined with data derived from a different (and external) set of obligors or
facilities, as mentioned in Section 3.2.

40. In accordance with Article 179(2)(a) of the CRR, where an institution uses data
that are pooled across institutions the rating systems and criteria of other
institutions in the pool must be similar to its own. To comply with this
requirement an institution should, among other things:

(a) ensure that there is a common definition of the key drivers and processes;

(b) ensure that policies and procedures considered for human judgement,
including overrides 21, can be applied in a comparable and similar manner
across all participating institutions.

20
The paragraphs below are also relevant in cases where institutions use pooled data from institutions
belonging to the same banking group.
21
Article 172(3) of the CRR.

ECB guide to internal models – Credit risk 68

3.5 Use of purchased rating systems or models (pool models 22 23)

41. In accordance with the last sentence of Article 144(1) of the CRR, the
requirements to use an IRB approach, including own estimates and CCFs,
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. To
comply with this provision, institutions should ensure in such cases that all
relevant internal information for model development and parameter calibration
is taken into account. In particular, LRAs of default rates, LGD and CCFs based
only on internal data should always be computed and considered for calibration.
The institution remains responsible for the performance of the rating system or
model.

42. In addition, to ensure the integrity of the rating systems or internal models when
institutions make use of pool models, and to comply with Article 144(1) of the
CRR, the principles set out below should be followed.

(a) If PD estimates are calculated using pooled data, institutions should verify
that the data used for risk quantification meet the data requirements for
default rate calculation as clarified in paragraph 80 below, or that the data
are adjusted accordingly.

(b) Where several institutions use a common pool model, each should ensure
that its rating process is aligned to the extent that all input risk drivers are
defined in the same way across all participating institutions. The
institutions should also ensure that all assessments of the qualitative
components of the rating model are performed in a comparable manner.

(c) If a pool model is used for the estimation of risk parameters and the
model-relevant parts of the process for managing distressed obligors
(including the strategy before and after default) of the participating
institutions are not aligned, these differences should be appropriately
taken into account within the model or through an appropriate adjustment,
in accordance with paragraph 37(a)(viii) of the EBA GL on PD and LGD. In
the case of a pool model 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.

(d) Institutions should ensure that all relevant internal information with respect
to the creditworthiness of an obligor is taken into account and the rating is
updated with new information in a timely manner.

22
A “pool model” is deemed to be a model where institutions develop a shared or common rating model
based on pooled data which is then applied by each participating institution to its portfolio(s).
Institutions which pool their data may work together very closely, disclosing to each other more
information than simply publicly available external data, and even sharing the same rating and
validation processes.
23
The paragraphs below are also relevant in cases where institutions use pooled data that are generated
from institutions belonging to the same banking group.

ECB guide to internal models – Credit risk 69

 (e) Each institution should remain responsible for the performance of the
rating model on its own portfolio.

43. To ensure that its ratings systems are operating properly on an ongoing basis, if
an institution introduces systematic adjustments to the outputs of the pool
model, the institution concerned should initiate internal procedures to analyse
whether significant weaknesses in the model exist and whether a model change
needs to be triggered.

3.6 Consistency in the definition of default

44. In accordance with Article 178(4) of the CRR, institutions that use external data
that are not in themselves consistent with the definition of default laid down in
paragraph 1 of that Article must make appropriate adjustments to achieve broad
equivalence with the definition of default. To comply with this requirement,
institutions should ensure that when they make use of external data or pooled
data they have a complete understanding of the definition of default applied to
these data and perform a comparison between the definition of default used
and the requirements of Article 178 of the CRR. If there are differences between
the definition of default applied in the external or pooled data and the
institution’s own definition of default, the institution should assess the
differences and describe the adjustments made to the risk estimates, in order to
achieve the required level of consistency with the internal definition of default. It
should also include an appropriate MoC to account for the adjustments
included. These adjustments should be appropriately documented and justified,
in particular by providing reasonable assurance that they do not undermine the
validity of the approach for the purposes of risk differentiation and risk
quantification.

3.7 Use of human judgement

45. In accordance with Article 171(1)(a) of the CRR, institutions must have specific
definitions, processes and criteria for assigning exposures to grades or pools.
The grade and pool definitions must be sufficiently detailed. To comply with this
provision, institutions should ensure that, when human judgement is used in the
assignment of exposures to grades or pools, there is a framework in place that
establishes clear and detailed guidelines and procedures on the application of
human judgement (e.g. through the use of pre-defined questionnaires). The use
of human judgement should be documented in a way that ensures the rating
assignment can be understood and replicated by a third party. 24 To ensure the
replicability and consistency of the rating assignment process, the principles set
out below should be followed.

24
Article 171(1)(b) of the CRR.

ECB guide to internal models – Credit risk 70

 (a) The basic structure of the model 25 should be applied consistently and not
modified by human judgement. It should, in particular, consist of:

(i) the minimum data sources to be used for the grade/pool assignment;

(ii) the complete list of pre-defined risk drivers which need to be

assessed individually and for which the assessments need to be
stored;

(iii) the importance of and the evaluation criteria for the risk drivers,
particularly the functional relationship 26 between risk drivers and the
assigned rating (before overrides);

(iv) the procedures and steps to be followed during the assessment,

including the steps where overrides can be applied.

(b) Institutions should assess the consistency of the rating assignment

process. To this end, they should define the tests and analyses in a way
that is proportionate to the degree of human judgement applied. In cases
where human judgement is the main component of the rating assignment
process, these tests should at least include an analysis of consistency in a
representative sample by having obligors re-rated independently by
different analysts. The results of the ratings consistency assessment
should be analysed against pre-defined thresholds as part of the
framework for review of estimates.

46. In accordance with Article 172(3) of the CRR, for grade and pool assignments
institutions must document the situations in which human judgement may
override the inputs or outputs of the assignment process and the personnel
responsible for approving these overrides. In the context of this requirement,
the ECB understands that an override is a judgement-based and discretionary
action that contributes to the assessment of the obligor’s creditworthiness as
obtained through pre-defined components of the rating system. These comprise
both automatic/quantitative components 27 and qualitative variables 28. An
override constitutes a direct intervention to an input or to an intermediate 29 or
final output of the rating assignment process, through an adjustment or
replacement of that input or intermediate/final output which is obtained under
the regular assignment process. Overrides should be limited to information
relevant to the obligor’s creditworthiness, if this is not captured well by the pre-
defined components of the model.

25
This concept is not intended to refer to pure statistical models and can encompass other methods for
assigning exposures to grades or pools.
26
This refers to a relationship between different variables in order to form an output (and not necessarily
a mathematical formula).
27
These components also include the “forcing rules”, i.e. automatic adjustments to the rating carried out
when certain conditions apply (e.g. more than 30 days past due).
28
As referred to in paragraph 201(a) of the EBA GL on PD and LGD.
29
Intermediate outputs should be defined in the model specifications (e.g. results of the financial
scorecard of a model).

ECB guide to internal models – Credit risk 71

47. Within a rating assignment process, the use of pre-defined risk drivers in the
form of qualitative model inputs should be distinguished from overrides. These
judgement-based or qualitative variables are expected to be considered under
the regular assignment process and as part of the pre-defined components of
the rating system. Therefore, they are not themselves considered as
overrides. 30

48. Whenever the functional relationship between the risk drivers and the assigned
rating as specified in paragraph 45(a)(iii) is circumvented, for example in cases
where the last-assigned rating deviates from the rating that would be assigned
on the basis of this functional relationship, it is considered to be an override. For
the purposes of Article 172(3) of the CRR, it should therefore be specified in the
policies and criteria for the use of overrides in the rating assignment process.

49. In accordance with Article 174(e) of the CRR, the results of the statistical model
must be complemented by human judgement, especially by taking into account
all information not included in the model. The higher the number of relevant
observations, the more the institution should rely on the outcomes of the
statistical model.

50. In addition, when human judgement is used for the purpose of model
development, for example in setting the model’s assumptions, the identification
of risk drivers and determination of their weights, or the identification and
combination of model components, there is a risk of the model-based
assignments being inaccurate. 31 To mitigate this risk, institutions should ensure
that the incorporation of human judgement is appropriately managed and
proportionate to the number of relevant available observations.

51. For the purposes of quantifying the risk parameters to be associated with
grades or pools, estimates must not be based purely on judgemental
considerations. 32 To this end, where human judgement is used to a greater
extent because of the low number of relevant available observations,
institutions should apply a higher MoC to their estimates to account for
additional uncertainty.

52. In addition, whenever human judgement is used in the estimation of risk

parameters (for either risk differentiation or risk quantification purposes)
institutions are expected to have in place a framework under paragraph 35 of
the EBA GL on PD and LGD.

30
However, these qualitative input variables can be subject to overrides in accordance with
paragraph 201(b) of the EBA GL on PD and LGD.
31
Article 174(e) of the CRR.
32
Article 179(1)(a) of the CRR.

ECB guide to internal models – Credit risk 72

4 Probability of default

4.1 Structure of PD models

4.1.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 144 (1)(a), (e)

161 (3)

169 (1), (2)

(1)(a) to (f), (2), (3)(a) to (c),

170
(4)

171 (2)

172 (1)(a), (d)

173 (1)(b)

174 (1)(a), (c)

179 (1)(a)

180 (1)(a), (g), (2)(a)

201, 202, 203, 236

EBA GL on PD and LGD 20/11/2017 20-27, 56-69, 96, 97, 98(b)

Other references

Final Draft RTS on assessment methodology for IRB 21/07/2016 24 (3)(c)

34 to 38, 41

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

53. In accordance with Article 179(1)(a) of the CRR, estimates must be based on
the material drivers of the risk parameters. The relevant material risk drivers
and rating criteria may be taken into consideration in several ways:

(a) when assigning exposures to different PD models;

(b) at a PD model level when assigning exposures to different ranking/scoring

methods;

(c) as explanatory variables in ranking/scoring methods;

(d) as drivers in the process for the assignment of PDs to grades or pools
(e.g. calibration segments).

54. When choosing the risk drivers for the models, there is a risk that risk drivers
that capture the characteristics of defaulted obligors could be inappropriately
inferred as relevant risk drivers for the portfolio. To mitigate this risk, institutions
should take appropriate measures against model misspecification with regard to

ECB guide to internal models – Credit risk 73

 overfitting. This is particularly relevant where default data for the development
of the model are scarce.

55. In accordance with Article 144(1)(a) of the CRR, institutions’ rating systems
must provide for a meaningful assessment of obligor and transaction
characteristics, a meaningful differentiation of risk and accurate and consistent
quantitative estimates of risk. To comply with this requirement, it is the ECB’s
understanding that PD models should perform adequately on economically
significant and material sub-ranges of application. 33 The sub-ranges are
identified by splitting the full range of application of the PD model into different
parts on the basis of potential drivers for risk differentiation, including the
following non-exhaustive list of drivers, 34 where relevant:

(a) for PD models covering exposures to small and medium-sized enterprises

(SMEs): country, industry (e.g. statistical classification of economic
activities in the European Community (abbreviated as NACE 35) code
section classification A to U), size of obligor (e.g. different buckets in terms
of total assets), past delinquency (e.g. obligors with delinquency events,
i.e. days past due, in the last 12 months);

(b) for PD models covering retail exposures: client type (e.g. high net
worth/private banking, other individuals, self-employed, SMEs), product
type (e.g. consumer credit, credit card, other), region (e.g. nomenclature of
territorial units for statistics (NUTS) 1, 2 or 3 as defined by Eurostat), past
delinquency (e.g. obligors with delinquency events, i.e. days past due, in
the last 12 months), maturity (e.g. original or remaining maturity);

(c) for PD models covering retail exposures secured by real estate: region
(e.g. NUTS 1, 2 or 3 as defined by Eurostat), type of real estate
(e.g. residential, commercial, other), past delinquency (e.g. obligors with
delinquency events, i.e. days past due, in the last 12 months), maturity
(e.g. original or remaining maturity);

(d) for PD models covering exposures to financial institutions: business model

(deposit-taking institutions, investment banking, insurance firms, other),
jurisdiction (or global region as appropriate) and size (defined buckets of
total assets);

(e) for PD models covering exposures to large corporates: industry

(e.g. NACE code section classification A to U), country (or global region as
appropriate) and size (defined buckets of total turnover).

33
Where an institution has approval for a PD model on a consolidated basis as well as on a sub-
consolidated and/or individual basis, it is the ECB’s understanding that this PD model should perform
adequately at these sub-consolidated and/or individual levels, as these are considered material sub-
ranges of application.
34
When external credit bureau scores or ratings are used as the main (or one of the main) driver(s) of the
internal rating, the set of all exposures for which the external score or rating is not available should also
be considered a significant sub-range of application.
35
Nomenclature statistique des activités économiques dans la Communauté Européenne.

ECB guide to internal models – Credit risk 74

 56. In accordance with Article 169(1) of the CRR, where an institution uses multiple
rating systems, the rationale for assigning an obligor or a transaction to a rating
system must be documented and applied in a manner that appropriately reflects
the level of risk. To comply with this requirement institutions should, in terms of
the range of application of a PD model:

(a) clearly describe its range of application (and sub-divisions into different
ranking/scoring methods and calibration segments) and also include an
explanation of the risk drivers which the institution considered when
designing the process but has decided not to use for the assignment of the
obligor to the rating system;

(b) ensure that there are no overlaps in the range of application of different PD
models and that each obligor or facility to which the IRB approach should
be applied can be clearly assigned to one particular PD model.

4.1.2 Risk differentiation

Principles for all model types

57. Article 170 of the CRR lays down requirements related to the structure of rating
systems. To comply with these requirements and with reference to Articles 36 to
38 of the Final Draft RTS on assessment methodology for IRB, institutions
should, among other things, ensure a meaningful differentiation of risk over time
which takes into account (i) the distribution of obligors or facilities; (ii) the
homogeneity of obligors or facilities assigned to the same grade or pool; and
(iii) the different levels of risk across obligors or facilities assigned to different
grades or pools to which a different PD is applied.

58. To ensure that the PD model performs adequately in terms of risk

differentiation, institutions should adopt the following approach.

(a) Define metrics (considering both their evolution over time and specific
reference dates) with well-specified targets, taking into account tolerance
levels that reflect the uncertainty of the metrics, and take action, where
necessary, to rectify any deviations from these targets that exceed the
tolerance levels. Separate targets and tolerances may be defined for initial
development and ongoing performance.

(b) Ensure that the tools used to assess risk differentiation are sound and
adequate considering the available data. The risk differentiation is
expected to be demonstrated on time series of realised default rates for
grades or pools under different economic conditions.

ECB guide to internal models – Credit risk 75

Principles specific for grades and pools

59. A grade or pool is understood by the ECB as the subset of obligors or facilities
to which the same PD is applied for the calculation of regulatory capital
requirements, irrespective of how this PD has been assigned (e.g. through the
use of masterscales).

Distribution of obligors or facilities across grades or pools

60. Articles 170(1)(c) and (d) and 170(3)(b) and (c) of the CRR require, among
other things, that the number of grades and pools is adequate to achieve
meaningful risk differentiation and quantification of the PD at the grade or pool
level. To comply with this requirement, institutions should:

(a) justify the criteria applied when determining the number of grades or pools
and the proportion of obligors or facilities assigned to each;

(b) ensure that the concentration of numbers of obligors or facilities is not

excessive in any grade or pool; any significant concentrations should be
supported by convincing empirical evidence of the homogeneity of risk for
those obligors or facilities;

(c) ensure that no grade or pool has too few obligors or facilities, unless this is
supported by convincing empirical evidence of the adequacy of the
grouping of the exposures in question.

Homogeneity within grades

61. Articles 170(1)(b) and (d) and 170(3)(b) and (c) of the CRR require, among
other things, that the structure of rating systems must ensure the homogeneity
of obligors or facilities assigned to the same grade or pool. In accordance with
this requirement and under paragraph 69 of the EBA GL on PD and LGD:

(a) homogeneity is understood as obligors or facilities assigned to a grade

having a reasonably similar default risk to ensure that the grade-level
default rate is representative of all obligors or facilities in that grade;

(b) in cases where it is found (through the use of additional drivers or a

different discretisation of the existing ones) that a material subset of
obligors or facilities within a grade/pool yields a significantly different
default rate to that of the rest of the grade or pool, this is considered to
indicate a lack of homogeneity.

ECB guide to internal models – Credit risk 76

 Risk differentiation across grades or pools

62. To comply with the requirement to ensure adequate risk differentiation across
grades or pools, 36 institutions should ensure that there are no significant
overlaps in the distribution of the default risk between grades or pools. This
should be ensured through a meaningful differentiation of the default rates of
each grade.

Principles specific for direct estimates

63. See paragraph 90.

4.1.3 Grade assignment dynamics

64. In order to ensure a meaningful assessment of obligor characteristics, 37 when

assigning obligors or facilities to a grade or pool institutions should follow
paragraphs 66 to 68 of the EBA GL on PD and LGD. Although the time horizon
used in PD estimation is one year, it is the ECB’s understanding that the
rating/grade/pool assignment process should also adequately anticipate and
reflect risk over a longer time horizon and take into account plausible changes
in economic conditions. In order to achieve this objective:

(a) all relevant information should be included in the rating/grade/pool

assignment process, giving an appropriate balance between drivers that
are predictive only over a short time horizon and drivers that are predictive
over a longer time horizon;

(b) a horizon of two to three years is considered to be appropriate for most

portfolios;

(c) in accounting for plausible changes in economic conditions, the institution

should consider at least past observed default patterns;

(d) the model should perform under different economic conditions.

As a consequence of the above, institutions’ grade assignment dynamics

should also adequately anticipate and reflect in the assignment of grades the
risk over the longer time horizon. For clarity, this does not mean that grades
remain stable over the longer time horizon in the event of changes in the risks
that are specific to the obligor.

65. Additionally, the following principles apply under the specific situations
considered in (a) to (c) below:

36
As required by Articles 170(1)(b) and (d) and 170(3)(b) and (c) of the CRR.
37
Article 144(1)(a) of the CRR.

ECB guide to internal models – Credit risk 77

 (a) when using external scores or ratings (e.g. from an external bureau or
external rating agency) as drivers for the purpose of risk differentiation
within a specific model, institutions should identify the grade assignment
dynamics embedded in the external rating and understand the effect on
their own grade assignment dynamics, considering the other risk drivers
used;

(b) when using external ratings as target variables for the purpose of risk
differentiation within a specific model (see Section 4.1.5), institutions
should preserve their own grade assignment dynamics by taking the
appropriate measures when necessary;

(c) when mapping internal grades to external grades in order to use external
default rates to estimate PD, institutions should ensure that the grade
assignment dynamics of the external ratings are sufficiently similar to their
own internal grade assignment dynamics, or perform the necessary
adjustments during risk quantification to compensate for any differences.

4.1.4 Use of ratings of third parties

66. In accordance with Article 172(1)(a) of the CRR, institutions must assign each
obligor to a grade or pool as part of the credit approval process. To comply with
this requirement, institutions should have a clear policy 38 specifying the
conditions under which the rating of a third party which has a contractual or
organisational relationship with an obligor of the institution (third-party support)
may be taken into account in the risk assessment of that obligor. This policy
should meet the following criteria.

(a) It should specify in which situations the rating of a parent entity could be
taken into account in the risk assessment of other entities of the group. In
particular, the policy should specify those situations in which obligors are
assigned to a better grade than their parent entities.

(b) It should include provisions on the use of ratings of third parties that
provide contractual support to more than one obligor. As a general rule,
the policy should include, but not be limited to, possible prioritisation,
eligibility, and the impact on the rating of the supporting third party.

67. Articles 201 to 203 of the CRR establish requirements for the eligibility of
unfunded credit protection. To comply with these requirements, institutions may
recognise the guarantee by applying the risk weight of the guarantor under the
standardised approach to the covered part of the exposure, provided that no
own estimates of LGD and CCFs are used (foundation IRB (F-IRB)). This
applies when an obligor is guaranteed by a third party that is not in the range of
application of a PD model and the guarantee fulfils all requirements for credit

38
As part of the policies mentioned in paragraph 62 of the EBA GL on PD and LGD.

ECB guide to internal models – Credit risk 78

 risk mitigation (CRM), consistently with paragraph 44 of the EBA Report on the
CRM Framework. In such situations, under paragraph 74 of the EBA GL on PD
and LGD, the guaranteed obligor should be included in the calculation of the
one-year default rate of the grade the obligor is assigned to, before the
recognition of the guarantee.

68. In addition, when the institution reflects substitution effects 39 arising from CRM
in the ratings assigned to a material number of exposures within a rating
system, there is a risk that the process of assigning exposures to grades or
pools might not provide for a meaningful differentiation of risk, as a result of the
inclusion of obligors with significantly different risk levels within the same rating
grade. 40 To mitigate this risk, institutions should verify that obligors guaranteed
by a third party do not carry a significantly different level of risk from those in the
same rating grade without such a guarantee, and that no separate calibration
segment as referred to in paragraph 97 of the EBA GL on PD and LGD is
required.

69. When, under paragraph 62(a) of the EBA GL on PD and LGD, an institution
performs a rating transfer across different rating systems that do not share the
same obligor rating scale, it should ensure that the mapping between rating
scales is performed in such a way that the final PD estimate (including MoC)
assigned to the guaranteed exposure amount is not better than the final PD
estimate (including MoC) being transferred from a third party. Article 171(2) of
the CRR establishes that information used to assign obligors and facilities to
grades or pools must be current. To comply with this requirement, if a material
proportion of exposures or obligors within a rating system receives a rating from
another IRB rating system as a result of rating transfers, institutions should
ensure that the transferred ratings are automatically updated when the rating of
the third party changes or when the PDs of the rating system to which the third
party belongs are re-estimated.

70. In accordance with Article 179(1)(a) of the CRR, estimates must be plausible
and intuitive and must be based on the material drivers of the respective risk
parameters. To comply with this requirement, institutions should have sufficient
empirical evidence to justify situations where an obligor has an equal or better
PD estimate than the third party providing support as a consequence of the
treatments specified in paragraph 62(c) of the EBA GL on PD and LGD. In
addition, differences between the various forms of contractual support should
be considered in the PD models, unless there is sufficient empirical evidence
that these differences are not relevant risk drivers. This understanding should

39
Substitution effects are understood as: the application of the treatment set out in Article 236 of the CRR
(i.e. the possibility to replace the PD of the obligor with the PD of the protection provider, or with a PD
between that of the borrower and that of the guarantor); or the recognition of a guarantee by applying
the risk weight of the guarantor under the standardised approach to the covered part of the exposure,
as described in paragraph 67 of this chapter.
40
In accordance with Article 170(3)(c) of the CRR, the process of assigning exposures to grades or pools
must provide for a meaningful differentiation of risk, for a grouping of sufficiently homogenous
exposures, and must allow for accurate and consistent estimation of loss characteristics at grade or
pool level.

ECB guide to internal models – Credit risk 79

 also be taken into account if the rating of the third party is being considered as
an indication for an override under paragraph 62(b) of the EBA GL on PD and
LGD.

71. In addition, in the situation described in paragraph 62(b) of the EBA GL on PD

and LGD, where a rating of a third party is being taken into account as an
indication for an override of the assignment of the relevant obligor to a grade or
pool, institutions should not assign a rating to an obligor that is better than the
rating of the third party as a consequence of an override resulting solely from
the existence of this third-party support.

72. Furthermore, when third-party support is used extensively in the scope of

application of a PD model, institutions should consider its existence as a
potential relevant driver for risk differentiation, in accordance with Section 4.1.2.

4.1.5 Use of shadow rating models

73. The ECB understands a shadow rating model (SRM) to be an internal rating
approach that selects and weighs the risk drivers to be used for risk
differentiation purposes by identifying the main factors that explain external
ratings provided by an external credit assessment institution or similar
organisation, rather than internal directly observed defaults.

74. In accordance with Article 144(1)(e) of the CRR, institutions must document the
rationale for their rating systems. To comply with this requirement, institutions
should justify and document the rationale for the use (and the continued use) of
the SRM, instead of the internal default prediction model, and also document
the alternative approaches that have been considered, in accordance with
Article 41 of the Final Draft RTS on assessment methodology for IRB. In
addition, and without prejudice to the risk differentiation requirements, when
developing the model institutions should set explicit threshold criteria in terms of
capacity to explain the target ratings and take appropriate action when those
thresholds are not met.

75. Assignment criteria and processes must be periodically reviewed to determine

whether they remain appropriate for the current portfolio and external
conditions. 41 To comply with this requirement, as part of the review of estimates
institutions should take all reasonable steps to demonstrate how the model
performs on the application population in terms of predicting defaults or, if that
is not possible (when there are not enough internal default data), at least in
terms of predicting the target ratings.

76. In accordance with Article 170(1)(b) of the CRR, institutions’ rating systems
must have an obligor rating scale which reflects exclusively the quantification of
the risk of obligor default. To this end, institutions should adjust the ratings used

41
Article 169(2) of the CRR.

ECB guide to internal models – Credit risk 80

 as targets for their SRMs if they do not solely embed default risk. They should
also document such adjustments.

77. In accordance with Article 174(1)(a) of the CRR, when an institution uses a
statistical model and other mechanical methods to assign exposures to obligors
or facilities, the input variables must form a reasonable and effective basis for
the resulting predictions. To comply with this requirement, when the institution
uses an SRM external ratings should not be used as risk drivers in addition to
target variables.

78. When assigning obligors and facilities to grades or pools institutions must take
all relevant information into account. 42 To comply with this requirement, when
different information sources are used institutions should ensure that they
understand the impact of any differences between these sources and establish
adequate procedures to ensure that these differences are adequately
addressed.

79. Furthermore, the data used to build the model must be representative of the
population of the institution's actual obligors or exposures. 43 To comply with this
requirement, institutions should analyse and provide evidence of the
representativeness of the data used for model development consistently with
paragraphs 20 to 27 of the EBA GL on PD and LGD.

42
Article 171(2) of the CRR.
43
Article 174(c) of the CRR.

ECB guide to internal models – Credit risk 81

4.2 PD risk quantification

4.2.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 4 (1)(78)

144 (1)(a)

169 (3)

170 (1)(b)

179 (1)(b)

180 (1)(a), (f), (h), (2)(a), (e)

185 (b)

EBA GL on PD and LGD 20/11/2017 70-99

Other references

Final Draft RTS on assessment methodology

21/07/2016 48, 49
for IRB

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

80. In accordance with Article 180(1) of the CRR, for exposures to corporates,
institutions, central governments and central banks and for equity exposures,
institutions must estimate PDs by obligor grade from the LRA of one-year
default rates. In accordance with Article 180(2) of the CRR, for retail exposures,
institutions must estimate PDs by obligor/facility grade or pool from LRAs of
one-year default rates. To comply with these requirements, institutions should
follow Sections 4.2.2 to 4.2.6 below.

4.2.2 Calculation of observed average default rates

81. For the calculation of the one-year default rate and observed average default
rates, institutions should follow paragraphs 73 to 81 of the EBA GL on PD and
LGD, also considering the following.

(a) It is the understanding of the ECB that, for retail exposures and when the
definition of default is applied at facility level, paragraphs 73 to 81 of the
EBA GL on PD and LGD can be applied facility level.

(b) Exposures for which there is no commitment (considering on-balance

sheet exposures, off-balance sheet items and unadvised limits) at
reference date should be excluded from the calculation of the default rate.
Conversely, if there is an EAD estimate, then these exposures should be
included in the calculation of the default rate.

ECB guide to internal models – Credit risk 82

 82. For the purpose of choosing an appropriate approach under paragraph 80 of
the EBA GL on PD and LGD, it is the ECB’s understanding that overlapping
one-year time windows should preferably be used when the analysis performed
by the institution under paragraphs 80(a), 80(b) and/or 80(d) of the EBA GL on
PD and LGD reveals any of the following:

(a) the proportion of short-term and terminated contracts and/or the respective
distribution of default rates is not stable over time;

(b) the observed average default rate using overlapping one-year time
windows is significantly different from the observed average default rate
using non-overlapping one-year time windows;

(c) there is a significant variation between the observed average default rates
calculated using non-overlapping one-year time windows on different
reference calculation dates within a year.

83. Institutions should estimate PDs taking their own internal data into
consideration. In cases where institutions use external or pooled data series to
complement their internal data for the purpose of PD estimation, the more
internal default experience an institution has, the less importance it needs to
give to external data. In addition, institutions should ensure that these external
or pooled data series are representative in accordance with Section 3.2 of this
guide. 44 To comply with the requirement for the data to be representative and as
part of their representativeness analysis, institutions should also ensure that the
average observed default rates from external data or from the external part of
the pooled data are calculated separately from, and compared with, those
based on internal data. 45 This comparison should be made at the levels at
which the default rate is to be calculated. This means, in accordance with
paragraph 79 of the EBA GL on PD and LGD, that the comparison should be
made for each rating grade or pool and for the type of exposures covered by
the relevant PD model, as well as for any relevant calibration segment. The
direction and magnitude of the differences between these averages should be
properly analysed and documented when calibrating the model, including the
need and adequacy of the category A MoC considered with regard to
paragraph 37(a)(viii) of the EBA GL on PD and LGD, and duly followed up in the
review of estimates.

4.2.3 Calibration to the LRA default rate

84. To calculate the LRA default rate, institutions should follow paragraphs 82 to 86
of the EBA GL on PD and LGD. If an institution makes adjustments to the

44
See Article 179(1)(d) of the CRR.
45
If the internal data constitute just a small fraction of the pooled data and are not considered material in
relation to the pooled data, for the purposes of this analysis the institution may perform a separate
calculation of the average observed default rates with pooled data and a comparison with those
calculated based on internal data only.

ECB guide to internal models – Credit risk 83

 observed average default rates in order to obtain LRA default rates under
paragraph 85(b) of the EBA GL on PD and LGD, these adjustments should be
based on (external) default rates, or – if no appropriate default rates are
available – on other observed indicators relevant for the type of exposures
considered.

85. For the purpose of assessing the representativeness of the historical

observation period used for the likely range of variability of one-year default
rates under paragraph 83 of the EBA GL on PD and LGD, the following should
be taken into account.

(a) Where the scarcity of internal exposures and/or defaults might unduly
influence the variability of internally observed default rates (i.e. where the
variability driven by statistical uncertainty is so high in comparison with the
structural variability of default rates that it hampers any analysis of them),
institutions should assess whether external or pooled default rate series
can be used to identify the relevant historical observation period for the
likely range of variability of one-year default rates. The external or pooled
default series used should be relevant for the specific portfolio at least in
terms of geographical composition, sectoral distribution and other relevant
risk drivers. When no relevant default rate series can be identified, the
items described in paragraphs 83(b) and 83(c) of the EBA GL on PD and
LGD should play a crucial role in the assessment.

(b) When taking into account the existence of one-year default rates relating
to bad years as reflected in economic indicators that are relevant for the
considered types of exposures within the historical observation period as
referred to in paragraph 83(b) of the EBA GL on PD and LGD, institutions
should ensure that such indicators are relevant for the portfolio at least in
the terms of geographical composition, sectoral distribution and other risk
drivers relevant to the portfolio.

86. With respect to calibration to the LRA default rate, institutions should follow
paragraphs 87 to 99 of the EBA GL on PD and LGD. If an institution chooses
the approach referred to in paragraph 92(b) of the EBA GL on PD and LGD, it
should perform additional tests as part of the development and ongoing
monitoring of its models to ensure that the final (post-calibration) PDs reflect the
LRA default rate of each grade. Specifically, institutions should ensure that
there are no systematic deviations when comparing the estimated PDs with the
LRA default rate of the grades, i.e. the direction of divergences across grades
should be random.

4.2.4 Weighting for retail exposures

87. Notwithstanding paragraph 84, for retail exposures institutions need not give
equal importance to historical data if more recent data are a better predictor of

ECB guide to internal models – Credit risk 84

 loss rates. 46 In the understanding of the ECB an institution may consider that
the more recent data are a better predictor of loss rates and may give more
importance to recent historical data if the following apply.

(a) There is a significant improvement in the predictive power when using the
more recent data with respect to the predictive power resulting from the
use of an arithmetic average under paragraph 81 of the EBA GL on PD
and LGD. This improvement should be evidenced by comparing the
estimated PDs for each grade with the realised default rates covering as
long a period as possible, in accordance with Article 185(b) of the CRR.

(b) Older data are considered as non-representative as a result of specific

policy or business changes in the bank, but not in order to reflect current
trends in default rates directly related to macroeconomic conditions.

(c) The weighting approach is used in a consistent manner over time and any
change in the applied weights of historical data is appropriately justified.

4.2.5 PD quantification based on mapping to external grades

88. The ECB interprets the possibility for institutions to attribute the default rate
observed for the grades of a rating agency or similar organisation to its own
grades in accordance with Article 180(1)(f) of the CRR as being equivalent to
the use of external data for PD quantification at a more aggregated level
(external grade) rather than at the obligor/facility level. Accordingly, Sections 3.2
and 4.2.2 of this chapter are relevant for institutions that do so.

89. In accordance with Article 180(1)(f) of the CRR, mappings must be based on a
comparison of internal rating criteria with the criteria used by the external
organisation and on a comparison of the internal and external ratings of any
common obligors. Biases or inconsistencies in the mapping approach or
underlying data must be avoided. To comply with these requirements,
institutions should follow the paragraphs listed below.

(a) Institutions should ensure that the quality of the mapping between internal
and external rating scales at a given date and over time is consistent and
provides for an adequate level of predictive ability.

(b) When mapping internal grades to external grades, institutions should

document and analyse any differences between the external and internal
rating criteria.

(c) The use of common obligors as a basis for the mapping should take into
account their representativeness for the application portfolio.

46
Article 180(2)(e) of the CRR.

ECB guide to internal models – Credit risk 85

 (d) Institutions should adjust the external rating scale if such rating scale does
not solely embed default risk. 47 They should also document such
adjustments.

(e) When mapping internal grades to external grades and using the default
rates of the external grades provided by the organisation, if the latter has a
material number of entities for which it no longer provides a rating
(withdrawn rating), the institution should take this into account. It should
adjust the external default rates accordingly, if necessary, and take into
consideration the provisions of paragraph 75 of the EBA GL on PD and
LGD. In the event that an adjustment is performed, the institution should
add the necessary MoC.

4.2.6 Specific requirements for direct PD estimates

90. In order to use direct PD estimates for the calculation of own funds
requirements in accordance with Article 169(3) of the CRR, institutions should
follow paragraphs 96 and 98(b) of the EBA GL on PD and LGD. To assess
whether the theoretical assumptions of the probability model underlying the
estimation methodology are satisfied to a sufficient extent in practice under
paragraph 96 of the EBA GL on PD and LGD, institutions should do the
following.

(a) Ensure good risk differentiation properties across the full PD range of the
rating system.

(b) Have an adequate and documented concept in place specifying the

calibration function currently implemented (concrete functional form),
including the underlying theoretical assumptions and the established
processes to conduct the PD calibration. It is the ECB’s understanding that
institutions should ensure consistency between the score-inferred PDs and
the observed default rates and should understand and justify the
transformation of the scores into PD values.

(c) Ensure that any transformation of the scores resulting from the probability
model that is applied during the calibration does not change the ranking of
the obligors/facilities (in other words, co-monotonicity between scores/raw
PDs and PD values should be ensured). Moreover, institutions should
avoid any undue influence of extreme values of score-inferred PDs on the
shape of the calibration function. Additionally, and when institutions use
different calibration functions for different sub-ranges, they should ensure
that this mix is appropriate (both in terms of the functional forms used and
the cut-offs selected) and that it is appropriately justified.

47
In accordance with Article 170(1)(b) of the CRR, institutions’ rating systems must have an obligor rating
scale which reflects exclusively the quantification of the risk of obligor default.

ECB guide to internal models – Credit risk 86

 (d) Ensure that there is a relevant number of observations across the whole
range of score-inferred PDs. Particular interest should be paid to situations
where the probability model is extended to ranges of PD values where
there are not enough defaulted observations.

(e) Ensure that there are no excessive concentrations of exposures or

obligors within the PD range of the rating system. In addition, high
concentrations of observations in a specific range of score-inferred PDs
should be properly analysed and justified in terms of homogeneity.

(f) For the purpose of performing the additional tests at grade level referred to
in paragraph 92(b) of the EBA GL on PD and LGD, grades should be
understood as sub-ranges of PD values. These sub-ranges should be
defined in a way that:

(i) represents sufficiently narrow ranges of PD values;

(ii) contains a sufficient number of observations to ensure a meaningful

calculation of the LRA default rate of the sub-range.

91. In cases where institutions map the PDs to a masterscale (defined in terms of
PD bounds) as a final step in the PD estimation process (using masterscale
discrete PDs for the purpose of RWEA calculation), there is a risk that the
mapping process could distort RWEAs. To mitigate this risk, institutions should
verify that deviations between the masterscale PDs and the average of the
direct PDs assigned to obligors in each grade do not show a systematic or
material bias towards underestimation of PD per grade over time. This analysis
should be provided for both the portfolio and for each grade.

ECB guide to internal models – Credit risk 87

5 Loss given default

5.1 Realised LGD

5.1.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 4 (1)(55)

5 (2)

144 (1)(e)

161 (2), (3)

164 (2)

166 (1)

174 (c)

175 (1), (4)(a)

176 (4), (5)

(1)(a), (c), (d) and second

179
sub-paragraph

(1)(a), (h), (i), (j), (2)(b) and

181
second sub-paragraph

182 (3)

183 (2)

185 (a)

191

EBA GL on PD and LGD 20/11/2017 37(a)(viii), 100-103, 131-146

Other references

Final Draft RTS on assessment methodology

21/07/2016 3 (1), (2)(d), (3)
for IRB

11 (2)(a)

17 (1)(a)

30 (1)(a)

31, 32

39 (a)

40 (2)(a), (b)

45 (1)(a), (c), (d), (2)(a), (b)

50

51 (b), (c), (d), (g), (h)

52 (b), (c)
48
EBA Guidelines on the definition of default 18/01/2017 44, 51, 71, 72, 77

48
EBA Guidelines on the application of the definition of default under Article 178 of Regulation (EU)
No 575/2013 (EBA/GL/2016/07), referred to in this guide as the “EBA Guidelines on the definition of
default”.

ECB guide to internal models – Credit risk 88

 Once adopted by the European Commission, the Final Draft RTS on
assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

5.1.2 Reference dataset

92. Under paragraph 102 of the EBA GL on PD and LGD, institutions should
estimate LGDs on the basis of their own loss or recovery experience.
Institutions may supplement their own historical data on defaulted exposures
with external data. The more own loss experience (i.e. the more internal
defaults) an institution has, the less importance it needs to give to external data.
Institutions should ensure that their own historical experience contains a
minimum number of defaults in order to determine whether external data are
sufficiently representative.

93. To ensure that LGD estimations are accurate and are not underestimated as a
result of different external and internal recovery processes, institutions should
place greater importance on comparisons of internal recovery processes with
the recovery processes underlying the external data, in cases where a high
weight is assigned to external data. Where limited representativeness of
external data is found, a category A MoC should be considered, in order to
reflect the uncertainty of the estimation under paragraph 37(a)(viii). Its
magnitude should also be quantified in relation to the weight assigned to the
external data.

94. When institutions use information derived from the market price of defaulted
financial instruments to supplement their internal loss or recovery experience
data, there is a risk of misspecification of their LGD estimates. To mitigate this
risk, institutions should ensure the following:

(a) institutions should verify whether the development sample is

representative of the application portfolio at least in terms of regions and
product type, even when those variables have not been identified as
relevant risk drivers;

(b) losses derived from market prices should be increased to reflect indirect
costs, as specified in paragraph 146 of the EBA GL on PD and LGD.

5.1.3 Calculation of realised LGD

95. Article 4(1)(55) of the CRR defines LGD as the ratio of the loss on an exposure
due to the default of a counterparty to the amount outstanding at default. For
the purposes of Article 181(1)(a) of the CRR, institutions are required to
calculate realised LGD. To comply with this requirement, it is the ECB’s
understanding that institutions should calculate realised LGD under
paragraphs 100 to 103 and 131 to 146 of the EBA GL on PD and LGD. In

ECB guide to internal models – Credit risk 89

 addition, when performing this calculation institutions should follow the
observations in the succeeding paragraphs.

96. In accordance with Article 181(1)(h) of the CRR, institutions must have
estimates of LGD in-default and ELBE on defaulted exposures. To comply with
this requirement, it is the ECB’s understanding that all principles regarding the
calculation of realised LGD should be applied for the estimation of LGD on non-
defaulted exposures and for the estimation of LGD in-default and ELBE on
defaulted exposures, unless mentioned otherwise (that is, if the reference date
is considered instead of the date of default).

97. Where, in the case of retail exposures and purchased corporate receivables,
institutions derive LGD estimates from realised losses and appropriate
estimates of PDs in accordance with Articles 161(2) and 181(2)(a) of the CRR
and under paragraph 103 of the EBA GL on PD and LGD, all the principles
regarding realised LGD should apply to realised losses.

98. Institutions must document the specific definitions of default and loss used
internally and ensure that they are consistent with the definitions set out in the
CRR. 49 To comply with these requirements, institutions should have in place
sufficiently detailed policies and procedures to ensure that the realised LGD is
calculated consistently and accurately, including the implementation of the
definition of economic loss. These policies and procedures should include
sufficiently detailed documentation to allow third parties to replicate the
calculation of realised LGD. To ensure that the policies and procedures are
implemented in an appropriate and adequate manner, the calculation process
should be regularly reviewed by an independent unit.

99. In accordance with Article 181(1)(a) of the CRR and under paragraph 100 of the
EBA GL on PD and LGD, institutions should calculate the realised LGD at
facility level for each default. In exceptional cases, the ECB considers
institutions to be compliant with the requirement to calculate realised LGD at
facility level if they can prove that the recovery is not performed at individual
facility level but at a more aggregated level (for example, several facilities of the
same or different types secured by the same collateral). The realised LGD can
therefore be calculated at a more aggregated level than individual facility level.
For this exceptional deviation from the calculation of realised LGD at facility
level to be acceptable, institutions should:

(a) provide evidence that recovery at aggregated level is legally enforceable.

(b) on a regular basis (as often as review of estimates is performed or more

often), provide evidence that recovery at a more aggregated level than
single facility level is in practice enforced. This evidence should be based
on the institution’s historical practice and data and demonstrate that both
the recovery process and its outcomes in terms of realised loss or
recovery are the same for all the facilities considered at the aggregated

49
Article 175(3) of the CRR.

ECB guide to internal models – Credit risk 90

 level. Specifically, institutions should be able to prove that all collateral
within an aggregation is called irrespective of the product triggering default
(thus, for a current account as for a home loan) and that realised loss or
observed recovery is the same for all types of facility within the
aggregation.

(c) for retail exposures where institutions use definition of default at facility
level in accordance with the last sentence of Article 178(1) of the CRR,
ensure that the default is triggered for all aggregated facilities.

In addition, institutions following this approach should:

(d) ensure that the parameters are applied in a manner that is consistent with
how they have been estimated, i.e. across aggregated facilities;

(e) ensure that the counting unit used for the purposes of risk quantification is
at this aggregated level;

(f) ensure that no bias results from the aggregation of facilities, by validating
the estimates (PD, LGD, CCF) at the more aggregated level also.

100. As mentioned in paragraph 95, for the purposes of Article 181(1)(a) of the CRR
institutions are required to calculate realised LGD, which is defined by
Article 4(1)(55) of the CRR as the ratio of the loss on an exposure due to the
default of a counterparty to the amount outstanding at default. Furthermore,
Article 5(2) of the CRR defines loss as an economic loss, including material
discount effects, and material direct and indirect costs associated with collecting
on the instrument. In accordance with these provisions, it is the ECB’s
understanding that institutions should calculate realised LGD as a ratio of the
economic loss to the outstanding amount of the credit obligation at the moment
of default, including any amount of principal, interest or fee (hereinafter
outstanding amount at default). To calculate realised LGD, institutions should
follow paragraphs 131 to 146 of the EBA GL on PD and LGD. In addition, they
should pay particular attention to the following points.

(a) Outstanding amount at default includes any part of the exposure that has
been forgiven or written off before or at the date of default (paragraph 134
of the EBA GL on PD and LGD). This amount is equal to the accounting
value gross of credit risk adjustment (i.e. “provisions”) (Article 166(1) of the
CRR). This amount also includes interest and fees capitalised in the
institution’s income statement before the moment of default. However,
interest and fees capitalised after the moment of default are not
considered (paragraphs 137 to 138 of the EBA GL on PD and LGD).
Where institutions include additional drawings after the moment of default
to estimate CCFs, these additional drawings discounted to the moment of
default are added to the outstanding amount at default in the denominator
(paragraphs 139 to 142 of the EBA GL on PD and LGD). In other words,
institutions should ensure that the exposure used for CCF estimation,
where additional drawings after default are discounted with the same

ECB guide to internal models – Credit risk 91

 discount factor as applied for LGD, is consistent with the denominator of
the LGD.

(b) Economic loss is calculated under paragraph 132 of the EBA GL on PD

and LGD. This also applies in the specific case of facilities that return to
non-defaulted status, where losses arising from payment delays are
expected to be accounted for as well as the “artificial cash flow” envisaged
by paragraph 135 of the EBA GL on PD and LGD.

(c) When recoveries are not directly observed but calculated on the basis of
the difference between exposure values at two consecutive dates or
derived, even partially, from some other treatment, all assumptions should
be duly justified and clearly documented in order to adequately replicate
the recovery flows that occur during the recovery process in accordance
with letters a) and b) above. Institutions are expected to pay particular
attention to the treatment of interest and fees capitalised after default, the
treatment of additional drawings and the treatment of write-offs.

101. The economic loss as defined in Article 5(2) of the CRR also includes material
discounts. The understanding of the ECB is as follows.

(a) Paragraph 134 of the EBA GL on PD and LGD refers to all losses incurred
through forgiveness or write-off.

(b) Where institutions open new facilities to replace previously defaulted

facilities as part of restructuring or for technical reasons, the economic loss
should include the decrease in the degree of financial obligation arising
from changes in the contractual conditions (i.e. material forgiveness or
postponement of payment of principal, interest, or fees). The amount by
which the financial obligation has diminished should be calculated under
paragraph 51 of the EBA Guidelines on the definition of default. However,
no double-counting of debt forgiveness is intended (for example, in the
case of exposures that return to non-defaulted status, a possible double-
counting through the diminished financial obligation and the artificial cash
flow is not intended).

102. Realised LGD for individual facilities may be zero or lower when it is the actual
result of the recovery process (for example, where additional recoveries offset
the discounting effect and costs). Institutions should, however, pay particular
attention to no-loss exposures, since they may reveal issues with the
calculation of realised losses – for example, costs not being adequately
allocated to recovery processes, or inadequate treatment of amounts forgiven
or written off.

5.1.4 Treatment of multiple defaults

103. For the purpose of LGD estimation and in order to ensure an appropriate
measurement of economic loss as defined in Article 5(2) of the CRR, institutions

ECB guide to internal models – Credit risk 92

 should consider an exposure that returns to normal status and subsequently
defaults in a short period of time as being constantly defaulted from the moment
the first default occurred. This treatment should be applied under
paragraph 101 of the EBA GL on PD and LGD. In addition, institutions should
follow the observations in the following paragraphs.

(a) Paragraph 101 of the EBA GL on PD and LGD envisages the definition of
a period longer than nine months when this is appropriate for the specific
type of exposures and reflects the economic meaning of the default
experience. It is the ECB’s understanding that a longer period is adequate
when the proportion of subsequent defaults occurring on individual
facilities over a period of more than nine months is significant, unless
institutions are able to provide evidence that the second (or subsequent)
default is unconnected with the original default event. This evidence may
include analysis of the curing process.

(b) Time considered between two defaults is conditional upon the existence
and length of probation periods. Where institutions have not adopted the
minimum three-month probation period, or any equivalent or longer
probation period, on non-distressed restructured facilities under
paragraph 71 of the EBA Guidelines on the definition of default, they
should consider a 12-month period for the application of paragraph 101 of
the EBA GL on PD and LGD. Where they have not adopted the minimum
12-month probation period, or any equivalent or longer probation period,
on distressed restructured facilities under paragraph 72 of the EBA
Guidelines on the definition of default, they should consider a 21-month
period for the application of paragraph 101 of the EBA GL on PD and LGD.

(c) In the particular case of an institution opening new facilities to replace

previously defaulted facilities as part of restructuring or for technical
reasons, it should be able to make or trace a connection between the
restructured facility and the facility (or facilities) previously advanced and
which it is restructuring.

ECB guide to internal models – Credit risk 93

5.2 LGD structure

5.2.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 143 (3) second sub-paragraph

144 (1)(a), (f), (h)

169 (3)

170 (1)(e), (f), (3)(b), (c), (4)

174 (d)

175 (1), (4)(b)

179 (1)(a)

185 (a), (b), (c)

190 (1)

EBA GL on PD and LGD 20/11/2017 121, 122, 123

Other references

Final Draft RTS on assessment methodology

21/07/2016 11 (1), (2)(a)
for IRB

12 (a), (f)

16 (3)(c)

32 (2)(b), (5)(b)

34, 36, 37, 38

39 (b)

41 (a)

43

51 (i)

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

104. In order to comply with the requirements regarding the structure of LGD models
as set out in Article 170(1)(e) and (f) and 170(3)(b) and (c) of the CRR,
institutions should follow the observations below.

105. LGD estimates must be based on the material drivers of risk. 50 To comply with
this requirement, institutions should identify and analyse potential risk drivers
under paragraphs 121 to 123 of the EBA GL on PD and LGD. When selecting
the risk drivers, institutions should take into consideration any changes in
product mix or characteristics between the reference and default dates.

106. Institutions’ rating systems must provide for a meaningful assessment of obligor
and transaction characteristics, a meaningful differentiation of risk and accurate
and consistent quantitative estimates of risk. 51 It is the ECB’s understanding
50
Article 179(1)(a) of the CRR.
51
Article 144(1)(a) of the CRR.

ECB guide to internal models – Credit risk 94

 that to comply with this requirement institutions should demonstrate that, in
terms of the range of application of LGD models, the model performs
adequately (in terms of discriminatory power and predictive power) on
economically significant and material sub-ranges of application of the rating
systems. The sub-ranges are identified by splitting the full range of application
of the LGD model into different parts on the basis of potential drivers for risk
differentiation, among which, where relevant, the drivers referred to in
paragraph 121 of the EBA GL on PD and LGD.

107. The number of grades and pools must be adequate for a meaningful risk
differentiation and for the quantification of the LGD at the grade or pool level. 52
To comply with this requirement, institutions should ensure the following

(a) an adequate distribution of facilities across grades or pools in the datasets

used for development and (initial and regular) validation. For this purpose:

(i) any unusually low number of facilities in a grade or pool is expected

to be supported by empirical evidence of the adequacy of isolating
those facilities in a specific grade or pool;

(ii) any unusually high concentration of facilities in a grade or pool is

expected to be supported by empirical evidence of homogeneity
within these grades or pools (for example by analysing whether some
potential risk drivers (e.g. exposure size) that could further
differentiate between riskier and less risky facilities have not been
considered).

(b) sufficient homogeneity of the risk within each grade or pool by providing
empirical evidence that the grade-level LGD is adequate for all facilities in
that grade. For this purpose, in cases where it is found (through the use of
additional drivers or a different discretisation of the existing ones) that a
material subset of facilities within a grade or pool yields a significantly
different average realised LGD to that of the rest of the grade or pool, this
is considered to indicate a lack of homogeneity.

(c) sufficient heterogeneity of the risk across grades or pools by providing

empirical evidence that the average realised LGD is different across
consecutive grades or pools, for subsets for which there is a meaningful
order.

108. Where an institution uses direct estimates of risk parameters, these may be
seen as estimates assigned to grades on a continuous rating scale. 53 In this
case, in the ECB’s understanding the same requirements apply when an
institution uses direct estimates of risk parameters as apply to grade-based
models. To comply with these requirements, institutions are expected to ensure
risk differentiation in accordance with the following principles:

52
Article 170(1)(e) and (f) and 170(3)(b) and (c) of the CRR.
53
Article 169(3) of the CRR.

ECB guide to internal models – Credit risk 95

 (a) if the LGD estimates used to calculate the RWEAs are based on default
weighted LRAs of realised LGDs for grades or pools, irrespective of
whether at some point direct LGD estimates may have been used to
define such grades or pools, this grade or pool level is the relevant one for
the application of the principles set out in paragraph 107;

(b) when the situation described in point (a) above does not apply and,
instead, several components are estimated separately and then combined
in order to obtain the direct LGD estimates at facility level, institutions
should adequately analyse and reflect in the model possible dependencies
between the components (e.g. through relevant risk drivers);

(c) in the case of other direct LGD estimates (i.e. where no components are
defined) the principles above are expected to be applied where relevant.

109. In addition to paragraph 108 above and when institutions split the facilities into
different components (for example secured and unsecured), there is a risk that
a meaningful differentiation of risk will not be achieved at facility level. To
mitigate this risk, institutions should ensure that no bias is introduced in the risk
differentiation when combining the different components in order to obtain the
final LGD estimate at facility level. Specifically:

(a) the allocation of recovery flows to these components should be adequately

documented and implemented in a consistent way;

(b) risk differentiation (analogous to risk quantification) should be ensured with

respect to facility level.

ECB guide to internal models – Credit risk 96

5.3 Risk quantification

5.3.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 179 (1)(a), (1)(d)

(1)(a), (b), (e), (f) and (2)

181
second sub-paragraph

185 (b)

EBA GL on PD and LGD 20/11/2017 100, 116-118, 147-164

EBA Guidelines for the estimation of LGD

appropriate for an economic downturn 06/03/2019
(“Downturn LGD estimation”) 54

Other references

Final Draft RTS on assessment methodology

21/07/2016 50 (c)
for IRB

51 (a) to (f), (i)

Final Draft RTS on the specification of the

nature, severity and duration of an economic 16/11/2018
downturn

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

5.3.2 Observed average LGD

110. To comply with the requirement of obtaining an LRA LGD in accordance with
Article 181(1)(a) of the CRR, institutions should calculate the observed average
LGD under paragraphs 147, 148, 154-157 and 160 of the EBA GL on PD and
LGD. When performing this calculation, institutions should follow the principles
set out below.

111. Under paragraph 147 of the EBA GL on PD and LGD, default observations that
are triggered close to the time of the LGD estimation process (i.e. observations
with a recent default when the LGD is being estimated) are part of the historical
observation period and should be included in the RDS. Since for these recent
defaults only limited information is available regarding the full recovery process,
the treatment of incomplete recovery processes envisaged in paragraph 158 of
the EBA GL on PD and LGD is more complex and could add uncertainty to the
LGD estimates. It is the ECB’s understanding that to mitigate this risk
institutions may establish a minimum period of time during which the default
should be observed in order for it to be considered in the calculation of the
observed average LGD. This minimum period should be adequately justified
54
EBA Guidelines for the estimation of LGD appropriate for an economic downturn (“Downturn LGD
estimation”) (EBA/GL/2019/03), referred to in this guide as the “EBA Guidelines for the estimation of
LGD appropriate for an economic downturn (“Downturn LGD estimation”).

ECB guide to internal models – Credit risk 97

 and institutions should ensure that all relevant information regarding defaults
observed for a shorter period (e.g. a change in the characteristics of defaults) is
considered in the LGD estimates. In any case this period should not be longer
than 12 months.

112. For the purposes of LGD estimation (and validation), long recovery processes
are expected to be considered as closed under paragraph 156 of the EBA GL
on PD and LGD. The objective of defining the maximum period of the recovery
process (“time-to-workout”) is to avoid situations where institutions give
consideration to overly optimistic recoveries from open exposures that are
already at a very advanced stage of the recovery process. To achieve this, the
specification of the “time-to-workout” should be supported by evidence of the
observed pace of recoveries and be consistent with the nature of the products
concerned, the type of exposures and the operational recovery process. In
addition, the institution should substantiate and clearly document the studies
that support the formulation of the time-to-workout and should pay particular
attention to the following.

(a) The specific moment after the date of default at which nearly nil evolution
of the average cumulative recovery rates is observed. For example, when
the cumulative recovery curves show a pronounced increase after which
they flatten out, the time spent in default after the significant increase
occurs could be used directly as the time-to-workout, especially in the
case of unsecured exposures.

(b) The period of time after the date of default where the cumulative
percentage of closed/recovered exposures flattens.

(c) The number of exposures used to construct the curves referred to at

letters (a) and (b) above, in order to identify situations where only a few
cases contribute to the shape of the curves.

(d) The expected recovery rate conditioned to vintages higher than the time-
to-workout.

(e) For secured exposures, the share of exposures for which recoveries from
collateral have not yet been realised.

5.3.3 Treatment of incomplete recovery processes

113. In order to obtain an LRA LGD in accordance with Article 181(1)(a) of the CRR,
institutions should ensure that the relevant information from incomplete
recovery processes is taken into account in a conservative manner. For this
purpose, institutions should analyse their incomplete recovery processes and
extract the information relevant for LGD estimation under paragraphs 153 to
159 of the EBA GL on PD and LGD. In addition, institutions should:

ECB guide to internal models – Credit risk 98

 (a) justify and document their methodology for the treatment of incomplete
recovery processes, and in particular how they take into account
paragraph 159 of the EBA GL on PD and LGD;

(b) for the purpose of paragraph 159(a) of the EBA GL on PD and LGD in
particular, base the extrapolation of future recoveries on defaults arising
from vintages (i.e. group of exposures which defaulted in a given period of
time) for which, during the period already observed, similar average past
recoveries have been realised on similar exposures;

(c) in order to ensure transparency regarding the impact from the treatment of
incomplete recovery processes, assess the sensitivity of the treatment with
respect to the main assumptions.

5.3.4 Recovery processes where collateral has been repossessed and

not yet sold

114. In specific cases where institutions have taken possession of but not yet sold
the collateral, there is a risk that the value of repossession might not adequately
reflect the value of the repossessed collateral. To mitigate this risk, institutions
should estimate haircuts to the value of the collateral under paragraphs 116 to
118 of the EBA GL on PD and LGD. In order to ensure transparency regarding
the impact from the treatment of repossessed collateral, institutions should:

(a) compare the estimated haircuts with the available observations regarding
the repossession and subsequent sale of similar types of collateral;

(b) assess the impact on the LRA LGD of the inclusion of the repossessed
collateral by performing sensitivity analyses based on the application of
different haircuts to the value of the collateral (at least, by applying a
haircut of 100% to cases where collateral has been repossessed but not
yet sold).

5.3.5 Long-run average

115. To comply with the requirement of obtaining an LRA LGD by facility grade or
pool in accordance with Article 181(1)(a) of the CRR, institutions should
estimate LGDs under paragraphs 100 and 149 to 164 of the EBA GL on PD and
LGD. When performing this estimation, institutions should follow the
observations below.

116. Under paragraph 150 of the EBA GL on PD and LGD, institutions should
calculate the LRA LGD as an arithmetic average of realised LGDs over a
historical observation period weighted by a number of defaults. When
performing this calculation, institutions should observe the following points.

ECB guide to internal models – Credit risk 99

 (a) In the event of definition of default applied at obligor level, where two
facilities of the same obligor are assigned to the same facility grade or
pool, institutions are expected to calculate the average realised LGD as
follows: first take the exposure-weighted average realised LGD at the
obligor level and then take the arithmetic average LGD weighted by the
number of defaulted obligors within the LGD grade. If institutions use a
different approach, they should demonstrate that there are no systematic
deviations from the approach referred to above.

(b) Under paragraph 160 of the EBA GL on PD and LGD, the realised LGD of
each observation should be floored at zero for the purpose of LGD
estimation. In cases where LGD estimates for specific facility grades or
pools are low or even zero (in exceptional cases), in order to ensure that
these estimates are accurate and not driven by (systematic) errors or
distortions institutions should ensure that their estimation process is
accurate. In particular, they should ensure that there is a sufficient number
of observations supporting the estimate and that these outcomes are
carefully monitored and scrutinised.

(c) Under paragraph 162 of the EBA GL on PD and LGD, institutions should
apply an appropriate treatment to extremely high values of realised LGDs
much above 100%, at the level of data quality, risk drivers, assignment to
grades or pools or assignment to calibration segments. To ensure that the
estimates are accurate, institutions are not expected to cap realised LGD
values (i.e. to replace the observed value by a pre-defined value when the
observed value is above the pre-defined one).

117. Institutions can calibrate LGD estimates to the LRA LGD calculated at the level
of the calibration segment under paragraph 161(b) of the EBA GL on PD and
LGD. When calibration segments are used for the purpose of LGD estimations,
institutions are expected to base their decision on a sound rationale, in
particular on quantitative evidence. It is the ECB’s understanding that, to
comply with Article 181(1)(a) of the CRR, institutions should also calculate the
LRA LGD at a more granular level than the calibration segment (i.e. individual
LGD grades or pools if estimation is discrete or ranges of LGD values if the
estimation is continuous). The level should be appropriate for the application of
the model. In addition, institutions should ensure that there are no systematic
deviations when comparing the estimated LGDs with the LRA of realised LGDs
at this more granular level, i.e. the direction of divergences should be random.

118. Where the LGD is the result of a combination of different components (for
example, secured and unsecured components), the calibration step according
to paragraph 117 (ensuring that the average realised LGD and the average
estimated LGD across facilities within the same calibration segment/range of
LGD values are aligned) is expected to be performed after the aggregation of
the components. In addition, there is a risk that systematic deviations could be
introduced to the estimation when combining these different components. In this
case, the direction of divergences would not be random. To mitigate this risk,
institutions should do the following.

ECB guide to internal models – Credit risk 100

 (a) For defaults in the RDS which are closed or considered closed, compare
the realised LGD at facility level with the estimates of LGD. Separate tests
should be performed for the LGD applied to the performing portfolio and
the LGD in-default. Analogous tests should be performed at component
level.

(b) In the case of models based on components with underlying data covering
time windows with different lengths and/or periods for each of the
components, ensure that no bias is introduced in the LGD estimates at
facility level with respect to the LRA. The analysis referred to in point (a)
should be performed, at least, for the available common time period.

119. Notwithstanding paragraph 116, for retail exposures institutions need not give
equal importance to historical data if more recent data are a better predictor of
loss rates. 55 It is the ECB’s understanding that an institution may consider the
more recent data to be a better predictor of loss rates and may give more
importance to recent historical data if its methodology is in line with
paragraphs 150 to 152 of the EBA GL on PD and LGD and if the following
apply.

(a) There is a significant improvement in predictive power when using the

more recent data with respect to the predictive power resulting from the
use of an arithmetic average under paragraph 150 of the EBA GL on PD
and LGD. This improvement would be evidenced by comparing the
estimated LGDs for each grade with the average realised LGD covering as
long a period as possible in accordance with Article 185(b) of the CRR.

(b) The oldest data are considered as non-representative as a result of

specific policy or business changes in the bank, but not in order to reflect
current trends in loss rates directly related to macroeconomic conditions.

(c) The weighting approach is used in a consistent manner over time and any
change in the applied weights of historical data is appropriately justified.

120. In accordance with Article 179(1)(a) of the CRR, an institution’s own estimates
must incorporate all relevant data and must be derived using both historical
experience and empirical evidence. To comply with these requirements, when
institutions use external or pooled data to complement their own loss or
recovery experience, they should ensure that LRA LGDs derived from external
or pooled data are also calculated separately from those based on internal data.
In addition, the direction and magnitude of the differences between these
averages should be properly analysed and documented when calibrating the
model, including the adequacy of the MoC considered, and duly followed up in
the review of estimates.

121. Article 179(1)(d) of the CRR requires, among other things, that the population of
exposures represented in the data used for estimation, the lending standards

55
Article 181(2), last paragraph, of the CRR.

ECB guide to internal models – Credit risk 101

 used when the data were generated and other relevant characteristics must be
comparable with those of the institution's exposures and standards.
Paragraph 164 of the EBA GL on PD and LGD further specifies that institutions
should take into account not only the current characteristics of the portfolio but
also, where relevant, any changes to the structure of the portfolio that are
expected to happen in the foreseeable future. When institutions perform
adjustments to their LGD estimates in order to comply with these requirements,
it is the ECB’s understanding that the following principles should apply.

(a) The adjustment should be based on a comparison of the data used in risk
quantification with the institution’s application portfolio. In many
circumstances (for example where a type of product has been
discontinued by the institution), the addition of these characteristics as risk
drivers for LGD estimation is the most simple and effective way of dealing
with issues of non-representativeness.

(b) In the event of changes in lending or recovery policies, institutions should

make only conservative adjustments until they are able to provide
empirical evidence concerning the impact of the new policies. Such
evidence should be based on the inclusion in the RDS of data from
periods more recent than the change of policy.

(c) All economic and market conditions experienced in the past and reflected
in historical observations should be considered by institutions as part of
foreseeable economic and market conditions (paragraph 147 of the EBA
GL on PD and LGD). They are not, therefore, a reason to perform
adjustments.

5.3.6 Downturn LGD

122. To obtain LGD estimates that are appropriate for an economic downturn in
accordance with Article 181(1)(b) of the CRR, institutions are expected to:

(a) characterise an economic downturn in accordance with the Final Draft

RTS on the specification of the nature, severity and duration of an
economic downturn;

(b) derive LGD estimates which are appropriate for the downturn conditions
specified, in accordance with the EBA Guidelines for the estimation of LGD
appropriate for an economic downturn (“Downturn LGD estimation”).

ECB guide to internal models – Credit risk 102

5.4 Estimation of ELBE and LGD in-default

5.4.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 179 (1)(c)

(1)(a), (h), (j) and (2) second

181
sub-paragraph

185 (a) to (c)

EBA GL on PD and LGD 20/11/2017 165 to 193

Other references

Final Draft RTS on assessment methodology

21/07/2016 11 (3)(a)
for IRB

46 (d)

50

51 (c)

54

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

123. In accordance with Article 181(1)(h) of the CRR, for the specific case of
exposures already in default institutions must use the sum of their ELBE for each
exposure, given current economic circumstances and exposure status and their
estimate of the increased loss rate caused by possible additional unexpected
losses during the recovery period. To comply with these requirements,
institutions should estimate ELBE and LGD in-default under paragraphs 165 to
193 of the EBA GL on PD and LGD. In this process, institutions should follow
the observations below.

124. In accordance with Article 181(1)(h) of the CRR, the ELBE must represent the
best estimate of expected loss given current economic circumstances and
exposure status. To comply with this requirement, it is the ECB’s understanding
that institutions should take into consideration the economic conditions
expected over the period of the recovery process, and in particular reflect
downturn conditions in the ELBE, if and only if current economic conditions are
in a downturn or a downturn is expected over the period of the recovery
process. This can be done either by adding the relevant macroeconomic and
economic factors as drivers of the ELBE model under paragraph 183 of the EBA
GL on PD and LGD, or alternatively through an adjustment to the LRA as
referred to in paragraph 184 of the EBA GL on PD and LGD.

125. Under paragraph 193 of the EBA GL on PD and LGD, LGD in-default can be
estimated directly or as the sum of ELBE and an add-on capturing the
unexpected loss related to the exposures in default that may occur during the
recovery period. In particular, the following should be taken into consideration.

ECB guide to internal models – Credit risk 103

 (a) The use of a constant value for unexpected losses for all defaulted
exposures is not risk sensitive. In the ECB’s understanding, therefore, it
does not allow an accurate assessment of risk. Where an institution does
use a constant value, it should justify this. It should demonstrate that the
constant value in question is an adequate estimate of all the components
of unexpected loss envisaged in paragraph 193(b) of the EBA GL on PD
and LGD during the remaining recovery period, i.e. between the date for
which estimates are being applied and the final closure of the recovery
process. This analysis should be performed at least for every calibration
segment.

(b) LGD in-default estimates are generally expected to be higher than ELBE
estimates and only equal for duly justified individual exposures, which are
expected to be very limited.

6 Conversion factors

6.1 Commitments, unadvised limits and scope of application

6.1.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 4 (1)(56)

151 (7), (8)

166 (1) to (8)(a), (b), (c), (d), (10)

126. In accordance with Article 151(7) and (8) of the CRR, institutions must use own
estimates of CCFs for the retail exposure class. If they have received
permission to use own estimates for the corporate, institutional, central
government and central bank exposure classes, they must again use own
estimates of CCFs, instead of the conversion factors set out in in
Article 166(8)(a) to (d) of the CRR. In both cases (retail and non-retail exposure
classes), the scope of CCF modelling is, in the ECB’s understanding, limited to
the off-balance sheet items referred to in Article 166(8) of the CRR. 56 The
treatment of off-balance-sheet items other than those mentioned in
Article 166(1) to (8) of the CRR is specified in Article 166(10). In accordance
with Article 166(10) of the CRR, an exposure value must be a specific
percentage of an off-balance-sheet item’s value, based on the classification of
off-balance-sheet items established in Annex I of the CRR.

56
The understanding of the ECB is also supported by EBA Single Rulebook Q&A,
Question ID: 2014_1263.

ECB guide to internal models – Credit risk 104

127. Conversion factor means the ratio of the currently undrawn amount of a
commitment that could be drawn and that would therefore be outstanding at
default to the currently undrawn amount of the commitment. The extent of the
commitment is determined by the advised limit, unless the unadvised limit is
higher. 57 The exposure value for the items listed in Article 166(8) of the CRR
must be calculated as the committed but undrawn amount multiplied by a
CCF. 58 To calculate the exposure value as required by Article 166(8) of the
CRR, institutions should adopt the following approach.

(a) Treat a facility as an exposure from the earliest date at which the facility is
recorded in the institution’s systems in a way that would allow the obligor
to make a drawing. An unadvised limit is any credit limit defined by the
institution (i) that is above the limit the obligor has been informed of by the
institution; and (ii) according to which additional drawings are possible, at
least temporarily. This higher (unadvised) credit limit may be disregarded if
its availability is subject to a further credit assessment by the institution, as
long as this additional assessment includes a re-rating or a confirmation of
the rating of the obligor.

(b) Consider as “commitment” any contractual arrangement that has been

offered by the institution and accepted by the obligor to extend credit,
purchase assets or issue credit substitutes.

(c) Consider as “conditionally cancellable commitment” any such arrangement

that can be and will be cancelled by the institution if the obligor fails to
meet conditions set out in the facility documentation, including conditions
that must be met by the obligor prior to any initial or subsequent drawdown
under the arrangement.

(d) Consider as “credit lines” all lines including products such as facilities
granted for construction where the payments to the obligor are made
according to the progress of the construction. Products such as
guarantees are not, however, included in the concept of credit lines.

128. For institutions not using own estimates of CCFs for exposures to corporates,
institutions, central governments and central banks, Article 166(8) of the CRR
defines the CCFs to be used for the purpose of calculating RWEA. In
accordance with Article 166(8)(a) and (c) of the CRR, institutions not using their
own estimates of CCFs for non-retail exposures are permitted to apply a 0%
CCF, under certain conditions.

It is the ECB’s understanding that, to comply with the conditions established

under Article 166(8)(a) and (c) of the CRR and to apply a 0% CCF, institutions
should have in place internal control systems that allow them to monitor the
obligor’s financial condition and to act in the event that a deterioration in the
obligor’s credit quality is detected. They should also be able to provide evidence

57
Article 4(1)(56) of the CRR.
58
Article 166(8) of the CRR.

ECB guide to internal models – Credit risk 105

 that the internal control systems work effectively. For this purpose, institutions
should demonstrate that there is only a very limited number of exposures of a
particular type observed during the previous year for which the EAD is higher
than the drawn amount at the reference date. This analysis should be
performed on a regular basis. The ECB considers it good practice when
institutions perform this analysis on an annual basis.

6.2 Realised CCFs

6.2.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 175 (3)

176 (4)

178 (1)

179

182 (1)(a)

Other references

Final Draft RTS on assessment methodology

21/07/2016 57
for IRB

EBA Guidelines on the definition of default 18/01/2017 16

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

129. To ensure that a consistent and accurate approach is adopted to calculate the
realised CCFs, institutions should have in place sufficiently detailed policies and
procedures. For institutions to comply with the data-related requirements set out
in Article 176(4) of the CRR, their RDS:

(a) should not be capped at the principal amount outstanding or at facility

limits;

(b) should include all credit obligations (paragraph 16 of the EBA Guidelines
on the definition of default), especially accrued interest, other due
payments (e.g. fees) and limit excesses.

130. In accordance with Article 182(1)(a) of the CRR, institutions must calculate the
realised CCF at facility level for each default. In cases where realised LGD is
calculated at a more aggregated level than single facility level, as described in
paragraph 99 above, CCF estimation can be performed at facility level or at the
LGD aggregation level. One such example is where CCF is estimated by facility
while several facilities are aggregated for LGD purposes, since they are all
secured by the same collateral. In this case, institutions should:

ECB guide to internal models – Credit risk 106

 (a) for retail exposures where they use the definition of default at facility level
in accordance with the last paragraph of Article 178(1) of the CRR, apply
full contagion of default across aggregated facilities;

(b) ensure consistency between estimation and application of the parameters;

(c) ensure that no bias results from the aggregation of facilities by validation
of the estimates (PD, LGD, CCF) also at the more aggregated level.

131. For the purposes of Article 182(1)(a) of the CRR, institutions must compute
realised CCF. To comply with this requirement, in the understanding of the ECB
institutions should adopt the following approach.

(a) Calculate realised CCF as the ratio of the difference between the EAD and
the exposure at the reference date in the numerator, and the difference
between the limit at reference date and the exposure at reference date
(i.e. the amount available to be drawn at the reference date) in the
denominator. This does not mean that, to address the issues with the
“region of instability”, institutions may not use direct EAD realisation (as
referred to in paragraph 137(a) of this chapter).

(b) Ensure that the definition of exposure is identical to the one used for LGD
estimation. In particular, treatment of post-default drawings should be
identical for the exposures used in both the LGD and CCF estimations.
Discounted additional drawings are expected to be included in CCF when
they are included in the LGD denominator. See also paragraph 100(a)
above.

(c) For each reference date and in cases where the same facility defaults
more than once during the observation period, consider as the date of
default relevant for CCF purposes the date of the first default.

ECB guide to internal models – Credit risk 107

6.3 CCF structure

6.3.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 170

182

Other references

Final Draft RTS on assessment methodology

21/07/2016 170
for IRB

172 (1)(b), (2)

174

175 (4)

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

132. To comply with the requirements for the structure of the CCF models
established in Article 170 of the CRR, and particularly when considering the risk
drivers envisaged by paragraph (4) of that Article, institutions should follow
these principles:

(a) Demonstrate a detailed understanding of the impact on the CCF model of

changes in customer product mix or characteristics that take place
between reference and default dates and the materiality of that impact. If
the impact is material, institutions should address it within their own
estimation process. This is because changes in exposure characteristics
(e.g. a change in the value of the limit) or “product profile transformations”
(e.g. a revolving loan that has been converted into a term loan or vice
versa) which commonly occur between reference and default dates can
impact the CCF model.

(b) Although the reference date for the calculation of realised CCF should be
up to 12 months prior to default, institutions should analyse risk drivers
considering information not only at the reference date but also before that
date whenever relevant. In assessing whether the use of information from
before the reference date is relevant, institutions should take into account
the volatility of the risk driver over time.

(c) Ensure that the models reflect the institution’s current policies and
strategies regarding account monitoring, including limit monitoring, and
payment processing.

ECB guide to internal models – Credit risk 108

6.4 CCF risk quantification

6.4.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 144 (1)(a)

166 (8)

169 (3)

182

Other references

Final Draft RTS on assessment methodology

21/07/2016 56 to 58
for IRB

Final Draft RTS on the specification of the

nature, severity and duration of an economic 16/11/2018
downturn

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

133. The exposure value for undrawn commitments is calculated as the committed
but undrawn amount multiplied by a CCF. 59 CCFs can also be derived from
direct estimates (for example by modelling total facility EAD) in accordance with
Article 169(3) of the CRR. In this case, and in order to comply with
Article 182(1)(a) of the CRR, it is the ECB’s understanding that institutions
should also calculate the LRA CCF at a level more granular than calibration
segment that is appropriate for the application of the model, namely using
individual CCF values if estimation is discrete or sub-ranges of CCF values if
estimation is continuous. In addition, institutions should ensure that there are no
systematic deviations when comparing the estimated CCFs with the LRA
realised CCFs in sub-ranges. In other words, the direction of divergences
should be random.

134. In accordance with Article 182(1)(a) of the CRR, institutions are required to
calculate the default weighted LRA CCF separately for each facility grade or
pool. To comply with this requirement, institutions should adopt the following
approach.

(a) Ensure that the historical observation period is as broad as possible and
contains data from different periods characterised by different economic
circumstances, including bad years as reflected in economic indicators
that are relevant for the type of exposures considered.

(b) Calculate the observed average CCF for each facility grade or pool on all
defaults observed in the historical observation period. Institutions should

59
Article 166(8) of the CRR.

ECB guide to internal models – Credit risk 109

 apply an appropriate treatment to extremely high values of realised CCF
much above 100%, at the level of data quality, risk drivers, assignment to
grades or pools or assignment to calibration segments. To ensure that the
estimates are accurate, it is the ECB’s understanding that institutions are
not expected to cap realised CCF values. In other words, they are not
expected to replace the observed value by a pre-defined value when the
observed value is above the pre-defined one.

(c) When the historical observation period is considered to be representative

of the LRA, the average realised CCFs should be computed as the
arithmetic average of the yearly averages of realised CCFs in that period.

(d) When the historical observation period is not considered to be

representative of the LRA:

(i) if bad years are under-represented in the historical observation

period, the observed average CCF should be adjusted upwards in
order to estimate an LRA CCF;

(ii) if bad years are over-represented in the historical observation period,

the observed average CCF may be adjusted to estimate an LRA CCF
where there is a significant correlation between the economic
indicators referred to in paragraph 122(a) and the available observed
CCF.

(e) It is the ECB’s understanding that, where CCF estimates for specific facility
grades or pools are low or even zero (in exceptional cases) before the
MoC is applied, and in order to ensure that these estimates are accurate
and not driven by (systematic) errors or distortions, institutions should
ensure that their estimation process is pertinent and accurate. In particular,
they should ensure that, in these specific facility grades or pools, there is
only a very limited number of exposures for which the exposure at the
moment of default is higher than the drawn amount at the reference date,
and that these outcomes are carefully monitored and scrutinised.

135. Notwithstanding paragraph 134, for retail exposures an institution need not give
equal importance to historical data if more recent data are a better predictor of
drawdowns. In the ECB’s understanding, an institution may consider that the
more recent data are a better predictor of drawdowns and may give more
importance to recent historical data if the following apply.

(a) There is a significant improvement in predictive power when using the

more recent data compared with the predictive power resulting from the
use of an arithmetic average. This improvement can be evidenced by
comparing the estimated CCFs for each grade with the average realised
CCF covering as long a period as possible, as set out in Article 185(b) of
the CRR.

(b) The oldest data are considered as non-representative as a result of

specific policy or business changes in the bank, but not in order to reflect

ECB guide to internal models – Credit risk 110

 current trends in realised CCFs directly related to macroeconomic
conditions.

(c) The weighting approach is used in a consistent manner over time and any
change in the applied weights of historical data is appropriately justified.

136. To comply with the requirements of Article 182(1)(b) and to have CCF estimates
that are appropriate for an economic downturn, institutions should characterise
an economic downturn in accordance with the Final Draft RTS on the
specification of the nature, severity and duration of an economic downturn. To
this end, an impact assessment should be performed to identify which identified
downturn period is most strongly evidenced by elevated levels of realised
CCFs. Any lag between the downturn period and the date of the impact on the
realised CCFs should be taken into account. This means that where high levels
of realised CCFs are not experienced simultaneously with the downturn
periods, but nevertheless result from it, these high CCFs should be considered
as the CCFs appropriate for the economic downturn.

137. In order to ensure a meaningful assessment of transaction characteristics, as

required by Article 144(1)(a) of the CRR, particular attention should be given to
the following observations.

(a) A common issue in estimating CCFs concerns facilities close to being fully
drawn at the relevant reference date, as a result of the instability that may
be observed in the estimates (also known as “region of instability”). To
mitigate this risk, institutions should ensure that their CCF model is robust
and provides estimates that are effectively protected against undesirable
issues caused by the “region of instability” and/or that their estimates are
adjusted adequately.

(b) Article 182 of the CRR lays down the requirements for CCF estimates. In
cases where institutions apply a fixed yet conservatively specified CCF
(e.g. 100%), the ECB considers them to be compliant with the
requirements when these estimates are applied in specific circumstances,
such as scarcity of data and low materiality of the scope of application, for
example in the event that the facility has no on-balance exposure until the
first disbursement and estimates are monitored and validated.

ECB guide to internal models – Credit risk 111

7 Model-related MoC

7.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 179 (1)(f)

EBA GL on PD and LGD 20/11/2017 36 to 52, 92(b)

Other references

Final Draft RTS on assessment methodology

21/07/2016 47
for IRB

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

138. Institutions must add to their estimates an MoC that is related to the expected
range of estimation errors. 60 To comply with this requirement, institutions are
expected to follow paragraphs 36 to 52 of the EBA GL on PD and LGD.

139. Since the MoC requirements laid down by the CRR also apply in cases where
institutions estimate CCFs, paragraph 138 is also relevant in such cases.

140. In the understanding of the ECB, to reflect the dispersion of the statistical
estimators as set out in paragraph 43(b) of the EBA GL on PD and LGD,
institutions should adopt the following approach.

(a) For PD, estimate an MoC to account for statistical uncertainty/sampling

error affecting the LRA estimate at grade/pool level. This MoC should be
based on the distribution of the estimator, which is the average of one-year
default rates of the grade/pool across time (i.e. the distribution of
(ΣDR t )⁄T, considering that the uncertainty is primarily driven by the
statistical uncertainty of each one-year default rate and the length of the
time series. As a result, it is expected that the lower the number of
observations per grade and the shorter the time series are, the higher the
MoC of the grade should be.

Institutions need to be aware of and deal adequately with the dependency

between default rates over time on the quantification of the MoC,
e.g. when using overlapping windows for the calculation of default rates.

The above principles also apply for institutions using direct PD estimates
and for institutions calibrating the LRA default rate at the level of the
calibration segment, as referred to in paragraph 92(b) of the EBA GL on
PD and LGD. When using direct PD estimates, the MoC is based on the
distribution of this direct PD estimator (which includes the risk

60
Article 179(1)(f) of the CRR.

ECB guide to internal models – Credit risk 112

 differentiation function), implicitly reflecting the uncertainty of the LRA.
When calibration is performed at calibration segment level, the general
estimation error may be computed at that level when the statistical
uncertainty/sampling error is not significantly different across grades or PD
sub-ranges.

(b) Similarly, for LGD and CCF, estimate an MoC to account for statistical
uncertainty/sampling error affecting the final estimates. This MoC should
be defined on the basis of the distribution of the estimators, considering
that their uncertainty is primarily driven by the statistical uncertainty of the
observations used to compute the long-run and downturn estimates and
the length of the time series.

8 Review of estimates

8.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 172 (3)

174 (e)

179 (1)(a), (c)

EBA GL on PD and LGD 20/11/2017 217 to 221

Other references

Final Draft RTS on assessment methodology

21/07/2016 37 part 2
for IRB

46

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

141. Institutions must review their estimates whenever new information comes to
light but at least on an annual basis. 61 To comply with this requirement, they are
expected to have in place a framework under paragraphs 217 to 221 of the EBA
GL on PD and LGD.

142. Since the review of estimates requirements under the CRR also apply in cases
where an institution estimates CCFs, paragraph 141 is also relevant to such
cases.

143. In the ECB’s understanding and for the purposes of paragraph 141, the
following principles apply.

61
Article 179(1)(c) of the CRR.

ECB guide to internal models – Credit risk 113

 (a) For PD models and regarding the analysis of the predictive power
envisaged by paragraph 218(c) of the EBA GL on PD and LGD:

(i) the analysis should be performed at grade level; for institutions using
direct PD estimates, it should be performed at a sufficient level of
granularity;

(ii) institutions should use a range of metrics to assess predictive ability,

including statistical tests and graphical analysis of the evolution of
default rates and PD.

(b) The analysis referred to in paragraph 218(c)(i) of the EBA GL on PD and

LGD should also consider, for CCFs, whether including the most recent
data leads to a significant change in the LRA CCF or downturn CCF.

(c) For LGD models that result from a combination of different components
(for example, secured and unsecured components), the back-testing
analysis referred to in paragraph 218(c)(ii) of the EBA GL on PD and LGD
should be run at both component and facility level.

(d) In addition, institutions should consider in their frameworks for the review
of estimates the availability of data for different exposure types, taking into
account the specificities of the model architecture, including the existing
and potential risk drivers, under paragraph 220 of the EBA GL on PD and
LGD. When data are scarce, they should use complementary analyses for
those exposure types where quantitative measures prove inconclusive as
a result, for example, of the low number of exposures available.

(e) Where internal data are not considered sufficient to establish fixed targets
and tolerances for defined metrics and tools to assess the performance of
the PD model in terms of risk differentiation, institutions should define and
put in place the appropriate actions to address this. 62 These actions could
encompass, for example, the use of complementary analyses for those
cases where the results for the application of metrics and tools are proven
to be inconclusive.

(f) When external credit bureau scores or ratings are used as the main (or
one of the main) driver(s) of the internal rating, in cases where significant
changes are applied to the credit bureau scoring institutions should
consider the possibility of adjusting their internal data following the
changes applied to the score, and whenever the input variables are no
longer considered appropriate in their credit rating process.

144. In the case of material models where the assignment of the grade is based on a
statistical model and where there is a risk that slight changes in the ranking of
the obligors, or in the boundaries between grades, could lead to significant
changes in the RWEA in that portfolio, the framework referred to in

62
As set out in Article 37(2) of the Final Draft RTS on assessment methodology for IRB.

ECB guide to internal models – Credit risk 114

 paragraph 141 should also include an analysis of whether the inclusion of the
most recent data in the RDS used for model development would lead to
materially different model outcomes. This analysis should be conducted on a
three-yearly basis, or more often, depending on the materiality of the model.
The analysis should consider, in particular, whether the discriminatory power of
the PD, LGD or CCF models would be materially increased when re-estimating
the model parameters on the basis of the updated RDS. Portfolios should be
considered as falling into this category when, for example: (i) a limited number
of obligors represent an important share of the total exposure; or (ii) exposures
are concentrated near the boundaries between two grades.

145. When the number of default observations is low, to analyse whether the main
drivers of the observed defaults are appropriately reflected in the model in
accordance with Article 179(1)(a) of the CRR 63 institutions should analyse
individual defaults (or at least a sample of them where the number of defaults
makes analysing all of them unduly burdensome). However, the model should
not be adapted simply to fit singular events from the institution’s file review.

146. In accordance with Article 172(3) of the CRR, for grade and pool assignments
institutions must document those situations in which human judgement may
override the inputs or outputs of the assignment process. In addition, institutions
must complement the statistical model by human judgement and human
oversight to review model-based assignments and ensure that the models are
used appropriately. 64 Furthermore, review procedures must be designed to find
and limit errors associated with model weaknesses. 65 To comply with these
requirements, institutions should assess the impact of the application of human
judgement on risk differentiation capability (e.g. on discriminatory power), under
paragraph 218(b) of the EBA GL on PD and LGD.

63
This article requires that estimates be based on the material drivers of the respective risk parameters.
64
Article 174(e) of the CRR.
65
Article 174(e) of the CRR.

ECB guide to internal models – Credit risk 115

9 Calculation of maturity for non-retail exposures

9.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 162 (2)(f), (3)

Other references

Final Draft RTS on assessment methodology 21/07/2016 73 (d)(i), (ii)

for IRB

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IRB will become an additional relevant legal
reference. Currently that document only exists in a final draft version.

147. For the cases described in Article 162(2)(f) of the CRR, the maturity parameter
(M) must be the maximum remaining time (in years) that the obligor is permitted
to take to fully discharge its contractual obligations. In the ECB’s understanding,
M should be calculated using the expiry date of a facility. The repayment date of
a current drawn amount should not be used.

148. To ensure that the calculation of the maturity parameter is correct and to avoid
any possible errors, for the purposes of Article 162(3) of the CRR institutions
should adequately justify and document any exemptions from the one-year
maturity floor.

ECB guide to internal models – Credit risk 116

 Market risk

1 Scope of the market risk chapter

1. The purpose of this chapter is to provide transparency on how the ECB
understands a number of topics related to internal models used in the
calculation of own funds requirements for market risk. It is important to note that
this chapter does not aim to cover exhaustively all topics that could be subject
to review during internal model investigations (such as, for example, model
governance). The topics covered in the market risk chapter have been selected
taking into account the requirements of the Capital Requirements Regulation
(CRR) and focus on certain modelling aspects relating, for example, to
regulatory back-testing of value-at-risk (VaR) models, to VaR and stressed VaR
(sVaR) methodologies, and to the incremental default and migration risk charge
(IRC) methodology.

ECB guide to internal models – Market risk 117

2 Scope of the internal model approach

2.1 Regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 4 (1)

6 (1)

11 (1)

18

92 (3), (4)

104, 106, 325, 349

350 (1), (2)

352 (2), (3)

353 (3)

362, 363

364 (2)

367 (2), (3)

368, 370, 372

386 (1), (3)

Commission Delegated Regulation 04/03/2015

on materiality of extensions and
changes of the IMA 1

SSM Regulation 2 15/10/2013 10

Other references

EBA Guidelines on the IRC 3 16/05/2012 4, 7

Final Draft RTS on assessment 22/11/2016 Section 2, recital (20)

methodology for IMA and significant
share 4

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IMA and significant share will become an
additional relevant regulatory reference. Currently that document only exists in
a final draft version.

1
Commission Delegated Regulation (EU) 2015/942 amending Delegated Regulation (EU) No 529/2014
supplementing Regulation (EU) No 575/2013 of the European Parliament and of the Council as regards
regulatory technical standards for assessing the materiality of extensions and changes of internal
approaches when calculating own funds requirements for market risk (OJ L 154, 19.6.2015, p. 1)
referred to in this guide as the “Commission Delegated Regulation on materiality of extensions and
changes of the IMA”.
2
Council Regulation (EU) No 1024/2013 of 15 October 2013 conferring specific tasks on the European
Central Bank concerning policies relating to the prudential supervision of credit institutions (OJ L 287,
29.10.2013, p. 63), referred to in this guide as the “SSM Regulation”.
3
EBA Guidelines on the Incremental Default and Migration Risk Charge (IRC) (EBA/GL/2012/3), referred
to in this guide as the “EBA Guidelines on the IRC”.
4
EBA Final Draft Regulatory Technical Standards on the specification of the assessment methodology
for competent authorities regarding compliance of an institution with the requirements to use internal
models for market risk and assessment of significant share under points (b) and (c) of Article 363(4) of
Regulation (EU) No 575/2013 (EBA/RTS/2016/07), referred to in this guide as the “Final Draft RTS on
assessment methodology for IMA and significant share”.

ECB guide to internal models – Market risk 118

2.2 Delimitation of the regulatory trading book

2. According to Article 4(1)(86) of the CRR, “trading book” means all positions in
financial instruments and commodities held by an institution either with trading
intent, or in order to hedge positions held with trading intent.

3. In accordance with Article 104(1) of the CRR, institutions must have clearly
defined policies and procedures for determining which positions to include in
the trading book for the purpose of calculating their capital requirements
(referred to in this guide as the “regulatory trading book”). The ECB
understands that, in this context, “positions” refers to instruments or
transactions, and not to risk positions as referred to in the glossary included in
this guide. The ECB understands that positions that are classified as “held for
trading” for accounting purposes are presumed to be included in the regulatory
trading book. Therefore, institutions should be able to list all positions that are
classified as “held for trading” for accounting purposes but not included in the
regulatory trading book, and should be able to justify these exclusions.

4. As the instruments and transactions are included either in the regulatory trading
book or in the non-regulatory trading book (referred to in this guide as the
“banking book”), the ECB understands that the policies required by
Article 104(1) of the CRR should also encompass rules for moving instruments
between the regulatory trading book and the banking book.

5. In order for the ECB to assess the appropriateness and implementation of the
policies and procedures for determining which positions to include in the
regulatory trading book, the ECB can, on the basis of Article 10 of the SSM
Regulation, require institutions to provide a list of types of positions and
instruments allocated to the regulatory trading or the banking book, identify all
related transactions including their relevant characteristics, and justify such
allocation.

6. In view of their nature in terms of trading intent, the ECB considers that the
following types of instruments and transactions are expected to be included in
the regulatory trading book:

(a) instruments in the correlation trading portfolio;

(b) instruments resulting from securities underwriting commitments;

(c) instruments held as accounting trading assets or liabilities (“held for

trading” assets and liabilities); 5

(d) instruments resulting from market-making activities;

(e) listed equities (other than equity investment funds);

5
Under IFRS 9, these instruments would be held within a trading business model and would be
accounted for at fair value though the profit and loss (P&L) account.

ECB guide to internal models – Market risk 119

 (f) trading-related repo-style transactions (repo-style transactions that are
(i) entered into for liquidity management purposes and are (ii) valued at
accrual for accounting purposes, are not presumed to be trading-related);

(g) instruments that would give rise to net short risk positions 6 for equity risk
or credit risk in the banking book;

(h) options including bifurcated embedded derivatives 7 from instruments

issued out of the banking book that relate to credit or equity risk.

7. In view of their nature in terms of trading intent, the ECB considers that the
following types of instruments and positions are expected be included in the
banking book:

(a) unlisted equities;

(b) instruments designated for securitisation warehousing;

(c) real estate holdings;

(d) retail credit and credit to small and medium-sized enterprises (SMEs);

(e) other types of credit;

(f) equity investments in a fund for which the institution cannot obtain daily
price quotes; 8

(g) derivative instruments that have any of the types of instrument mentioned
in points (a) to (f) as an underlying asset;

(h) instruments held for the purpose of hedging a particular risk of a position in
any of the types of instrument mentioned in points (a) to (g).

8. In addition, for each category listed in paragraphs 6 and 7, institutions should

be able to indicate to what extent the corresponding positions are included
within the scope of the internal model approach (IMA).

9. In accordance with Article 106 of the CRR, internal hedges (as defined in
Article 4(1)(96) of the CRR) must be properly documented and not be primarily
intended to avoid or reduce own funds requirements. Therefore, institutions
should be able to identify 9 all internal hedges and should document their

6
An institution will have a net short risk position for equity risk or credit risk if the present value of the
position increases when an equity price decreases or when the credit spread of an issuer or group of
issuers of debt increases.
7
Bifurcation means the separation of a derivative that is embedded in a hybrid security and that has to
be separated according to accounting rules from the host security, and which has to be accounted for
using the accounting rules for derivatives.
8
Where an institution is aware of the underlying investments of the fund on a daily basis, the underlying
investments might be assigned to the trading or banking book depending on their characteristics,
irrespective of the availability of daily price quotes for the fund itself.
9
To “identify” here means to be able to spot these trades among the institution’s transactions. The
institution is not required to be able to segregate internal hedges in dedicated portfolios on which
specific analysis is carried out.

ECB guide to internal models – Market risk 120

 treatment for the purpose of calculating own funds requirements for market risk.
In accordance with Article 106(1)(d) of the CRR, the market risk that is
generated by an internal hedge must be dynamically managed in the regulatory
trading book within the authorised limits. For this reason, the ECB considers
that proper documentation should distinguish between:

(a) hedges of a banking book credit risk exposure (or counterparty credit risk
exposure) using an internal risk transfer with the regulatory trading book;

(b) hedges of a banking book equity risk exposure using a hedging instrument
purchased from the market through the regulatory trading book;

(c) hedges of a banking book interest rate risk exposure using an internal risk
transfer with the regulatory trading book;

(d) hedges of a banking book foreign exchange risk exposure using an

internal risk transfer with the regulatory trading book;

(e) hedges of a banking book commodity risk exposure using an internal risk
transfer with the regulatory trading book;

(f) eligible hedges (as defined in Article 386(1) of the CRR) that are included
in the credit valuation adjustment (CVA) capital charge.

Additionally, institutions should be able to identify internal transactions which

are in the regulatory trading book 10 and within the scope of the internal model,
and show that these transactions do not contribute to the own funds
requirements obtained using the internal model.

10. In accordance with Article 386(3) of the CRR, eligible CVA hedges in the
regulatory trading book must not be included in the calculation of the own funds
requirements for specific risk of debt instruments. The ECB therefore considers
that they should be included in the scope of calculation of own funds
requirements for general risk (for example, included in the VaR and sVaR, or
treated through the framework for risks not in the VaR or sVaR engine – see
Section 7). Additionally, other (i.e. non-eligible) CVA hedges in the regulatory
trading book should be included in the calculation of own funds requirements for
market risk (i.e. general and specific risk).

11. Back-to-back transactions in the regulatory trading book (i.e. transactions

exactly matched with a third-party transaction) are generally included in the
calculation of own funds requirements for market risk. The ECB considers that
back-to-back transactions included in the scope of the internal model may be
excluded from the calculation of own funds requirements, provided that
institutions are able to document them and demonstrate that there are no
residual market risks stemming from these transactions. However, potential
profit and loss (P&L) generated by these back-to-back transactions should be

10
For example, transactions within the scope of the IMA made between two trading units.

ECB guide to internal models – Market risk 121

 considered in the back-testing (for those P&L components that are not excluded
from the actual or hypothetical P&L). This is because, although they do not
carry residual market risks, such back-to-back transactions could generate P&L
(for example, at the inception of the trade, or where the transaction is closed
before maturity).

2.3 Treatment of banking book positions

12. In accordance with Article 92(4)(a) of the CRR, for foreign exchange (FX) risk
and commodities risk the own funds requirements must include those arising
from all the business activities of an institution. Therefore, the ECB understands
that for FX risk and commodities risk, the requirements for the calculation of
own funds and, in particular, the internal models approach are not limited only
to regulatory trading book positions but also include the positions in the banking
book.

13. For institutions that have approval to use the IMA for FX risk, the ECB is aware
that the modelling of banking book FX positions in the internal model may be
challenging owing to different trade booking systems and different market data
processes for the banking book and for the regulatory trading book. In
accordance with Article 363(2) of the CRR, permission to use internal models
for market risk will be granted only if the internal model covers a significant
share of the positions of a certain risk category. 11 Therefore, institutions may
exclude banking book FX positions from the scope of the internal model,
provided that they can demonstrate to the satisfaction of the ECB that the
scope of the approved internal model nevertheless covers a significant share of
the positions of the FX risk category. If that is the case, the banking book FX
exclusions should be treated in the same way as those positions excluded from
the regulatory trading book (see Section 2.5).

14. In accordance with Article 92(3)(c) of the CRR, the own funds requirements for
foreign exchange risk must be determined in accordance with the CRR
provisions for market risk (using either the standardised approach or the IMA).
Therefore, where excluded from the internal model, the banking book FX
positions must be subject to own funds requirements calculated according to
the standardised approach. The ECB considers a prudent approach to be that
for the purpose of this own funds requirement calculation, banking book FX
positions are not netted with regulatory trading book FX positions.

15. In accordance with Article 368(1)(e) of the CRR, institutions must have
established procedures for monitoring and ensuring compliance with a
documented set of internal policies and controls concerning the overall
operation of their internal models. To satisfy the requirements of
Article 368(1)(e) of the CRR, institutions should have documented processes
and methodologies in place for determining FX positions. The ECB considers

11
See also Section 2 of the Final Draft RTS on assessment methodology for IMA and significant share.

ECB guide to internal models – Market risk 122

 that in order to adequately cover the overall operation of the internal model,
such documentation should include, in particular, the intermediate steps
followed for calculating the FX positions, beginning with each individual
subsidiary and proceeding to the group level (for example, before and after
netting, the treatment of intragroup deals, the methodology applied to derive the
FX position of banking book items including whether the institution applies the
provisions of Article 352(3) of the CRR).

If an institution excludes any balance sheet items denominated in foreign

currency from the FX positions in accordance with the provisions of
Article 352(2) of the CRR, it should document in detail which positions are
covered by the exclusion – including details on the materiality of each of them –
and the justification for the exclusion, so that the institution can demonstrate
that the provisions of the article have been complied with. In particular, this also
applies at the consolidated and sub-consolidated levels to balance sheet items
in foreign currencies that stem from consolidated subsidiaries and is without
prejudice to the extent and manner of prudential consolidation prescribed in
Article 18 of the CRR.

16. In order for the ECB to assess whether banking book FX positions have been
excluded from the scope of application of the IMA for the sole purpose of
reducing the own funds requirements for market risk, the ECB can, on the basis
of Article 10 of the SSM Regulation, require an institution to estimate the
difference between the own funds requirements calculated under the
standardised approach and the own funds requirements calculated under the
IMA 12 for those banking book FX positions. 13

17. The ECB is aware that (in contrast to FX positions) it is not common practice to
include commodities positions in the banking book. However, where an
institution has permission to use the internal model for commodities risk,
commodities positions in the banking book should not be systematically
excluded from the scope of application of the internal model.

2.4 Partial use models

18. If an institution does not have permission to use an internal model to calculate
the own funds requirements for market risk for all of the six risk categories listed
in Article 363(1) of the CRR, but only for some of them (for the purposes of this
guide, referred to as “partial use”), the institution must apply the standardised
approach in accordance with Article 363(2) of the CRR to calculate the own
funds requirements for the risk categories for which it has not been granted
permission.

12
For estimating the own funds requirements under the IMA, the calculation can be based on only one
date, rather than the average over the last 60 business days.
13
FX positions excluded from the scope of the calculation of the own funds requirements for market risk
in accordance with Article 352(2) of the CRR would not need to be considered for this comparison.

ECB guide to internal models – Market risk 123

 19. In the case of portfolios for which the bulk of the risks fall outside the scope of
the approved risk categories of an internal model, 14 institutions may completely
carve out such portfolios from the scope of application of the internal model,
provided that the internal model still covers a significant share of the relevant
risk categories. The ECB considers it best practice to carve out such portfolios
only if the overall own funds requirements for market risk after the carve-out are
higher than they would have been if the carve-out had not been performed.
Institutions should duly notify the ECB of such exclusions in accordance with
the Commission Delegated Regulation on materiality of extensions and
changes of the IMA. Institutions should, nevertheless, ensure that the risks of
these portfolios are adequately managed. Institutions should determine the own
funds requirements for the carved-out portfolios according to the standardised
approach (for all risk categories, including those for which the institution has
permission to use the internal model).

20. Portfolios for which the bulk of the risks fall within the scope of the model
approval should be included in the calculation of own funds requirements using
the internal model (for the risk categories within the scope of the internal model
permission). The own funds requirements for the risks not included in the scope
of the internal model permission should be determined according to the
standardised approach.

21. In accordance with Article 362 of the CRR, general risk of debt instruments
refers to the risk of a price change due to a change in the level of interest rates.
The ECB considers that this is a reference to risk-free interest rates. In
accordance with Article 362 of the CRR, general risk of equity instruments
refers to the risk of a price change due to a change of a broad equity-market
movement unrelated to any specific attributes of individual securities. The ECB
considers that this is a reference to index movements. An institution without
permission to use the IMA for the specific risk of debt or equity instruments, and
which applies a different definition of general risk of debt or equity instruments
(as applicable) should be able to demonstrate that the definition/s it applies lead
to at least the same level of own funds requirements that would result from
applying the principles of this paragraph.

2.5 Exclusion of positions in the regulatory trading book from the scope
of application of the IMA

22. In accordance with Article 363(2) of the CRR, an internal model must cover a
significant share of the positions of each risk category for which the permission
is granted. The ECB understands that this requirement applies not only on the
date on which the permission is granted, but on an ongoing basis. The ECB is
aware that institutions have a certain margin of discretion not to include all
positions exposed to the relevant approved risk categories, provided that the

14
A typical example would be a portfolio of equity options for an institution that has no permission to
model equity risk, so that it only models the position risk of debt instruments.

ECB guide to internal models – Market risk 124

 internal model still covers a significant share of those positions. In the ECB’s
view, exclusions would be justified where the inclusion of those positions in the
internal model is operationally challenging (for example, in the case of products
requiring a more sophisticated modelling approach). Exclusions should never
be made for the sole purpose of reducing the overall own funds requirements
for market risk.

The ECB considers that to be able to demonstrate that the internal model
covers a significant share of positions, institutions should monitor the exclusion
of market risk positions, including the materiality of those positions. In the case
of FX and commodities risks, this monitoring should also extend to exclusions
of banking book positions. This monitoring should consider all excluded
positions in accordance with each relevant scope of approval of the internal
model, which could be at individual (“solo”), sub-consolidated, or consolidated
level. Institutions should be able to justify such exclusions and demonstrate that
the risk of the excluded positions is adequately managed.

23. The own funds requirements for the positions deliberately excluded by the
institution from the internal model (including any banking book positions) should
be determined according to the standardised approach. For these positions,
institutions should be able to demonstrate that the level of own funds
requirements under the standardised approach is commensurate with their risks
and that the exclusion was not made for the sole purpose of reducing the
overall own funds requirements for market risk. 15

24. The ECB considers that an appropriate approach to calculating the own funds
requirements for derivatives on unusual underlyings (such as temperature,
weather or mortality) 16 could be to include them in the scope of the IMA (the
“exotic” risks might be treated under the “risks not in the model engines
(RNIME) 17” framework, where necessary). However, institutions may use the
standardised approach for these positions, provided that they can demonstrate
that the level of own funds requirements under the standardised approach is
commensurate with the risks of such positions.

25. The ECB considers that a materiality criterion at transaction level (for example,
a notional amount lower than a certain EUR amount) is not an appropriate
criterion by itself for an exclusion from the scope, because the cumulative effect
of these transactions may be a material position. Therefore, this type of
exclusion should not be applied.

26. If back-to-back transactions are excluded from the calculation of own funds
requirements using the internal model (in the circumstances set out in
paragraph 11), it is not necessary – as it is with other excluded positions – to

15
This understanding is also supported by Article 13(b) of the Final Draft RTS on assessment
methodology for IMA and significant share.
16
The EBA considers that certain “unusual” underlyings, such as freight rate, weather derivatives or
emission certificates can be considered as, or assimilated to, commodities (see the EBA Single
Rulebook Q&A, Question ID: 2014_934).
17
See Section 7 for more details.

ECB guide to internal models – Market risk 125

 calculate the own funds requirements for these transactions under the
standardised approach, as no residual market risks stem from them.

2.6 Treatment of specific positions

2.6.1 Own-debt exposures

27. For the purposes of this guide, the meaning of “own debt” requires clarification,
given that the CRR does not provide a definition. As defined in Article 4(1)(47)
of the CRR, “consolidated situation” means the situation that results from
applying the requirements of the CRR to an institution as if that institution
formed, together with one or more other entities, a single institution. As defined
in Article 4(1)(49) of the CRR, “sub-consolidated basis” means “… on the basis
of the consolidated situation of a parent entity … that is not the ultimate parent
entity”. Therefore, the ECB considers an acceptable approach to be that when
determining their own-debt positions, institutions using an IRC model also take
into account the debt positions in their subsidiaries within the scope of
prudential consolidation, depending on the level within the group of the
institution using the IRC model.

28. By way of illustration, the positions described in the following situations can be
considered as own-debt exposures. Figures 1, 2 and 3 below each illustrate an
example, as indicated.

Figure 1: Where the institution is the group EU parent institution –

all positions in debt exposures to institutions within the prudential consolidation
scope should be considered as own-debt exposures.

Figure 2: Where the institution is the sub-consolidating institution or financial

holding company that is not the ultimate EU parent institution –
all positions in debt exposures to institutions within the relevant prudential sub-
consolidation scope should be considered as own-debt exposures.

Figure 3: Where the institution is a subsidiary without dependent subsidiaries

(solo) – all debt exposures of an institution to itself at the solo level should be
considered as own-debt exposures.

This is without prejudice to other definitions of own debt that institutions may
apply and that the ECB will assess on a case-by-case basis to take into account
specific circumstances.

ECB guide to internal models – Market risk 126

Figure 1
Own-debt positions at the consolidated level

EU parent institution
P

Sub-consolidating institution Sub-consolidating institution

or financial holding company or financial holding company
SC1 SC2

Subsidiary Subsidiary Subsidiary Subsidiary

S1 S2 S3 S4

Own-debt positions at the consolidated level are shown with a blue background.

Figure 2
Own-debt positions at the sub-consolidated level of the sub-consolidating institution
or financial holding company SC1

EU parent institution
P

Sub-consolidating institution Sub-consolidating institution

or financial holding company or financial holding company
SC1 SC2

Subsidiary Subsidiary Subsidiary Subsidiary

S1 S2 S3 S4

Own-debt positions at the sub-consolidated level of the sub-consolidating institution or financial holding company SC1 are shown with
a blue background, while non-own-debt positions at the sub-consolidated level of SC1 are shown with a red background.

ECB guide to internal models – Market risk 127

 Figure 3
Own-debt positions at the solo level of subsidiary S1

EU parent institution
P

Sub-consolidating institution Sub-consolidating institution

or financial holding company or financial holding company
SC1 SC2

Subsidiary Subsidiary Subsidiary Subsidiary

S1 S2 S3 S4

Own-debt positions at the solo level of subsidiary S1 are shown with a blue background, while non-own-debt positions at the solo level
of S1 are shown with a red background.

29. Under paragraphs 7.1 and 7.2 of the EBA Guidelines on the IRC, for long or
short positions in an institution’s own debt which may arise from trading or
market-making activity in its own-debt issuances, or from trading protection in
the institution’s own name (for example, via an index), the institution should only
model the migration risk. The default risk of these positions should not be
modelled in the IRC approach.

To ensure consistency with the IRC model when modelling the specific risk of
debt instruments in the VaR and sVaR models, institutions should include their
own creditworthiness. 18 The ECB considers it best practice to model such own
creditworthiness as (a) separate risk factor(s) in the VaR and sVaR models. 19

30. As regards the general risk of debt instruments for own debt, and in the
absence of any specific provision in the CRR or the EBA Guidelines on the IRC,
the ECB understands that the general risk of own-debt instruments should be
accommodated in the internal model if the institution has the relevant approval.

2.6.2 Positions in defaulted debt

31. Under paragraph 4.5 of the EBA Guidelines on the IRC, institutions should
include in the calculation of the IRC the positions in defaulted debt that are held
in the regulatory trading book, where material. In order for the ECB to assess
the appropriateness of the treatment of such positions, the ECB can, on the
basis of Article 10 of the SSM Regulation, require institutions to provide a list of

18
This refers exclusively to position risk taken against debt issued by the institution or derivatives
referencing that debt and does not refer to debit valuation adjustments (DVA).
19
This understanding is also supported by Recital (20) of the Final Draft RTS on assessment
methodology for IMA and significant share.

ECB guide to internal models – Market risk 128

 all positions in defaulted debt that are held in the regulatory trading book, along
with the following information:

(a) the market value of the exposure;

(b) notional value of the exposure;

(c) a Boolean variable indicating whether the positions are included in the
VaR/sVaR;

(d) a Boolean variable indicating whether the positions are included in the
IRC.

32. The ECB considers that to reflect paragraph 4.5 of the EBA Guidelines on the
IRC, material positions in defaulted debt should be included in the scope of the
IMA. It also considers it best practice that non-material positions are included in
the IMA; either in the VaR, sVaR (and IRC) engines or under the RNIME
framework for the VaR, sVaR (and IRC).

33. Under paragraph 4.5 of the EBA Guidelines on the IRC, positions in defaulted
debt held in the regulatory trading book should, where material, be included in
the calculation of the IRC. In accordance with the requirement in Article 372(d)
of the CRR as it relates to Article 370(e), the IRC model must be sensitive to
material idiosyncratic differences between similar but not identical positions.
Therefore, the ECB considers that when modelling the risk of price changes of
positions in defaulted debt in the IRC, and to account for those idiosyncratic
differences, institutions should apply a specific calibration of the recovery rates
which is appropriate for the positions in defaulted debt. If an institution does not
have the capacity to model price changes of positions in defaulted debt in the
IRC model (for example, owing to IT constraints), the ECB considers that an
appropriate way to account for those differences is to apply a stressed price of
the positions in defaulted debt that is proven to be adequately conservative
given the quantile and holding period of the IRC.

34. The ECB considers that default should be an absorbing state (i.e. once a
position has defaulted it does not migrate to a different state). Therefore, the
ECB considers it best practice that no migration from default to non-default
states is considered in the IRC model.

2.6.3 Collective investment undertakings

35. This paragraph, and the following paragraphs 36 to 41 inclusive, apply to all
exposures that institutions may have in collective investment undertakings
(CIUs), as referred to in Article 4(1)(7) of the CRR (including exchange-traded
funds (ETFs), equity investments in hedge funds, etc.) – which for the purposes
of these paragraphs are referred to as “positions in CIUs” – and also apply to
derivative instruments that have these instruments as underlying assets.

ECB guide to internal models – Market risk 129

36. In accordance with Articles 368(1)(e) and 104(2) of the CRR, institutions must
have established procedures for ensuring compliance with a documented set of
internal policies and controls concerning the overall operation of their internal
models and the regulatory trading book. Therefore, the ECB considers that in
order to ensure that an appropriate treatment is applied to positions in CIUs in
their internal models for market risk, institutions should have a documented set
of policies and controls in place to identify, for each CIU, the following:

(a) the risk categories, as listed in Article 363 of the CRR, to which the CIU is
subject;

(b) whether the criteria as defined in Article 349 of the CRR are fulfilled;

(c) whether daily look-through is possible, as referred to in Article 350(1) of

the CRR;

(d) where daily look-through is not possible, whether the requirements as laid
down in Article 350(2) of the CRR are satisfied – in particular, institutions
should define and document the methodology used to assess the
correlation between the CIU and the index/basket that it tracks;

(e) the extent to which the CIU can be marked-to-market daily by reference to
an active, liquid two-way market as referred to in Article 104(2)(b) of the
CRR (for example, whether a daily liquid price is available).

37. The above information in relation to each CIU should be documented. In the
ECB’s view it is necessary to update the information regularly, in order to
ensure that the documented information is based on the current characteristics
of the positions in CIUs and fully reflects the market risk to which the positions
are exposed. The ECB considers that an appropriate update frequency for
existing positions in CIUs is at least annual, as one year is a reasonable time in
which significant changes in the market or in the positions in CIUs could occur.
In addition, this time frame would allow institutions to use the outcomes of the
updates in the review of their overall risk management process, as referred to in
Article 368(2) of the CRR. For new positions in CIUs, the ECB considers that
the procedure should take place before the investment in a new CIU is
approved internally, in order to ensure that institutions compute own funds
requirements for the positions in CIUs in compliance with the CRR
requirements.

38. For the foreign exchange market risk related to CIUs, the CRR contemplates a
specific treatment. In accordance with Article 367(2)(b) of the CRR, the actual
foreign exchange positions of a CIU must be taken into account in the internal
model. Institutions may rely on third-party reporting of the foreign exchange
positions of a CIU where the correctness of that reporting is adequately
ensured. If an institution is not aware of the foreign exchange positions of a
CIU, this position must be carved out and treated in accordance with
Article 353(3) of the CRR.

ECB guide to internal models – Market risk 130

 39. As regards general and specific risk of equity instruments, general and specific
risk of debt instruments, and commodities risk: for those positions in CIUs for
which the conditions for either the look-through approach (as referred to in
Article 350(1) of the CRR) or the representation approach (as referred to in
Article 350(2) of the CRR) are met, the ECB considers that in order to ensure
an accurate risk measurement, the own funds requirements for the general and
specific risk of equity instruments, the general and specific risk of debt
instruments and the commodities risk should be calculated by incorporating the
underlying investments of the relevant CIU – or the index/basket that it tracks –
into the internal model for the risk categories for which the institution has
permission to use internal models.

Where an institution with internal model approval for specific risk of debt
instruments includes listed equity positions in the IRC, it should be consistent in
including in the IRC either the underlying listed equity positions of the CIU, or
those of the index/basket that the CIU tracks.

40. For positions in CIUs where neither the conditions for the look-through
approach as referred to in Article 350(1) of the CRR nor the conditions for the
representation approach as referred to in Article 350(2) of the CRR are met, the
following apply.

(a) In accordance with Article 364(2)(a) of the CRR, institutions using an

internal model to calculate their own funds requirements for specific risk of
debt instruments must fulfil an additional own funds requirement for
specific risk of debt instruments in accordance with the standardised
approach for positions in CIUs under Articles 348 to 350 of the CRR.

(b) The ECB considers that the positions in CIUs can be incorporated into the
VaR and sVaR models as a single risk factor to account for the general
and specific risk of equity, the general risk of debt instruments, and the
commodities risk of the positions in CIUs. As is the case for any other
position, sufficient objective information on market risk should be available.
The ECB considers that a suitable approach is to use the daily liquid price
of the CIU.

41. For those risk categories in respect of which the institution does not use an
internal model to compute own funds requirements for market risk, or for
positions in CIUs to which none of the above-mentioned provisions apply, the
institution should compute the own funds requirements for the respective
positions in CIUs according to the standardised approach for CIUs.

2.7 Aggregation requirements

42. In accordance with Article 363 of the CRR, competent authorities may grant
permission to institutions to calculate their own funds requirements for market
risk by using their internal models instead of, or in combination with, the
standardised approach – provided that the internal model covers a significant

ECB guide to internal models – Market risk 131

 share of the positions of a certain risk category. In accordance with Articles 6(1)
and 11(1) of the CRR, institutions must meet their own funds requirements on
an individual basis (unless a derogation in the circumstances set out in Article 7
of the CRR has been granted), and the parent institution in a Member State
must comply with the own funds requirements on a consolidated basis.

Therefore, the institution at the highest level of consolidation operating within

the framework of the Single Supervisory Mechanism should be able to specify:

(a) which legal entities within the group have been granted permission to use
an internal model for calculating own funds requirements for market risk;

(b) the scope of application of each model permission (i.e. individual, sub-
consolidated or consolidated);

(c) the risk categories listed in Article 363(1) of the CRR for which each
permission has been granted.

The ECB also understands that, for institutions to calculate own funds
requirements at the consolidated level by using their internal models, a
permission to use their internal models at consolidated level is required under
Article 363(2) of the CRR.

43. In those cases in which the scope of the permission applies at the consolidated
or sub-consolidated level, institutions should be able to provide a list of legal
entities included in the scope of consolidation or sub-consolidation, specifying
which of those entities effectively contribute to the market risk own funds
requirements determined by using the internal model. In order for the ECB to
assess how the own funds requirements are determined, institutions should
also be able to provide information on how each legal entity is integrated into
the information system infrastructure of the risk management system and
whether impediments exist to such integration.

If impediments exist to the integration of the risk numbers from individual legal
entities, the ECB considers it best practice to integrate the risk numbers from
these entities into the model-based own funds requirements by applying a
simple sum aggregation.

44. In accordance with Article 368(1)(a) of the CRR, the internal model must be
closely integrated into the daily risk management process of the institution and
serve as the basis for reporting risk exposures to senior management.
Therefore, the ECB understands that institutions should ensure an integrated
and harmonised risk management across all legal entities included in the scope
of the model permission.

45. In accordance with Article 325 of the CRR, institutions may use positions in one
institution or undertaking to offset positions in another institution or undertaking
only for the purpose of calculating net positions and own funds requirements for
market risk on a consolidated basis, and only subject to the permission of the
competent authorities. The ECB understands that this requirement applies to all

ECB guide to internal models – Market risk 132

 positions, in particular to intragroup transactions. 20 The offsetting of positions
can be performed irrespective of the calculation approach that is applied for
market risk own funds requirements (i.e. IMA or standardised approach). In
order for the ECB to assess how these requirements are fulfilled, institutions
should document how the offsetting of positions is performed.

46. In accordance with Article 367(3) of the CRR, an institution may use empirical
correlations within risk categories and across risk categories only if the
institution's approach to measuring correlations is sound and implemented with
integrity. The ECB understands that for empirical correlations to be sound they
should be based on reliable and objective data. If this cannot be ensured, an
institution should use the simple sum aggregation of stand-alone risk numbers
within risk categories or across risk categories.

In order for the ECB to assess the soundness and integrity of the
implementation of the use of empirical correlations, the ECB can, on the basis
of Article 10 of the SSM Regulation, require institutions to provide the stand-
alone VaR and sVaR corresponding to each of the following risk classes: 21
interest rate risk; equity risk; commodity risk; foreign exchange risk; and credit
spread risk.

3 Regulatory back-testing of VaR models

3.1 Regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 34

105 (10)

106 (3)

352 (2)

366

368 (1)

386 (3)

SSM Regulation 15/10/2013 10

Other references

Final Draft RTS on assessment methodology 22/11/2016 40 (4), (5), (10)

for IMA and significant share
43 (4)

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IMA and significant share will become an

20
This interpretation is also supported by the explanation in footnote 7 of the Final Draft RTS on
assessment methodology for IMA and significant share.
21
If this is not possible, the calculation should be based on the approved risk categories in accordance
with Article 363(1) of the CRR.

ECB guide to internal models – Market risk 133

 additional relevant regulatory reference. Currently that document only exists in
a final draft version.

3.2 Scope of application of regulatory back-testing

47. The scope of application of regulatory back-testing, as referred to in Article 366

of the CRR, should be clearly documented.

48. In accordance with Article 366(3) of the CRR, regulatory back-testing compares
the hypothetical and actual changes in the portfolio’s value (“hypothetical P&L”
and “actual P&L”) with the related one-day value-at-risk number generated by
the institution's model. Therefore, the changes in value of all of (and only) the
instruments and transactions entailing positions included in the scope of
calculation of the VaR model should be considered in the calculation of the
hypothetical P&L and the actual P&L.

49. In particular, if the institution is authorised to apply the IMA for foreign exchange
and/or commodities risk positions, and the banking book positions in these risk
categories are included in the scope of the internal model, the institution should
include these banking book positions in the back-testing and should clearly
document how the actual and the hypothetical P&L of these positions are
calculated.

50. Regarding positions in the banking book that are included in the IMA, only the
changes in value of market data pertaining to FX risk and commodity risk
should be taken into account in the calculation of the actual and the
hypothetical P&L. Only for those instruments or transactions where the
separation of the positions stemming from risk categories other than FX risk or
commodities risk (for example, risk of debt instruments) is operationally
challenging or its effect on the total P&L is immaterial, an institution may include
changes in value of market parameters pertaining to all risk categories in the
actual and the hypothetical P&L. Institutions should be able to justify the
immateriality or the operational challenge, if applicable. In particular,
considering only the effect of FX risk in the actual and the hypothetical P&L is
not deemed to be operationally challenging.

51. Positions excluded from the calculation of the own funds requirements for
market risk on the basis of a permission granted by the competent authorities
under Article 352(2) of the CRR (i.e. (i) positions taken in order to hedge against
the adverse effect of the exchange rate on the institution’s capital ratios, or
(ii) positions which an institution has which relate to items that are already
deducted in the calculation of own funds) should also be excluded from the
scope of application of the back-testing.

52. Positions that are excluded from the regulatory trading book for the purpose of
calculating capital requirements on the basis of Article 106(3) of the CRR
(i.e. internal or external credit derivative hedges for banking book credit risk

ECB guide to internal models – Market risk 134

 exposure or counterparty risk exposure) should also be excluded from the
scope of application of the back-testing.

53. In accordance with Article 386(3) of the CRR, eligible hedges that are included
in the calculation of own funds requirements for CVA risk must not be included
in the calculation of the own funds requirements for specific risk. Therefore, the
change in value of those positions that are attributable to specific risk of those
eligible hedges should also be excluded from the actual and the hypothetical
P&L. However, if the own funds requirements for general risk of these eligible
hedges are calculated using the VaR model (see paragraph 10), the change in
value of those positions that are attributable to general risk should be included
in the actual and in the hypothetical P&L.

54. Hedges which under Article 386(3) of the CRR are not eligible hedges for
regulatory CVA should be included in the VaR calculation and in the scope of
calculation of the hypothetical P&L and the actual P&L for back-testing.

3.3 Historical period used to perform back-testing, definition of

business days, and documentation

55. In accordance with Article 366(2) of the CRR, the addend to the multiplication
factors must depend on the number of overshootings for the most recent 250
business days.

56. For the purpose of paragraph 55, institutions should define and document local
and global business days according to the guidance set out in paragraphs 57 to
59.

57. The ECB considers that when the business trading unit of an institution is
conducting planned business operations on the risk positions (even with a
reduced number of staff) on a given day in a given location, this constitutes a
local business day for the institution in that location. Therefore, as it is a
business day, it requires actual and hypothetical P&L calculation, VaR
calculation and market risk monitoring and reporting. Institutions should
consistently define their business days, and therefore should be able to justify
any non-business days. Unchanged risk positions are a necessary but not
sufficient condition to demonstrate adequately that a particular day constitutes a
non-business day.

58. The ECB considers that global business days should be defined at the
consolidated level (or for a national sub-group, at the sub-consolidated level),
and that for the purpose of defining global business days the institution’s most
important trading location (the “reference location”) should be used, to ensure
that the most important trading activity is adequately captured in back-testing.
Global business days should include at least the local business days of the
reference location. If there are two or more important trading locations (of
approximately the same size), the institution should choose one reference
location and is allowed to add additional global business days based on the

ECB guide to internal models – Market risk 135

 local business days of the other important trading locations. In such a case, the
rationale underlying this choice should be documented.

59. For every global business day, actual and hypothetical P&L calculation, VaR
calculation and market risk monitoring and reporting are required. 22 However, if
such calculations are carried out on non-global business days, these should not
be used for the purpose of regulatory back-testing. The actual and hypothetical
P&L used for back-testing should always be the P&L between two consecutive
global business days, and should be compared with the related one-day VaR
forecast for a one-day holding period between those two global business days,
and be based on the composition of the portfolio on the first of those global
business days.

60. Based on Article 368(1)(e) of the CRR, the ECB understands that institutions
should have a documented policy and procedure describing how they calculate
the actual and hypothetical P&L. The ECB considers that to be fit for purpose,
the policy and procedure should include, at least, the following key information:

(a) how the actual P&L is calculated and, in particular, the differences
between the economic 23 and actual P&L;

(b) the fees, commissions and net interest income excluded from the actual
P&L;

(c) how the hypothetical P&L is calculated and, in particular, the differences
between the actual and hypothetical P&L;

(d) the valuation adjustments not updated every day and whether or not they
are included in the P&L time series.

In order for the ECB to assess the appropriateness and implementation of the
policy and procedure for the calculation of the actual and hypothetical P&L, the
ECB can, on the basis of Article 10 of the SSM Regulation, require institutions
to provide, for a sample of transactions or portfolios, detailed decompositions of
economic, actual and hypothetical P&L into their elements.

22
All positions of trading units in a location with local non-business days should be included in the
calculation of the consolidated figures.

23
As defined in the glossary.

ECB guide to internal models – Market risk 136

3.4 Calculation of actual P&L

3.4.1 General rules

61. In accordance with Article 366(3) of the CRR, actual P&L must be based on
actual changes in the portfolio's end-of-day value. Therefore, the ECB
understands that the actual P&L should correspond closely to the daily
economic P&L as reflected in the books and records of the institution, with the
exception of certain elements as specified in the following paragraphs.

62. Actual P&L should include the profit and loss stemming from intraday activities,
as they change the portfolio’s value. In accordance with Article 366(3) of the
CRR, fees, commissions and net interest income must be excluded from the
actual P&L. The definitions and methods used to apply this exclusion should be
clearly documented.

63. In accordance with Article 366(3) of the CRR, fees, commissions, and net
interest income must be excluded from the portfolio's actual value when
computing the actual P&L. The ECB understands that these exclusions from the
portfolio’s actual end-of-day value are required by the CRR so that fees,
commissions, and interest gains or losses are removed from the daily actual
P&L in order to ensure that the back-testing assesses whether P&L fluctuations
driven by market risk are accurately captured by the VaR model. Furthermore,
the ECB acknowledges that the net interest income definition applied to banking
book items accounted for at amortised cost cannot easily be transferred to the
fair value items in the regulatory trading book.

The ECB considers that it is acceptable for an institution to define the net
interest income in the regulatory trading book as equal to zero; this leads to no
P&L component being excluded as net interest income.

The ECB considers that where an institution uses another definition for net
interest income it should be able to duly justify this approach, taking into
account its trading strategy. In any case, theta effects (for example, options
theta) and P&L contributions of unearned credit spreads should not contribute
to the net interest income, because they are susceptible to market risk.

64. The actual P&L is calculated for instruments and transactions entailing
positions 24 in the regulatory trading book and banking book which are within the
scope of the IMA. In accordance with Article 366(3) of the CRR, the actual P&L
must be based on the actual value at the end of the subsequent day. Therefore,

24
For the purposes of this guide, a “position” is defined as a risk position (as stated in the glossary). For
example, for a bond denominated in FX and where the scope of the IMA approval does not include FX
risk, the FX risk position is not in the scope of the VaR model, while the FX risk is reflected in the actual
P&L.

ECB guide to internal models – Market risk 137

 the ECB considers that the change in value of all market risk parameters (even
those that are not modelled in the VaR) should be taken into account in the
actual P&L.

65. In order to ensure that the actual P&L corresponds closely to the daily
economic P&L as reflected in the books and records of the institution, the ECB
considers that the pricing methods, model parametrisations and market data
should be the same as those used to compute the daily economic P&L.

3.4.2 Valuation adjustments

66. Because the actual P&L should correspond closely to the daily economic P&L
as reflected in the books and records of the institution, all valuation adjustments
or reserves made in the economic P&L are also relevant for the calculation of
the actual P&L. Therefore, institutions should clearly document all such
valuation adjustments and reserves (methodology, frequencies, calculation
process, etc.).

67. Except for the elements referred to in paragraphs 68 to 70 below, fair value
adjustments 25 and all other valuation adjustments or reserves (hereinafter
referred to as “adjustments”) made in the economic P&L should be included in
the actual P&L even if they are not computed on a daily basis – provided that
they are in the scope of market risk.

68. The ECB considers that credit valuation adjustments (CVA) should be excluded
from the actual P&L, 26 because they receive a specific regulatory treatment.
The same applies to debit valuation adjustments (DVA), due to their nature as
the reverse side of CVA.

69. Additional valuation adjustments (AVA) 27 that are calculated to obtain the
prudent value of the positions in the regulatory trading book should also be
excluded from the actual P&L, as they receive a specific regulatory treatment
under Article 34 of the CRR as an additional layer of prudence.

70. Changes in portfolio value generated by the default of a counterparty should be

excluded from the actual P&L, provided that the institution demonstrates that
they are related solely to counterparty credit risk; this is because the
corresponding profits or losses are taken into account in the institution’s
counterparty credit risk framework. Conversely, profits or losses due to the

25
This refers at least, but is not limited to, the examples of potential valuation adjustments listed in
Article 105(10) of the CRR.
26
This understanding is also supported by Article 40(5)(d) of the Final Draft RTS on assessment
methodology for IMA and significant share.
27
As set out in Commission Delegated Regulation (EU) 2016/101 of 26 October 2015 supplementing
Regulation (EU) No 575/2013 of the European Parliament and of the Council with regard to regulatory
technical standards for prudent valuation under Article 105(14) (OJ L 21, 28.1.2016). Those categories
may overlap with fair value adjustments (e.g. market price uncertainty, close-out costs, etc.).

ECB guide to internal models – Market risk 138

 default of a bond or other fixed income security are not in the scope of
counterparty credit risk and should therefore be included in the actual P&L.

71. In accordance with Article 366(3) of the CRR, back-testing must be performed
daily on the portfolio's end-of-day value. It could therefore be understood that
changes in valuation adjustment figures should be computed daily to obtain an
accurate portfolio end-of-day value. However, the ECB is aware that some
valuation adjustments are not calculated daily by some institutions. In such
cases, the ECB considers that changes in valuation adjustment figures should
be taken into account in the actual P&L on the business day which is taken as
the reference date for the calculation of the valuation adjustment. As a
consequence, institutions should not apply any kind of smoothing or distribution
over several dates in relation to changes in valuation adjustment figures.

3.5 Calculation of hypothetical P&L

72. In accordance with Article 366(3) of the CRR, the hypothetical P&L is based on
changes in the portfolio's value assuming unchanged positions at the end of the
subsequent day. The ECB understands that the term “unchanged positions”
refers to an unchanged composition of the portfolio in terms of instruments and
transactions. Therefore, the P&L generated by intraday trading and by new
transactions entered (or maturing transactions) during the day is not taken into
account. The ECB understands that the term “subsequent day” could imply a
passage of time of one business day and that this could lead to a change in the
risk positions due solely to this passage of time. Therefore, institutions may
choose to include the passage of time of one business day in the hypothetical
P&L.

The ECB understands from Article 366(3) of the CRR that back-testing on the
hypothetical P&L is intended to focus on detecting deficiencies in the internal
model. Therefore, back-testing on hypothetical P&L should be used as a
statistical test of the integrity of the VaR measure, allowing for a more “pure”
testing of the model. 28

73. In accordance with Article 366(3) of the CRR, the hypothetical P&L is to be
based on the portfolio's value, assuming unchanged positions, while the actual
P&L is to be based on the portfolio’s actual value. Therefore, the ECB considers
that any adjustments taken into account in order to obtain the actual value of
the portfolio should not be considered in the hypothetical P&L, unless they are
part of the VaR model. 29 Consequently, any other profit and loss element – such
as credit valuation adjustments (CVA), debit valuation adjustments (DVA),
additional valuation adjustments (AVA) and any other valuation adjustments –

28
This understanding is also supported by Section 2.3.2 of the report of the Final Draft RTS on
assessment methodology for IMA and significant share.
29
This understanding is also supported by Article 40(5)(d) of the Final Draft RTS on assessment
methodology for IMA and significant share.

ECB guide to internal models – Market risk 139

 as well as fees, commissions and net interest income, should not be included in
the hypothetical P&L. 30

74. As both the actual and the hypothetical P&L are based on the portfolio’s value,
they should be calculated using the same pricing framework. Therefore, the
hypothetical P&L should be computed using the same pricing methods, model
parametrisations and market data as those used to compute the daily economic
P&L. Where an institution computes the hypothetical P&L in a system that is
different from the one that is used to produce the daily economic P&L, the risk
is that differences in the computations could occur. To minimise this risk, the
institution should ensure that differences in market value computations at
instrument or transaction level and at the total hypothetical P&L level are
negligible, and should monitor the alignment frequently.

75. The back-testing on the hypothetical P&L should be used as a statistical test of
the integrity of the VaR measure. Therefore, paragraph 74 applies to partial use
models so that only the changes in market value due to changes in pricing risk
factors within the risk categories in the scope of the model are considered, and
the other pricing risk factors outside the scope of the model are held fixed. 31 For
example, (i) if the institution is authorised to use an internal model for general
interest rate risk only, the hypothetical P&L should include the changes in value
of market parameters pertaining to general interest rate risk only; or (ii) if FX
risk is not in the scope of the model, market value changes due to changes in
the FX rate should not be reflected in the hypothetical P&L. 32

In the case of partial use models, only for those instruments or transactions
where the exclusion of the P&L stemming from risk categories not included the
scope of the internal model is operationally challenging or its effect on the total
P&L is immaterial, an institution may include in the hypothetical P&L those
changes in value of market parameters pertaining to all risk categories. 33
Institutions should be able to justify the immateriality or the operational
challenge, if applicable. In particular, excluding the effect of FX risk in the
hypothetical P&L is not automatically deemed to be operationally challenging.

76. The passage of time effect (theta effect) should be considered (or not) in the
VaR and in the hypothetical P&L in a consistent manner. However, if institutions

30
This understanding is also supported by Article 40(4)(d) of the Final Draft RTS on assessment
methodology for IMA and significant share.
31
The requirement to use the market quote or pricing methods and model parametrisations used for the
economic P&L takes precedence over the requirement to change only the risk factors within the risk
categories in scope of the model in this case.
32
For example, let pv(𝑡𝑡; 𝑝𝑝; 𝑔𝑔𝑡𝑡 , 𝑠𝑠𝑡𝑡 , 𝑒𝑒𝑡𝑡 , 𝑓𝑓𝑡𝑡 , 𝑥𝑥𝑡𝑡 ) be the present market value of a position at time 𝑡𝑡 used in the
economic P&L calculation, depending on some parameter set 𝑝𝑝 (not explicitly time dependent), and risk
factor sets 𝑔𝑔𝑡𝑡 , 𝑠𝑠𝑡𝑡 , 𝑒𝑒𝑡𝑡 , 𝑓𝑓𝑡𝑡 , 𝑥𝑥𝑡𝑡 for all risk categories at time 𝑡𝑡. If the specific interest rate risk 𝑠𝑠 and the FX risk
𝑥𝑥 are not in the scope of the model, the risk factor values in those categories do not change from the
previous time 𝑡𝑡0, and the hypothetical P&L at time 𝑡𝑡 should be calculated as:
HypoP&L(𝑡𝑡) = pv(𝑡𝑡𝑥𝑥 ; 𝑝𝑝; 𝑔𝑔𝑡𝑡 , 𝑠𝑠𝑡𝑡0 , 𝑒𝑒𝑡𝑡 , 𝑓𝑓𝑡𝑡 , 𝑥𝑥𝑡𝑡0 ) − pv(𝑡𝑡0 ; 𝑝𝑝; 𝑔𝑔𝑡𝑡0 , 𝑠𝑠𝑡𝑡0 , 𝑒𝑒𝑡𝑡0 , 𝑓𝑓𝑡𝑡0 , 𝑥𝑥𝑡𝑡0 ), where
𝑡𝑡𝑥𝑥 = 𝑡𝑡0 if VaR uses an instantaneous shock; or 𝑡𝑡𝑥𝑥 = 𝑡𝑡 if VaR includes theta for consistency.
33
In cases where the exclusion of the P&L stemming from risk categories not included in the scope of the
internal model is operationally challenging or its effect on the total P&L is immaterial, if a market price
that incorporates all risks is used in the economic P&L it should also be used in the hypothetical P&L.

ECB guide to internal models – Market risk 140

 include the passage of time in the P&L and not in the VaR, or vice versa, they
should be able to demonstrate that the effect of this inconsistency is not
material. 34

3.6 Counting of overshootings

77. In accordance with Articles 366(2) and 366(3) of the CRR, the back-testing
addend is determined as the higher of the number of overshootings under
hypothetical and actual changes in the value of the portfolio for the most recent
250 business days.

78. In accordance with Article 366(5) of the CRR, institutions must notify the
competent authorities promptly, and in any case no later than within five
working days, of overshootings that result from their back-testing programme.
The ECB understands that the period of five working days should start on the
first working day after the “subsequent day” as referred to in the second and
third paragraphs of Article 366(3) of the CRR.

79. If either a P&L or the VaR is not available or cannot be computed within five
working days, the ECB considers that there is a risk that an overshooting may
have occurred, and that in order to ensure that the number of overshootings is
not misrepresented, a prudent approach would be to consider such an instance
as an overshooting under hypothetical or actual changes, respectively.

80. If an overshooting has occurred due to malfunctions in the calculation of a P&L

or the VaR and is notified to the ECB, and the institution demonstrates to the
satisfaction of the ECB that the overshooting was caused by an acceptable
reason, the institution may withdraw the overshooting notification. The
explanation of the malfunction should be supported by clear and complete
documentation. If malfunctions leading to erroneous calculations and
overshooting notifications are recurrent, this may indicate that the internal
model is not implemented with integrity as required in Article 368(1) of the CRR,
and the ECB may require the institution to present a remediation plan.

81. The ECB considers that examples of acceptable reasons for withdrawing an
overshooting notification could include:

(a) errors in the calculation of the actual P&L, hypothetical P&L or VaR due to
IT issues or incorrect data;

(b) errors in the scope of positions for the calculations of the P&L or the VaR;

(c) false or missing bookings, or incorrect positions included in the scope of

the calculations;

(d) delayed reserve releases;

34
This understanding is also supported by Article 43(4) of the Final Draft RTS on assessment
methodology for IMA and significant share.

ECB guide to internal models – Market risk 141

 (e) temporary transmission problems between different business locations.

82. However, the ECB considers the following to be a non-exhaustive list of

reasons for withdrawing an overshooting notification which would not be
acceptable, because they are not considered as malfunctions in the calculation
of the P&L or the VaR:

(a) differences in pricing functions between the VaR engine and the actual and
hypothetical P&L calculations (typically using front-office pricing functions);

(b) losses due to the trading or transfer of large positions at a price that
deviates from the market price as a result of trading volumes;

(c) the overshooting corresponds to a small difference between VaR and a

P&L;

(d) unexpected market movements;

(e) a model deficiency that has caused an overshooting in the past has
already been addressed (there is no backward adjustment of
overshootings);

(f) a change in the P&L calculation method, pricing function or

parameterisation of a pricing function or a change in the market data input
used in the P&L calculation.

83. In accordance with Article 368(1)(a) of the CRR, the internal model must be
closely integrated into the daily risk management process. In order for
institutions to be able to meet this requirement, the ECB considers that the VaR
numbers should be available within three business days. In addition, this would
enable institutions to fulfil the requirement to notify back-testing overshootings
within five business days.

If delays in the VaR computation are recurrent, this may indicate that the
internal model is not implemented with integrity as required by Article 368(1) of
the CRR, and the ECB may require the institution to justify such delays or to
present a remediation plan.

3.7 Analysis of overshootings

84. In accordance with Article 368(1)(b) of the CRR, the risk control unit must
produce and analyse daily reports on the output of any internal model, including
overshootings. The ECB considers that such an analysis of overshootings
should include at least the following areas, as they are the most relevant drivers
of the VaR number:

(a) identification of the set of positions responsible for the overshooting

(portfolio analysis);

ECB guide to internal models – Market risk 142

 (b) identification, description and analysis of the market moves contributing to
the overshooting (market analysis);

(c) identification of possible weaknesses in the internal model in the light of

(a) and (b) above (analysis of the internal model).

Paragraphs 85 to 88 explain what the ECB considers are best practices in order
to analyse each of the three areas referred to in (a), (b) and (c) above.

The ECB considers it best practice that for every regulatory back-testing
overshooting a detailed analysis should be performed by the institution and
provided to the competent authority within one month. 35

In accordance with Article 368(1)(f) of the CRR, any internal model for market
risk must have a proven track record of reasonable accuracy in measuring
risks. In order to assess the track record of reasonably accuracy in measuring
risk, the ECB can, on the basis of Article 10 of the SSM Regulation, request a
detailed analysis (in accordance with its specific instructions) of overshootings
and reporting of time series related to back-testing.

3.7.1 Portfolio analysis

85. The analysis of the back-testing overshooting should include a detailed

description of the trading portfolio for which the one-day VaR forecast
calculated was exceeded by the one-day change in the portfolio’s value. If the
overshooting was notified for the actual P&L, the intraday changes in the
portfolio that affected the actual change should also be analysed.

86. The analysis of back-testing overshooting should be performed not only at the
overall portfolio level, but also at lower portfolio levels, to identify the main
positions that caused the overshooting. If specific sub-portfolios can be
identified, they should be mentioned and analysed.

3.7.2 Market analysis

87. The analysis of the market should describe the market moves contributing to
the cause of the overshooting and explain them on the basis of objective market
data (for example, asset prices, indices, interest rates, FX rates, implied
correlations and volatilities). To assess the significance of the market data
movements, the market data, including those that are risk factors in the VaR,
should be analysed in a historical context. The significance of the change in
market data that are risk factors in the VaR, and which contributed to the P&L,
should be tested against the historical 99% confidence interval of risk factor
changes. Changes in the structure of correlations between the risk factors

35
This understanding is also supported by Article 40(10) of the Final Draft RTS on assessment
methodology for IMA and significant share.

ECB guide to internal models – Market risk 143

 should also be analysed. In addition, the analysis should, as far as possible,
include the economic reasons for the market movements.

3.7.3 Analysis of the internal model

88. The suitability of the internal model should be assessed on the basis of the two
previous analyses. Where positions contributing to the back-testing
overshooting can be identified, the appropriateness of the model for these
particular positions should be assessed. To do this, the part of the P&L that can
be explained by the model (i.e. risk factors and pricing functions) should be
distinguished from the part which cannot. In addition, the reliability of the VaR
calculation and of the actual and hypothetical changes in the portfolio should be
evaluated. The analysis of the internal model should focus on:

(a) the appropriateness of risk factors used;

(b) the modelling of risk factors;

(c) the suitability of the processes for calculating VaR, hypothetical P&L and
actual P&L.

4 Aspects of internal validation of market risk models

4.1 Regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 368 (1), (2)

369

SSM Regulation 15/10/2013 10

Other references

Final Draft RTS on assessment methodology 22/11/2016 23 (2)

for IMA and significant share

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IMA and significant share will become an
additional relevant regulatory reference. Currently that document only exists in
a final draft version.

4.2 Frequency of internal validation

89. Institutions are required under Article 369(1) of the CRR to conduct a validation
of their internal models on a periodic basis. The ECB considers that an
appropriate frequency is at least annually, as one year is a reasonable time in
which significant changes in the market or in the composition of the portfolio

ECB guide to internal models – Market risk 144

 could occur. 36 In addition, a validation conducted at least annually would allow
the institution to use the results in the review of its overall risk management
process, as referred to in Article 368(2) of the CRR.

4.3 Internal back-testing of VaR models

4.3.1 Granularity of internal back-testing

90. In accordance with Article 369(1)(b) of the CRR, institutions must, in addition to
the regulatory back-testing programmes, carry out their own internal model
validation tests – including back-testing – in relation to the risks and the
structure of their portfolios. The ECB considers that to satisfy the requirement
regarding internal back-testing in relation to the risks and the structure of the
portfolios, institutions should perform separate back-testing at more granular
levels than the top-of-the-house level 37 on at least the hypothetical P&L
(i.e. counting and analysing of overshootings under the hypothetical P&L).

(a) The ECB understands that, at a minimum, internal back-testing should be

performed:

(i) at one level below the top-of-the-house level;

(ii) for each portfolio that is subject to a separate VaR limit established by
the institution's management body.

(b) Where an institution intends to apply the revisions to the IMA as set out in
the fundamental review of the trading book (FRTB) in the future, the ECB
considers that it would be beneficial if the institution identified the sub-
portfolios within the current scope of the internal model that would most
likely satisfy the requirements for becoming FRTB trading desks, and
performed separate internal back-testing on them.

This is without prejudice to the requirement for the internal model validation
function to perform back-testing on both actual and hypothetical P&L under
Article 369(2) of the CRR, which the ECB understands as relating to the top-of-
the-house level.

36
This understanding is also supported by Article 23(3) of the Final Draft RTS on assessment
methodology for IMA and significant share.

37
See the glossary.

ECB guide to internal models – Market risk 145

 The ECB considers it best practice that the internal back-testing defined above
in this paragraph is performed on a daily basis in order align it with the
regulatory back-testing programme. 38

91. The ECB understands that Article 369(2) of the CRR requires that the back-
testing performed in internal validation complies with the same requirements as
the regulatory back-testing regarding the calculation of actual and hypothetical
P&L. Therefore, the requirements described in Section 3 regarding the
calculation of actual and hypothetical P&L should also be applied to internal
back-testing, in order to ensure consistency. In verifying compliance with this
provision of the CRR, the ECB will take into account the specific circumstances
of the institution.

4.3.2 Tests to be performed in internal back-testing

92. In accordance with Article 369(1)(b) of the CRR, institutions must carry out their
own internal model validation tests, including back-testing. The ECB considers
it best practice that the periodic internal validation tests include the following (or
their equivalent) for at least the top-of-the-house level:

(a) statistical tests on the overshootings, such as the Kupiec (1995) 39 and
Christoffersen (1998) 40 tests, including an analysis of the validity of the
hypotheses underlying those statistical tests;

(b) a test on the uniformity of the distribution of the p-values 41 of the daily
actual P&L and the hypothetical P&L in the daily forecasts of P&Ls of the
VaR 42 engine, at least for the daily data of the last year. For example, a
P&L value equal to the VaR at confidence level of 99% corresponds to a
p-value of 0.01.

38
This understanding is also supported by Article 23(2)(b) of the Final Draft RTS on assessment
methodology for IMA and significant share.
39
See Kupiec, P.H., “Techniques for verifying the accuracy of risk measurement models”, Journal of
Derivatives, Vol. 3, Issue 2, 1995, pp. 73-84.
40
See Christoffersen, P., “Evaluating interval forecasts”, International Economic Review, Vol. 39, Issue 4,
1998, pp. 841-862.
41
The probability integral transformation states that for a continuous random distribution 𝑋𝑋, applying the
cumulative distribution function (CDF) of 𝑋𝑋, 𝐹𝐹𝑋𝑋 , on 𝑋𝑋 yields a uniform distribution. By negation, if the
resulting distribution is not uniform, 𝐹𝐹𝑋𝑋 is not the CDF of 𝑋𝑋.
Given a vector of simulated P&L (used to estimate the VaR) sorted in ascending order, 𝑥𝑥1 , … , 𝑥𝑥𝑛𝑛 , the
p-value, 𝑝𝑝(𝑥𝑥), corresponding to a given P&L, 𝑥𝑥, should be obtained in the following way:
𝑥𝑥𝑘𝑘+1 −𝑥𝑥 𝑥𝑥−𝑥𝑥𝑘𝑘
⋅ 𝑝𝑝𝑘𝑘 + ⋅ 𝑝𝑝𝑘𝑘+1 if 𝑥𝑥𝑘𝑘 ≤ 𝑥𝑥 ≤ 𝑥𝑥𝑘𝑘+1 for 𝑘𝑘 = 1, … , 𝑛𝑛 − 1
⎧𝑥𝑥𝑘𝑘+1 −𝑥𝑥𝑘𝑘 𝑥𝑥𝑘𝑘+1 −𝑥𝑥𝑘𝑘
⎪ 𝑥𝑥
⎪ 𝑝𝑝
� 1 �𝑥𝑥1
⎪ 1−𝑝𝑝1
𝑥𝑥 if 𝑥𝑥 < 𝑥𝑥1 𝑘𝑘
𝑝𝑝(𝑥𝑥) = 𝑝𝑝1 𝑥𝑥1
1+�1−𝑝𝑝 � where 𝑝𝑝𝑘𝑘 =
⎨ 1 𝑛𝑛+1
𝑥𝑥
⎪ 𝑝𝑝
� 𝑛𝑛 �𝑥𝑥𝑛𝑛
⎪ 1−𝑝𝑝𝑛𝑛
if 𝑥𝑥 > 𝑥𝑥𝑛𝑛
⎪ 𝑝𝑝
𝑥𝑥

⎩ 1+� 𝑛𝑛 �𝑥𝑥𝑛𝑛
1−𝑝𝑝𝑛𝑛
42
If an institution has a model based on a mixture of approaches in several VaR model components, the
most material approach should be used.

ECB guide to internal models – Market risk 146

 93. In order to assess whether the periodic internal validation tests used by an
institution are adequate and fit for their purpose, the ECB can, on the basis of
Article 10 of the SSM Regulation, require the institution to provide the following
information:

(a) for the top-of-the-house level, the complete economic P&L, hypothetical
P&L, actual P&L, and VaR time series of at least one year, but preferably
three years;

(b) for the top-of-the-house level, the number of overshootings and the
corresponding dates when they occurred over at least the last year, but
preferably over the last three years;

(c) for the top-of-the-house level, for the overshootings (i.e. −P&L𝑡𝑡+1 > VaR 𝑡𝑡
where VaR 𝑡𝑡 > 0 by convention), the time series of at least one year, but
preferably three years, of the loss overshooting ratio (LOR) defined as:
−P&L𝑡𝑡+1 −VaR 𝑡𝑡
LOR = , where −P&L𝑡𝑡+1 > VaR 𝑡𝑡 ;
VaR 𝑡𝑡

(d) for the top-of-the-house level, the time series of p-values of the daily actual
P&L and the hypothetical P&L in the daily forecasts of P&Ls of the VaR
engine of at least one year, but preferably three years;

(e) for the more granular levels referred to in paragraph 90:

(i) complete economic P&L, hypothetical P&L and VaR time series of at
least one year, but preferably three years;

(ii) an analysis of all overshootings, including an explanation of the

cause of the overshooting over the hypothetical P&L and an
assessment of the model adequacy on the relevant level.

4.4 Validation on hypothetical portfolios

94. In accordance with Article 369(1) of the CRR, institutions must have processes
in place to ensure that all their internal models for market risk have been
adequately validated. Therefore, the requirement of Article 369(1)(c) to use
hypothetical portfolios in the internal model validation refers in particular to VaR,
sVaR, and IRC models.

95. In accordance with Article 368(1)(e) of the CRR, institutions must have a
documented set of internal policies and controls in place concerning the overall
operation of their internal models, including the internal validation. Therefore,
institutions should have a policy in place that governs the overall processes
related to the validation of their internal models for market risk using
hypothetical portfolios. The ECB considers that in order to cover the overall
process, such a policy should comprise the following aspects:

(a) portfolio definition – the processes for defining hypothetical portfolios;

ECB guide to internal models – Market risk 147

 (b) analysis – the processes for analysing the performance of the model
based on the results of the tests performed on hypothetical portfolios,
including:

(i) an assessment of the ability of the models to capture the risk of the
hypothetical portfolios;

(ii) verification that the insights gained by the analysis of hypothetical

portfolios are reflected in the models;

(iii) in particular, for back-testing on hypothetical portfolios, an

identification of the market movements and parameters causing
overshootings;

(c) reporting – the processes to ensure that the results of validation on

hypothetical portfolios are reported to a management body with sufficient
authority in respect of internal models.

96. In accordance with Article 369(1)(c) of the CRR, institutions must conduct
validation exercises using hypothetical portfolios in order to ensure that a model
is able to account for particular structural features. 43 The ECB understands that
these hypothetical portfolios should have targeted compositions so that the
model can be tested at a level of granularity that enables the identification and
isolation of specific model performance for those structural features (for
example, related to specific business lines, instrument features, and/or trading
strategies).

97. As they should ensure that the risk model is validated for the institution’s risk
management purposes, such hypothetical portfolios should be designed in line
with the business model of the institution. For example, it is not necessary to
include products that are not covered by trader mandates, nor to test specific
features that are not relevant for potential positions according to the institution’s
approved trading strategy. Consequently, an institution should review the
hypothetical portfolios in the event of a change in its business model or trading
strategy.

98. For the same reason, the number of hypothetical portfolios should be
commensurate with the nature, scale and complexity of the activities of the
institution.

43
These hypothetical portfolios should not be limited to portfolios defined in the benchmarking exercises
for market risk conducted by the EBA or the Basel Committee on Banking Supervision, as those
portfolios cannot account for all relevant particular structural features. Participation in such
benchmarking exercises is thus not sufficient to meet the requirements of this section of the guide.

ECB guide to internal models – Market risk 148

4.5 Validation based on hypothetical portfolios for VaR models by
internal back-testing

99. The ECB considers that the validation requirements of Article 369(1)(c) of the
CRR for VaR models can be fulfilled by internal back-testing where an
institution can demonstrate that it has set up internal back-testing for the VaR
model using sub-portfolios at a level which is sufficiently granular to account for
the particular structural features that may arise in its portfolios.

100. Where an institution performs internal back-testing on hypothetical portfolios for

the VaR model, the P&L calculations for this back-testing of hypothetical
portfolios should not differ from the P&L calculations for regulatory VaR back-
testing as described in Section 3, in order to ensure consistency. This back-
testing can be conducted based on the hypothetical P&L only, as hypothetical
portfolios are not part of the daily trading activity and therefore the actual P&L is
not relevant.

101. As the purpose of such internal back-testing is the internal validation of the VaR
model, the ECB considers that in order to ensure consistency:

(a) the comparison should be carried out using the daily hypothetical P&L and
the one-day VaR;

(b) back-testing periods for hypothetical portfolios should cover at least the
period used to calibrate the VaR as of the validation date, to ensure that
the results are relevant for the model at that date;

(c) institutions should ensure that the particular structural feature, as referred
to in paragraph 96, for which each hypothetical portfolio was selected,
continues to be in place over time and during the entire historical period for
which the back-testing is performed.

ECB guide to internal models – Market risk 149

5 Methodology for VaR and stressed VaR

5.1 Regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 365

366 (2)

367, 368, 369, 370

SSM Regulation 15/10/2013 10

Other references

EBA Guidelines on sVaR 44 16/05/2012 6,10

Final Draft RTS on assessment methodology 22/11/2016 38 (1), (2)

for IMA and significant share
49 (2)

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IMA and significant share will become an
additional relevant regulatory reference. Currently that document only exists in
a final draft version.

5.2 General requirements

102. In accordance with Article 369(1) of the CRR, the appropriateness of any
assumptions made within the internal model must be demonstrated. Therefore,
institutions should demonstrate the appropriateness of any assumptions about
the distribution of risk factors included in the VaR and sVaR models on the
basis of objective data. 45 In order to assess the appropriateness of the
distribution assumptions, the ECB can, on the basis of Article 10 of the SSM
Regulation, require institutions to consider different plausible distribution
assumptions and compare the VaR and sVaR amounts calculated according to
those different assumptions to demonstrate that the selected assumption is
appropriate.

103. In accordance with Article 368(1)(f) of the CRR, an internal model must be
reasonably accurate in measuring risks. Therefore, when using Monte Carlo
simulations, institutions should be able to demonstrate that the number of
simulations used to compute the VaR and sVaR is sufficient to produce
accurate and stable VaR and sVaR numbers.

44
EBA Guidelines on Stressed Value At Risk (Stressed VaR) (EBA/GL/2012/2), referred to in this guide
as the “EBA Guidelines on sVaR”.
45
Where an institution applies historical returns in its model, the requirement refers to the choice of the
specific methodology to determine the returns (for example, the use of relative or absolute returns).

ECB guide to internal models – Market risk 150

104. An institution may apply different methodologies (i.e. the absolute, relative or
mixed approach 46) to calculate returns used to calibrate the VaR and sVaR
models for different risk factors. The ECB has observed that the best practices
used in the VaR and sVaR models are the following methodologies:

Risk factor category Methodology used to calculate returns

Interest rate curves Absolute or mixed approach

Bond spread Absolute or mixed approach

Credit default swap (CDS) spread Absolute or mixed approach

Foreign exchange rate Relative approach

Equities spot Relative approach

Commodities Relative approach

As institutions are required under Article 368(3) of the CRR to apply best
practices, they should be able to explain deviations from these methodologies
and quantify the impact of those deviations.

Article 365(2) of the CRR requires that the calculation of the sVaR is made in
accordance with the requirements for calculation of the VaR. Therefore, for a
given risk factor, where a specific methodology is used in the VaR, the same
methodology is expected to be used for the same risk factor in the sVaR.

As regime changes could occur between the VaR effective historical

observation period and the sVaR historical period, the method should be
suitable for both periods. In accordance with Article 369(1) of the CRR, the
appropriateness of any assumptions made within the internal model must be
demonstrated. The ECB considers that this choice of method is one of the
assumptions whose appropriateness should be demonstrated.

105. In accordance with Article 368(1) of the CRR, the internal model must be
conceptually sound and implemented with integrity. Therefore, it is expected
that the returns are calculated on the basis of one single holding period (for
example, one day or ten days) for all risk factors. 47

106. Under paragraph 10.3(c) of the EBA Guidelines on sVaR, institutions should be
able to prove that on the day of the week chosen for the sVaR calculation their
portfolio is representative of the portfolio held during that week, and that the
chosen portfolio does not lead to a systematic underestimation of the sVaR
numbers when computed weekly; shown, for example, by using sensitivities or

46
Either of the two examples following could be considered as a “mixed approach”: (i) the case where
some risk factors within a given risk factor category are calculated via absolute returns while others
within the same risk factor category are calculated via relative returns (e.g. interest rate curves with low
interest rate levels calculated via absolute returns and interest rate curves of other currencies with
higher levels via relative returns); or (ii) the case where a single methodology takes into account
different regimes (e.g. return close to absolute for low levels of interest rates and close to relative for
higher levels).
47
Uniform use of a one-day holding period in VaR and a 10-day holding period in sVaR might be
permissible if adequately justified by an institution.

ECB guide to internal models – Market risk 151

 by proving that the VaR is not systematically lower on the day of the week
chosen for sVaR.

107. In order to assess that the day of the week when the sVaR amounts are
calculated does not lead to material bias, the ECB can, on the basis of
Article 10 of the SSM Regulation, require an institution to recalculate the sVaR
for 15 consecutive business days (including three reporting days). If it is not
possible to perform this calculation in the production environment, it can be
performed in a test environment that replicates the calculation of the regulatory
sVaR.

108. In accordance with Article 368(1)(e) of the CRR, institutions must ensure
compliance with a documented set of internal controls. In order for the ECB to
assess compliance with this requirement, an institution should be able to
provide an inventory of all open validation findings in relation to the VaR and
sVaR models including, for each finding, a description thereof, the envisaged
remedial action and the target date for closure of the finding. In addition,
institutions should retain closed validation findings for at least one year after the
closure date and should be able to provide a description of the remedial action
implemented.

5.3 Data inputs, length of the time series used to calibrate VaR and
sVaR, and quantile estimation

109. In accordance with Article 365(1)(d) of the CRR, institutions must use an
effective historical observation period of at least one year for the calculation of
the VaR, except where a shorter observation period is justified by a significant
upsurge in price volatility. The ECB considers that this requirement can be
fulfilled by taking returns referring to 250 consecutive business days 48 in order
to allow alignment with the time frame referred to in Article 366(2) of the CRR.
The ECB understands that an effective historical observation period of at least
one year means that the average time lag 49 of the scheme used by an

48
The observation period corresponds to the time frame between the first day of calculation of returns
and the last day of calculation of returns. Where an institution uses 10-day returns, the minimum
observation period is 250 business days but the time frame between the first day of calculation of the
first 10-day return and the end date of the last 10-day return is 260 days.
49
A weighting scheme is considered to be the set of weights directly or implicitly applied to observations
of a risk factor, {𝑤𝑤𝑡𝑡 }𝑡𝑡=1,2,…, where typically for giving more weight to more recent observations 𝑤𝑤𝑡𝑡 ≥
𝑤𝑤𝑡𝑡+1 ∀𝑡𝑡, and 𝑡𝑡 is the lag in number of business days between the VaR or sVaR computation date and
the historical observation date. The average time lag using the most recent 𝑛𝑛 observations is defined
as:
∑𝑛𝑛
𝑡𝑡=1 𝑡𝑡⋅𝑤𝑤𝑡𝑡
average time lag(𝑛𝑛, 𝑤𝑤) = ∑𝑛𝑛
𝑡𝑡=1 𝑤𝑤𝑡𝑡

This definition can be extended to equally weighted schemes by setting 𝑤𝑤𝑖𝑖 = 1 for all dates. For an
equally weighted scheme on 250 continuous business days, the formula results in 125.5 days.
The ECB may, after analysing the particular circumstances of an institution, also consider other
methods of processing market data or risk factors to be a weighting scheme, and assess whether such
methods provide an effective historical observation period of at least one year in accordance with
Article 365(1)(d) of the CRR.

ECB guide to internal models – Market risk 152

 institution is at least the average time lag for an equally weighted observation
period of one year (i.e. 125.5 days for 250 business days). 50

110. Where the institution uses a shorter effective historical observation period (for
example, by applying a weighting scheme) due to a significant upsurge in price
volatility, the ECB considers that this should not lead to a lower VaR risk
number. Therefore, the institution should use the higher of the following:

(a) the VaR amount calculated with an equally weighted historical observation
period of at least one year;

(b) the VaR amount calculated with the shorter effective historical observation
period. 51

In accordance with Article 365(1)(e) of the CRR, this comparison should be

performed at least monthly, and the institution should continue to apply the
resulting calibration method until the next comparison.

111. For the purpose of the regulatory back-testing conducted under Article 366 of
the CRR, the higher of the two metrics mentioned in paragraph 110 should be
used for consistency with the own funds requirement calculation.

If the institution always uses a VaR calculated with the shorter effective
observation period as mentioned in paragraph 110 for its risk management, the
institution is allowed to perform internal back-testing of the VaR under
Article 369(1)(b) of the CRR using this shorter effective observation period, in
order to ensure that the model is closely integrated into the daily risk
management process.

112. In accordance with Article 365(2) of the CRR, the institution must calculate the
sVaR calibrated to historical data from a continuous 12-month period of
significant financial stress relevant to the institution's portfolio. The ECB
considers that this requirement can be fulfilled by taking returns referring to 250
consecutive business days in order to allow alignment with the time frame
referred to in Article 366(2) of the CRR.

Under paragraph 6.8 of the EBA Guidelines on sVaR, no weighting of historical

data should be applied when determining the relevant historical period or when
calibrating the sVaR model. 52 The ECB considers that a calibration to historical
data from a continuous 12-month period implies that no weighting scheme
should be applied to the historical data used to calibrate the sVaR. 53

50
This understanding is also supported by Article 38(1) of the Final Draft RTS on assessment
methodology for IMA and significant share.
51
This understanding is also supported by Articles 38(1) and 38(2) of the Final Draft RTS on assessment
methodology for IMA and significant share.
52
Under paragraph 10.10 of the EBA Guidelines on sVaR, this does not contradict the requirement that
the sVaR methodology should be based on the current VaR methodology.
53
This understanding is also supported by Article 49(2)(h) of the Final Draft RTS on assessment
methodology for IMA and significant share.

ECB guide to internal models – Market risk 153

113. In order to ensure that the approach for measuring empirical correlations is
sound and implemented with integrity as required by Article 367(3) of the CRR,
the institution should use one single observation period (i.e. with the same start
date and end date, and consequently the same length of observation period) for
all risk factors modelled in the VaR. This also applies to the sVaR.

114. Consequently, if a new instrument (e.g. a single stock or credit index series) is
issued, the time series corresponding to this instrument should not be used on
its own for the calibration of the VaR and sVaR models until the length of the
available time series reaches the length of the observation period used by the
institution. In this case, because, at the least, the missing portion needs to be
completed, a risk factor calibrated to this time series is considered to be proxied
and the requirements for proxies should be observed (see Section 5.5).

115. In accordance with Article 365 of the CRR, the VaR and sVaR are calculated as
the 99th percentile, one-tailed confidence interval, and In accordance with
Article 367(1)(a) of the CRR, the model must accurately capture all material
price risks. In accordance with Article 368(1)(f) of the CRR, any internal model
must have a proven track record of reasonable accuracy in measuring risks.
Therefore, the ECB considers that for institutions using a simulation approach
(either historical or Monte Carlo) in their VaR (or sVaR) model, the percentile
estimation method used to obtain the 99th percentile should be based on
reasonable statistical properties that ensure its accuracy – that is, it should be
statistically unbiased, distribution-free, and assume that the probability of
experiencing a P&L lower (or higher) than the lowest (or highest) simulated
value is strictly greater than zero. For these reasons the ECB considers that
both the method proposed by Harrell and Davis 54 55 and the simplified method
proposed below 56 are appropriate methods to ensure that price risks are
accurately captured when using the percentile estimation method.

Consider the vector of simulated P&L of length 𝑛𝑛 for the VaR (or sVaR)
percentile estimation (P&L1 to P&L𝑛𝑛 ) in ascending order. 57 The result 𝑄𝑄(0.99) is
obtained as the weighted average of the two subsequent P&L values −P&LInt(𝑚𝑚)
and −P&LInt(𝑚𝑚)+1 , computed as

𝑄𝑄(0.99) = �𝑚𝑚 − Int(𝑚𝑚)� ⋅ (−P&LInt(𝑚𝑚)+1 ) + (Int(𝑚𝑚) − 𝑚𝑚 + 1) ⋅ (−P&LInt(𝑚𝑚) ) ,

𝑛𝑛+1
with 𝑚𝑚 = .
100

For example:
for 𝑛𝑛 = 250, the percentile result of this method is
0.51 × (−P&L3 ) + 0.49 × (−P&L2 );

54
See Harrell, F.E. and Davis, C.E., “A new distribution-free quantile estimator”, Biometrika, Vol. 69,
1982, pp. 635-640.
55
Corresponding to estimator 9 in Dielman, T., Lowry C. and Pfaffenberger, R., “A comparison of quantile
estimators”, Communications in Statistics. Simulation and Computation, Vol. 23(2), 1994, pp. 355-371.
56
Corresponding to definition 6 in Hyndman, R.J. and Fan, Y., “Sample quantiles in statistical packages”,
American Statistician, Vol. 50, 1996, pp. 361-365.
57
P&L1 is the lowest P&L (i.e. the highest loss).

ECB guide to internal models – Market risk 154

 for 𝑛𝑛 = 260, the percentile result of this method is
0.61 × (−P&L3 ) + 0.39 × (−P&L2 ).

5.4 Data quality

116. In accordance with Article 367(2)(e) of the CRR, an internal model must meet
minimum data standards. This applies in particular to risk factor time series,
which are fundamental inputs to a VaR model. For each risk factor time series
used to calibrate the shocks of the VaR model, the institution should have a
process in place to regularly check the quality of the time series. The ECB
considers that an appropriate minimum frequency of such checks is quarterly,
as this allows alignment with the regulatory reporting cycle. This is without
prejudice to the discretion of institutions to perform certain checks on a more
frequent basis if needed to meet minimum data standards. Where an institution
uses different data sources for its VaR model and the daily economic P&L
calculation reflected in the inventory referred to in paragraph 129, the ECB may
request that the institution explain the differences between the two sets of data
sources, in order to verify that they meet minimum data standards.

117. The ECB understands that the minimum data standards should ensure that the
true volatility of a position or portfolio is captured. Therefore, the quality checks
on the risk factor time series should, at the minimum, identify for each time
series:

(a) the number of days for which data points were initially missing and then
filled using a particular methodology (e.g. interpolation and
extrapolation); 58

(b) the number of days for which data points were initially available and were
replaced using a particular methodology (e.g. interpolation and
extrapolation);

(c) the number of days with no daily changes;

(d) the maximum number of consecutive days with no daily change.

118. Material or large numbers of changes in the time series may affect the ability to
capture the true volatility of a position or portfolio. In order to monitor and
ensure that the operation of the internal model is not adversely affected,
institutions should maintain up-to-date documentation 59 describing any changes
in the risk factor time series, including in particular any methodology for the
replacement of missing data, and the list of tasks that may be performed during
manual adjustments. This documentation should contain the following:

58
There should be no missing data points for the final time series of shocks used to calibrate the model.
59
In accordance with Article 368(1)(e) of the CRR.

ECB guide to internal models – Market risk 155

 (a) a description of the methodology followed to introduce the adjustment –
the description should be sufficiently detailed so that any staff member of
the unit in charge is able to produce the same outcome;

(b) a description of the processes in place to ensure the appropriate

implementation of a manual process in accordance with the
documentation.

119. In accordance with Article 367(2)(e) of the CRR, the institution's internal model
must conservatively assess the risk arising from less liquid positions and
positions with limited price transparency and must meet minimum data
standards. Therefore, the ECB considers that institutions should have in place
documentation which defines the minimum data quality standards that risk
factor time series should meet, and be able to provide justification for the use of
time series that include an elevated number of consecutive business days with
missing data or no daily changes. Moreover, the ECB considers that when
using time series with only a low number of available data points per year
institutions should provide justification that the number of data points is
sufficient to reflect the true volatility of a position or portfolio.

120. In order to ensure that changes in the risk factor time series do not affect the
ability to capture the true volatility of a position or portfolio, institutions should
analyse how the replacement of missing data affects the volatilities and
correlations of the IMA. This applies particularly where time series are used that
have:

(a) the value of the same data of the previous day in the case of automatic
and systematic replacement of missing data points;

(b) an elevated number of consecutive business days with missing data or no

daily changes;

(c) only a low number of data points per year before any data cleaning or
treatment.

Filtering of data or exclusions of outliers should not be performed unless the

institution can demonstrate that the excluded data points correspond to
erroneous or stale data and do not represent the real market volatility of the risk
factors. As part of the requirement under Article 368(1)(e) of the CRR to have
procedures for ensuring compliance with controls concerning the overall
operation of internal models, the ECB considers that institutions should keep
track of any exclusion made in the risk factor time series used to calculate VaR
and sVaR.

121. Conversely, automatic and systematic filtering of data leading to exclusions of

high or low returns should not be performed without further analysis and
documentation.

ECB guide to internal models – Market risk 156

5.5 Proxies, beta approximation and regressions

122. For the purposes of this guide, the ECB understands that market data are
proxied in the calculation of the VaR or sVaR 60 when market data that are used
as inputs in the pricing model to compute the economic P&L for an IMA position
are replaced with other market data 61 (or a weighted average of market data)
for the purpose of calibrating the VaR or sVaR (respectively) for that position.

Where for the economic P&L a certain market data input (for example, the
directly observable price of an instrument) is used, while the VaR or sVaR
model (respectively) uses other market data that would lead to an equivalent
price, the ECB considers that these data should not be considered as proxies. 62
Conversely, if the data would not lead to an equivalent price, they should be
considered as proxies.

In accordance with Article 367(1)(b) of the CRR, where a risk factor is

incorporated into the institution's pricing model but not into the risk
measurement model, the institution must be able to justify such an omission to
the satisfaction of the competent authority. Therefore, the ECB considers that in
the event of reduced granularity of market data inputs for curves or surfaces in
the VaR or sVaR model, as compared with the economic P&L, an institution
should duly justify why the data points interpolated owing to the reduced
granularity should not be considered as proxies (e.g. by means of equivalent
price as mentioned above).

123. In accordance with Article 370(e) of the CRR, an internal model used for
specific risk must capture name-related basis risk, and in particular be sensitive
to material idiosyncratic differences between similar, but not identical, positions.
The ECB therefore considers that the use of each single stock price (where
available) as a risk factor in the VaR or sVaR is the best practice for modelling
specific risk for equity instruments. Similarly, the direct use of idiosyncratic
market data (where available) (for example, the idiosyncratic bond spread or
each single-name credit default swap) as a risk factor is considered to be the
best practice for modelling specific risk for debt instruments.

However, the use of beta approximations or regressions could be accepted if

they are documented and regularly validated (i.e. they are shown to lead to
good model performance) as required by Articles 368(1)(e) and 369 of the CRR.

60
For the purpose of partial use models this proxy definition should be applied only to market data inputs
that fall into the approved risk categories of the model. Market data inputs giving rise to risks that are
out of the scope of the approved risk categories should not automatically be counted as proxied. For
example, an equity price that is modelled by a regression to an index for a general equity risk model
(no approval for specific equity risk) should not be counted as proxied.
61
If market data used for pricing and VaR calculation only differ in the source (for example, P&L pricing
uses one source and VaR calculations are based on another source), the data used for risk calculation
should not be considered as a proxy. In any case, the quality of the data should be checked and the
sources justified by the institution.
62
Examples that could lead to equivalent prices are: (i) where the economic P&L is computed by market
instrument (yield) rates, while the VaR/sVaR is computed based on zero coupon rates; and (ii) where a
price-based economic P&L is used (for example, listed options, or the direct bond price), while the VaR
uses a model-based P&L.

ECB guide to internal models – Market risk 157

 Institutions with internal model approval for specific risk should be able to
demonstrate that the idiosyncratic volatility of equity or debt instruments with
specific risk is correctly taken into account in the VaR and sVaR models.

124. The ECB considers that in order to demonstrate that the model captures
accurately all material price risks as required by Article 367(1)(a) of the CRR,
the institution should document and make available upon request an inventory
of time series of risk factors that are proxied for the calibration of the VaR and
sVaR models, together with the materiality of the corresponding risk factors.

125. The ECB considers that the requirement to have a documented set of internal
policies and controls also applies to the use of proxies, as they are part of the
overall operation of internal models. Therefore, institutions should have a policy
in place that defines clear processes for deriving and validating each proxy for
VaR and sVaR. The policy should further define a set of controls (for example,
statistical analysis or comparison against alternative proxies) that should be
performed to ensure the appropriateness of proxies.

126. As a control to ensure that the proxies are appropriately conservative and are
reflective of the true volatility where sufficient market data are available,
institutions should perform analyses to show that the proxy market data (i) are
highly correlated with the market data used for economic P&L, and (ii) show a
similar level of volatility for VaR and sVaR. Where analyses based on market
data used for economic P&L are not feasible, institutions should at least assess
alternative plausible proxy choices.

In particular, institutions which have approved internal models for specific risk of
debt instruments or specific risk of equity instruments should demonstrate that
the use of a proxy enables the idiosyncratic risk to be appropriately captured as
required by Article 370(e) of the CRR.

127. As proxies are part of the internal models, any proxy should also be validated
for VaR and sVaR at least annually in accordance with Section 4.2 on the
frequency of regular internal validation.

128. In order to assess that proxies are appropriately conservative and reflective of
the true volatility, the ECB can, on the basis of Article 10 of the SSM Regulation,
require an institution to provide, for a selection of sub-portfolios, business days,
and material proxies:

(a) the hypothetical P&L used for regulatory back-testing;

(b) the hypothetical P&L calculated on the same unchanged positions but
replacing, for the positions for which proxies are used in the VaR, the
market data with the market data of their proxies;

(c) the hypothetical P&L calculated on the same unchanged positions but
replacing, for the positions for which proxies are used in the sVaR, the
market data with the market data of their proxies.

ECB guide to internal models – Market risk 158

 The specific information required will depend on the results of the institution’s
analysis of the appropriateness of the proxies.

5.6 Risk factors in the model

129. In accordance with Article 367(1)(b) of the CRR, the VaR and sVaR models
must capture a sufficient number of risk factors, depending on the level of
activity of the institution in the respective markets. Where a risk factor is
incorporated into the institution’s pricing model (referred to as “market data
input” for the purposes of this guide) but not into the risk measurement model,
the institution must be able to justify such an omission to the satisfaction of the
competent authority.

So that it can assess compliance with this requirement, the ECB considers that
an institution should be able to provide an inventory of all the market data inputs
to the economic P&L and of all the risk factors used in the VaR and sVaR
models. This inventory should include a comprehensive mapping between the
market data used to calculate the economic P&L and the corresponding risk
factors included in the VaR and sVaR model. The inventory should contain at
least the following information:

(a) a list of the market data inputs used in the calculation of the institution’s
economic P&L; 63

(b) for each market data input, information confirming:

(i) whether the market data input is directly modelled in the VaR engine
(i.e. whether it is a risk factor of the model and involves no use of a
proxy);

(ii) whether the market data input is proxied in the VaR calculation;

(iii) whether the market data input is modelled (or not) in the VaR engine;

(iv) where relevant, how the market data input is proxied in the VaR
calculation (for example, by one market data input or by a
combination of several market data inputs in a regression approach).

A similar inventory should be provided for the sVaR model where relevant.

130. Identical underlyings should always be mapped to the same risk factor in order
to ensure consistency within the model.

131. In order to assess whether VaR and sVaR models capture a sufficient number
of risk factors and to assess the materiality of missing risk factors, the ECB can,
on the basis of Article 10 of the SSM Regulation, require institutions to provide,
for a selection of sub-portfolios, business days, and missing risk factors:

63
For example, inputs in institutions’ pricing models for economic P&L.

ECB guide to internal models – Market risk 159

 (a) the hypothetical P&L used for regulatory back-testing as defined in
Section 3.5;

(b) the P&L calculated assuming simultaneously:

(i) unchanged positions and omitting the changes in value of the missing
risk factors of the VaR and sVaR;

(ii) use of the pricing method and model parametrisation used to

compute the economic P&L. 64

5.7 Pricing functions and methods in the model

132. In accordance with Article 367(1)(a) of the CRR, models must capture
accurately all material price risks. In addition, in accordance with
Article 368(1)(e) and (f) of the CRR, institutions must have a set of documented
procedures and controls concerning the overall operations of their internal
models, and those models must have a proven track record of reasonable
accuracy in measuring risks. Therefore, institutions should be able to produce
and update, on a regular basis, an inventory of all the VaR and sVaR pricing
functions and methods, and the pricing functions and methods used in the
economic P&L. This inventory should include a comprehensive mapping
between the pricing functions and methods used in VaR and sVaR and the
pricing functions and methods used for the daily economic P&L. It should
include the following information at the relevant level of granularity:

(a) the pricing functions and methods, and pricing functions and methods
parametrisation (for example, the number of Monte Carlo simulations)
used to calculate the daily economic P&L;

(b) the scope of instrument types covered by each pricing function and
method used to calculate the daily economic P&L;

(c) a meaningful indication of the materiality of positions priced with the

corresponding pricing function and method, for example the number of
individual positions, as well as the total amount of outstanding notional and
market value covered by each pricing function and method used to
calculate the daily economic P&L;

(d) corresponding pricing functions and methods as well as the pricing

functions and methods parametrisation (for example, the number of Monte
Carlo simulations) used in the VaR engine;

(e) a self-assessment by the institution, including a scorecard indicator (green,

amber, red) 65 of the appropriateness of VaR pricing methods (VaR engine
pricing versus daily economic P&L pricing).
64
This should be identical to the pricing function used to calculate the hypothetical P&L under
paragraph 74.

ECB guide to internal models – Market risk 160

 The criteria for assessing this scorecard indicator should be described in an
internal policy.

A similar inventory should be available for the sVaR model, where relevant.

The ECB considers that an appropriate frequency for updating this inventory is
at least annually, so that it can be used in the annual review of the institution’s
overall risk management process as referred to in Article 368(2) of the CRR.

133. This inventory should be reviewed at least annually by a unit independent of the
one that produces it (for example, the internal audit function or internal
validation function). This review should check the quality, reliability and
comprehensiveness of the information provided in the inventory.

134. As for any other assumption in an internal model, the differences in the pricing
functions and methods used for the calculation of the VaR and sVaR, compared
with those used for the calculation of the economic P&L, should be subject to
validation 66 in accordance with Article 369(1) of the CRR. This validation should
include any simplifications of pricing functions and methods introduced for use
for VaR or sVaR-related purposes (for example, a reduced number of
parameters or simulations). The validation should be performed at least initially
when a pricing method is introduced into the VaR or sVaR calculation that is not
identical to the one for economic P&L purposes, and should assess the impact
of the use of different pricing methods. Additionally, a regular validation should
be performed in order to check that this impact remains low. The scorecard
indicator mentioned above should be based on the results from this (initial and
regular) validation. The institution should develop a work plan to mitigate the
risk or improve the quality of any pricing functions or methods that are deemed
inadequate according to the institution’s assessment in the scorecard (i.e. a red
indicator).

135. In order to assess the accuracy and appropriateness of the pricing functions
and methods in the VaR and sVaR models, the ECB can, on the basis of
Article 10 of the SSM Regulation, require institutions to provide, for a selection
of sub-portfolios, business days, and pricing functions/methods:

(a) the hypothetical P&L used for regulatory back-testing as defined in

Section 3.5;

(b) the P&L, calculated on the same unchanged positions, by using the pricing
functions and methods used to compute the VaR and sVaR numbers with
the market data input used for the hypothetical P&L.

This information allows assessment of the isolated impact on the hypothetical

P&L of using the pricing functions and methods in VaR and sVaR calculations,
instead of those in the economic P&L.

65
The scorecard indicators are: green – fully appropriate; amber – acceptable; red – weakness detected.
66
The validation of pricing functions used for economic P&L purposes is expected to be regularly
performed by an institution and thus is the basis for this additional requirement.

ECB guide to internal models – Market risk 161

6 Methodology for IRC models focusing on default risk

6.1 Regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/07/2013 4 (1)

153, 336, 368

369 (1)

370

372 (a), (d)

373

374 (2), (4)

375 (1)

376 (2), (3), (6)

SSM Regulation 15/10/2013 10

Other references

EBA Guidelines on the IRC 16/05/2012 17, 25, 29

Final Draft RTS on assessment methodology 22/11/2016 63 (4)(b)

for IMA and significant share
65 (3)

70 (3)

Once adopted by the European Commission, the Final Draft RTS on

assessment methodology for IMA and significant share will become an
additional relevant regulatory reference. Currently that document only exists in
a final draft version.

6.2 General requirements

136. Under paragraph 29.2 of the EBA Guidelines on the IRC, institutions should be
able to prove that, on the day of the week chosen for the IRC calculation, their
portfolio is representative of the portfolio held during the week and that the
chosen portfolio does not lead to a systematic underestimation of the IRC
numbers when computed weekly (for example, by using sensitivities or jump-to-
default).

137. In order to assess that the day of the week when the IRC numbers are
calculated does not lead to material bias, the ECB can, on the basis of
Article 10 of the SSM Regulation, require an institution to recalculate the IRC for
15 consecutive business days (including three reporting days). If it is not
possible to perform this calculation in the production environment, it can be
performed in a test environment replicating the calculation of the regulatory
IRC.

138. In accordance with Article 374(4) of the CRR, the IRC model must be based on
the assumptions of a constant level of risk over the one-year time horizon or,

ECB guide to internal models – Market risk 162

 alternatively, on the assumption of a one-year constant position. As with any
other modelling assumption, an institution should be able to demonstrate that
the chosen assumption appropriately captures the risk of its portfolio.

In order to assess the appropriateness of that choice, the ECB can, on the
basis of Article 10 of the SSM Regulation, require an institution that uses a
constant level of risk assumption and liquidity horizons shorter than one year to
calculate the impact of using a constant position assumption on the IRC and the
default risk in the IRC 67 amounts.

139. In accordance with Article 375(1) of the CRR, hedging or diversification effects
associated with long and short positions may only be recognised by explicitly
modelling gross long and short positions in the different instruments, and
institutions must reflect the potential for significant basis risks in hedging
strategies, in particular by maturity. Therefore, the ECB considers that
irrespective of whether a one-year constant position assumption or a constant
level of risk assumption is used, institutions should not overestimate
diversification or hedging effects, and in particular should ensure that maturity
mismatches between long and short positions occurring within the liquidity
horizon or within the one-year risk modelling horizon do not lead to an
underestimation of risk. 68

In accordance with Article 376(3)(c) of the CRR, as part of the annual

independent review and the initial and periodic validation of its IRC model, an
institution must apply appropriate quantitative validation. Therefore, the ECB
considers that institutions should, as part of the annual independent review and
the initial and periodic validation of their IRC models, assess quantitatively how
maturity mismatches – that may lead to imbalanced positions within the
modelling horizon – impact the IRC and the default risk in the IRC amounts. 69

140. In accordance with Article 372(d) in conjunction with Article 368(1)(f) of the
CRR, an IRC model must be reasonably accurate in measuring risks.
Therefore, an institution should be able to demonstrate that the number of
simulations used in its model to compute the IRC and the default risk in the IRC
is sufficient to ensure accurate and stable IRC amounts.

141. In order to assess the accuracy of the IRC calculations, the ECB can, on the
basis of Article 10 of the SSM Regulation, require institutions to calculate a
confidence interval of IRC estimation with a confidence level of 95%.

67
For the purposes of this document, default risk in the IRC means the risk charge calculated with the
institution’s IRC methodology and on the institution’s current IRC portfolio, but without taking the effect
of rating migrations into account. Thus, default risk in the IRC is a stand-alone risk number and not the
default risk contribution to the IRC amount.
68
This understanding is also supported by Article 63(4)(b) of the Final Draft RTS on assessment
methodology for IMA and significant share.
69
A simple way of testing the impact of maturity mismatches leading to imbalanced positions may be to
scale down the PDs and migration of maturing positions, taking into account the reduced time horizon
until maturity.

ECB guide to internal models – Market risk 163

 142. Under paragraph 17.2 of the EBA Guidelines on the IRC, institutions should use
one (or, where relevant data are available, more than one) migration matrix that
is specific to sovereign issuers (where relevant). Therefore, institutions should
use a separate migration matrix for other types of issuers. Where an institution
uses only one matrix for all types of issuers, it should be able to demonstrate
that this leads to conservative IRC amounts. 70

143. In accordance with Article 372(d) in conjunction with Article 368(1)(e) of the
CRR, institutions must ensure compliance with a documented set of internal
controls related to their IRC model. So that the ECB can assess compliance
with this requirement, an institution should be able to provide an inventory of all
open validation findings in relation to its IRC model, including a description of
the finding, the envisaged remedial action and the target date for closure of the
finding. In addition, institutions should retain closed validation findings for at
least one year after the closure date and should be able to provide a description
of the remedial action implemented.

6.3 Data inputs

144. In accordance with Article 376(6) of the CRR, an IRC model must meet
minimum data standards. This applies in particular to time series used to
calibrate the IRC model, for which the institution should have a process in place
to check the quality of the time series regularly. The ECB considers that an
appropriate minimum frequency for checking the quality of the time series is
quarterly, as this allows alignment with the regulatory reporting cycle. Therefore,
the data quality requirements for VaR and sVaR models indicated in
paragraphs 117 to 118 and 120 to 121 also apply to the market data used for
calibration of the IRC model.

6.4 Distribution and correlation assumptions

145. In accordance with Article 376(3)(a) of the CRR, an institution must, as part of
the annual independent review and the initial and periodic validation of its IRC
model, validate that its modelling approach for correlations and price changes is
appropriate for its portfolio, including the choice and weights of its systematic
risk factors. The ECB understands that this provision requires institutions in
particular to justify (i) the choice of systematic factor types (for example, region
and industry) and, for each type of systematic factor, its granularity, and (ii) the
full correlation structure and its calibration for the entire set of risk factors used.

146. An institution that does not calibrate the correlations of its IRC model to market
data, but instead uses internal ratings based (IRB) correlations, should
demonstrate their appropriateness in relation to its portfolio. The ECB considers

70
The impact on the default risk in the IRC should also be provided upon request.

ECB guide to internal models – Market risk 164

 that owing to the nature of the regulatory trading book, the correlations as
defined in Article 153 of the CRR should be used for this purpose.

147. In accordance with Article 374(2) of the CRR, correlation assumptions must be
supported by analysis of objective data in a conceptually sound framework. In
accordance with Article 372(d) in conjunction with Article 369(1) of the CRR,
institutions must perform tests to demonstrate that any assumptions made
within the internal model are appropriate. In view of those two provisions, the
ECB considers that any assumption for correlation modelling made by the
institution should be supported by objective market data (for example, credit
default swap data, equities data or rating migrations data) and justified by a
quantitative analysis as part of its initial and periodic validation process. In
particular, this quantitative analysis should compare the level of correlation
between issuers that is derived from the institution’s IRC correlation model and
from objective market data. The ECB understands that this requirement also
applies to those institutions using an IRB-based methodology and to those
using a vendor model.

148. In order to assess the appropriateness of the modelling approach for

correlations, the ECB can, on the basis of Article 10 of the SSM Regulation,
require institutions to provide correlations for all issuer pairs, and all relevant
correlation values according to their factor model, in particular for the
systematic factors.

149. In accordance with Article 374(2) of the CRR, concentrations under stressed
conditions must be reflected in the correlation assumptions of the IRC model.
The ECB understands that the use of a short period of data for calibrating the
correlations implies the risk that stressed conditions are not appropriately
reflected. Therefore, institutions should be able to justify that stressed
conditions have been adequately captured and to quantify the impact of using
sufficiently long time series capturing a relevant stressed period for calibrating
correlations in the IRC and the default risk in the IRC. In accordance with
Article 370(c) of the CRR in conjunction with Article 372(d) of the CRR, the IRC
model must be robust to an adverse environment. Therefore, the ECB
considers that, in order to also ensure a robust calibration of the IRC model, a
time series of at least 10 years, capturing a relevant stressed period, is
appropriate. 71

150. In accordance with Article 376(3)(b) of the CRR, institutions must perform
sensitivity analysis and scenario analysis to assess the qualitative and
quantitative reasonableness of the internal model, particularly with regard to the
treatment of concentrations. Because the weights of the systematic risk factors
are relevant for the modelling of concentrations, the ECB considers that
institutions should – as part of the independent review, and in the initial and
periodic validation process – perform sensitivity analyses for the IRC and the
default risk in the IRC. In particular, the ECB considers it best practice that this

71
By using proxies if, and where, necessary.

ECB guide to internal models – Market risk 165

 sensitivity analysis includes, as a minimum, the following basic analysis, where
systematic risk factor weights or correlations of risk factors 72 in the model are
shifted up or down by a fixed value or set to generic values:

(a) all weights of the systematic factors per issuer, for each issuer, 73 are
shifted by +10% in absolute value (not going beyond 100%); 74

(b) all weights of the systematic factors per issuer, for each issuer, are shifted
by −10% in absolute value (not going below 0%);

(c) all weights of the systematic factors per issuer, for each issuer, are set to
0;

(d) all weights of the systematic factors per issuer, for each issuer, are set to
1;

(e) all correlations between systematic factors are set to 100% (weights of
issuers to their respective systematic factors remain unchanged);

(f) all correlations between systematic factors are set to 0% (weights of

issuers to their respective systematic factors remain unchanged).

In order to assess the appropriateness of the sensitivity analysis and scenario

analysis performed to validate the reasonableness of the internal model, the
ECB can, on the basis of Article 10 of the SSM Regulation, require institutions
to provide the results of the sensitivity analysis as described in this paragraph,
points (a) to (f).

151. In order to assess the appropriateness of the modelling approach for

correlations, and in particular the choice and weights of the systematic risk
factors, the ECB can, on the basis of Article 10 of the SSM Regulation, require
institutions to calculate the IRC and the default risk in the IRC amounts based
on a one-factor Merton Model (using one single global systematic risk factor)
and one flat correlation with different correlation assumptions: 0%, 5%,10%,
15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, 60%, 70%, 80%. All other inputs
into institutions’ IRC models remain unchanged.

152. In accordance with Article 374(2) of the CRR, correlation assumptions must be
supported by analysis of objective data in a conceptually sound framework. In
accordance with Article 376(3)(a) of the CRR, an institution must, as part of the
annual independent review and the initial and periodic validation of its IRC
model, validate that its modelling approach for correlations is appropriate for its
portfolio. In accordance with Article 376(3)(c) of the CRR, institutions must
apply appropriate quantitative validation. Under paragraph 25.2(iii) of the EBA
72
The latent variables of the model that determine the correlation of migration and default events of the
issuers.
73
If the asset value 𝐴𝐴𝑖𝑖 of an obligor 𝑖𝑖 is written as follows in a factor model: 𝐴𝐴𝑖𝑖 = �𝜌𝜌 i 𝑋𝑋𝑖𝑖 + �1 − 𝜌𝜌 i 𝜉𝜉𝑖𝑖
where 𝑋𝑋𝑖𝑖 is driven by systemic contributions and 𝜉𝜉𝑖𝑖 is the idiosyncratic noise term, the weights of the
issuers to their respective systematic factors correspond to �𝜌𝜌 i .
74
Which reduces the idiosyncratic weight accordingly.

ECB guide to internal models – Market risk 166

 Guidelines on the IRC, in the validation process the impact of different copula
assumptions should be analysed, for example by testing the impact of different
distributional assumptions. Because the copula choice is a key assumption of
the modelling approach for correlations, the ECB understands that these
provisions require, in particular, that institutions demonstrate the
appropriateness of and validate the copula choice of the modelling approach for
correlations. The copula choice refers to the copula of the joint multivariate
distribution 75 of the risk factors for migration and default and of the joint
systematic risk factors, where relevant.

153. In accordance with Article 376(3)(b) of the CRR, institutions must perform
sensitivity analysis and scenario analysis to assess the qualitative and
quantitative reasonableness of the internal model, particularly with regard to the
treatment of concentrations. Because the copula choice is a key assumption of
the modelling approach for correlations and is relevant for the modelling of
concentrations, the ECB considers that institutions should – as part of the
independent review, and in the initial and periodic validation process – perform
sensitivity analyses for different copula assumptions. The ECB considers that
the following are suitable choices for comparing the impact of different copulas
on the IRC and the default risk in the IRC with respect to the approved model:

(a) using a Student-t copula for all issuer risk factors with 8 degrees of
freedom;

(b) where relevant, using a Student-t copula for the systematic risk factors
with 8 degrees of freedom;

(c) using a Student-t copula for all issuer risk factors where the degrees of
freedom have been calibrated to market data;

(d) where relevant, using a Student-t copula for the systematic risk factors
calibrated to market data.

This list is without prejudice to the discretion of an institution to perform

additional analyses on copula choices that it deems more fitting for its particular
circumstances.

75
Sklar’s theorem (in Sklar, A., “Fonctions de répartition à n dimensions et leurs marges”, Publications de
l’Institut de Statistique de L’Université de Paris, Vol. 8, 1959, pp. 229-231) states that every multivariate
cumulative distribution function of a random vector can be expressed in terms of its marginals and a
copula.

ECB guide to internal models – Market risk 167

6.5 Ratings, probabilities of default and recovery rate assumptions

6.5.1 Documentation requirements

154. In accordance with Article 372(d) in conjunction with 368(1)(e) of the CRR, an
institution must have a documented set of internal policies and controls
concerning the overall operation of its internal models. The ECB considers that
for the IRC model institutions should have in place, in particular:

(a) methodology and process documents for the determination of probabilities

of default (PDs) and recovery rates (RRs), including a process and
documentation concerning the fallback approaches applied;

(b) validation documents demonstrating that the assumptions relating to PDs

and RRs are appropriate;

(c) a documented hierarchy of preferred sources for the determination of PDs

and RRs, which are applied to all issuers and instruments within the scope
of the IRC model.

155. In order to assess the appropriateness and implementation of the policies and
procedures for determining PDs and RRs, the ECB can, on the basis of
Article 10 of the SSM Regulation, require institutions to provide on request a
complete list of positions in the IRC model, together with the respective issuer
or obligor ratings, the PDs of the issuer or obligor and the RRs for the positions.
If the ratings, PDs or RRs have been adjusted or have not been assigned using
the usual automated process (for example, by manual intervention or deviation
from the usual automated process), the institution should maintain a complete
list of such ratings, PDs and RRs, and provide the rationale for the adjustment,
or (for example) manual intervention or deviation from the usual automated
process (as applicable) in each case.

6.5.2 Validation requirements

156. In accordance with Article 376(3)(b) of the CRR, institutions must perform
sensitivity analysis and scenario analysis to assess the qualitative and
quantitative reasonableness of the internal model. Therefore, the ECB
considers that institutions should – as part of the independent review and in the
initial and periodic validation process – perform sensitivity analyses with respect
to the PDs and RRs that are applied to assess the quantitative impact in terms
of the IRC and the default risk in the IRC. In particular, the ECB considers it
best practice that such sensitivity analysis include, as a minimum, the following
basic analyses on the main drivers of the IRC model:

(a) a simultaneous 10% (absolute) up and down shift (not going beyond 0% or
above 100%) of the RRs used in the portfolio. For models using stochastic

ECB guide to internal models – Market risk 168

 RRs, institutions are expected to incorporate this impact by adjusting the
mean of the RR distributions;

(b) for models using stochastic RRs, a simultaneous 30% (relative) up and
down shift of the standard deviation of RRs;

(c) for all the PDs used in the IRC calculation:

(i) a minimum value of 0.01% for all PDs;

(ii) a minimum value of 0.03% for all PDs;

(iii) a simultaneous 10% relative upshift of all PDs;

(iv) a simultaneous 10% relative downshift of all PDs;

(v) a simultaneous 1bp absolute upshift of all PDs;

(vi) a simultaneous 1bp absolute downshift 76 of all PDs.

The change in PD should be compensated for by proportionally increasing (or

decreasing) all the migration probabilities belonging to the same initial rating
class to maintain the cumulative 100% migration and default probability. 77

As the sensitivity analyses listed above are part of the model validation,
institutions should take them into account in detail when assessing and
justifying their PDs and RRs parameters. The assessment should encompass
an analysis of how the most important issuers and groups of issuers are
affected by the altered PD and RR values.

In order to assess the appropriateness of the sensitivity analysis and scenario

analysis performed to validate the reasonableness of the internal model, the
ECB can, on the basis of Article 10 of the SSM Regulation, require institutions
to provide the results of the sensitivity analysis described in paragraph 156(a) to
(c).

157. In accordance with Article 376(2) of the CRR, the IRC model must be based on
data that are objective and up-to-date. Therefore, the ECB considers that
institutions should demonstrate, on the basis of objective data, that the PD
estimates 78 are appropriate.

158. In accordance with Article 373 of the CRR, the IRC model must cover all
positions that are subject to own funds requirements for specific interest rate
risk, including those with a 0% specific risk capital charge under Article 336 of
the CRR. In accordance with Article 372(a) of the CRR, the IRC model must

76
Not going below zero.
77
Given an initial set of migration probabilities, 𝑝𝑝1 , … , 𝑝𝑝𝑛𝑛 , where 𝑝𝑝𝑛𝑛 corresponds to the PD, the
′
𝑝𝑝𝑛𝑛 −𝑝𝑝𝑛𝑛
probabilities 𝑝𝑝𝑖𝑖 for 𝑖𝑖 = 1, … , 𝑛𝑛 − 1 should be adjusted by 𝑝𝑝𝑖𝑖′ = 𝑝𝑝𝑖𝑖 ⋅ �1 + � where 𝑝𝑝𝑛𝑛′ corresponds to
1−𝑝𝑝𝑛𝑛
the modified PD.
78
The same requirements apply to the rating agency data.

ECB guide to internal models – Market risk 169

 give a meaningful differentiation of risk, and accurate and consistent estimates
of incremental default and migration risk. Therefore, the ECB understands that
all annual PDs should be risk sensitive and greater than zero 79 for all obligors.
In this context, the term “greater than zero” is interpreted to mean greater than,
or equal to, one basis point.

159. In accordance with Article 376(2) of the CRR, the IRC model must be based on
data that are objective and up-to-date. Therefore, the ECB considers that
institutions should demonstrate, based on objective data, that the RR estimates
are appropriate.

Based on its observations of the practices of the industry, the ECB considers it
best practice that the RRs do not exceed the following values:

(a) 25% for subordinated debt;

(b) 55% for senior unsecured debt;

(c) 88.75% for covered bonds;

(d) 75% for any other product.

This does not exclude the possibility that higher RRs may be used, where
institutions can justify them by objective and up-to-date data in accordance with
Article 376(2) of the CRR. This best practice also applies to positions under the
fallback approach for the RRs (i.e. for which no direct data sources are
available).

6.5.3 Consistency requirements

160. In accordance with Article 372(a) of the CRR, the IRC model must give
accurate and consistent estimates of incremental default and migration risk.
Therefore, and in order to ensure that institutions do not use different sources
for PDs and RRs for the sole purpose of reducing their overall own funds
requirements for market risk, the ECB considers that they should apply
consistent sources for PDs and RRs in the IRC model. Therefore, institutions
using internal ratings should use the corresponding internal RRs, and those
using external ratings should use historical, market implied or market
convention RRs. 80

79
This understanding is also supported by Article 65(3) of the Final Draft RTS on assessment
methodology for IMA and significant share.
80
This understanding is also supported by Article 70(3) of the Final Draft RTS on assessment
methodology for IMA and significant share.

ECB guide to internal models – Market risk 170

6.5.4 Requirements for PD fallback values

161. In accordance with Article 376(6) of the CRR, proxies must be appropriately
conservative and may be used only where the available data are insufficient.
The ECB is aware that for positions where a reliable PD assignment is not
possible due to a lack of adequate data (for example, where no internal or
external ratings or liquid credit spread time series are available) institutions
apply a fallback PD value. As fallback PD values are used when the available
data are insufficient, the ECB considers them as proxies. In order to ensure that
fallback PDs are appropriately conservative, the ECB considers it best practice
that the fallback PD assigned to each of those issuers and positions is at least
equal to the higher of the following.

(a) The PD that is equivalent to the worst investment grade rating applicable,
according to the institution’s sources for the determination of PDs – those
institutions using internal rating approaches for the assignment of PDs
should use the PD that is equivalent to the worst investment grade rating
in their internal rating scales.

(b) The equally weighted 81 average PD 82 of those issuers included in the IRC
model which are not subject to the fallback approach. Institutions may
exclude defaulted issuers when calculating the equally weighted average
PD, provided that they can ensure that the fallback PD is not applied to
defaulted issuers.

162. In accordance with Article 372(a) of the CRR, the IRC model must provide a
meaningful differentiation of risk, and accurate and consistent estimates of risk.
In accordance with Article 376(2) of the CRR, the IRC model must be based on
data that are objective and up-to-date. Therefore, the ECB considers that
institutions should periodically assess the materiality of those issuers and
positions that are assigned a fallback PD in the IRC model. The ECB considers
that a suitable analysis for this purpose consists of the following:

(a) comparison of the jump-to-default risk (where applicable, by using the

average of the RRs in the case of a stochastic RR) of those positions that
are assigned fallback PDs with the jump-to-default risk of all positions in
the IRC model;

81
All issuers have the same weights.
82
The PD may be derived from the rating of the issuer by applying the migration matrix of the IRC model.
In addition, defaulted positions are included in the calculation of the average with a PD equal to 100%
as issuers subject to the fallback approach could be in default.

ECB guide to internal models – Market risk 171

 (b) calculation of the ratio of the incremental 83 IRC contributions and the
incremental default risk in the IRC contributions of the positions assigned
fallback PDs, to the IRC number as calculated by the IRC model:

IRC (full scope) - IRC (non-fallback for PD)

;
IRC (full scope)

Default risk in the IRC(full scope) - Default risk in the IRC (non-fallback for PD)
Default risk in the IRC(full scope)

163. In accordance with Article 372(a) of the CRR, the IRC model must provide a
meaningful differentiation of risk. The ECB considers that if a significant
percentage of the IRC is calculated using fallback PDs, there is a risk that the
IRC model may not provide the meaningful differentiation of risk required. The
ECB considers that if the percentage of the IRC calculated using fallback PDs is
larger than 10%, 84 the institution should investigate whether additional data
sources are available to reduce the percentage of issuers subject to the fallback
PD assignment.

In accordance with Article 376(3)(b) of the CRR, institutions must perform

sensitivity analysis and scenario analysis to assess the qualitative and
quantitative reasonableness of the internal model. In the event that the resulting
percentage of the IRC calculated using fallback PDs is larger than 10%, the
ECB considers that institutions should perform, as part of the independent
review and validation, a sensitivity analysis for the IRC and the default risk in
the IRC. In particular, the ECB understands that this requirement implies
assessing the sensitivity of the IRC and default risk in the IRC amounts by
assigning one rating grade higher and one rating grade lower than the one used
in the fallback PD assigned on the basis of paragraph 161. 85

6.5.5 Requirements for the calculation of losses based on recovery rates

164. In accordance with Article 372(a) of the CRR, the IRC model must provide
accurate and consistent estimates of incremental default risk. Therefore, the
ECB considers that the market value change following the default of an issuer
should be calculated as the difference between the current market value of the
position and the expected market value subsequent to default.

165. In accordance with Article 372(a) of the CRR, the IRC model must provide a
meaningful differentiation of risk. In accordance with Article 375(1) of the CRR,

83
Risk management literature is not uniform in the use of the terms “incremental” risk number
(e.g. incremental value-at-risk) and “marginal” risk number. We adopt the convention that an
incremental risk number refers to the exact finite change in a risk number when adding a finite position,
whereas “marginal” risk number refers to the derivative of a risk number with respect to a position
(infinitesimal change rate).
84
This threshold of 10% is set by analogy with the Commission Delegated Regulation on materiality of
extensions and changes of the IMA, which establishes 10% as the threshold for assessing materiality.
85
When the average PD is used as a fallback approach, institutions should, by analogy, apply this
requirement (i.e. identify the rating grade that is closest to the average PD and shift up and down
starting from this rating grade).

ECB guide to internal models – Market risk 172

 institutions must reflect the potential for significant basis risks in hedging
strategies by product, seniority in the capital structure, internal or external
rating, maturity, vintage and other differences in the instruments. The ECB
considers that in order to provide a meaningful differentiation of risk and to
reflect the potential for significant basis risks, recovery rates should at least
reflect the type of product, including the collateralisation of the position, and its
seniority in the capital structure.

166. In accordance with Article 372(a) of the CRR, the IRC model must provide
accurate and consistent estimates of incremental default risk. In accordance
with Article 372(d) in conjunction with Article 369(1) of the CRR, institutions
must demonstrate that any assumptions made within the IRC model are
appropriate. The ECB considers that this applies also to RRs. An RR measures
the expected market value subsequent to default of a position as a percentage
of the base value (for example, notional). For a long credit position (for
example, the holding of a long bond position) a default of the issuer would lead
to a loss. For RRs based on notional value, a negative RR indicates a negative
expected market value subsequent to default, whereas an RR above 100%
indicates that the expected market value subsequent to default is higher than
the notional value. The ECB is aware that RRs generally range between 0%
and 100%. RRs outside this range could indicate that the assumptions made
within the IRC model are not appropriate – because they could imply an
expected profit subsequent to default – and so institutions should be particularly
prudent in applying such RRs or be able to demonstrate that they are
conservative.

6.6 Treatment of groups of connected issuers

167. In accordance with Article 374(2) of the CRR, the IRC model must appropriately
reflect issuer concentrations. As defined in Article 4(1)(39) of the CRR, two or
more legal persons in the same group of connected clients constitute a single
risk, unless it is shown otherwise. The ECB considers that groups of connected
clients are relevant for modelling issuer concentrations. Therefore, such groups
should be appropriately reflected in the IRC model and their treatment in the
model is subject to the same requirements as any other component of the
model, in particular documentation and validation.

168. In accordance with Article 374(2) of the CRR, the IRC model must appropriately
reflect issuer concentrations. As defined in Article 4(1)(39) of the CRR, a group
of connected clients constitute a single risk, unless it is shown otherwise.
Therefore, the ECB considers that institutions should model issuers and
obligors in the same group of connected clients as a single risk (this means, for
example, that in an asset value model they should be modelled as a single
asset value). However, the existence of different rating grades within a group of
connected clients indicates the possibility that not all of those in the group
default or migrate simultaneously. Therefore, the ECB considers that a suitable
method of modelling is to distinguish within a group of connected clients by sub-

ECB guide to internal models – Market risk 173

 groups of issuers that have the same internal or external rating grade and
where the default or migration of each sub-group would occur simultaneously in
the IRC model – unless it is demonstrated that another treatment is more
appropriate in view of the definition in Article 4(1)(39) of the CRR.

169. In accordance with Article 376(3)(a) of the CRR, an institution must, as part of
the annual independent review and the initial and periodic validation of its IRC
model, validate that its modelling approach for correlations is appropriate for its
portfolio. Because modelling groups of connected clients is relevant for
modelling issuer concentrations and the correlations among them, the ECB
considers that validation of the modelling of groups of connected clients is part
of the annual independent review and the initial and periodic validation of the
institution’s IRC model.

7 Risks not in the model engines

7.1 Regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 99, 103, 105

363 (1)

366, 367, 368

369 (1)

372 (a)

377

SSM Regulation 15/10/2013 10

Commission Delegated Regulation on 04/03/2015 7a (1)(c)

materiality of extensions and changes of the
IMA

7b

Annex III Part II, Section 2(13)

Commission Implementing Regulation on 16/04/2014 5 (a)

supervisory reporting 86

7.2 The framework for risks not in the model engines

170. In accordance with Article 367(1)(a) of the CRR, any internal model used to
calculate capital requirements for market risk must capture accurately all
material price risks. For IRC models, the accuracy of the risk estimates is also
required by Article 372(a) of the CRR. In accordance with Article 367(1)(b) of

86
Commission Implementing Regulation (EU) No 680/2014 laying down implementing technical
standards with regard to supervisory reporting of institutions according to Regulation (EU) No 575/2013
of the European Parliament and of the Council (OJ L 191, 28.6.2014, p. 1), referred to in this guide as
the “Commission Implementing Regulation on supervisory reporting”.

ECB guide to internal models – Market risk 174

 the CRR, where a risk factor is incorporated into the institution's pricing model
but not into the risk measurement model, the institution must be able to justify
such an omission to the satisfaction of the competent authority. In accordance
with Article 368(2)(d) of the CRR, the annual review of an institution’s overall
risk management process must consider the scope of risks captured by the risk
measurement model. In accordance with Article 369(1) of the CRR, institutions
must have processes in place to ensure that all their internal models for market
risk have been adequately validated to ensure that they are conceptually sound
and adequately capture all material risks. In accordance with Article 368(1)(e) of
the CRR, the institution must have in place established procedures for
monitoring and ensuring compliance with a documented set of internal policies
and controls concerning the overall operation of its internal models.

Based on the provisions referred to above, the ECB considers that the
processes set out in detail in this section for risks not captured in the model 87
engines (also referred to in this guide as “risks not in the model engines”, or
“RNIME” 88) are an integral part of the overall processes of the IMA for market
risks. Therefore, institutions should develop an RNIME framework, the elements
of which are further elaborated on in the following paragraphs.

171. For the purposes of this guide and in relation to the RNIME framework, the
following diagram shows schematically different components of the market risk
own funds requirements and the RNIME framework. 89 The ECB considers that
an internal model comprises all of the required methods, processes, policies,
controls, and IT systems. Each internal model includes, inter alia, the following
constituent elements.

(a) An “engine” – that is, the calculation methodology for each risk number,
referred to collectively as “risk engines”. The ECB understands that
Articles 367 and 368 of the CRR refer to an engine as a “risk
measurement model”. There is one risk engine for each risk number and
the risk engine is used to compute the daily risk number. Typically, an
engine models and computes all risks in an integrated manner. However, it
may comprise several components, for example, a main component for
the bulk of the risks, and some “satellite” components. A satellite
component is part of a model engine, typically for a subset of products or
risk positions, meeting all applicable CRR requirements for risk
measurement models on an ongoing basis (for example, for particular
risks not modelled in the main component). In accordance with
Article 367(3) of the CRR, institutions may, in any internal model used for

87
In this section, the generic reference to “model” means a reference to the VaR, sVaR, IRC, and
comprehensive risk measure (CRM) models for correlation trading portfolios as referred to in
Article 377 of the CRR.
88
In this document, the abbreviation “RNIME” may be singular or plural depending on whether it refers to
a single risk, several risks, or collectively all risks not captured in the model engines.
89
In order to simplify the diagram, CRM is not explicitly included. It should be treated in the same way as
IRC.

ECB guide to internal models – Market risk 175

 market risk, use empirical correlations; where they are not used, the model
uses a simple sum aggregation of these components.

(b) An RNIME framework relating to all risk engines, in which RNIME are
identified, quantified, managed and, if appropriate, capitalised by RNIME
add-ons to the risk exposure amounts. The process for determining
RNIME add-ons is part of the RNIME framework. An RNIME add-on is
understood as a temporary risk exposure amount 90 that remains in place
until the corresponding RNIME is incorporated into the model engine(s) in
a manner compliant with the CRR. The ECB considers that the RNIME
add-ons are not part of the model engines, and are therefore not included
in the risk numbers. In particular, RNIME add-ons are not included in the
VaR number used for regulatory back-testing.

Figure 4
Components of market risk own funds requirements and risk exposure amounts
(blue filled boxes), internal models (green frames), and RNIME framework (black
frame)

In accordance with Article 363(1) of the CRR, institutions may calculate their
own funds requirements for market risk using their internal models instead of, or
in combination with, the methods of the standardised approaches for market
risk. Because the positions exposed to RNIME according to the process and
requirements described in this Section 7 are within the scope of the IMA, they
do not need to be accounted for under the standardised approaches (SA) for
market risk.

Exclusions of positions from the scope of the IMA in risk categories for which
the IMA is approved are subject to the requirements described in Section 2.5 of
this guide.

90
To be reported in COREP as “Additional risk exposure amount due to Article 3 CRR” (COREP C02.00
Row 760, Column 010) together with any other own-initiative capital buffers.
Moreover, there could be supervisory imposed add-ons related to market risk which are not shown in
the figure.

ECB guide to internal models – Market risk 176

 172. In accordance with Article 368(1)(e) of the CRR, institutions must have
established procedures for monitoring and ensuring compliance with a
documented set of internal policies and controls concerning the overall
operation of internal models. Therefore, the ECB considers that an institution
should have a policy and controls in place that govern the overall process for
the identification, quantification and management of RNIME. In order to enable
efficient monitoring of RNIME, the ECB considers that the documented policies
should include a description of the different tasks and responsibilities, and the
frequency of their execution. This policy and these controls constitute the
RNIME framework. The ECB considers that the RNIME framework should cover
the tasks described in the following paragraphs of this section.

173. In accordance with Article 368(1)(b) of the CRR, the risk control unit is
responsible for the overall risk management system. Because the RNIME
framework is an integral part of the overall IMA processes, the ECB considers
that the risk control unit is also responsible for the overall RNIME framework.

In accordance with Article 368(1)(b) of the CRR, the risk control unit must
conduct the initial and ongoing validation of any internal model for market risk.
Therefore, the ECB considers that the RNIME framework and methodologies
should be initially and periodically validated internally, and updated if necessary.

7.3 Identification of RNIME

174. A single RNIME identified refers to a distinct risk not accurately captured or
omitted, and related to positions or instruments within the approved risk
categories in the IMA in the VaR, sVaR, IRC or CRM models. This can refer to a
single risk factor, a set of risk factors (e.g. related to a yield curve), a particular
effect (e.g. volatility skew) or specific instruments.

The ECB considers that RNIME can emerge as a result of the following
circumstances.

(a) Differences in the positions, risk factors and pricing methods captured in
VaR, sVaR, and IRC (and CRM if applicable) engines, in comparison with
those of the end-of-day valuation process for the books and records of an
institution. 91 In particular, these may include risk factors that are taken into
account in the economic P&L, but not in the risk measurement model as
referred to in Article 367(1)(b) of the CRR.

Some examples could be: simplified pricing models or sensitivities based

P&L in the risk engines; use of proxies for risk factors; calibration of pricing
models in the risk engines; and risks not adequately modelled, such as
basis risk between two different classes of shares.

91
Those potential RNIME are different from valuation adjustments that an institution might have made in
order to satisfy the fair value and prudent valuation requirements under Article 105 of the CRR.

ECB guide to internal models – Market risk 177

 (b) Weaknesses and limitations in the stochastic modelling of risk factors in
the risk engines that are not linked to the valuation produced by the end-
of-day valuation process.

Some examples could be: distributional assumptions for risk factors of

both the marginal distributions and joint distributions (i.e. correlation
structure); jump risks; calibration of model parameters; regression
approach calibration and deviations; IRC factor model assumptions and
calibration; and insufficient or unreliable data for risk factors.

(c) Other factors leading to risks not being captured accurately or being
omitted from the risk engines.

Some examples could be: instruments on exotic underlyings in the IMA

scope that may be treated under the RNIME framework in the manner
referred to in paragraph 24 of this chapter; positions in defaulted debt, as
referred to in paragraph 32 of this chapter; some risks not accurately
captured due to position data not being updated daily.

175. In accordance with Article 368(1)(e), an institution must have established

procedures for monitoring and ensuring compliance with a documented set of
internal policies and controls concerning the overall operation of its internal
models. Therefore, the ECB considers that in order to ensure a comprehensive
coverage of such risks, the institution should clearly describe and document
each RNIME in an inventory, as part of its RNIME framework.

In order to properly monitor each RNIME, the ECB considers that institutions
should explain how each RNIME is identified and defined, and should, in
particular, be able to justify the cases where a single RNIME is defined across
portfolios or product classes. In order to properly assess materiality, the ECB
understands that the current portfolio composition and trading strategy of the
institution should be taken into account when assessing each RNIME. Unless
the institution can provide justification that the effect of an RNIME is negligible
in the current portfolio and will remain negligible taking into account the trading
strategy, it should take that RNIME into account in its RNIME framework. The
institution should be able to provide justification as to why any particular RNIME
is not included in its risk engines.

176. In accordance with Article 367(1)(a) of the CRR, any internal model must
capture accurately all material price risks. The ECB considers that in order to
ensure an accurate capture of risks, institutions should not rely solely on the
monitoring of current RNIME, but strive to identify RNIME on an ongoing basis,
and as early as possible, as part of the overall risk management. The ECB
considers it best practice to use existing processes efficiently to identify RNIME.

As part of such best practice, and in order to maximise efficiency, institutions

should, at a minimum, use the following processes to identify RNIME:

(a) a review of the institution’s trading strategy, as referred to in Article 103 of

the CRR, considering, in particular, the expansion and reorientation of the

ECB guide to internal models – Market risk 178

 trading business, given that expanding a particular business could lead to
RNIME becoming significant, or to additional risks that are not currently
covered in the RNIME process;

(b) the regulatory back-testing process, as referred to in Article 366 of the

CRR, as part of which the institution should review the results and
analyses of overshootings in order to identify RNIME;

(c) market data quality assurance processes for risk factors, as referred to in
Article 367(2)(e) of the CRR, where market data display insufficient quality;

(d) initial and ongoing internal validation of internal models, as referred to in

Articles 368(1)(b) and 369(1) of the CRR, at least where differences
between the institution's pricing model and risk measurement model are
identified (for example, risk factors that are used for the valuation of a
product for the end-of-day valuation process, but not for risk
measurement), and where internal back-testing shows a high number of
overshootings;

(e) introduction of new products, where the institution should analyse whether
the market risks inherent in the new products and their related trading
strategies can be adequately captured by the risk engines in order to
ensure that these new products – which may pose additional risk factors or
require methodological changes – are fully compatible with the
comprehensive risk control and validation by the risk control unit, as
required by Article 368(1)(b) of the CRR.

In accordance with Article 368(2)(d) of the CRR, the annual review of an

institution’s overall risk management process must consider the scope of risks
captured by the risk measurement model. Therefore, the ECB considers that a
review of the inventory of RNIME should be carried out at least once a year.

7.4 Quantification of RNIME

177. In accordance with Article 367(1)(a) of the CRR, any internal model used to
calculate capital requirements for market risk must capture accurately all
material price risks. In order to ensure a meaningful quantification of RNIME in
relation to the internal models, the ECB considers that the risk parameters for
RNIME quantification should be aligned to the regulatory specifications.
Therefore, the quantification of risks not in the VaR engine should aim to reflect
a loss at a 99% confidence level and a holding period of ten days. Similarly, the
quantification of risks not in the sVaR engine should aim to reflect a loss at a
99% confidence level and a holding period of ten days, and be calibrated to
historical data from the stressed period used to calibrate the sVaR model. The
quantification of risks not in the IRC engine (or CRM engine, if applicable)
should aim to reflect a loss at a 99.9% confidence interval over a time horizon
of one year.

ECB guide to internal models – Market risk 179

178. In accordance with Article 367(1)(a) of the CRR, any internal model used to
calculate capital requirements for market risk must capture accurately all
material price risks. Therefore, the ECB considers that in order to ensure that
the internal models capture all material price risks, institutions should quantify
RNIME in an appropriate way and document and duly justify the methodology
applied. The ECB understands that the quantification of the impact of the
identified i-th RNIME (denoted by RNIME𝑖𝑖 ) serves to assess the need to
incorporate the i-th RNIME into the engine.

The ECB considers it best practice that for each RNIME𝑖𝑖 identified, the impact
quantification 𝑀𝑀𝑖𝑖 should be estimated as the incremental risk number 92 where
RNIME𝑖𝑖 would be incorporated into the model engine; this is in comparison with
the current engine using the same portfolio as reference,

𝑀𝑀𝑖𝑖 ≝ risk number(engine with RNIME𝑖𝑖 incorporated)

− risk number(current engine),
risk number ∈ {VaR, sVaR, IRC, CRM}

where no RNIME add-ons (or other add-ons) are included in the risk numbers.

The impact quantification 𝑀𝑀𝑖𝑖 is a signed number and could be negative if

incorporating RNIME𝑖𝑖 were to be risk-reducing.

The estimation of 𝑀𝑀𝑖𝑖 should be as accurate as possible using reasonable effort.

Therefore, the ECB understands that the 𝑀𝑀𝑖𝑖 estimation methodology can use
appropriate approximations, assumptions, or a stress methodology when duly
justified and documented.

Because the impact quantification should allow the different RNIME to be

assessed individually, no diversification effect should be applied between
different RNIME when quantifying the individual RNIME.

179. The ECB considers that a more conservative impact quantification than
described in paragraph 178 could be used where this is duly justified. In
particular, where an appropriate impact quantification using an incremental risk
number cannot be performed, the ECB considers it a prudent approach to
resort to a stand-alone impact estimation for RNIME𝑖𝑖 ,

�𝑖𝑖 ≝ risk number(RNIME𝑖𝑖 as only source of risk),

𝑀𝑀
risk number ∈ {VaR, sVaR, IRC, CRM},

�𝑖𝑖 for the impact quantification.

and 𝑀𝑀𝑖𝑖 is set to 𝑀𝑀

�𝑖𝑖 should be as accurate as possible using reasonable effort.

The estimation of 𝑀𝑀
Therefore, the ECB understands that the 𝑀𝑀 �𝑖𝑖 estimation methodology can use
appropriate approximations, assumptions, or a stress methodology when duly
justified and documented.

92
See footnote 83 above for details.

ECB guide to internal models – Market risk 180

 As an illustration, in the case of the VaR, and where RNIME𝑖𝑖 can be well
described as a sensitivity 𝑝𝑝𝑖𝑖 to an additional risk factor (i.e. a risk position), the
impact quantification 𝑀𝑀𝑖𝑖 corresponds to its incremental VaR, i.e. the incremental
effect on VaR of adding the risk position 𝑝𝑝𝑖𝑖 to the existing set of risk positions.
Let 𝑝𝑝 denote the set of current risk positions, and let VaR(𝑝𝑝) denote the current
VaR, then the impact quantification 𝑀𝑀𝑖𝑖 of RNIME𝑖𝑖 interpreted as an additional
risk position 𝑝𝑝𝑖𝑖 is

𝑀𝑀𝑖𝑖 = VaR(𝑝𝑝 + 𝑝𝑝𝑖𝑖 ) − VaR(𝑝𝑝)

The impact quantification as incremental risk, M𝑖𝑖 , is different from the

assessment of the risk on a stand-alone basis as a sole source of risk, 𝑀𝑀 �𝑖𝑖 . In
the setting above, the stand-alone risk would be 𝑀𝑀 �𝑖𝑖 = VaR(𝑝𝑝𝑖𝑖 ), which in general
is different from VaR(𝑝𝑝 + 𝑝𝑝𝑖𝑖 ) − VaR(𝑝𝑝). If the sub-additivity property VaR(𝑝𝑝) +
VaR(𝑝𝑝𝑖𝑖 ) ≥ VaR(𝑝𝑝 + 𝑝𝑝𝑖𝑖 ) holds, the stand-alone risk VaR(𝑝𝑝𝑖𝑖 ) is a conservative
estimate of the incremental risk,

�𝑖𝑖 = VaR(𝑝𝑝𝑖𝑖 ) ≡ VaR(𝑝𝑝) + VaR(𝑝𝑝𝑖𝑖 ) − VaR(𝑝𝑝) ≥ VaR(𝑝𝑝 + 𝑝𝑝𝑖𝑖 ) − VaR(𝑝𝑝) = 𝑀𝑀𝑖𝑖 .

𝑀𝑀

Because VaR, sVaR, IRC and CRM are all value-at-risk-based risk measures,
the same applies for those, by analogy.

180. In accordance with Article 367(1)(a) of the CRR, any internal model used to
calculate capital requirements for market risk must capture accurately all
material price risks. The ECB considers that in order to ensure that the
quantification of RNIME is appropriately accurate, the quantification should,
where possible, make use of objective market data, even if the data quality is
not sufficient to model these risks in the model engine.

In order to ensure alignment with the internal models when quantifying an

RNIME – for example, by using sensitivities – the shocks applied in order to
quantify it should be based on the same holding period and, in principle, on the
same observation period as those for the shocks for the other risk factors used
in the relevant internal model. Differences in the observation period should be
duly justified. If scarce data are used to calibrate these shocks, the shocks
should be estimated conservatively. This may involve relying to some extent on
expert judgement.

7.5 Management of RNIME and implementation in an institution’s risk

engines

181. In accordance with Article 367(1)(a) of the CRR, any internal model must
capture accurately all material price risks. The ECB considers that in order to
ensure ongoing accurate risk capture, the risk control unit should carry out
regular impact quantification and monitoring of all RNIME.

If an institution can provide justification that an impact quantification of a VaR

RNIME also applies for sVaR, the sVaR impact quantification and monitoring

ECB guide to internal models – Market risk 181

 may be based on the VaR impact quantification. If it cannot provide such
justification, or where certain RNIME have been identified specifically for the
sVaR engine, a specific impact quantification and monitoring for those sVaR
RNIME should be performed. Monitoring of RNIME should include, in particular,
checking whether RNIME are above certain thresholds, as further detailed
below in this Section 7.5.

182. In accordance with Article 99 of the CRR in conjunction with Article 5(a) of the
Commission Implementing Regulation on supervisory reporting, institutions
must submit the information relating to own funds requirements with a quarterly
frequency. Therefore, the ECB considers that in order to assess the adequacy
of own funds, institutions should quantify and monitor the RNIME at least
quarterly.

The risk control unit should report the outcome of the quantification and
monitoring to the committee or persons responsible for deciding on the
management of RNIME in terms of identification, quantification, treatment,
limitation, reporting frequency, etc.

183. In accordance with Article 367(1)(a) of the CRR, any internal model must
capture accurately all material price risks. Therefore, the ECB considers that in
order to ensure that the models accurately capture all material price risks
including RNIME and thereby result in a sufficient level of own funds,
institutions should take into account all of the following points.

(a) An RNIME𝑖𝑖 , where 𝑀𝑀𝑖𝑖 < 0 does not allow the reduction of own funds
requirements until the related risk has been incorporated into the relevant
engine.

(b) Institutions should determine thresholds for assessing, at their own

discretion, the impact of individual RNIME above which an individual
RNIME is considered a “substantial” RNIME.

The ECB considers that if a single RNIME already has a 5% impact, there
is a risk that the risk engine might not capture accurately all material risks.
Therefore, the ECB considers as best practice that the i-th individual
RNIME is considered substantial if the impact quantification 𝑀𝑀𝑖𝑖
corresponds to more than 5% of the amount computed by the risk engine 93
(without taking into account any add-ons, as they are not included in the
relevant risk number).

That is, RNIME𝑖𝑖 is considered substantial if

𝑀𝑀𝑖𝑖
> 5% , risk number ∈ {VaR, sVaR, IRC, CRM}
risk number

93
The calculation should be made at the end of the quarter by comparing the impact quantification of the
RNIME, e.g. at the end of the quarter, with the previous 60-business day average of the VaR or sVaR,
or the previous 12-week average of the IRC or CRM amount (without any add-ons).

ECB guide to internal models – Market risk 182

 This is without prejudice to the discretion of an institution to set a lower
threshold than 5%.

The ECB considers it best practice and prudent that institutions should
include any substantial RNIME𝑖𝑖 in their total risk exposure amount by way
of an RNIME𝑖𝑖 add-on of size 𝑀𝑀𝑖𝑖 multiplied by 12.5, in order to calculate the
corresponding risk exposure amount. In so doing they should take into
account the multiplication factors (mc ) and (ms ) for VaR and sVaR as
referred to in Article 366 of the CRR without the back-testing addend, until
they have incorporated it into the engine affected. Consistent with the
impact quantification, the ECB considers that there should not be any
diversification effect between different RNIME add-ons.

(c) Institutions should determine, at their own discretion, thresholds above

which RNIME are incorporated into the model engines.

In accordance with Article 7a(1)(c)(ii) of the Commission Delegated

Regulation on materiality of extensions and changes of the IMA, a change
of 10% or more of a relevant risk VaR, sVaR, IRC, or CRM number is to be
considered a material change to the IMA. Therefore, the ECB considers,
by analogy, that if the cumulative RNIME impact corresponds to more than
10% of the amount computed by the risk engine, this indicates that an
engine might not capture accurately all material price risks, 94 as the
change needed to incorporate them into the engine could amount to a
material model change.

In order to assess whether that is the case, institutions should calculate

the cumulative impact quantification CIQ risk number per risk number by
adding the positive impacts of all RNIME related to that risk number,
including those subject to an RNIME add-on, without taking any
diversification among the different RNIME into account, and dividing by the
risk number computed by the model engine without taking any add-ons
into account. 95 If the resulting ratio is greater than 10%, the ECB considers
that the model engine might not accurately capture all material price risks,

∑all RNIME𝑖𝑖 related to risk number max[𝑀𝑀𝑖𝑖 , 0]

CIQ risk number = > 10%,
risk number

risk number ∈ {VaR, sVaR, IRC, CRM}

This is without prejudice to the discretion of an institution to set a lower

threshold than 10%.

94
This is without prejudice to a determination by the ECB, based on an assessment taking into account
the specific circumstances of the institution, that the model does not accurately capture all material
price risks.
95
The calculation should be made at the end of the quarter by comparing the sum of impact quantification
of the RNIME, e.g. at the end of the quarter, with the 60-business-day average of the VaR or sVaR, or
the 12-week average of the IRC or CRM amount of the preceding quarter.

ECB guide to internal models – Market risk 183

 If it is the case that the ratio as calculated above is greater than 10% (or a
lower threshold set by the institution), the institution should provide the
ECB with an implementation plan for the incorporation of some or all of
these RNIME in the model engine(s), such that the cumulative impacts are
reduced below the threshold.

(d) The ECB considers it a prudent approach that RNIME which are to be
incorporated into the relevant engine(s) are capitalised with RNIME
add-ons as part of the implementation plan, until they are incorporated into
the relevant engine(s). If the institution deems it convenient, the remaining
RNIME may also be capitalised with RNIME add-ons.

184. With reference to the previous paragraphs in this Section 7.5, the incorporation
of RNIME into the model engine should be performed so that the engine
complies with all relevant requirements of the CRR including, in particular,
internal validation. The term “incorporation” here means the integration of
RNIME into the relevant risk engine, and into its methodology and processes,
typically allowing for risk diversification. This is without prejudice to the
discretion of an institution not to use empirical correlations within risk categories
or across risk categories, as referred to in Article 367(3) of the CRR, by
applying instead a simple sum aggregation.

185. In accordance with Article 99 of the CRR in conjunction with Article 5(a) of the
Commission Implementing Regulation on supervisory reporting, institutions
must submit the information relating to own funds requirements with a quarterly
frequency. Therefore, the ECB considers that in order to ensure an accurate
quarterly reporting of own funds requirements and risk exposure amounts, the
RNIME add-ons should be updated at least quarterly.

In order to enable monitoring of RNIME add-ons, the ECB can, on the basis of
Article 10 of the SSM Regulation, require an institution to provide an overview
of RNIME add-ons in a suitable format chosen by the institution.

186. Because the RNIME framework is an integral part of the overall IMA processes,
a change to the RNIME framework – in particular one that relates to the RNIME
identification methodology, the consideration of new types of RNIME, the impact
quantification methodology, or the RNIME add-on methodology – constitutes an
IMA model change and should therefore be assessed in accordance with the
Commission Delegated Regulation on materiality of extensions and changes of
the IMA.

In accordance with Article 7b and Annex III, Part II, Section 2(13) of the
Commission Delegated Regulation on materiality of extensions and changes of
the IMA, any structural, organisational or operational change to the core
processes in risk management or risk controlling functions requires ex ante
notification to the competent authorities. The ECB considers that because the
RNIME framework is an integral part of the overall IMA processes, a change to
the RNIME framework should accordingly be notified ex ante to the competent
authorities.

ECB guide to internal models – Market risk 184

 However, changes within the existing RNIME framework which do not need new
methodologies or processes to be implemented should be notified to the
competent authorities through ex post notifications only.

187. Ceasing to capitalise an RNIME, or capitalising an RNIME with an RNIME add-

on according to the thresholds of the RNIME framework, does not constitute a
model change and does not need to be separately notified as a model change,
provided that it is based on the approved methodology of the RNIME
framework.

188. The incorporation of RNIME𝑖𝑖 into the model engine, irrespective of whether it
was previously treated as an RNIME add-on or not, and irrespective of whether
it is an RNIME identified previously or is newly identified, constitutes an IMA
model change and should therefore be assessed in accordance with the
Commission Delegated Regulation on materiality of extensions and changes of
the IMA. The materiality assessment, in accordance with Article 7a(1)(c)(ii) of
that Commission Delegated Regulation, should be based on the new risk
number, i.e. on the following ratio,

risk number(engine with RNIME𝑖𝑖 incorporated)

,
risk number(current engine)
risk number ∈ {VaR, sVaR, IRC, CRM}

For the sum of market risk requirements, the assessment of materiality in

accordance with Article 7a(1)(c)(i) of that Commission Delegated Regulation
should be made analogously.

189. Because the RNIME add-ons are not included in the VaR number, they should
not be taken into account when performing regulatory back-testing. However, all
VaR engine components that constitute the VaR engine (including, where
applicable, satellite components) should be taken into account in the regulatory
back-testing.

ECB guide to internal models – Market risk 185

 Counterparty credit risk

1 Scope of the counterparty credit risk chapter

1. The purpose of this chapter is to provide transparency on how the ECB
understands a number of topics related to the principles defined for the Internal
Model Method (IMM 1, as referred to in Part Three, Title II, Chapter 6, Section 6
of the CRR), which have been looked at in the targeted review of internal
models (TRIM). Consequently, this chapter does not contain an exhaustive list
of topics relevant for compliance with IMM requirements that could be subject to
review during future internal model investigations. 2

2. In the understanding of the ECB and in order to comply with Article 6(1) of the
CRR, all requirements of Part III Title 2 Chapter 6 Section 6 must also be met
by all legal entities that have approval to use the IMM for solo capital
requirement calculations. This is especially important when underlying portfolios
differ and the portfolio at consolidated level is not representative of that at the
solo level. In particular, these requirements are relevant for stress period
determination in accordance with Article 292(3) of the CRR and for all validation
requirements when selecting, for example, relevant risk factors and synthetic
portfolios for back-testing.

3. The following sections are structured in the same manner and cover those
issues relating to counterparty credit risk (CCR) for which the TRIM project was
intended to ensure the consistent application of regulatory requirements. For
each item the following apply.

(a) References are only made to the relevant CRR provisions that require
more guidance. Other relevant provisions of the CRR are therefore not
mentioned in the guide, but are not to be disregarded; this refers
specifically to paragraphs 7, 19, 25, 37, 42, 49, 54, 58, 73, 74 and 77.

(b) Principles are expressed following CRR requirements as they are

understood by the ECB.

1
Note that the advanced method for the CVA capital requirement is not in scope here.
2
A prominent example is data quality.

ECB guide to internal models – Counterparty credit risk 186

2 Trade coverage
4. For the purposes of this section, “IMM transactions” are transactions for which
the institution has approval to use the IMM to estimate the related exposure
value.

5. This section refers to transactions for which the institution does not have
approval to use the IMM, and IMM transactions, for which the related exposure
is not fully simulated in the IMM. 3

6. The section also addresses potential carve-outs of transactions from the IMM
scope to a non-IMM method, for example due to price differences compared
with benchmarking systems 4, and the consequences of the potential creation of
synthetic netting sets.

2.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 273 (6)

283 (1), (3)

284 (1)

293 (4)

294 (1)(d), (l), (o)

Other references

ECB Guide on options and discretions

11/2016 Section II, Chapter 3, paragraph 9
available in Union law

7. The regulatory provisions relating to the topic addressed in this section that
require further guidance are the following.

(a) Articles 283(1) (permission to use the IMM) and 283(3) (sequential
implementation of the IMM) of the CRR, further specified for banking
supervision in Section II, Chapter 3, paragraph 9 of the ECB Guide on
options and discretions available in Union law, form the basis for IMM
approval.

(b) Article 273(6) of the CRR requires, for all methods in Part Three, Title II,
Chapter 6, Sections 3 to 6 of the CRR (Articles 274 to 294), that the
exposure value for a given counterparty is calculated as the sum of
exposure values, calculated for each netting set with that counterparty.

3
“Fully simulated” in this context means that, for each of the simulated market data paths with a joint
dependency structure at the predefined grid points, a full revaluation of the transactions is performed.
All material risk drivers of the valuation routine are simulated and the pricing function is not
approximated compared with the benchmarking system.
4
See the definition in the Counterparty credit risk glossary.

ECB guide to internal models – Counterparty credit risk 187

 However, these provisions do not explicitly address the case of synthetic
netting sets arising from the splitting of a contractual netting set.

(c) Article 294(1)(d) of the CRR requires that actions be taken to address the
inaccuracy of the model if model validation indicates that the effective
expected positive exposure (EEPE) is underestimated.

(d) Article 294(1)(l) of the CRR requires that pricing functions be tested
against an appropriate independent benchmark.

(e) Article 294(1)(o) of the CRR requires, in a general way and without further
explanation, that validation “shall assess whether or not the counterparty
level and netting set exposure calculations … are appropriate”.

(f) According to Article 293(4) of the CRR, any “institution shall define criteria
with which to assess its CCR exposure models and the models that input
into the calculation of exposure and maintain a written policy that
describes the process by which unacceptable performance will be
identified and remedied”. However, this provision is drafted in a general
way and needs to be detailed further.

(g) Article 284(1) of the CRR requires that the exposure value at netting set
level be calculated for those transactions where the institution has the
permission to use the IMM in accordance with Article 283(1) of the CRR.

(h) Article 284(1)(a) of the CRR provides that the model used by the institution
must “specify the forecasting distribution for changes in the market value
of the netting set attributable to joint changes in relevant market variables,
such as interest rates, foreign exchange rates”.

(i) Article 284(1)(b) of the CRR provides that the model used by the institution
must “calculate the exposure value for the netting set at each of the future
dates on the basis of the joint changes in the market variables”. It is not
specific about excluding any exposure simulation for some transactions
outside the standard joint Monte Carlo simulations.

8. The CRR does not explicitly establish a requirement regarding how to handle
netting sets in cases where transactions which the institution has general
approval to treat with the IMM need to be carved out from the IMM to a non-
IMM method for any reason.

2.2 Principles for ECB banking supervision

9. With regard to the coverage of the IMM, institutions should comply with
Section II, Chapter 3, paragraph 9 of the ECB Guide on options and discretions
available in Union law, where the IMM coverage mentioned covers transactions
treated under the method described in Part Three, Title II, Chapter 6, Section 6
of the CRR. Transactions treated under any non-IMM method are, however,

ECB guide to internal models – Counterparty credit risk 188

 excluded. In particular, transactions which are carved out from the IMM are
excluded from the IMM coverage.

10. For cases where, for a given legally enforceable netting agreement as defined
in Part Three, Title II, Chapter 6, Section 7 of the CRR, one part of the
transactions is treated under the method described in Section 6 (IMM) and
another part is covered by one of the non-IMM methods 5, the ECB considers,
as a best practice, the creation of different synthetic netting sets, one per
method. Hence, one synthetic netting set covers all the transactions under the
IMM and the other synthetic netting sets cover all the transactions under each
non-IMM method (one per non-IMM method).

11. It is the ECB’s understanding that synthetic netting sets created for the
purposes described in paragraph 10 should cover only transactions under the
same contractual netting agreement; that is, Article 273(6) of the CRR (netting
set-specific application of any CCR method) is understood to apply also to
synthetic netting sets.

12. In relation to the requirement provided for by Article 294(1)(l) of the CRR 6 and
in accordance with Article 294(1)(d) of the CRR, it is the ECB’s understanding
that institutions should implement a framework that enables them to identify
significant pricing model deficiencies at transaction level. It is seen as good
practice to apply at least the following filter criteria to ensure the consistent
identification of such deficiencies (in accordance with Article 294(1)(o) of the
CRR) when comparing the IMM transaction’s 𝑡𝑡0 value and the respective
benchmark value:

(a) a threshold based on the absolute price difference;

(b) a threshold for differences expressed as a percentage of the notional

amount;

(c) a threshold for differences expressed as a percentage of the absolute

value of the respective benchmark.

The institution should be able to justify the setting of the above filter criteria,
which should be regularly validated and defined so that unacceptable model
performance as set out in Article 293(4) of the CRR can be assessed,
especially for pricing.

13. The ECB considers that appropriate measures to address identified model
weaknesses as referred to in the above assessment are as follows.

(a) A carve-out of transactions to one of the non-IMM methods, together with

the creation of synthetic netting sets to remedy unacceptable performance

5
This implies that not all transactions covered by the contractual netting agreement are treated under
the IMM.
6
See paragraph 69 with regard to how to detect value differences of transactions between the IMM and
the benchmarking system.

ECB guide to internal models – Counterparty credit risk 189

 of the CCR exposure model in accordance with Article 293(4) in
conjunction with Article 294(1)(d) of the CRR. This is proposed provided
that one of the identified price differences referred to in paragraph 12 is
observed for longer than the number of business days that is pre-defined
by the institution for this case. The ECB sees it as good practice that this
number is limited to ten business days during the reference quarter.

(b) Measures other than carve-outs and the creation of synthetic netting sets
that could be applied to address model deficiencies, provided that these
other measures (i) can be justified, (ii) are regularly validated, and (iii)
meet the purpose set out in Article 294(1)(d) of the CRR of not
systematically underestimating exposure, in conjunction with the purpose
of Article 293(4) of the CRR of identifying and remedying unacceptable
exposure model performance.

This includes, in the ECB’s understanding as further explained in paragraph 69,

that institutions should take all necessary remediation actions to solve the root
causes creating the most significant differences between the values of pricing
functions used for revaluation under the IMM and the respective benchmarking
value in a timely manner.

14. The ECB may see it as a violation of Article 292(1) CRR if the price differences
as identified in accordance with paragraph 12 (i) are persistent, (ii) do not lead
to remediation of model deficiencies and (iii) lead to a systematic
underestimation of exposure, In such cases, the ECB may consider supervisory
measures regarding affected transaction types.

15. In the ECB’s understanding, transactions carved out as a result, for example, of
price differences with a benchmarking system should not be considered as
contributing to the required IMM coverage explained in Section II, Chapter 3,
paragraph 9 of the ECB Guide on options and discretions available in Union
law.

16. For all transactions covered by and remaining in the IMM, it is the ECB’s
understanding that the differences between the 𝑡𝑡0 transaction values and the
benchmarks as referred to in Article 294(1)(l) of the CRR should be taken into
account. The ECB regards the following corrections as appropriate measures to
remedy unacceptable performance of exposure calculation as a result of pricing
model deficiencies in accordance with Article 293(4) of the CRR:

(a) At 𝑡𝑡0 , IMM transaction values are adjusted to match the respective
benchmark value.

(b) At future grid points, an adjustment to the modelling of the transaction’s

future values is applied in such a way that the EEPE of the netting set after
correction is not lower than that without any correction. Rather than using
the observed difference at 𝑡𝑡0 , the correction could be estimated using
more sophisticated methods, taking future market scenarios and
amortising transactions into account.

ECB guide to internal models – Counterparty credit risk 190

 The corrections should be regularly validated (see paragraph 69(b)).

17. The ECB considers it best practice to perform a full simulation in accordance
with paragraph 5 for all IMM transactions to directly comply with the
requirements of Article 284(1) of the CRR. In cases where this practice is not
feasible, the ECB considers that the following approach would still be compliant
with the CRR:

(a) if other (approximate) pricing methods are used, they should be subject to
the validation requirements described in paragraphs 69 and 71(a) to 71(c);

(b) if any alternative way to calculate exposures 7 is used, then the following
points should be taken into account along with the validation requirements
described in paragraphs 69 and 71(d).

(i) The institution should be able to demonstrate that the sole reason for
using this exposure calculation method is pricing performance, or a
performance issue related to calibrating certain transaction-specific
risk factors. For example, including these transactions in a full
simulation in accordance with paragraph 5 would delay regulatory
reporting by more than one business day.

(ii) Correlations with the other risk factors simulated in the CCR
exposure model should be taken into account when calculating or
calibrating such exposures. This would also hold in the case of new
or aggregated risk factors only used for this exposure calculation
method.

(iii) The underlying risk factor simulation should account for the exposure
time dependency, in particular for margined trading regarding the time
grid point to which the margin period of risk (MPOR) is attached.

(iv) This exposure calculation method should account for potential trade-
related cash flows (CFs) during the MPOR, either directly or in such a
way as to avoid systematic underestimation of the exposure.

(v) Pricing functions used for the purpose of calculating or calibrating the
current exposure of affected transactions should be an explicit part of
the IMM framework and governance.

(vi) The sum of the absolute 𝑡𝑡0 values of these transactions is below
20% 8 of the total sum of absolute 𝑡𝑡0 values from all transactions
covered by the IMM.

7
Examples could be scenario-independent, pre-defined time profiles per transaction starting at the 𝑡𝑡0
value, or scenario-independent, pre-defined value increases per transaction during MPOR, or new risk
factors aggregated from those used in Article 284(1)(a) of the CRR with an own stochastic process.
8
This percentage may decrease in the future.

ECB guide to internal models – Counterparty credit risk 191

3 Margin period of risk and cash flows
18. This section refers to the modelling of the MPOR, 9 including the following
aspects.

(a) Treatment of margin call and trade-related CFs in all currencies. The
trade-related CFs include here both intermediary flows and the settlements
at maturity related to trades, as well as flows in the form of a commodity or
precious metal or any other asset that may be paid/received during the
MPOR. Trade-related CFs paid by the institution to the counterparty result
in upward jumps of the exposure time profile (hereinafter called “spikes”).

(b) Taking the default management process (DMP) into account when
modelling CFs paid/received during the MPOR. The DMP refers to all legal
and operational actions performed by the institution upon counterparty
default before the institution stops paying margin call and trade-related
CFs to the defaulted counterparty.

(c) Interpolation techniques that may be applied to estimate the netting set
market value at MPOR time points that do not belong to the simulation
time grid used.

(d) Mapping between each time grid point t, for which EE(t) is calculated, and
the associated MPOR. 10

3.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 272 (9)

284 (1), (4)

285 (2), (3), (4), (5)

289 (5)

292 (1)(a), (b)

294 (1)(g), (i)

19. The regulatory provisions relating to the topic addressed in this section that
require further guidance are the following.

(a) Article 292(1)(a) of the CRR, which requires the model to reflect
transaction terms and specifications in a timely, complete and conservative
fashion, but does not make explicit mention of trade-related CFs.

9
Note that the modelling of collateral is addressed in Section 4.
10
Due to the small distance between the adjacent grid points (t), MPORs related to the two adjacent grid
points may overlap.

ECB guide to internal models – Counterparty credit risk 192

 (b) Article 292(1)(b) of the CRR, which requires that netting agreements
(including actions upon counterparty default or outstanding payments of
the counterparty as part of netting arrangements) be reflected.

(c) Article 289(5) of the CRR, which notably provides that an “institution shall
estimate EE along a time profile of forecasting horizons that adequately
reflects the time structure of future CFs and maturity of the contracts and
in a manner that is consistent with the materiality and composition of the
exposures”. However, modelling within the MPOR is not explicitly
mentioned.

(d) Article 272(9) of the CRR, which provides a definition of the MPOR:
“‘margin period of risk’ means the time period from the most recent
exchange of collateral covering a netting set of transactions with a
defaulting counterparty until the transactions are closed out and the
resulting market risk is re-hedged”. However, this definition does not
mention the trade-related CFs.

(e) Article 284(4) of the CRR, which specifies how to use the alpha parameter,
mentioning that competent authorities may require a higher one than 1.4.

(f) Article 294(1)(g) of the CRR, which notably requires the validation of
transaction-specific information to capture the effects of margining in the
model, and Article 294(1)(i) of the CRR, which requires the testing of key
assumptions of the CCR exposure model, without mentioning explicitly
advanced features in MPOR and CF modelling, such as the use of
“Brownian Bridge”-based interpolation for additional time grid points in the
MPOR.

(g) According to Article 284(1) of the CRR, the exposure value needs to be
calculated “on the basis of joint changes in relevant market variables”. It
does not give explicit details regarding the starting point in time of these
changes.

(h) Article 285(2) to (5) of the CRR sets the length of the MPOR. There is no
special provision for its length if the MPOR is attached to (i) time grid
points t after 𝑡𝑡0 but before 𝑡𝑡0 plus the MPOR length or (ii) time grid points
at the end of the exposure time axis with t plus MPOR being later than the
one-year horizon or later than the final maturity of the netting set.

(i) According to Article 284(4) of the CRR, the “model shall estimate EE at a
series of future dates t1, t2, t3, etc.” The article does not specify for
margined trading in which way an MPOR needs to be attached to these
future dates, in particular concerning attachments close to 𝑡𝑡0 and close to
the one-year future date or the final maturity of the netting set.

ECB guide to internal models – Counterparty credit risk 193

3.2 Principles for ECB banking supervision

20. The requirements of Articles 292(1)(a) and 289(5) of the CRR are also seen as
being applicable to the modelling of exposure changes of margined trading
within the MPOR. The term “margin arrangement”, as mentioned in
Article 292(1)(b) of the CRR, is understood as comprising all contractual
features, the margining mechanism with margin call triggers, grace periods and
close-out provisions, which, according to Article 292(1)(a) of the CRR, must be
reflected in the model.

(a) In the view of the ECB, with regard to the modelling of margin call and
trade-related CFs within the MPOR, Article 272(9) of the CRR should be
understood as requiring that none of these CFs be received from the
counterparty after the beginning of the MPOR. An institution may receive
trade-related CFs after the beginning of the MPOR only if it can justify that
its assumptions are consistent with its modelling of default time within the
MPOR, its DMP and its assumptions regarding non-payment of CFs. 11

(b) Furthermore, the counterparty is supposed to default at some time point

during the MPOR, and non-payment of trade-related CFs to the defaulting
counterparty may be assumed to the extent that this assumption is
consistent with:

(i) the DMP and the features of enforceable settlement mechanisms

(e.g. agreements to net CFs with related margin calls or analogues to
the Continuous Linked Settlement system);

(ii) the grace period and close-out requirements specified in the netting
agreement, and in particular how the close-out is affected by paid or
non-paid CFs.

It is seen as good practice and cautious modelling (for example, given that
watchlists of critical counterparties include only a subset of all potentially
critical counterparties) that trade-related CFs from the institution to the
counterparty that are due according to the underlying contract are
assumed to be paid at least for a time period after the beginning of the
MPOR corresponding to the re-margining period.

(c) If the institution has no defined DMP or the DMP is not taken into account
in the modelling, all trade-related CFs due by the institution should be
assumed to be paid to the counterparty during the whole MPOR.

(d) Assuming that there are documented and enforceable settlement netting
rules, the aggregation of netting set CFs with opposite signs falling due on
the same date from different legs of the same transactions and/or from
other transactions in the netting set could be integrated into the modelling

11
At the very least, while CFs are still modelled to be received from the counterparty, it should be
assumed that CFs are also paid to the counterparty.

ECB guide to internal models – Counterparty credit risk 194

 of CFs within the MPOR. The resulting net CF should be treated in
accordance with points 20(a) and 20(b).

(e) A modelling different from the expected modelling described above

showing discrepancies with the DMP could be accepted if it is shown that
the quantitative impact of this approach on the EEPE is not material.

21. If (i) an institution does not comply with the requirements of Articles 292(1)(a)
and 289(5) of the CRR as explained in sub-paragraphs (a)-(d) above and (ii)
there is a material impact as referred to in paragraph (e), the ECB has the
power to impose an appropriate and proportionate supervisory remediation
measure, which can consist – as provided for by Article 284(4) of the CRR – in
an increase of the alpha parameter.

22. MPOR modelling may require the estimation of netting set market values at
time points that do not belong to the simulation time grid. It is the ECB’s
understanding of Article 294(1)(g) of the CRR that interpolation/extrapolation
techniques 12 used by institutions to perform such estimations should be
validated by studies showing that impacts on the EEPE, compared with full
revaluation, are not material.

23. In the view of the ECB, backward 13 and forward 14 modelling of the MPOR
setting, as well as a mix of both 15, can be considered CRR-compliant. In
particular the following holds.

(a) Backward modelling approach:

For time grid points t falling within the interval [𝑡𝑡0 , 𝑡𝑡0 + 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀], institutions
should calculate expected exposure 𝐸𝐸𝐸𝐸(𝑡𝑡)) as required in Article 284(4) of
the CRR by modelling joint changes in relevant market variables
mentioned in Article 284(1)(b) of the CRR starting from 𝑡𝑡0 , since
Article 284(5) of the CRR defines this date as the earliest date for the
calculation of exposure.

(b) Forward modelling approach:

When using the forward modelling approach, institutions should calculate
expected exposure 𝐸𝐸𝐸𝐸(𝑡𝑡)) as required by Article 284(4) of the CRR by
taking into account close-out amounts that are determined after 𝑡𝑡 within
the MPOR period as given by Article 285(2) to (5) of the CRR. This also
applies when 𝑡𝑡 ∈ [1𝑦𝑦 − 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀, 1𝑦𝑦], i.e. for such a 𝑡𝑡, 𝑡𝑡 + 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 ≥ 1𝑦𝑦 holds.
Furthermore, if 𝑡𝑡 equals the maturity (𝑇𝑇), of the longest-lasting transaction
in the netting set, and if no collateral is modelled as held by the institution

12
For example, a Brownian Bridge-based interpolation.
13
In backward modelling, 𝐸𝐸𝐸𝐸(𝑡𝑡) is calculated on the basis of the evolution of exposure (as a result of the
evolution of transaction and collateral values) in the time interval [𝑡𝑡 − 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀, 𝑡𝑡], where MPOR denotes
the time length of the MPOR.
14
In forward modelling, 𝐸𝐸𝐸𝐸(𝑡𝑡) is calculated on the basis of the evolution of exposure in the time interval
[𝑡𝑡, 𝑡𝑡 + 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀]. In this approach, the MPOR starts at t and ends at 𝑡𝑡 + 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀.
15
This includes variants of attaching the MPOR to the 𝑡𝑡 of 𝐸𝐸𝐸𝐸(𝑡𝑡), where the 𝑡𝑡 is not at the border of the
time interval set by the MPOR.

ECB guide to internal models – Counterparty credit risk 195

 at 𝑇𝑇 for a given scenario, the effective length of the MPOR may shorten, as
no close-out or re-hedging is due after maturity of the last transaction in
the netting set.

The ECB understands that the effective length of the MPOR for these grid
points may be shortened and considers that this will not affect the formal length
of the MPOR as provided for by Article 285(2) to (5) of the CRR.

4 Collateral modelling
24. This section deals with the modelling of cash and non-cash margin collateral,
that is, its potential value changes from the time when the last margin call at the
beginning of the MPOR is settled up to the end of the MPOR. Initial margin (IM)
modelling is addressed in Section 5.

4.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 223, 224, 225, 226, 227

285 (1), (6), (7)

292 (1)

25. The regulatory provisions relating to the topic addressed in this section that
require further guidance are the following.

(a) Article 292(1) of the CRR requires an institution to ensure the integrity of
its modelling process by reflecting, among other things, transaction terms
and specifications, which also include margining arrangements. However,
it does not further specify how the future collateral composition or a
reflection of the actual collateral balance in the IMM’s estimated exposure
for a netting set should be reflected in the IMM’s assumptions.

(b) Article 285(1), (6) and (7) of the CRR clarifies the modelling options under
which the effects of margining can be directly recognised in the exposure
value calculation.

(i) However, these provisions do not determine whether institutions

should model margin collateral in a manner consistent with the
modelling of securities underlying over-the-counter (OTC) derivatives
and securities financing transactions (SFTs) 16, or whether they are
allowed to adopt a different modelling approach for margin collateral,
on the one hand, and securities underlying the different transactions,
on the other.

16
See the definition in the Counterparty credit risk glossary.

ECB guide to internal models – Counterparty credit risk 196

 (ii) The provisions of Article 285(6) of the CRR leave room for
interpretation regarding the term “jointly modelled”.

(iii) The wording of Article 285(7) 17 of the CRR leads to the conclusion
that this article provides an exemption to Article 285(6) of the CRR in
cases where an institution is not able to model collateral jointly with
the exposure. In this case, and in accordance with Article 285(7) of
the CRR, the institution is allowed to use volatility adjustments to
recognise the effects of margining in the exposure calculation directly
such that the institution does not have to apply one of the EEPE
calculation measures presented in Article 285(1)(a) or (b) of the CRR.

(iv) In addition, Article 285(1), (6) and (7) of the CRR is not specific about
whether a combination of the two options (use of volatility
adjustments and joint modelling) to account for margining effects is
possible. Thus, it is not clear if it is possible to use volatility
adjustments in line with Article 285(7) of the CRR, together with the
jointly modelled risk factors in accordance with Article 285(6) of the
CRR, for the collateral modelling in cases where, for some risk
factors, the institution is able to model some collateral components
jointly.

(c) Article 285(7) of the CRR refers to the standards of the Financial Collateral
Comprehensive Method (as set out in Article 223 of the CRR, which refers
to Articles 224 to 227 of the CRR) in cases where an institution wants to
make use of volatility adjustments to recognise the effect of margining
directly in its exposure calculation. Article 223(1) third sub-paragraph of
the CRR requires institutions to apply a volatility adjustment to reflect
mismatches between the collateral currency and the settlement currency
for OTC derivative transactions covered by recognised netting
agreements. However, Article 223(1) third sub-paragraph of the CRR
neither defines exactly what the settlement currency is, nor its relationship
with the currency in which the exposure is denominated in the context of
netting sets with attached margin agreements. In addition, the case of
these currencies being different from the reporting currency is not treated
explicitly.

4.2 Principles for ECB banking supervision

26. The ECB sees as a best practice that non-cash margin collateral is treated in a
manner that is consistent with the modelling of securities underlying OTC or
SFT transactions, provided that these transactions are within the scope of the
IMM. For example, if a certain type of security is fully simulated (or if a volatility

17
“If an institution is not able to model collateral jointly with the exposure [in accordance with
Article 285(6) of the CRR] it shall not recognise […] the effect of collateral […], unless it uses […]
volatility adjustments […].”

ECB guide to internal models – Counterparty credit risk 197

 adjustment is applied) in the security leg of an SFT, then it should also be fully
simulated (or a volatility adjustment should also be applied) if it occurs as
margin collateral within the IMM. In the case of inconsistent treatment, the ECB
sees it as beneficial that the institution is able to justify this choice and to
demonstrate that (i) its approach does not systematically underestimate
exposures and (ii) the quantitative impact on the final EEPE is not material.

27. In order to comply with the requirements laid down by Article 292(1)(a) and (b)
of the CRR with respect to the terms of margining and netting arrangements,
the ECB is of the view that the future composition of the collateral pool over the
lifetime of the netting set should reflect one or more of the following:

(a) the contractual arrangements in terms of eligible margin collateral;

(b) the institution’s policy as regards posted collateral types 18 being eligible;

(c) the composition observed historically; 19

(d) at least the current composition of the margin collateral pool.

This holds for the same or similar characteristics of these collateral types.

28. When a contractual margin agreement contains transactions treated under both
the IMM and a non-IMM method and therefore the contractual netting set is split
into different synthetic netting sets, and if collateral modelling uses the actual
collateral balance at 𝑡𝑡0 , the ECB considers that the actual margin collateral
should be assigned to the synthetic netting sets. This should be done in a way
that also reflects their respective current exposures, as defined in
Article 272(17) of the CRR, and does not lead to double-counting of collateral. If
the institution chooses a different approach (e.g. a full assignment of collateral
to only one synthetic netting set), it should be able to justify this choice and
demonstrate that its methodology does not systematically underestimate the
resulting exposure values.

29. The ECB considers the use of the “model-estimated collateral balance at 𝑡𝑡0 ” 20
to be compliant with Article 292(1)(b) of the CRR, if the resulting modelled
collateral balance is regularly benchmarked against the actual collateral
balance at 𝑡𝑡0 in accordance with Article 292(1)(g) of the CRR. In this case,
validated but still relevant differences between model-estimated and actual 𝑡𝑡0
collateral balances should be taken into account in the modelling of 𝑡𝑡0 so that
the exposure value is not systematically underestimated. Transactions that are

18
This requires the institution to have a clear and well-documented policy further limiting the contractually
eligible collateral to certain types of posted collateral.
19
This includes the use of a historically observed composition of collateral for counterparties with a
comparable behaviour in the case of new agreements, i.e. agreements with new counterparties without
their own history in the institution.
20
That is, when the 𝑡𝑡0 collateral balance is estimated as a function of the calculated netting set value as
of 𝑡𝑡0 – using IMM pricing functions and modelled features of the margin agreement – and is not set as
being equal to the actual collateral balance.

ECB guide to internal models – Counterparty credit risk 198

 potentially carved out from the IMM as well as collateral potentially not yet
settled at 𝑡𝑡0 should also be taken into account.

30. It can be inferred from Article 285(6) in conjunction with Article 285(7) of the
CRR that, in order to directly capture the effects of margining in the calculation
of exposure values, an institution can use either of the following:

(a) the option of joint modelling (Article 285(6) of the CRR) for the modelling of
all collateral;

(b) the volatility adjustment option (Article 285(7) of the CRR) for the
modelling of all collateral.

31. The ECB is of the view that using both options at the same time would only be
compliant with the above CRR articles if volatility adjustments for non-cash
collateral were used and the joint modelling for the treatment of FX risk were
applied in the collateral modelling only. In this context, the ECB considers it to
be good practice that the above combination can only be made by using jointly
modelled FX rates for all currencies that are simulated for the exposure
calculation under the IMM. In other words, the ECB would not consider a partial
application of FX volatility adjustments alongside jointly modelled FX rates for
the purpose of collateral modelling as being consistent.

32. In the context of Article 285(6) of the CRR, the ECB understands the provision
“an institution shall model collateral … jointly with the exposure in its exposure
value calculation” as requiring model integrity in accordance with
Article 292(1)(a) and (b) of the CRR. This means in particular that:

(a) the collateral value changes over time and during the MPOR are
determined by using the same model as for the calculation of the
transactions’ value changes;

(b) the use of the same model refers to the IMM’s general modelling features
(including simulated and non-simulated risk factors, the dependency
structure, pricing functions, etc.), which should be used for both the
calculation of the transactions’ value changes and the calculation of
collateral value changes applying the same generated scenarios;

(c) if some risk factors are not required for the calculation of the transactions’
value changes and are only used for the collateral modelling, these risk
factors should be modelled consistently with those for derivatives and
SFTs within the scope of the IMM, also regarding the dependency
structure.

33. Article 285(7) of the CRR provides that, if an institution is not able to model
collateral jointly with the exposure, it may use volatility adjustments to recognise
the effects of margining on the exposure itself, provided the institution complies
with the requirements of the Financial Collateral Comprehensive Method as per
Article 223 of the CRR. If an institution needs to apply volatility adjustments to
recognise the effects of margining on the exposure itself, these adjustments

ECB guide to internal models – Counterparty credit risk 199

 must be applied to reflect currency mismatches in accordance with
Article 223(1) third sub-paragraph of the CRR. In the light of Article 220(2)(d) of
the CRR, the ECB sees the identification of the currency that is potentially
different from the collateral currency as described below as best practice:

(a) when Article 223(1) second sub-paragraph of the CRR uses the term “the
currency in which the underlying exposure of the netting set is
denominated”, and when Article 223(1) third sub-paragraph of the CRR
(for OTC derivative transactions only) uses the term “settlement currency”,
it is the currency as determined in (b);

(b) it is the currency:

(i) agreed in the individual derivative contract if no netting has been

agreed upon; or

(ii) of the relevant governing master netting agreement if agreed without

a credit support annex; or

(iii) of the relevant credit support annex, if agreed; or

(iv) of the close-out amount if more than one credit support annex has
been defined for one master netting agreement.

34. In the view of the ECB, in order to comply with Article 292(1)(a) and (b) of the
CRR for both unmargined and margined cases, and with Article 292(1)(a) and
(b) in conjunction with Article 285(6) and (7) for margined cases, institutions
should take into account the potential FX risk arising from currency
mismatches. The potential mismatches are between (i) any of the various
currencies of the exposure components (e.g. various transactions with different
currencies, collateral types with different currencies) and (ii) the currency in
which the netting set’s total exposure is determined in the simulation (e.g. the
currency of the governing master agreement as explained in paragraph 33).
The ECB considers that potential FX risk is treated in compliance with
Article 292(1)(a) and (b) of the CRR by applying either of the following:

(a) simulation of FX rates for all exposure components at all relevant points in
time;

(b) FX volatility adjustments in accordance with Article 223(1) of the CRR

when making use of Article 285(7) of the CRR.

35. It is the understanding of the ECB that any potentially remaining FX risk that
arises from currency mismatches between (i) the currency in which the netting
set’s exposure is determined in the simulation and (ii) the reporting currency
should be taken into account in the institutions’ modelling process. The ECB
considers the treatment of this potentially remaining FX risk to be compliant with

ECB guide to internal models – Counterparty credit risk 200

 Article 292(1)(a) and (b) of the CRR if institutions apply the simulated FX rates
at the end of the MPOR. 21

5 Modelling of initial margin

36. IM is already applied in central clearing and currently carries over to bilateral
OTC agreements. The modelling issue with respect to CCR is that the IM
depends on the risk profile of the future netting set in terms of the levels and
volatility of simulated market risk factors and on transactions still alive, i.e. it is a
variable agreement parameter.

To be clear, it should be specified that “IM modelling” refers here to the

modelling of IM under the IMM and not, for example, to the implementation of
European Market Infrastructure Regulation (EMIR) 22 requirements in the
institutions’ collateral management in terms of calculating an appropriate level
of IM.

5.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 292 (1)(b) and (g)

293 (1)(b)

37. The regulatory provisions relating to the topic addressed in this section that
require further guidance are the following:

(a) Article 292(1)(b) of the CRR, which requires institutions to include, among
other transaction terms, margining and netting arrangements in the model;
Article 292(1)(g) of the CRR, which requires processes for formal
reconciliation between the model and source data systems;

(b) Article 293(1)(b) of the CRR, which notably requires the comparison of risk
measures generated by the model with realised risk measures.

5.2 Principles for ECB banking supervision

38. In relation to the requirements set out in Article 292(1)(b) of the CRR, and for
agreements subject to IM where both the transaction exposures and IM are

21
The additional use of simulated FX rates applies in particular to those cases where the currency as per
Article 223(1) of the CRR differs from the reporting currency, but also to cases of joint modelling where,
for example, the netting set’s currency and the collateral currency are different from each other and
from the reporting currency.
22
Regulation (EU) No 648/2012 of the European Parliament and of the Council of 4 July 2012 on OTC
derivatives, central counterparties and trade repositories.

ECB guide to internal models – Counterparty credit risk 201

 within the scope of the IMM, the ECB considers it good practice that institutions
have an IM modelling that adequately reflects contractual arrangements for the
respective netting set. 23 In particular, if contractual arrangements provide that
the IM should reflect forward variability and maturing transactions 24 inside
netting set value changes relevant for the contractual IM method, the IMM
modelling of the IM should take this feature into account unless the institution
demonstrates that its choice for IM modelling inside the IMM does not
systematically underestimate exposures.

39. The ECB considers that, in order to avoid the risk of non-compliance with
Articles 292(1)(b) and 293(1)(b) of the CRR, the level of the modelled IM at 𝑡𝑡0
should be benchmarked on a regular basis against the respective real margin at
𝑡𝑡0 in accordance with Article 292(1)(g) of the CRR. Validated but still relevant
differences should be taken into account in the modelling (e.g. by using some
corrective exposure level add-on if the modelling is not risk sensitive and/or
may lead to non-conservative exposures).

6 Maturity
40. This section refers to the estimation of the parameter M used in the calculation
of the risk weight for counterparties, towards which the institution has an IMM
exposure and for which the institution uses the IRB approach.

41. The section also refers to the treatment of contingent transaction maturities,
especially where there are early termination clauses (ETCs, also called break
clauses) for derivatives and SFTs, and to different CRR interpretations.

Note that transaction maturities (and their changes) affect (i) the M parameter of
Article 162 of the CRR, (ii) the shape of the EE(t) time profile, and (iii) the
maximum transaction maturity relevant for Article 284(6) of the CRR, where
(i) affects the calculation of risk-weighted assets (RWAs) for IRB institutions and
(ii) and (iii) affect the calculation of the EEPE and then (via the exposure value)
also RWAs.

23
In cases where IM agreements include discretionary or undisclosed elements, institutions are still
expected to consider all contractual arrangements to the extent possible when modelling the IM within
the IMM, potentially also taking the history of observed IMs into account. This includes potential
information and assumptions for past IM amounts.
24
A contractual IM specification includes IM changes resulting from newly contracted transactions,
whereas the IMM modelling starts with the portfolio at 𝑡𝑡0 and then has a “melting down” portfolio, since
the effect of new trades is modelled only through the effective EPE.

ECB guide to internal models – Counterparty credit risk 202

6.1 Relevant regulatory references
Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 162

284 (4), (6)

42. The regulatory provisions relating to the topic addressed in this section that
require further guidance are the following:

Article 162 (defining the maturity parameter M) and Article 284(4) 25 and (6)
(defining the remaining transaction maturity) of the CRR. Contingent transaction
maturities and contractual arrangements for early termination are not mentioned
in these articles.

6.2 Principles for ECB banking supervision

43. Article 162 of the CRR will be understood as outlined below for the exclusive
purpose of applying the IMM as specified by Part Three, Title II, Chapter 6,
Section 6 of the CRR. In the ECB’s understanding, the article should apply in
the following way:

(a) paragraph (2)(b) should apply to unmargined derivatives subject to a

master netting agreement if the longest-dated contract in the netting set
has a maturity of less than or equal to one year;

(b) paragraph (2)(c) should apply to fully or nearly fully collateralised

derivatives or fully or nearly fully collateralised margin lending transactions
subject to a master netting agreement if the longest-dated contract in the
netting set has a maturity of less than or equal to one year;

(c) paragraph (2)(d) should apply to unmargined and margined SFTs

(excluding margin lending transactions) subject to a master netting
agreement if the longest-dated contract in the netting set has a maturity of
less than or equal to one year;

(d) paragraph (2)(g) should apply to all transactions not subject to a master
netting agreement, on the one hand, and to multiple transaction netting
sets where the maturity of the transaction or the longest-dated contract
within the netting set is greater than one year, on the other hand, unless
the conditions for applying paragraph (2)(i) are satisfied;

(e) if the conditions of paragraph (2)(i) are fulfilled, setting M to one year
should apply only for those transactions or netting sets where the

25
Article 284(4) of the CRR stipulates: “The model shall estimate EE at a series of future dates t1, t2, t3,
etc.”

ECB guide to internal models – Counterparty credit risk 203

 application of paragraphs (2)(a), (b), (c), (d), (f) or (g) would result in M
being greater than one year;

(f) the provisions of paragraph (3) regarding the floor value for M should be
used in the following way: when applying paragraph (2)(c) or (2)(d) for the
purpose of estimating M for a given netting set, an institution should be
allowed to use the floor value provided by paragraph (3) (one business
day), if all paragraph (3) requirements are fulfilled.

44. The CRR is silent on the transaction maturity that should be considered for both
the calculation of the EEPE and the calculation of the M parameter in the case
of open term repos or, in general, SFTs without an explicitly fixed maturity.

(a) If the institution has the right to terminate the transaction, in the ECB’s
view the transaction maturity should be set at the higher of:

(i) the contractually agreed first date on which the transaction can be
terminated;

(ii) the applicable MPOR.

(b) If the institution does not have the right to terminate the transaction, the
ECB considers that the transaction maturity should be given by the longest
past lifetime of transactions with the same or comparable counterparties,
subject to a five-year cap.

45. For derivatives with ETCs:

(a) it is seen by the ECB as best practice that non-mandatory ETCs are not
used for the calculation of EE as used in Article 284(4) and (6) of the CRR,
and of M, as provided for in Article 162(2) of the CRR (this article aims to
deal only with a non-contingent maturity);

(b) the ECB also considers it as compliant with Article 162(2) and (3) of the
CRR to use mandatory ETCs for the calculation of EE and M instead of
the contractual maturity, because there cannot be any positive exposure
after that date due to the provisions of Article 284(4) of the CRR.

46. Internal analyses by the institutions should be able to justify, as the case may
be, choices of M values shorter than:

• five business days for netting sets consisting only of SFTs;

• ten business days for all other netting sets, including the derivative
instruments listed in Annex II of the CRR.

In accordance with Article 162(3) of the CRR, provisions for prompt liquidation
need to be in place in order to use short M values. If these provisions are
different for the M parameter referred to in Article 162 of the CRR than in the
IMM exposure modelling of margined trading, the appropriateness of shorter
close-out periods would also have to be demonstrated.

ECB guide to internal models – Counterparty credit risk 204

 47. The ECB sees it as best practice when the estimation of the maturity of
physically settled options on derivatives (e.g. swaptions, used for the purpose of
calculating the parameter M and for the calculation of the EEPE) is made on the
basis of the maturity date of the underlying derivatives (e.g. the swap underlying
the swaption), unless the institution is able to justify the use of a different
maturity for specific products. 26

7 Granularity, number of time steps and scenarios

48. This section refers to the chosen time grid for the future exposure calculation
and the number of scenarios generated. More specifically:

(a) the number and density of time grid points have an impact on the accuracy
of EE profiles and thus also on the accuracy of the EEPE;

(b) the number of scenarios and the type of random number generator
determine the numerical accuracy of the calculations and thus the
statistical error of expected exposures.

7.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 284 (4), (5), (6)

292 (1)(a), (b)

293 (1)(a), (c)

294 (1)

368 (1)(f)

49. The regulatory provisions relating to the topic addressed in this section that
require further guidance are the following.

(a) In accordance with Article 292(1)(a) of the CRR, an institution must ensure
that the model reflects transaction terms and specifications in a timely,
complete and conservative fashion. Article 292(1)(b) of the CRR specifies
that these terms must include at least the contract notional amounts,
maturity, reference assets, margining arrangements and netting
arrangements. However, it stays silent on the number of grid points
necessary to take CFs resulting from these terms into account.

(b) In accordance with Article 284(4) of the CRR, EE must be calculated for
time grid points 𝑡𝑡𝑖𝑖 = 1, 2, 3, … The output, 𝐸𝐸𝐸𝐸(𝑡𝑡𝑖𝑖 ), is used in the EEPE

26
One example is when an underlying swap becomes subject to central clearing and thus the
counterparty changes.

ECB guide to internal models – Counterparty credit risk 205

 calculations (Article 284(5) and (6) of the CRR). However, there is no
specific requirement as to how to set these 𝑡𝑡𝑖𝑖 values.

(c) Article 293(1)(c) of the CRR requires an institution to carry out initial and
ongoing validation of its CCR exposure model, while Article 294(1) states
the requirements that need to be met by the institution’s validation
programme. While there is no explicit requirement regarding the number of
scenarios in Article 294 of the CRR, Article 368(1)(f) of the CRR (which is
included in the reference to Part Three, Title IV, Chapter 5 of the CRR
made by Article 293(1)(a) of the CRR) requires the internal model to have
a proven track record of “reasonable accuracy” in measuring risks.

7.2 Principles for ECB banking supervision

50. Since the modelling process has to reflect transaction terms, as required by
Article 292(1)(a) and (b) of the CRR, in the understanding of the ECB the
density and location of grid points as defined in Article 284(4) of the CRR
should capture intermediate and final trade-related CFs depending on notional
amounts, maturities, etc. that influence the shape of the exposure profile. The
ECB also considers that, if the EEPE calculated with a very dense time grid 27 is
more than 5% above 28 the EEPE as calculated by the institution using its
standard set of grid points under the standard configuration of the IMM for the
whole portfolio, then the ECB can increase the alpha parameter following the
process described in Section 11. 29 Institutions can conduct this impact
assessment on representative sub-portfolios as defined in the Counterparty
credit risk glossary.

51. The ECB is of the view that, in order to fulfil the requirements set out in
Article 368(1)(f) of the CRR, the estimation and monitoring of the numerical
error of the EEPE due to the number of scenarios should be part of the regular
validation programme mentioned in Articles 293(1)(c) and 294(1)(d) of the CRR.
If the numerical error 30 is more than 5% of the EEPE for the whole portfolio, the
ECB can increase the alpha parameter following the process described in
Section 11. 31 Institutions can conduct this impact assessment on representative
sub-portfolios as defined in the Counterparty credit risk glossary.

27
The expression “very dense time grid” means here a daily grid, unless the institution can show that all
CFs are captured with a coarser time grid.
28
However, the impact of any numerical error (see paragraph 51) could also be considered in order to
avoid potential double counting if the institution can demonstrate a corresponding overlap.
29
For example, if the difference is more than 5%, the alpha parameter could be increased by at least
0.05, etc.
30
See the Annex for a description of how to derive the statistical error at a 95% confidence level.
31
For example, if the error is more than 5%, the alpha parameter could be increased by at least 0.05, etc.

ECB guide to internal models – Counterparty credit risk 206

8 Calibration frequency and stress calibration
52. The calibration frequency is relevant both for regulatory reporting and for
internal risk management (line consumption, etc.) as part of the use test
requirements set by Article 289 of the CRR:

(a) for Pillar 1 purposes, Article 292(2) of the CRR requires that the minimum
quarterly frequency is increased to reflect (important) changes in market
conditions;

(b) for internal risk management purposes, the calibration frequency also
affects the quality of exposure numbers used for the institution’s day-to-
day risk management process.

53. To compute own funds requirements for CCR, Article 284(3) of the CRR
requires that institutions use two different calibrations: one based on current
market data and one based on a stress period.

8.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013

284 (3)(b)

289 (1), (4), (5)

292 (2), (3), (4)

54. The regulatory provisions relating to the topic addressed in this section that
require further guidance are the following.

(a) Article 289(1) of the CRR requires among other things “that the distribution
of exposures generated by the model used to calculate effective EPE is
closely integrated into the day-to-day CCR management process of the
institution”, without further specifying the meaning of “closely integrated”.
Article 289(4) of the CRR requires that institutions measure and manage
current exposures; Article 289(5) requires them to have system capabilities
to estimate EE daily if necessary, unless they can demonstrate to the
satisfaction of their competent authorities that their exposures to CCR
warrant less frequent calculation.

(b) In accordance with Article 284(3)(b) of the CRR, institutions must compute
the EEPE using a stress calibration. This provision should be read in the
light of Article 292(2) of the CRR, which sets out the requirements of the
stress calibration, and Article 292(3) of the CRR, which sets out the
requirements for the stress period determination.

(c) In accordance with Article 292(4) of the CRR, the EPE model must use
data – implied or historical – that include the data from the stressed credit

ECB guide to internal models – Counterparty credit risk 207

 period and must use such data in a manner consistent with the method
used for the calibration of the EPE model to current data. It does not
further specify the meaning of “consistent”.

8.2 Principles for ECB banking supervision

55. The ECB considers that Article 289(1) of the CRR should be understood as
implying that the exposure distribution used for internal risk measurement in the
day-to-day CCR management process is sufficiently up-to-date for daily line
consumption calculations. Accordingly, the revaluation of current exposure 32 for
internal risk management purposes should also be performed on a daily basis
in compliance with Article 289(4) and (5) of the CRR. The ECB regards a
monthly or higher frequency for the recalibration of the parameters of the
underlying stochastic processes (such as drift, volatility and correlation) for
internal risk management as good practice to minimise the risk of non-
compliance with Article 292(2) and Article 289(5) of the CRR, since an outdated
calibration may no longer reflect market conditions or adequately reflect the
exposure profile.

56. In order to comply with Article 292(2) and (4) of the CRR, the following holds:

(a) The ECB considers as best practice that parameters underlying the
stochastic processes of the EPE model are calibrated with the data from
the identified stress period (i.e. with the exact three years of data defining
the stress period in the case of historical data) using the same estimation
method that is applied for the current calibration. This comprises, in
particular, the parameters needed for the simulation of market risk factors,
the pricing of transactions and collateral valuation. It is expected that the
adequacy of expert-set parameters for the identified stress period is
assessed and, if applicable, their values are adjusted accordingly.

(b) The ECB would accept an alternative stress calibration method for
parameters other than volatilities and correlations if the institution is able to
demonstrate that its approach is consistent with its current calibration (for
example, regarding the length of the calibration window or boundary
conditions that need to be satisfied) and does not systematically
underestimate exposures.

9 Validation
57. This section refers to the validation framework set up by institutions to assess
the performance of the CCR exposure model, in particular back-testing
methodologies, the validation of pricing functions and further checks on key
modelling assumptions.

32
Current exposure is the starting point at 𝑡𝑡0 for every EE time profile.

ECB guide to internal models – Counterparty credit risk 208

9.1 Relevant regulatory references
Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 287 (2)

292 (1)(a), (b), (g), (6)(a)

293 (1)(b), (c), (4)

(1)(c), (d), (e), (g), (h), (i), (j),

294
(l), (o)

58. The regulatory provisions relating to the topic addressed in this section that
require further guidance are as follows.

(a) Article 287(2) of the CRR states that the risk control unit is expected to be
responsible for the initial and ongoing validation of the model.
Furthermore, Article 293(1)(c) of the CRR provides that the validation and
review must be conducted independently of model development, which
needs to be reconciled with Article 287(2) of the CRR given that model
development is usually also done within the risk control unit.

(b) Article 293(4) of the CRR requires, among other things, that institutions
“maintain a written policy that describes the process by which
unacceptable performance will be identified and remedied”, without further
describing what constitutes unacceptable performance and what the
remedies might be.

(c) According to Article 292(6)(a) of the CRR, “an institution shall subject the
model to a validation process that specifies the kind of testing needed to
ensure model integrity and identify conditions under which the
assumptions underlying the model are inappropriate and therefore result in
an understatement of EPE”. However, the CRR does not further specify
which assumptions should form part of the validation process.

(d) Article 293(1)(b) of the CRR requires institutions to conduct “a regular

programme of back-testing”, but does not further specify the frequency of
the back-testing.

(e) Regarding the requirements laid out with respect to back-testing levels 33
and methodologies, Article 294(1)(c) of the CRR provides that “an
institution shall back-test the performance of its CCR exposure model and
the model’s relevant risk measures as well as market risk factor
predictions”, without mentioning any restrictions. Article 294(1)(h) of the
CRR requires the model validation process to “include static, historical
back-testing on representative counterparty portfolios that are actual or
hypothetical”, not specifying whether the “or” in this sentence is an
inclusive or exclusive “or”. According to Article 294(1)(i) and (j) of the CRR,

33
Back-testing levels refer to the risk factor level, the transaction level, and the actual and/or hypothetical
portfolio level.

ECB guide to internal models – Counterparty credit risk 209

 “back-testing shall be designed to test the relevant risk measures” and
furthermore “be appropriate and capable of identifying poor performance in
an EPE model’s risk measure”.

(f) Article 294(1)(g) of the CRR states that “as part of the initial and ongoing
validation of its CCR exposure model and its risk measures, an institution
shall ensure that the CCR exposure model includes transaction-specific
information to capture the effects of margining”, without specifying any
further details of the expected validation tasks.

(g) Article 292(1)(a) of the CRR requires that the model reflect transaction
terms and specifications in a timely, complete and conservative fashion
(also regarding pricing and the market data to be used). Article 292(1)(b)
of the CRR specifies that those terms include, at least, notional amounts,
maturity, reference assets, margining arrangements and netting
arrangements. Article 292(1)(g) of the CRR requires ongoing processes for
reconciliation between the model and source data, which verify that
transaction terms and specifications are reflected correctly or at least
conservatively.

(h) Article 294(1)(e) of the CRR provides that, as part of the initial and
ongoing validation process, an institution “shall test the pricing models
used to calculate CCR exposure for a given scenario of future shocks to
market risk factors”, as well as regularly testing these pricing models
against appropriate independent benchmarks in accordance with
Article 294(1)(l) of the CRR.

(i) As outlined in Article 294(1)(o) of the CRR, “the initial and ongoing
validation of CCR exposure models shall assess whether or not the
counterparty level and netting set exposure calculations of exposure are
appropriate”. Article 294(1)(d) of the CRR provides that “if the model
validation indicates that effective EPE is underestimated, the institution
shall take the action necessary to address the inaccuracy of the model”.
Both requirements are set out in a general way and therefore need further
guidance.

(j) Article 292(1)(a) of the CRR requires the model to reflect transaction terms
which must be ensured by formal reconciliation processes between the
model and source data in accordance with point (g) of the same article.

9.2 Principles for ECB banking supervision

59. In accordance with Article 293(1)(c) of the CRR, model validation/review and
model development must be conducted independently, that is, the validation
function must be effectively separated from model development. Hence, the
ECB considers that for cases where certain parts of the validation framework,
e.g. back-testing or the benchmarking of IMM pricing functions, are conducted
by staff also responsible for model design and development, the above-

ECB guide to internal models – Counterparty credit risk 210

 mentioned requirement provided for by Article 293(1)(c) of the CRR would be
fulfilled if all of the following practices were implemented:

(a) a regular, independent and effective challenging of the underlying

methodological aspects of the respective validation task comprising at
least scope, data samples, tools/statistical tests (including, if applicable,
test statistics and thresholds 34 in use) is performed by the validation
function;

(b) the assessment of the outcomes of the analysis (e.g. the evaluation of
back-testing traffic lights or pricing deficiencies detected in the
benchmarking) and the judgement regarding the respective remediation
measures are reviewed by the validation function, which may require
further analyses and/or changes to the actions concerned.

Moreover, the ECB considers that the organisational requirements of the risk
control unit (see Article 287(2) of the CRR) should be regarded as fulfilled when
(part of) the initial or ongoing validation of the model is conducted by staff not
belonging directly to the risk control unit, but for instance to a separate
validation unit. 35

As part of the process by which unacceptable performance will be identified and

remedied in accordance with Article 293(4) of the CRR, the ECB considers that
it is good practice to ensure a comprehensive view of all the findings, problems,
weaknesses and limits of the exposure model, identified by all staff contributing
to the validation and review of the exposure model.

60. The validation framework is expected to cover the kind of testing needed to
ensure model integrity and the appropriateness of assumptions underlying the
model in accordance with Article 292(6)(a) of the CRR. The ECB considers as
best practice the inclusion of various types of analyses on the key modelling
assumptions in a regular validation schedule. In particular, it is the ECB’s
understanding that the key modelling assumptions contain:

(a) the grid point setting;

(b) the chosen stochastic processes 36;

(c) the Monte Carlo error of the EEPE (see paragraph 51);

(d) expert-set parameters and boundaries in use (such as caps and floors for
risk factor paths);

(e) all pricing functions used in the IMM;

34
This refers, for example, to thresholds used for back-testing traffic lights or for the benchmarking of
pricing functions.
35
Please refer to Section 1.7 of the general topics chapter of this guide regarding the principle that the
internal audit function should not be responsible for validation.
36
At least in the event of poor back-testing results, the chosen stochastic processes should be thoroughly
challenged.

ECB guide to internal models – Counterparty credit risk 211

 (f) modelling features regarding margining, i.e. the MPOR setting, agreement-
dependent variation margin and IM mechanisms and the modelling of
collateral value changes during the MPOR.

61. In order to comply with Article 293(1)(b) of the CRR, the ECB sees it as best
practice if back-testing is performed and reported on at least once a year.

62. In accordance with Article 294(1)(c) of the CRR, back-testing at risk factor level
is mandatory. In the ECB’s view, not all key assumptions of the CCR exposure
model (mentioned in Article 294(1)(i) of the CRR) can be captured when back-
testing is only conducted on hypothetical portfolios, in particular when
considering non-plain vanilla transactions or margined netting sets. Hence, in
the light of Article 294(1)(h) of the CRR, the ECB regards back-testing at both
actual and hypothetical portfolio level as good practice.

63. In order to support the analysis of portfolio back-testing and mitigate the risk of
breaching Article 294(1)(e) and (i) of the CRR, it is recommended and seen as
good practice to include back-testing at single transaction level in the regular
framework.

64. According to Article 294(1)(h) of the CRR, back-testing samples 37 must be

representative and chosen on the basis of their sensitivity to material risk
factors as well as their combinations. As stated in point (j) of the same
paragraph in Article 294, the institution’s back-testing programme must be
capable of identifying poor performance of an EPE model’s risk measures. As a
result, the ECB considers that back-testing samples should allow for a
meaningful assessment of the CCR exposure model and that institutions should
ensure a comprehensive coverage of their back-testing framework by
calculating back-testing coverage ratios (i.e. shares of back-tested risk factors
or portfolios), at least at risk factor and, if applicable, at actual portfolio level. In
particular:

(a) next to a simple number-based 38 approach, institutions should take into

account different weighting schemes like sensitivities and exposure
metrics;

(b) at risk factor level, in addition to the full risk factor set 39, coverage ratios
should also be calculated by asset class; 40

(c) institutions should be able to provide an explanation justifying the level of

the respective coverage ratio.

37
This refers to samples comprising the subset of risk factors, transactions or portfolios, including margin
agreements used for the purpose of back-testing.
38
This means, for example, the number of risk factors, the number of portfolios that are covered, etc.
39
Note that the set of risk factors should include all underlying risk factors/drivers that are integrated into
the IMM exposure model (not differentiating between whether risk factors are directly or implicitly
diffused).
40
It should be noted that for a sensitivity-based approach, coverage ratios by asset class only (such as
interest rate, foreign exchange, equity, etc.) are sufficient.

ECB guide to internal models – Counterparty credit risk 212

 Such coverage ratios should form part of the back-testing reports to ensure that
the scope of the back-testing is transparent.

Furthermore, it is the ECB’s understanding that in order to comply with the

representativeness requirements stated in Article 294(1)(h) of the CRR, SFTs
should be included in the back-testing samples if they are within the IMM scope.

65. The ECB considers that statistical tests used for back-testing should be
adapted when back-testing samples contain forecasts over fully or partially
overlapping time periods 41 (compared with standard versions of statistical tools
applicable for the case of non-overlapping forecasts) to account for
dependencies in the sample and therefore serve as a proper indicator of the
model performance. In the view of the ECB, this practice would avoid the risk of
being in breach of Article 294(1)(j) of the CRR.

66. Where back-testing relies only on IMM pricing functions for both predictions and
realisations (i.e. realised prices derived from benchmarking systems are not
taken into account), the attention given to the assessment of the adequacy of
IMM pricing functions (as provided for by Article 294(1)(e) and (l) of the CRR) is
seen to be even more important. Consequently, in the view of the ECB,
institutions should strengthen their validation/review of IMM pricing functions
accordingly.

67. In order to ensure appropriate back-testing practices as required by

Article 294(1)(j) of the CRR, the ECB sees it as good practice to pay special
attention to the consistency of predictions and realisations in the case of actual
portfolio back-testing; in other words, changes of the portfolio composition
during the observation period (e.g. due to new or closed-out transactions)
should be handled accordingly.

68. In accordance with Article 294(1)(c), (e) and (g) of the CRR as understood by
the ECB in paragraphs 62 and 63 of this chapter, the ECB sees benefit in back-
testing different relevant risk measures, including the market value 42 at
transaction level, the market value of netting sets 43 as well as the exposure 44 at
netting set level. If direct back-testing of the exposure of margined netting sets
is not feasible, institutions should have a separate validation of the margining
process, of collateral value changes and of netting set market value changes
over the relevant time horizons.

69. In order to reduce the risk of breaching Article 292(1)(a) and (g) of the CRR,
and based on the requirements of Article 294(1)(l) of the CRR, institutions

41
For instance, distinct variables over the same forecasting period are tested simultaneously or tests are
built on a single variable and different successive but overlapping observation periods.
42
Market values can be either positive or negative.
43
This means the sum of all transaction market values within that netting set. This sum can be positive or
negative.
44
Exposure should always take into account the collateral balance and the margin mechanism. In the
case of unmargined netting sets, the collateral is zero. Combining Articles 272(14) and 272(17) of the
CRR, exposure is understood as an inherently non-negative value.

ECB guide to internal models – Counterparty credit risk 213

 should compare the values of pricing functions used for revaluation under the
IMM with values from a benchmarking system on a regular basis.

(a) The ECB understands Article 294(1)(o) of the CRR as requiring a

validation of the identification process for significant price differences (see
paragraph 12) as well as a corresponding full analysis of these differences
and their root causes.

(b) The action taken to address the inaccuracy of the model in accordance
with Article 294(1)(d) of the CRR, including price corrections also for future
grid points in accordance with paragraph 16, should be validated. 45

70. The ECB considers it compliant with Article 292(1)(b) and (g) of the CRR when
the following benchmark comparisons are conducted:

(a) The model-estimated collateral balance at 𝑡𝑡0 is benchmarked against the

actual collateral balance at 𝑡𝑡0 , if applicable (see paragraph 29).

(b) Institutions benchmark the IM (see paragraph 39).

A full analysis of the differences detected, their root causes and the action taken
to address the inaccuracy of the model should be conducted regularly in order
to reduce the risk of breaching Article 294(1)(d) and (o) of the CRR.

71. For the purposes of Article 294(1)(e), (l) and (o) of the CRR, in accordance with
the understanding of the ECB described in paragraph 17 and in addition to
paragraph 69, the ECB views it to be best practice if institutions assess the
following within their validation framework.

(a) Whether deviations from a full simulation 46 are documented and justified.

(b) Whether the effect of using approximated pricing functions instead of those
from any reliable benchmarking system is not significant.

(c) Whether, for all approximated pricing functions, the value changes due to
risk factor changes occurring in IMM simulated paths are reliable
compared with value changes from non-approximated pricing functions
(from any reliable benchmarking system) for the same transaction type.

(d) If any alternative way to calculate exposures is used, the ECB considers
that the items mentioned in paragraph 17(b) of this chapter should also be
met. Furthermore, validation should ensure that the respective methods
are applied in a way that does not lead to a systematic underestimation of
exposures compared with the full simulation (as described in paragraph 5)
for the transactions affected.

45
The validation for future grid points may consider market data scenarios deviating significantly from 𝑡𝑡0
and potential changes in sign and absolute value of the detected price difference.
46
As described in paragraph 5.

ECB guide to internal models – Counterparty credit risk 214

 In the ECB’s understanding, transactions treated with alternative exposure
calculation methods should also be included in the back-testing
framework. In order to fulfil the requirements of Article 294(1)(o) of the
CRR, the ECB sees it as beneficial to also analyse affected transactions
separately rather than mixing effects when back-testing is only conducted
at actual portfolio level.

In addition, the netting benefits (numerical impact) when using any type of
alternative method to calculate exposures in the IMM should be assessed
by comparing the resulting exposure with those obtained after:

(i) splitting the transactions into synthetic netting sets differentiating

between transactions treated using the “standard” IMM calculation
and those where exposures are calculated in an alternative way;

(ii) carving-out the affected transactions into a standardised method.

10 Effective expected positive exposure

72. This section refers to the normalisation of the weights ∆𝑡𝑡𝑘𝑘 that are used in the
calculation formula for the EEPE.

10.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 47 284 (6)

73. The regulatory provision relating to the topic addressed in this section that
requires further guidance is the calculation formula for the EEPE, which
appears in Article 284(6) of the CRR.

74. The corrigendum of 25 January 2017 amends the formula in Article 284(6) of
the CRR by dividing the weighted sum of the Effective EEs by the applicable
time horizon (1 year or the maturity of the longest-dated transaction belonging
to a netting set if this is below 1 year):
𝑚𝑚𝑚𝑚𝑚𝑚{1 𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦,𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚}
1
𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝐸𝐸𝐸𝐸𝐸𝐸 = ∙ � 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝐸𝐸𝐸𝐸𝑡𝑡𝑘𝑘 ∆𝑡𝑡𝑘𝑘
𝑚𝑚𝑚𝑚𝑚𝑚{1 𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦, 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚}
𝑘𝑘=1

However, the units of the weights ∆𝑡𝑡𝑘𝑘 and 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 still need to be defined.

47
See corrigendum to the CRR of 25 January 2017.

ECB guide to internal models – Counterparty credit risk 215

10.2 Principles for ECB banking supervision

75. In the understanding of the ECB, Article 284(6) of the CRR should be
understood as requiring that the weights ∆𝑡𝑡𝑘𝑘 and the parameter 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 are
expressed in units of one year.

11 Alpha parameter
76. The alpha multiplier affects all netting sets and thus all counterparties and
should be considered as intending to capture extra risk arising, for example,
from the fact that exposures are correlated with credit drivers (e.g. PD, LGD)
and to address general deficiencies in the IMM framework. Alpha is the only
parameter besides capital buffers that can be increased explicitly to account for
such deficiencies.

11.1 Relevant regulatory references

Date of issue Article Paragraph/Point

Legal background

CRR 26/06/2013 284 (4)

293 (1), (2)

77. The regulatory provisions relating to the topic addressed in this section that
require further guidance are the following:

(a) Article 284(4) of the CRR defines the exposure value as the product of
alpha and the EEPE with “α = 1.4, unless competent authorities require a
higher α or permit institutions to use their own estimates in accordance
with paragraph 9 [of Article 284 of the CRR]”;

(b) Article 293(2) of the CRR, based on Article 284(4) of the CRR, links the
level 48 of the supervisory alpha setting to the degree with which the
institution meets the requirements for the risk management system as set
out in Article 293(1) of the CRR;

(c) Article 293(1) of the CRR refers in particular to overall validation, adequate
processes, integration into the day-to-day risk management process and
limit utilisation (use test), documentation and independent reviews.

48
This refers to levels higher than the floor value of 1.4 for the non-modelled and 1.2 for the modelled
alpha parameter in accordance with Article 284(4) and (9) of the CRR.

ECB guide to internal models – Counterparty credit risk 216

11.2 Supervisory actions

78. In accordance with Article 284(4) of the CRR, the ECB can increase the alpha
parameter in a proportionate and appropriate way for either an interim or an
undefined period to address model, risk management or governance
deficiencies identified by the ECB. In particular, targeted deficiencies may
include (i) model deficiencies, which lead or may lead to an underestimation of
the EEPE as defined in Article 284(5) and (6) of the CRR and Article 285 of the
CRR for margined trading, or (ii) deficiencies in the validation framework.

It should be noted that:

(a) both supervisory alpha increases related to an interim period and those
related to an undefined period require explicit supervisory decisions;

(b) if alpha is increased for an interim period, the decision will specify the
length of the interim period or the condition when it ends.

79. The ECB can base the amount of a potential alpha increase above the floor
values to the extent possible on an available impact analysis.

(a) The analysis assesses 49 the impact on the EEPE as calculated without the
identified model deficiency.

(b) As this deficiency is obviously related to the standard configuration of the

IMM, which contains this deficiency, an impact calculation based only on a
subset of the relevant portfolio could be accepted for this purpose. This
calculation can be performed in a well-defined developer area for
representative sub-portfolios (as defined in the Counterparty credit risk
glossary).

(c) Some non-exhaustive examples of how identified model deficiencies can

increase alpha are discussed in this document (see for example
paragraphs 21, 50 and 51), where the general alpha increase (applied to
all netting sets) reflects whether the identified deficiencies possibly affect
only a part of the netting sets (e.g. only the margined ones).

(d) The ECB considers that increases should be in multiples of half a decimal
point. For example, if alpha = 1.4, alpha becomes at least 1.45 if an
increase is deemed necessary.

If no impact calculations are available, the ECB may estimate the amount of the
alpha increase in a conservative way using all other available information.

49
This assessment can also include less precise estimations, where needed.

ECB guide to internal models – Counterparty credit risk 217

Annex

This annex outlines two examples of a technique for assessing the confidence
interval of the estimated effective EPE (EEPE) referred to in paragraph 51, assuming
that the EEPE is calculated using a Monte Carlo method and a pseudo random
number generator.

In the examples, the MC error on the EEPE is defined as an aggregation of the MC

error on the different netting sets. At netting set level, the MC error on the EEPE is
defined as half the length of the 95% two-sided confidence interval centred around
the sample estimated EEPE.

Two methods are proposed for the calculation performed at the netting set level.
These are described in the “Method 1” and “Method 2” sections. How the MC error
should be inferred for a whole portfolio consisting of several netting sets is detailed
below in the “Aggregation” section.

Note that the methods below apply to banks that use a pseudo Monte Carlo
simulation method and not to banks that apply a quasi Monte Carlo simulation
method. In this context, a pseudo Monte Carlo simulation method is defined as a
method that utilises a random number generator based on an algorithm creating a
sequence of desired length N of numbers that mimic independent samples drawn
from a uniform distribution. A quasi Monte Carlo simulation method is defined as a
method that utilises a low-discrepancy sequence of numbers, which is
deterministically uniformly distributed (e.g. Sobol).

Irrespective of the numerical method implemented for its estimation of the EEPE
(e.g. types of random number generators), the institution should provide an analysis
as part of its validation framework demonstrating that its approach has a reasonable
accuracy as required by Article 368(1)(f) of the CRR (as referenced by
Article 293(1)(a) of the CRR). This analysis should include an assessment of
convergence and an error estimation.

In the following, “MC run” refers to a pseudo Monte Carlo simulation with N
scenarios calculated with one particular set of random numbers.

Method 1
�𝑁𝑁 (𝛼𝛼) denote the estimator of the EEPE for one given netting set 𝛼𝛼 obtained
Let 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
from one MC run with 𝑁𝑁 simulations (e.g. 𝑁𝑁 = 2000).

�𝑁𝑁 (𝛼𝛼), on the basis of a 95%

The institution can estimate an MC error on 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
confidence level, by using a set of several MC runs. In what follows, notations are
�𝑁𝑁 (𝛼𝛼) is replaced by 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
simplified: 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � ; 𝛼𝛼 and 𝑁𝑁 are dropped, since the
calculations detailed below are performed on the same netting set 𝛼𝛼 and with the
same number of simulations per MC run, 𝑁𝑁.

ECB guide to internal models – Annex 218

Furthermore, let 𝑚𝑚 denote the size of the set of MC runs (e.g. 𝑚𝑚 = 50). The different
MC runs are obtained by running the MC simulation with different random numbers
(e.g. by using different seeds).

� calculated with method 1 is defined as:

The MC error on 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑟𝑟 � �: = 𝛷𝛷−1 (0.975) ∙ 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚) ∙ �𝑣𝑣𝑣𝑣𝑟𝑟

�𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �

≃ 1.96 ∙ 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚) ∙ �𝑣𝑣𝑣𝑣𝑟𝑟

� �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � ,

with

1 1 2
𝑣𝑣𝑣𝑣𝑟𝑟
� �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �: = ∑𝑚𝑚 � 𝑘𝑘 ∑𝑚𝑚 � 𝑙𝑙
𝑚𝑚−1 𝑘𝑘=1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 − 𝑚𝑚 𝑙𝑙=1 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � ;

• � 𝑘𝑘 denoting the estimation of 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 using the k-th run of the MC run set;
𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

• 𝛷𝛷−1 standing for the inverse cumulative function of a standard normal

distribution.

By using 𝛷𝛷−1 (0.975) ≈ 1.96, we arrive at the following error formula:

2
1 𝑚𝑚
� 𝑘𝑘 − 1 � 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑚𝑚
� 𝑙𝑙 �
�𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑟𝑟 � �: = 1.96 ∙ 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚) ∙ � � �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑚𝑚 − 1 𝑘𝑘=1 𝑚𝑚 𝑙𝑙=1

The rationale of this formula is as follows.

� follows a normal distribution, 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑟𝑟

If we assume that 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � � can be
�𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
interpreted as half of the length of the 95% two-sided confidence interval centred
� . More precisely, it is estimated through a three-step approach:
around 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

1. � � is half of the length of the 95% two-sided confidence

𝛷𝛷−1 (0.975)�𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
� , since we have:
interval centred around 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

� − 𝛷𝛷−1 (0.975)�𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑃𝑃 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 ∈ �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � �,

� + 𝛷𝛷−1 (0.975)�𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � ��� = 95% .

2. � � being unknown, it is approximated by �𝑣𝑣𝑣𝑣𝑟𝑟

�𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �. The length

� �, is then
of the two-sided 95% confidence interval, 𝛷𝛷−1 (0.975)�𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

approximated by 𝛷𝛷 −1 (0.975)�𝑣𝑣𝑣𝑣𝑟𝑟
� �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �.

3. However, one must take into account that whenever 𝑚𝑚 is too small (e.g. 𝑚𝑚 <
�
50), 𝑣𝑣𝑣𝑣𝑟𝑟 � �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � may not have properly converged to 𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �. Finally,

� � is estimated by
𝛷𝛷−1 (0.975)�𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

𝛷𝛷−1 (0.975) ∙ 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚) ∙ �𝑣𝑣𝑣𝑣𝑟𝑟

� �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �,

ECB guide to internal models – Annex 219

 where∙ 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚) takes into account the fact that 𝑣𝑣𝑣𝑣𝑟𝑟
� �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � may not have
�
properly converged to 𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �.

Details of 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚):

The parameter 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚) is chosen such that

� � < 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚) �𝑣𝑣𝑣𝑣𝑟𝑟

𝑃𝑃 ��𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �� = 95%

� has a normal distribution,

holds. More precisely, still under the assumption that 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
one can write:

𝑚𝑚 − 1 (1)
𝑣𝑣𝑣𝑣𝑟𝑟
� � 2
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � ~ 𝜒𝜒𝑚𝑚−1
��
𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

2
where 𝜒𝜒𝑚𝑚−1 denotes a standard chi-squared distribution with 𝑚𝑚 − 1 degrees of
freedom.

� � < 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚)2 𝑣𝑣𝑣𝑣𝑟𝑟

From (1), we get 𝑃𝑃 �𝑣𝑣𝑣𝑣𝑣𝑣�𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �� = 95% with

𝑚𝑚−1
• 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑚𝑚) = � ,
𝑞𝑞(𝑚𝑚−1;97.5%)

2
• 𝑞𝑞(𝑚𝑚 − 1; 97.5%) is such that 𝑃𝑃(𝑞𝑞(𝑚𝑚 − 1; 97.5%) ≤ 𝑍𝑍) = 97.5% with 𝑍𝑍~𝜒𝜒𝑚𝑚−1 .

Method 2
�𝑁𝑁 (𝛼𝛼 ) as the estimator of the EEPE for
As in the previous section, we denote 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
one given netting set 𝛼𝛼 obtained from one MC run with N simulations (e.g. N = 2000)
�𝑁𝑁 (𝛼𝛼 ) as 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
and, as in the previous section, we simplify the notation 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �.

� is a method where only one MC

The second method to estimate the error on 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
run is needed (contrary to method 1 where a set of 𝑚𝑚 MC runs was needed).

Before presenting the method for the estimation of the MC error, let us detail some
definitions and notations. For any time point 𝑡𝑡𝑘𝑘 of the time grid used for exposure
calculations, we denote 𝐸𝐸(𝑡𝑡𝑘𝑘 ) as the netting set exposure at time 𝑡𝑡𝑘𝑘 and 𝐸𝐸𝐸𝐸(𝑡𝑡𝑘𝑘 ) as its
� (𝑡𝑡𝑘𝑘 ) be the estimator of 𝐸𝐸𝐸𝐸(𝑡𝑡𝑘𝑘 ) based on the MC run, i.e.
expected value. Let 𝐸𝐸𝐸𝐸

1 𝑁𝑁
� (𝑡𝑡𝑘𝑘 ) =
𝐸𝐸𝐸𝐸 � 𝐸𝐸𝑗𝑗 (𝑡𝑡𝑘𝑘 ),
𝑁𝑁 𝑗𝑗=1

where 𝐸𝐸𝑗𝑗 (𝑡𝑡𝑘𝑘 ) stands for the netting set exposure level at time 𝑡𝑡𝑘𝑘 for scenario j.

The following equations holds if the EEPE is not dominated by the current
exposure 𝐸𝐸(𝑡𝑡0 ), meaning there is at least one 𝑡𝑡𝑘𝑘 below one year with 𝐸𝐸(𝑡𝑡0 ) <
� (𝑡𝑡𝑘𝑘 ) – otherwise the numerical error of the EEPE is in any case zero. For the
𝐸𝐸𝐸𝐸
� (𝑡𝑡1 ).
sake of simplicity, it is also assumed that 𝐸𝐸𝐸𝐸(𝑡𝑡0 ) < 𝐸𝐸𝐸𝐸

ECB guide to internal models – Annex 220

The effective reference dates are the subset of dates 𝑡𝑡𝑘𝑘 among the simulation dates
(𝑡𝑡ℎ )ℎ>0 such that

𝐸𝐸𝐸𝐸(𝑡𝑡𝑘𝑘 ) > 𝑚𝑚𝑚𝑚𝑚𝑚 𝐸𝐸𝐸𝐸(𝑡𝑡ℎ ) .

0≤ℎ<𝑘𝑘

Let us denote (𝑠𝑠𝑢𝑢 )𝑢𝑢 these effective reference dates with

𝑠𝑠1 < 𝑠𝑠2 < ⋯ < 𝑠𝑠𝑢𝑢 < ⋯ < 𝑠𝑠𝑝𝑝 ≤ 𝑡𝑡1𝑦𝑦 ,

i.e. p dates.

For the given MC run, the estimated effective reference dates are the subset of
dates 𝑡𝑡𝑘𝑘 among the simulation dates (𝑡𝑡ℎ )ℎ>0 such that:

� (𝑡𝑡𝑘𝑘 ) > 𝑚𝑚𝑚𝑚𝑚𝑚 𝐸𝐸𝐸𝐸

𝐸𝐸𝐸𝐸 � (𝑡𝑡ℎ )
0≤ℎ<𝑘𝑘

Let us denote (𝑠𝑠�𝑢𝑢 )𝑢𝑢 these estimated (i.e. as resulting from an MC simulation)
effective reference dates with:

𝑠𝑠�1 < 𝑠𝑠�2 < ⋯ < 𝑠𝑠�𝑢𝑢 < ⋯ < 𝑠𝑠�𝑝𝑝� ≤ 𝑡𝑡1𝑦𝑦

i.e. 𝑝𝑝̂ dates.

� depends only on the �𝐸𝐸𝐸𝐸

𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � (𝑠𝑠�𝑢𝑢 )� and the time profile of effective EE values as
𝑢𝑢
� (𝑠𝑠�𝑢𝑢 ),
defined in Article 284(5) of the CRR. More precisely, it is fully determined by 𝐸𝐸𝐸𝐸
as can be seen by the following schematic graph:

EE

Effective EE (implied
from CRR)

application period of

t0 t3 time
t1 t2 = t4 t5
= = =

The method below relies on the assumption, which should be checked by the
institution when applying the method, that N is large enough such that all 𝐸𝐸𝐸𝐸 � (𝑡𝑡𝑘𝑘 ) are
“sufficiently close” to their true values 𝐸𝐸𝐸𝐸(𝑡𝑡𝑘𝑘 ) and that, as a consequence, the
effective reference dates are properly identified, i.e. (𝑠𝑠�𝑢𝑢 )𝑢𝑢 = (𝑠𝑠𝑢𝑢 )𝑢𝑢 .

ECB guide to internal models – Annex 221

Under the complementary assumptions that 𝑠𝑠𝑝𝑝 ≠ 𝑡𝑡1𝑦𝑦 , and considering, as previously
� is given by: 1
mentioned, that (𝑠𝑠�𝑢𝑢 )𝑢𝑢 = (𝑠𝑠𝑢𝑢 )𝑢𝑢 , 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑝𝑝−1

𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � �𝑠𝑠𝑝𝑝 �
� (𝑠𝑠𝑢𝑢 ) + �𝑡𝑡1𝑦𝑦 − 𝑣𝑣𝑝𝑝−1 �𝐸𝐸𝐸𝐸
� = �(𝑣𝑣𝑢𝑢 − 𝑣𝑣𝑢𝑢−1 )𝐸𝐸𝐸𝐸
𝑢𝑢=1
𝑁𝑁 𝑝𝑝−1
1
= � ��(𝑣𝑣𝑢𝑢 − 𝑣𝑣𝑢𝑢−1 )𝐸𝐸𝑗𝑗 (𝑠𝑠𝑢𝑢 ) + �𝑡𝑡1𝑦𝑦 − 𝑣𝑣𝑝𝑝−1 �𝐸𝐸𝑗𝑗 �𝑠𝑠𝑝𝑝 �� .
𝑁𝑁
𝑗𝑗=1 𝑢𝑢=1

Where (𝑣𝑣𝑢𝑢 )𝑢𝑢 are the “application period dates”: they are such that [ 𝑣𝑣𝑢𝑢−1 , 𝑣𝑣𝑢𝑢 ] is the
� (𝑠𝑠𝑢𝑢 ) is applied to. For instance, for the case illustrated in the graph above,
period 𝐸𝐸𝐸𝐸
�
𝐸𝐸𝐸𝐸 2 is applied on [𝑡𝑡1 , 𝑡𝑡3 ], and thus 𝑣𝑣1 = 𝑡𝑡1 and 𝑣𝑣2 = 𝑡𝑡3 .
(𝑠𝑠 )

Let us define, for each scenario 𝑗𝑗 from 1 to 𝑁𝑁:

𝑝𝑝−1

𝐷𝐷𝑗𝑗 : = �(𝑣𝑣𝑢𝑢 − 𝑣𝑣𝑢𝑢−1 )𝐸𝐸𝑗𝑗 (𝑠𝑠𝑢𝑢 ) + �𝑡𝑡1𝑦𝑦 − 𝑣𝑣𝑝𝑝−1 �𝐸𝐸𝑗𝑗 �𝑠𝑠𝑝𝑝 �

𝑢𝑢=1

1
� = ∑𝑁𝑁
By definition of 𝐷𝐷𝑗𝑗 , we have 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑗𝑗=1 𝐷𝐷𝑗𝑗 . 𝑁𝑁

𝑝𝑝−1
For 𝐷𝐷: = ∑𝑢𝑢=1(𝑣𝑣𝑢𝑢 − 𝑣𝑣𝑢𝑢−1 )𝐸𝐸(𝑠𝑠𝑢𝑢 ) + �𝑡𝑡1𝑦𝑦 − 𝑣𝑣𝑝𝑝−1 �𝐸𝐸�𝑠𝑠𝑝𝑝 �, the variance of 𝐷𝐷 can be
estimated by:
2
1 𝑁𝑁 1 𝑁𝑁 1 𝑁𝑁
� (𝐷𝐷) =
𝑣𝑣𝑣𝑣𝑣𝑣 � �𝐷𝐷𝑗𝑗 − � 𝐷𝐷𝑘𝑘 � = � �2 .
� �𝐷𝐷𝑗𝑗 − 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑁𝑁 − 1 𝑗𝑗=1 𝑁𝑁 𝑘𝑘=1 𝑁𝑁 − 1 𝑗𝑗=1

� (𝑡𝑡1 ) and/or 𝑠𝑠𝑝𝑝 = 𝑡𝑡1𝑦𝑦 are not derived in this annex.

Note: cases where 𝐸𝐸(𝑡𝑡0 ) ≥ 𝐸𝐸𝐸𝐸
However similar equations can be obtained.

� is then given by:

An estimator of the variance of 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

1 1 𝑁𝑁
𝑣𝑣𝑣𝑣𝑟𝑟
� �
𝑀𝑀2 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � = � (𝐷𝐷) =
𝑣𝑣𝑣𝑣𝑣𝑣 � �2 .
� �𝐷𝐷𝑗𝑗 − 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑁𝑁 𝑁𝑁(𝑁𝑁 − 1) 𝑗𝑗=1

As mentioned in the first footnote of paragraph 51 requiring a statistical error at a

� should be calculated
95% confidence level, the estimation of the MC error on 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
according to the following formula:

𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 � � ∶= 𝛷𝛷−1 (0.975)�𝑣𝑣𝑎𝑎𝑟𝑟

� 𝑀𝑀2 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � �
𝑀𝑀2 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 �

1 𝑁𝑁
≃ 1.96 ∙ � � �2 .
� �𝐷𝐷𝑗𝑗 − 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑁𝑁(𝑁𝑁 − 1) 𝑗𝑗=1

The rationale of the formula is the same as that outlined in method 1, with a different
� and without a convergence adjustment. If we
estimator of the variance of 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
�
assume that 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 follows a normal distribution, then 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 � � can be
� 𝑀𝑀2 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
interpreted as half of the length of the 95% two-sided confidence interval centred

1
This assumes the longest-lasting transaction in the netting set has a maturity equal to or higher than
one year and all time differences in the above formulas are expressed in units of a year (not dividing by
the minimum between 1y and the netting set maturity for simplicity) – otherwise the normalised
weighting as described in paragraph 74 needs to be applied.

ECB guide to internal models – Annex 222

 � . No adjustment (similar to 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(. ) in the first method) is needed,
around 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
since for usual values of 𝑁𝑁, we have 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝑁𝑁) close to 1, e.g. 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(500) ≃
1.067 and 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(1000) ≃ 1.046.

Aggregation across netting sets

a) When risk factors are simulated all together (no “silo”), the MC error of the
estimator of the EEPE for the full scope should be calculated in a similar way to that
� 𝑁𝑁 should be understood as the
described for a single netting set, except that 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
sum of the estimators of the EEPE related to all netting sets belonging to the
institution’s portfolio. Assume that a set of n netting sets 𝐴𝐴 = {𝛼𝛼1 , … , 𝛼𝛼𝑛𝑛 } is available
for the MC error analysis.

This means for method 1 that

2
1 𝑚𝑚 1 𝑚𝑚
𝑣𝑣𝑣𝑣𝑟𝑟
� �
𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝑁𝑁 � = � �� � 𝑁𝑁𝑘𝑘 (𝛼𝛼𝑖𝑖 ) − � �
𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 � 𝑁𝑁𝑙𝑙 (𝛼𝛼𝑖𝑖 )�
𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑚𝑚 − 1 𝑘𝑘=1 𝛼𝛼𝑖𝑖 ∈𝐴𝐴 𝑚𝑚 𝑙𝑙=1 𝛼𝛼𝑖𝑖 ∈𝐴𝐴

should be inserted into the equation for 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 � �.

� 𝑀𝑀1 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸

For method 2, the addition needs to happen at the netting set-specific D term.

𝐷𝐷𝑗𝑗 = � 𝐷𝐷𝑗𝑗 (𝛼𝛼𝑖𝑖 )

𝛼𝛼𝑖𝑖 ∈𝐴𝐴

should be inserted into the equation for 𝑣𝑣𝑣𝑣𝑟𝑟

� �
𝑀𝑀2 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝑁𝑁 � to calculate the variance, then
� 𝑁𝑁 �.
�𝑀𝑀2 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
this should be inserted into the equation for 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑟𝑟

b) When risk factors are not simulated all together (in cases where exposures are
estimated through “silos”, e.g. one per asset class), the MC error should be derived
� 𝑁𝑁 per silo. Using either method 1 or 2 for computing the
from the MC errors of 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
MC error per silo as explained immediately above (item a), the error on the total
portfolio is then given by:

𝑆𝑆
2
� 𝑀𝑀1/𝑀𝑀2 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 � 𝑁𝑁 𝑜𝑜𝑜𝑜 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 � = �� �𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑟𝑟
� �
𝑀𝑀1/𝑀𝑀2 �𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝑁𝑁 𝑜𝑜𝑜𝑜 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑖𝑖 �� ,
𝑖𝑖=1

where

• 𝑆𝑆 is the total number of silos,

• 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑖𝑖 is a sub-portfolio of the institution’s total portfolio corresponding to all the

netting sets simulated in silo 𝑖𝑖.

ECB guide to internal models – Annex 223

Acronyms

BCBS
Basel Committee on Banking Supervision
CCF
Conversion factor
CCR
Counterparty credit risk
CEBS
Committee of European Banking Supervisors
CF
Cash flow
EAD
Exposure at default
EBA
European Banking Authority
ECB
European Central Bank
EL
Expected loss
ELBE
Expected loss best estimate
EU
European Union
GL
Guidelines
IRB
Internal ratings-based
IT
Information technology
LGD
Loss given default
M
Maturity parameter
OTC derivative
Over-the-counter derivative
SME
Small and medium-sized enterprise
PD
Probability of default
RR
Recovery rate
RTS
Regulatory Technical Standards
RWA
Risk-weighted asset
RWEA
Risk-weighted exposure amount
SSM
Single Supervisory Mechanism
TRIM
Targeted review of internal models

ECB guide to internal models – Acronyms 224

Glossary

General topics
Basel Committee on Banking Supervision (BCBS) 328
Basel Committee on Banking Supervision “Guidelines: Corporate governance principles for banks”
Basel Committee on Banking Supervision (BCBS) - Newsletter No. 4
Basel Committee Newsletter No. 4 (January 2005), “Update on work of the Accord Implementation
Group related to validation under the Basel II Framework”
Basel Committee on Banking Supervision (BCBS) - Newsletter No. 9
Basel Committee Newsletter No. 9 (September 2006), “The IRB Use Test: Background and
Implementation”
Basel Committee on Banking Supervision (BCBS) - Regulatory consistency assessment
programme (RCAP)
Basel Committee on Banking Supervision “Regulatory Consistency Assessment Programme
(RCAP) – Analysis of risk-weighted assets for credit risk in the banking book”
CEBS Guidelines on Outsourcing
Committee of European Banking Supervisors “Guidelines on Outsourcing”
Commission Delegated Regulation (EU) No 529/2014
Commission Delegated Regulation (EU) No 529/2014 of 12 March 2014 supplementing Regulation
(EU) No 575/2013 of the European Parliament and of the Council with regard to regulatory
technical standards for assessing the materiality of extensions and changes of the Internal Ratings
Based Approach and the Advanced Measurement Approach (OJ L 148, 20.5.2014, p.36)
CRCU
Credit risk control unit
CRD IV
Directive 2013/36/EU of the European Parliament and the Council of 26 June 2013 on access to
the activity of credit institutions and the prudential supervision of credit institutions and investment
firms, amending Directive 2002/87/EC and repealing Directives 2006/48/EC and 2006/49/EC (OJ L
176, 27.6.2013, p. 338)
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). For the purposes of this document, the reader’s attention
is also drawn to the corrigendum published on 30 November 2013 (OJ L 321, 30.11.2013, p. 6)
EBA Consultation Paper 2014/10
Consultation Paper “Draft regulatory technical standards on the sequential implementation of the
IRB Approach and permanent partial use under the Standardised Approach under Articles 148(6),
150(3) and 152(5) of Regulation (EU) No 575/2013 (CRR)” (EBA/CP/2014/10)
EBA Guidelines on internal governance
Guidelines on internal governance (EBA/GL/2017/11)
Final Draft RTS on assessment methodology for IRB
Final Draft Regulatory Technical Standards on the specification of the assessment methodology for
competent authorities regarding compliance of an institution with the requirements to use the IRB
approach in accordance with Articles 144(2), 173(3) and 180(3)(b) of Regulation (EU) No 575/2013
(EBA/RTS/2016/03)
Final Draft RTS on assessment methodology for IMA and significant share
Final Draft Regulatory Technical Standards on the specification of the assessment methodology for
competent authorities regarding compliance of an institution with the requirements to use internal
models for market risk and assessment of significant share under points (b) and (c) of Article
363(4) of Regulation (EU) No 575/2013 (EBA/RTS/2016/07)
G-SII
Global systemically important institution
KPI
Key performance indicator
IFRS
International Financial Reporting Standards

ECB guide to internal models – Glossary 225

O-SII
Other systemically important institution
PPU
Permanent partial use
RDS
Reference dataset
RORAC
Return on risk-adjusted capital
SLA
Service level agreement
SSM Regulation
Council Regulation (EU) No 1024/2013 of 15 October 2013 conferring specific tasks on the
European Central Bank concerning policies relating to the prudential supervision of credit
institutions (OJ L 287, 29.10.2013, p. 63)
EBA Guidelines on SREP
Guidelines on common procedures and methodologies for the supervisory review and evaluation
process (SREP) (EBA/GL/2014/13)

Credit risk
Basel Committee on Banking Supervision (BCBS) 239
Basel Committee on Banking Supervision “Principles for effective risk data aggregation and risk
reporting”
CRM
Credit risk mitigation
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). For the purposes of this document the reader’s attention is
also drawn to the corrigendum published on 30 November 2013 (OJ L 321, 30.11.2013, p. 6)
EBA GL on PD and LGD
EBA Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures
(EBA/GL/2017/16)
F-IRB
Foundation IRB
Final Draft RTS on assessment methodology for IRB
Final Draft Regulatory Technical Standards on the specification of the assessment methodology for
competent authorities regarding compliance of an institution with the requirements to use the IRB
approach in accordance with Articles 144(2), 173(3) and 180(3)(b) of Regulation (EU) No 575/2013
(EBA/RTS/2016/03)
GDP
Gross domestic product
LRA
Long-run average
MoC
Margin of conservatism
NACE
Nomenclature statistique des activités économiques dans la Communauté Européenne
NUTS
Nomenclature of territorial units for statistics
SRM
Shadow rating model

ECB guide to internal models – Glossary 226

Market risk
Actual P&L
The daily actual changes in the portfolio’s value, as defined in Article 366(3) of the CRR.
AVA
Additional valuation adjustment
CIU
Collective investment undertaking
CVA
Credit valuation adjustment
CRM
Comprehensive risk measure
DVA
Debit valuation adjustment
Economic P&L
The daily changes in the portfolio’s value (or profit and loss, P&L) calculated on the basis of end-of-
day mark-to-market or mark-to-model (depending on the instruments) values of the books and
records of the institution, taking into account the independent price verification (IPV) process. It is
generally calculated using front-office systems (position data, pricing models, valuation methods,
pricing parameters, end-of-day market data, etc.).
ETF
Exchange-traded fund
FX
Foreign exchange
Fundamental review of the trading book (FRTB)
The document entitled “Minimum capital requirements for market risk” issued by the Basel
Committee on Banking Supervision (BCBS) in January 2019.
Hypothetical P&L
The daily hypothetical changes in the portfolio’s value, as defined in Article 366(3) of the CRR.
IMA
The internal model approach for the calculation of own funds requirements for market risk.
IRC
Incremental default and migration risk charge
P&L
The daily changes in the portfolio’s value (or profit and loss).
Position
Understood to be a risk position. A risk position is a non-identically-zero sensitivity to a risk factor.
Holding securities or entering into transaction contracts entails having a position. When defining a
position, neither hedging nor netting should be considered.
Top-of-the-house level
Both (i) the legal entity for which an approval for the IMA approach has been granted, and (ii) (within
the scope of the IMA) the highest level of the portfolio structure.
RNIME
Risk(s) not in the model engines as set out in detail in Section 7 of the market risk chapter. In this
document, the abbreviation “RNIME” may be singular or plural depending on whether it refers to a
single risk, several risks, or collectively all risks not captured in the model engines.
VaR
Value-at-risk
sVaR
Stressed VaR

Counterparty credit risk

Benchmarking system
In the context of pricing functions mentioned in the guide, this means the respective front-office
pricing functions, pricing functions of accounting systems or other benchmarks with which front-office
prices are frequently compared (at least quarterly, as for CCR purposes). Values taken from such

ECB guide to internal models – Glossary 227

benchmarking systems are values after independent price verification (see Article 4(70) of the CRR)
without any valuation adjustments beyond the default-free value (such as the credit valuation
adjustment).
DMP
Default management process
EEPE
Effective expected positive exposure
ETC
Early termination clause
IM
Initial margin
IMM
Internal Model Method for counterparty credit risk.
MPOR
Margin period of risk
Pricing function
A dedicated implementation of a pricing model taking into account:
• the input data used in this particular implementation (e.g. the input market data needed, day-
count conventions, etc.);
• the parametrisation of the implemented pricing model including the method for its calibration;
• the numerical method used (e.g. binomial tree, finite difference, Monte Carlo, etc.).
Pricing model
The quantitative, mathematical model (e.g. a Black 76 swaption) that is used to determine the market
value of a transaction for a given (current or future) date and specified market conditions/scenarios.
Representative sub-portfolios
A subset of all counterparties or netting sets that is representative of the full set at least in terms of:
• transaction types and their “moneyness”;
• underlying risk factors;
• the ratio of the value of short positions to the value of long positions;
• margin agreement types;
• the ratio of margined to unmargined netting sets;
and for which the institution is able to demonstrate to supervisors that the chosen sub-portfolios are
sufficiently representative in terms of the above criteria and meaningful regarding the purpose for
which the portfolio has been selected.
Securities financing transactions (SFTs)
This term covers repurchase agreements, margin lending and borrowing agreements, as well as
securities and commodities lending and borrowing agreements. It thus encompasses all products
covered by Article 272(25)(a) and (b) of the CRR.
𝑡𝑡0
The first date of the simulation time grid in the IMM and the reporting date for which the EEPE is
calculated. It is thus equal to the “current date” referred to in Article 284(5) of the CRR.
Systematically underestimated exposures
This expression means a progressive, aggressive or non-conservative modelling of exposures in
almost all cases compared with a precise treatment without approximations, which may refer to
almost all cases of:
• simulated scenarios;
• portfolio configurations;
• market conditions at t 0;
• market conditions during the period used for calibration.
This holds to the extent that “almost all cases” can be anticipated from past experience or historic
time series. Otherwise, this expression refers to an a priori estimation, for example resulting from a
mathematical consideration. Example: A model implementation approximates the true value of a
bought vanilla call option inside a netting set by its intrinsic value (the value if exercised). Since the
true value is always more than the intrinsic value before exercising the option, this modelling would
lead to too low a transaction value and thus to too low a netting set value; hence, in this specific
example, “almost all cases” changes even to “always”.
If the expression is used in the context of the netting set value, it means that modelling/pricing leads
in almost all cases to too low an overall netting set value, i.e. after applying the netting rules. If it is
used in the context of single transaction values, it means that transactions with a positive value have
in almost all cases too low a value and that transactions with a negative value (if they are also inside
a netting set) have in almost all cases too high an absolute value compared with a precise treatment.

ECB guide to internal models – Glossary 228

© European Central Bank, 2019
Postal address 60640 Frankfurt am Main, Germany
Telephone +49 69 1344 0
Website www.ecb.europa.eu
All rights reserved. Reproduction for educational and non-commercial purposes is permitted provided that the source is acknowledged.
For specific terminology please refer to the SSM glossary.
PDF ISBN 978-92-899-3715-3, doi:10.2866/849746, QB-02-19-097-EN-N