Requirements for IRB Models in Banks, Credit Institutions and Finance Companies

FINANSTILSYNET
Postboks 1187 Sentrum
0107 Oslo

Circular
Requirements for IRB models in banks, credit institutions and finance companies

CIRCULAR: 3/2021
DATE: 09.06.2021
THE CIRCULAR APPLIES TO:
Banks
Credit institutions
Finance companies

Requirements for IRB Models in Banks, Credit Institutions and Finance Companies

Table of Contents

1. Introduction 3
2. Background and General Requirements 3
   2.1 The Financial Supervisory Authority's Application Processing and Assessments 3
   2.1.1 Model Changes 4
   2.1.2 Portfolio Size 4
   2.2 Data History and Safety Margins 5
   2.3 Board of Directors' Responsibility 5
   2.4 Validation 6
   2.5 Application Requirements 6
3. Requirements for Probability of Default (PD) Models 6
4. Requirements for Loss Given Default (LGD) Models 7
   4.1 Economic Downturns 7
   4.2 Requirements for Collateral 8
   4.3 Calculation of Observed Loss Rates 8
   4.4 Validation of Estimates for the Performing Portfolio 8
   4.5 Reference Model for Mortgages 9
5. Other Parameters Included in the Calculation of Risk Weights 10
   5.1 Maturity Parameter (M) 10
   5.2 Size Parameter (S) 11

---

1 Introduction

The IRB approach (Internal Ratings-Based Approach) implies that risk weights and capital requirements for credit risk are calculated based on the bank's own estimates of risk parameters such as probability of default (PD), loss given default (LGD), assumed utilization of credit lines and commitments (CfI), as well as calculated maturity (M) and size (S).

Use of the IRB approach requires permission from the Financial Supervisory Authority. IRB permission can only be granted on the condition that the IRB system provides a sound measurement of risk, is used in the bank's approval and follow-up of loans, and in risk management and measurement. The IRB system must also be regularly verified (validated).

With this circular, the Financial Supervisory Authority clarifies the Authority's practice regarding the IRB regulatory framework. Chapter 2 covers the Financial Supervisory Authority's application processing and general requirements for the IRB system. Chapters 3 and 4 describe specific requirements for models for probability of default (PD) and loss given default (LGD). Chapter 5 covers the maturity and size parameters (M and S).

This circular replaces Circular 8/2014 on IRB models for mortgages, Circular 3/2015 on changes in IRB models, and Circular 9/2016 on discount rate for calculation of realized LGD.

2 Background and General Requirements

2.1 The Financial Supervisory Authority's Application Processing and Assessments

The requirements for an IRB system are set out in Part 3, Section II, Chapter 3 of the Capital Requirements Regulation. According to Article 144 of the Capital Requirements Regulation, IRB permission shall only be granted if the bank can document to the authorities that its models provide a basis for classifying and quantifying risk, that the bank uses the system in the approval and follow-up of individual exposures and in risk management, and that it has good risk management systems, including systems and procedures for verifying (validating) the models.

The regulatory framework requires judgment in a number of areas and places high demands on the assessments and practices of banks and authorities. The European Commission has established detailed rules (technical standards) based on proposals from the European Banking Authority (EBA).

Technical standards become applicable in Norwegian law by reference in statutory or regulatory text (incorporation), after they have been included in the EEA Agreement. In addition, EBA has drawn up extensive guidelines for the use and approval of IRB models. The Financial Supervisory Authority bases its understanding and practice of the regulatory framework on EBA's guidelines. An updated overview of the guidelines that underpin the Authority's practice is available on the Financial Supervisory Authority's website.

1. https://www.finanstilsynet.no/regelverk/eba-retningslinjer/eba-retningslinjer/

2.1.1 Model Changes

Significant changes to the models or areas of application require permission from the supervisory authority, cf. Article 143 of the Capital Requirements Regulation. A separate technical standard² for model changes sets out qualitative and quantitative criteria for what shall be considered significant changes, as well as criteria and notification procedures for other changes. Significant changes include changes in the classification and quantification of risk, changes in the definition of default³, changes in validation procedures, and changes in the model's area of use, i.e., use of the model in new business areas, products, or customer groups.

Other changes that result in significant changes to the capital requirement shall also be considered significant. According to the technical standard, changes shall be considered significant if they are estimated to reduce the capital requirement for credit risk by 1.5 percent at the consolidated level, or by 15 percent for the portfolio on which the model is used. If the bank makes several changes simultaneously or in connection with each other, the effect shall be assessed collectively. The Financial Supervisory Authority expects the bank to have criteria for assessing what constitutes a foreseeable variation in the level of risk parameters compared to the level that formed the basis for the permission. A reduction in the average level of the portfolio beyond foreseeable variation is to be considered a model change that requires permission.

Model changes deemed necessary to correct underestimation shall be implemented immediately and notified to the Financial Supervisory Authority.

The bank must have systems to capture changes that affect the IRB system and assess whether the changes, collectively or individually, require permission. All changes shall be logged. Changes that do not require permission from the Financial Supervisory Authority shall be notified to the Financial Supervisory Authority no later than two months before they are implemented, cf. Article 5 of the technical standard. If the Financial Supervisory Authority believes that the change requires permission, the bank will receive a warning to this effect, normally within one month of submission.

2.1.2 Portfolio Size

The requirements for the use and quality assurance of the IRB system, including the requirements for data basis, competence, and resources for the development and validation of models, imply that the bank's portfolios should be of a certain size. The Financial Supervisory Authority has experienced that the IRB system is used to a lesser extent in the approval and follow-up of mortgages and other loans to the mass market, and that it is very difficult for the bank to obtain data for such loans from bad times. The Financial Supervisory Authority has also experienced that it is particularly difficult for banks with smaller corporate portfolios to meet the requirements for IRB systems. Based on these experiences, it is the Financial Supervisory Authority's assessment that banks that have loans to companies of less than NOK 30 billion normally cannot expect to receive IRB permission.

2. Commission Delegated Regulation (EU) No 529/2014 of 12 March 2014
3. Cf. Financial Supervisory Authority Circular 4/2020: "Identification of Defaulted Exposures"

2.2 Data History and Safety Margins

All relevant information shall be taken into account in the estimates. The estimates for PD shall reflect long-term outcomes, and the estimates for LGD shall reflect downturns.

The bank's data history underlying the estimation must normally be longer than the minimum requirement of five years, cf. Articles 180–182 of the Capital Requirements Regulation. A particular challenge for Norwegian banks is that the data basis for the models generally reflects good economic upturns. In the Financial Supervisory Authority's assessment, the banking crisis in the early 1990s is the last serious economic downturn in the Norwegian economy to date. The coronavirus crisis has caused a serious economic setback, but the further course of the crisis is highly uncertain, and it will take time before complete data history from this crisis will be available. The Financial Supervisory Authority therefore requires that the experiences from the banking crisis are reflected in the bank's estimates. This is elaborated further in Chapters 2 and 3.

To account for uncertainty in the estimates, safety margins shall be added, cf. Article 179 of the Capital Requirements Regulation. The safety margins shall take into account statistical uncertainty and uncertainty and deficiencies in the data basis and modeling. In its guidelines for PD and LGD estimation,⁴ point 4.4, EBA points to the following factors as indicating extra safety margins:

• deficiencies in the data basis, for example missing information on risk drivers over time
• missing data basis from downturns and weaknesses in methodology for adjusting for this
• changes in the definition of default over time
• changes in the bank's approval practice, risk tolerance, exposure follow-up, and collection practice
• changes in market conditions or regulatory and legal framework that make the data basis less representative

Furthermore, an extra safety margin shall be added for the general uncertainty associated with model use.

2.3 Board of Directors' Responsibility

The Board of Directors is responsible for ensuring that the bank's IRB system functions in accordance with the requirements and shall approve all significant changes to the IRB system, cf. Article 189 of the Capital Requirements Regulation. To fulfill this responsibility, the Board of Directors must process validation and internal audit reports at least annually. The Board of Directors must also ensure that the bank's control functions are sufficiently independent.

4. EBA GL 2017/16

2.4 Validation

The bank shall validate the models at least annually by comparing the estimates with observed defaults and losses, cf. Article 185 of the Capital Requirements Regulation. The comparisons shall be based on all available and relevant history, and should be made under equal assumptions for observed values and estimates. This means that observed values must be viewed in light of economic cycles, and model estimates that are set under the assumption of downturns should be adjusted to reflect the current economic situation. Model estimates should be validated without safety margins. In addition, the bank must validate the level of the safety margins and review the assumptions underlying the adjustment of model estimates to account for downturns.

Furthermore, the bank must assess whether the data basis is representative for the current portfolio, for example by comparing the impact of the most important explanatory variables in today's portfolio with the dataset used to estimate the model. If there are deviations, beyond the foreseeable, between average levels of risk parameters and the levels that formed the basis for the permission, the bank must upgrade the model estimates or apply to change the model, cf. point 2.1.1.

2.5 Application Requirements

The IRB system shall be applied in the credit processes and in risk management. The risk parameters used in credit processes and risk management may, however, differ from the parameters used in the capital requirement calculation if these differences are justified and documented. According to the guidelines for PD and LGD estimation, the bank may for internal purposes use estimates without downturn adjustments or special safety margins required to provide sound levels of capital requirements. The Financial Supervisory Authority clarifies that in such cases, the uncertainty in the estimates must be evident, and it must be evident what the regulatorily approved models say about the risk. Furthermore, internal estimates must be validated and followed up in the same way as the approved model.

3 Requirements for Probability of Default (PD) Models

It follows from Article 180 of the Capital Requirements Regulation that PD estimates shall reflect average default rates measured over a long period. The authorities must in their assessments and follow-up of IRB applications ensure that the data history covers foreseeable variation, cf. Article 46 in the European Commission's draft technical standard on model assessments.⁵ EBA clarifies in its guidelines for PD and LGD estimation, Chapter 5.3.4, that the data basis must include bad years, and that the bank must adjust PD estimates if it lacks data from bad years.

5. Cf. EBA's "Regulatory Technical Standards on assessment methodology for IRB approach":
   https://www.eba.europa.eu/regulation-and-policy/credit-risk/regulatory-technical-standards-on-assessment-methodology-for-irb-approach

In the Financial Supervisory Authority's assessment, at least 20 percent of the data basis must reflect an economic downturn corresponding to the banking crisis in the early 1990s. If the bank does not have sufficient data from a serious economic downturn, it must estimate default frequencies for downturns that are representative for the bank's portfolio, and weight these at least 20 percent in the estimation of long-term defaults. These estimates shall be added with safety margins to account for missing data basis and uncertainty in the estimates. For cyclical-sensitive industries, there is considerable uncertainty related to the weighting of good and bad years and to the relevance of data history. The bank must assess which industries and segments are cyclical-sensitive and should have special safety margins.

For loans with collateral in real estate, there is particular uncertainty related to data quality from the banking crisis. The uncertainty is exacerbated by the fact that the bank's other data basis is characterized by very good years in the Norwegian economy and high housing price growth. The bank must assume a default rate of at least 3.5 percent in a serious economic downturn. The level is calculated from observed problem loan shares during the banking crisis adjusted, among other things, for the relationship between the stock of problem loans and the share of new defaults. To account for weak data basis in segments with few defaults, the bank must add a safety margin to particularly low probabilities of default, such that all mortgages have a PD of at least 0.2 percent.

In the validation of PD models, discriminatory power and level shall be tested separately. To assess the models' predictive power over time, discriminatory power must be tested over a period of at least two years. The PD level in each risk class shall be compared with observed default rates both in the validation year and measured over a longer period.

4 Requirements for Loss Given Default (LGD) Models

4.1 Economic Downturns

Estimates for LGD and CfI shall reflect loss rates and utilization in downturns if these are assumed to be higher than in normal times, cf. Articles 181 and 182 of the Capital Requirements Regulation. According to these articles, EBA shall draw up a proposal for a technical standard that sets out criteria for identifying downturns. EBA assumes that downturn periods shall be identified based on real economic factors such as development in GDP and unemployment, defaults and losses on loans, as well as sector-specific indicators such as property prices, commodity prices, industrial indices, household debt, and disposable income.⁶ The bank shall use the most serious downturn period in the last 20 years, but shall look further back in time if necessary.

6. https://www.eba.europa.eu/sites/default/documents/files/documents/10180/2459703/3136b895-0dfb-454f-8984-beddb888b8cc/EBA%20BS%202018%20xxx%20%28Final%20draft%20RTS%20on%20economic%20downturn%29_final%20%28002%29.pdf?retry=1

The Financial Supervisory Authority considers the banking crisis in the early 1990s to be the last downturn period in the Norwegian economy associated with serious, coinciding decline in macroeconomic indicators such as GDP growth, unemployment, and property prices, combined with subsequent significant defaults and losses on loans in the banking sector. In more recent downturn periods, setbacks in GDP growth have been relatively short-lived, and increases in unemployment, declines in property prices, and defaults and losses in the banking sector have been moderate.

If the bank does not have sufficient data basis from downturns, the bank must estimate declines in collateral values and impact on other model parameters in a downturn or extend the data history back in time with estimates of how losses would have been in a downturn.

4.2 Requirements for Collateral

Where collateral is taken into account in the estimate for LGD, the bank must ensure that internal requirements for the treatment of collateral meet the requirements in Articles 205 to 217 of the Capital Requirements Regulation, cf. Article 181. The estimates shall not only take into account the value of the collateral, but also take into account the bank's ability to take over and realize the collateral. In the case of insufficient information on the collateral object, limited data basis for realization of comparable objects, or a less liquid market for the object type, values of collateral must be taken into account with caution, cf. EBA's guidelines for PD and LGD estimation (point 6.1.3). The Financial Supervisory Authority will point out that for buildings under construction and other collateral that are not completed, there can be considerable uncertainty related to any realization, especially in a downturn, cf. EBA's guidelines for PD and LGD estimation (point 6.2.3). The Financial Supervisory Authority believes that such collateral normally cannot be assigned value in the LGD estimates.

4.3 Calculation of Observed Loss Rates

To reflect economic loss, the recovered amount after default must be discounted back to the time of default, cf. point 6.3.1.1 in EBA's guidelines for PD and LGD estimation. If the bank can time the payments related to each individual exposure, it can use the interbank rate plus 5 percentage points as the discount rate. If there is uncertainty related to the data basis for payments, such that the bank uses accounting losses or other approximations to estimate the payments, the discount rate must be at least 9 percent.

Payments from collateral that are not taken into account in the LGD estimate shall be considered recovery of the unsecured part of the exposure. Such payments shall be documented and monitored separately.

Defaults that are not concluded shall be included in the data basis for estimation and validation of LGD models. It is stated in the guidelines for PD and LGD estimation (point 6.3.2.3) that further recovery for non-concluded defaults must be assessed conservatively. In the Financial Supervisory Authority's assessment, this implies that the bank cannot assume recovery, except for exposures awaiting recovery in accordance with the quarantine provision in point 2.3 of the Financial Supervisory Authority's Circular 4/2020 on the definition of default. Further recovery must be limited to the realization value of approved collateral, which must be assessed conservatively.

4.4 Validation of Estimates for the Performing Portfolio

In many cases, there will be systematic biases in LGD estimates between defaulted and performing exposures. In such cases, comparison of predicted and observed LGD for defaulted exposures will not provide sufficient basis for assessing LGD estimates for the performing portfolio. The bank must therefore compare average LGD values in the portfolio with average observed loss rates. If the data basis allows it, such comparisons can be made per industry, type of collateral object, or sub-portfolio.

Comparison of predicted and observed LGD for defaulted exposures is relevant at the exposure or LGD class level. The bank must place special emphasis in its assessments on the classes where the majority of the performing portfolio is located.

4.5 Reference Model for Mortgages

The data basis for estimating losses on mortgages in downturns is very weak. To ensure sound LGD estimates at the portfolio level, the Financial Supervisory Authority has created a reference model for LGD for mortgages, based on loan-to-value ratio and assumptions about price declines, recovery values, recovery rates, and costs in downturns. The average LGD for the bank's portfolio shall not be lower than the level the reference model gives for a portfolio with a corresponding distribution on loan-to-value ratio.

The reference model is given by the following formula:

LGD = (1 - tf)(a · (1 - gjv) + (1 - a) · k)

In the formula, tf stands for recovery share, a is the unsecured share of the exposure (a = max [0, 1 - (1 - LTV) / Loan-to-Value]), gjv is the share of recovery of the unsecured part of the loan, and k represents the loss share of the secured part of the exposure. The technical assumptions for the reference model are 10 percent recovery, 10 percent recovery rate, 5 percent loss share of the secured part, and 55 percent downward adjustment of realization value compared to market value.

The reference model is used for each bank's portfolio, including mortgages that have been transferred to mortgage credit institutions. The bank is expected to assess the calibration against the reference model in connection with the annual validation and carry out measures if the institution's model gives a lower LGD than the reference model. The assessment must be evident in the validation report.

The parameter for decline in collateral values may be adjusted upwards if housing prices grow faster over time than households' incomes. After a housing price decline, the height of the decline may be reduced. In forecasts and stress tests used in internal capital assessments (ICAAP, "Internal Capital Adequacy Assessment Process"), the bank may assume that changes in loan-to-value ratios resulting from a housing price decline are offset by reduced downward adjustment in the reference model, such that LGD levels are not affected.

5 Other Parameters Included in the Calculation of Risk Weights

For exposures that are not mass market exposures, the maturity and the size of the counterparty are also included in the calculation of the risk weight. These parameters are not included in the banks' modeling, but are set in accordance with provisions in the regulatory framework.

5.1 Maturity Parameter (M)

The maturity parameter, M, shall reflect that short-term loans may be less risky than long-term loans. Banks that have permission to use their own estimates for LGD shall, in accordance with Article 162 no. 2 of the Capital Requirements Regulation, set M based on the contractual repayment profile (letter a) or the maximum period the customer has to fulfill the obligation (letter f), limited down and up to respectively one and five years. For banks that do not have permission to use their own LGD estimates, M shall be 2.5 years, cf. Article 162 no. 1 of the Capital Requirements Regulation.⁷

7. [Link omitted in source]

When asked whether the institution's ability to extend contracts should be taken into account in the maturity calculation, EBA responded that contractual maturity may be used where the customer is not in a position to extend the contract or demand changes that extend the maturity, and that the regulatory framework does not require banks to take into account their own ability to extend in the maturity calculation.⁸ The Financial Supervisory Authority will point out that even if the customer cannot demand an extended maturity, the customer's situation may imply that the bank must effectively extend the maturity. This applies, for example, to loans with a significant payment at the end of the contractual period, where it is assumed that the loan is renewed or that the borrower obtains new financing. Obtaining new financing can be difficult, especially if the risk is high or has increased since the previous approval. There is thus a risk that the customer will not be able to refinance at the contractual time, so that the loan must be extended. Corresponding conditions apply to credit lines that are renewed regularly, since this financing is assumed to be central to the borrower's further operations and servicing capacity.

8. [Link omitted in source]

In cases where it is practically difficult to terminate the loan relationship at the expiration of the contract, the bank's actual risk exposure will last beyond the contractual maturity. This can typically apply to financing of real estate and other long-term investments. On this basis, the bank should consider setting M to the maximum value of five years for fina...