EBA欧洲银行-EBA-Report-on-IRB-modelling-practices_157页_3mb

EBA Report on IRB Modelling Practices Summary

Core Content

This report summarizes the internal ratings-based (IRB) modelling practices of 102 institutions from 22 EU Member States, based on the responses to the EBA's IRB survey conducted in the context of the Guidelines (GLs) on Probability of Default (PD), Loss Given Default (LGD), and the treatment of defaulted exposures. The report provides an impact assessment of the GLs on these risk parameters and outlines the key findings and policy choices made in the final guidelines.

Key Information

• Survey Scope: The survey covers both high-default and low-default portfolios, with a focus on PD and LGD models. It includes all COREP exposure classes, although some are better represented than others.
• Participation: 102 institutions participated in the survey, representing 64% of EU institutions' total credit risk-weighted exposures.
• Model Types: A total of 252 PD models and 202 LGD models were reported. The median institution completed 3 PD and 2 LGD models.
• Coverage: PD models cover 17% of all PD models used by institutions, while LGD models cover 20% of all LGD models used.

Main Views and Findings

1. PD Models

• Rating Systems: The survey revealed the use of both continuous and discrete rating scales. The majority of institutions use a discrete rating scale.
• Calibration: Institutions use different calibration methods, and the final GLs include a list of allowed calibration types under the CRR. Calibration is necessary to ensure that PD estimates align with the long-run average default rate (DR).
• Default Rates: Institutions calculate one-year DRs at different frequencies. The CP proposed a quarterly frequency, which was already in use by 45% of PD models, with higher usage in retail exposures.
• Margin of Conservatism (MoC): MoC is applied in PD models, with some institutions using it before or after calibration. The final GLs require that institutions must justify any MoC application.
• Model Changes: Around 54% of PD models will need to change their practice to align with the final GLs, particularly regarding the frequency of DR calculation.

2. LGD Models

• Recovery from Collateral: The survey showed significant variation in how institutions treat collateral in LGD estimation. Some institutions do not consider certain types of collateral, such as guarantees and credit derivatives.
• Economic Loss Calculation: Most institutions assume that economic loss for a cured case is zero, which is considered imprudent. The final GLs retained the approach of discounting additional recovery cash flows.
• Discounting Rate: The methodologies for determining the discounting rate vary, with some using the funding rate or risk-free rate plus add-on. The GLs provide clarity on the specification of the discounting rate.
• Historical Observation Period: There is considerable heterogeneity in how institutions define and use the historical observation period for LGD estimation. The GLs propose a benchmark based on the average of one-year DRs over the most recent five years and the whole available observation period.
• Incomplete Recovery Processes: The GLs clarify that incomplete recovery processes should be treated as closed after a defined maximum period, and institutions must justify any deviation from this.
• Downturn Adjustment: The GLs require that LGD estimates should reflect economic downturns. The methodology for downturn adjustment is specified, and institutions must consider the impact of economic conditions on their LGD models.

3. ELBE Estimation

• ELBE (Expected Loss Best Estimate): The survey provided insights into how institutions estimate ELBE, including the use of specific credit risk adjustments (SCRA) and the inclusion of economic conditions in the estimation process.
• Model Changes: The final GLs require that institutions re-estimate existing models to reflect current economic conditions. This will have an impact on the capital requirements of institutions.

Structure of the Report

1. Background and Rationale

• The report is based on the IRB survey responses and provides an impact assessment of the GLs on PD, LGD, and the treatment of defaulted exposures.

2. Introduction

• Details on the survey sample, PD and LGD estimates, and data quality are discussed.

3. General Estimation Requirements

• Principles for specifying the range of application of rating systems, data requirements, and the use of MoC are outlined.

4. PD Models

• Characteristics of the survey sample, data requirements, default rates, and calibration methods are detailed.

5. LGD Models

• Focuses on recovery from collateral, inclusion of collateral, economic loss calculation, discounting rate, and downturn adjustment.

6. Estimation of Risk Parameters for Defaulted Exposures

• Covers the estimation of LGD in-default and ELBE, including the use of SCRA and the impact of economic conditions.

7. Application of Risk Parameters

• Discusses how PD and LGD estimates are applied in practice.

8. Review of Estimates

• Provides an overview of the review process for PD and LGD estimates.

Key Policy Choices

• One-Year DR Frequency: Institutions should evaluate observed one-year DRs at least quarterly.
• Calibration: Institutions must perform calibration tests at the relevant calibration segment level.
• Economic Loss Calculation: Artificial cash flows (unpaid late fees and capitalised interest) should not increase the economic loss or amount outstanding at the moment of default.
• Discounting Rate: The discounting rate should be specified at the appropriate level of granularity.
• Historical Observation Period: The long-run average DR should be calculated based on a representative historical observation period, with adjustments if necessary.
• Downturn Adjustment: LGD estimates must reflect economic downturns, and the methodology for this is clarified in the GLs.

Conclusion

The GLs are based on the current practices of surveyed institutions, aiming to provide a neutral impact on the models. However, the final impact on capital requirements will only be known after model redevelopment and recalibration. The report emphasizes the need for monitoring the impact of the GLs due to the complexity and variability of internal models.