This report presents the results of the 2025 market risk benchmarking exercise pursuant to Article 78 of the Capital Requirements Directive. It summarizes the conclusions drawn from a hypothetical portfolio exercise conducted by the EBA, aimed at assessing the variability of risk-weighted assets. The results include analyses of risk measures, assessments by competent authorities, and recommendations for the future.
This report presents the results of the 2025 benchmarking exercise on market risk (Market Risk Benchmarking Exercise - IMA), conducted by the European Banking Authority (EBA) in 2026 in accordance with Article 78 of the Capital Requirements Directive (CRD) and the associated regulatory and implementing technical standards. The exercise focuses on assessing the variability of risk-weighted assets (RWA) for market risk calculated by internal models of European banks. It covers 43 banks from 13 European jurisdictions, with data on 105 instruments recombined into 168 market portfolios, including the main asset classes: equities (EQ), interest rates (IR), foreign exchange (FX), commodities (CO), credit spreads (CS), and correlation trading products (CTP). The exercise took place in early 2025, with data collection until June 2025. The scope includes initial market valuation (IMV), risk measures VaR, sVaR, IRC, and APR according to approved internal models (IMA). (p. 1-20)
- Subject: The report analyzes the results of the 2025 benchmarking exercise on internal market risk models of European banks, aiming to measure the variability of calculated capital requirements.
- Importance: The variability of RWAs affects comparability and robustness of internal models, essential for prudential supervision and financial stability.
- Key findings:
- The dispersion of initial market valuations (IMV) has generally improved compared to 2024, with low IQDs for equities (2%), interest rates (3%), and credit spreads (1%). Foreign exchange (FX) saw a strong reduction in dispersion to 3% thanks to better clarification of instructions on FX forwards. Commodities (CO) remain more dispersed (14%) due to a small number of instruments and limited submissions.
- The variability of risk measures is stable or slightly improved: average VaR IQD at 14%, sVaR more dispersed at 27%, IRC very variable at 43%. These levels are among the lowest since the start of the exercises.
- The higher dispersion of sVaR is partly explained by the use of non-harmonized stress periods among banks. An analysis over a homogeneous stress period shows a reduction in this dispersion.
- Variability factors include methodological choices (majority HS modeling, lookback period choice, data weighting), supervisory actions (VaR multipliers above the regulatory minimum in 71% of banks), and model approval levels.
- The quality of submitted data has improved, although some minor errors persist, notably on "clean price" vs "dirty price" valuation for certain bond instruments.
- Competent authorities (CAs) identified and addressed most discrepancies and outliers, with enhanced monitoring for unexplained cases.
- Conclusions: The exercise confirms continuous improvement in the quality and consistency of internal market risk models, with controlled RWA variability. Residual discrepancies mainly relate to methodological choices and heterogeneous supervisory practices.
- Recommendations: Continue improving data quality and instructions, strengthen monitoring of unexplained outliers, harmonize stress periods for sVaR, and prepare for integration of future regulatory changes (notably FRTB and ASA). (p. 8-15)
- Context: Since 2016, the EBA regularly conducts benchmarking exercises on internal market risk models of European banks, in line with regulatory requirements (CRD, CRR). These exercises aim to assess the dispersion of results produced by internal models, identify sources of variability, and improve comparability and model quality.
- Objectives:
- Define a homogeneous set of hypothetical instruments and portfolios (HPE) for all participants.
- Collect initial valuation data and risk measures (VaR, sVaR, IRC, APR).
- Analyze result variability, detect outliers, and understand causes.
- Provide competent authorities with tools and analyses to supervise internal models.
- Gather authorities’ feedback on actions planned to correct discrepancies.
- Limitations:
- Hypothetical portfolios differ from real portfolios in size and composition.
- Complementary supervisory measures (add-ons, diversification restrictions) are not fully assessable in this exercise.
- sVaR variability is affected by the absence of a common stress period.
- Results from banks with partial approvals are treated separately to avoid bias.
- Scope: 43 European banks with approved internal models, 105 instruments, 168 portfolios, all major asset classes. (p. 16-33)
Definition and data collection:
- Hypothetical portfolios composed of 105 instruments recombined into 168 portfolios, covering EQ, IR, FX, CO, CS, and CTP, with aggregated portfolios to measure diversification.
- Data collected via competent authorities, with initial valuation as of 6 February 2025 and risk measures calculated over the submission period (until June 2025).
- Precise instructions to exclude funding costs and counterparty risk.
Data quality:
- Notable improvement in data quality compared to 2024, especially for FX due to clarifications on forwards.
- Persistence of some minor errors, notably on "clean price" vs "dirty price" valuation for certain bond instruments.
- Anomaly control processes and resubmissions to correct errors.
Variability of initial valuations (IMV):
- Low average IQD for EQ (2%), IR (3%), CS (1%), FX (3%), higher for CO (14%) due to a small number of instruments and limited submissions.
- Clustering mainly observed on certain specific instruments (e.g., instrument 310).
Variability of risk measures:
- VaR: stable average IQD at 14%, with lower dispersion for FX and IR, higher for EQ and CO.
- sVaR: higher IQD at 27%, impacted by diversity of stress periods used by banks.
- IRC: very high IQD at 43%, reflecting complexity and varied methodological choices.
- P&L VaR and VaR HS show lower dispersion, confirming method impact on variability.
- Average sVaR/VaR ratio around 2.12, with atypical cases indicating potential errors.
Variability factors:
- Methodologies: 67% of banks use historical simulation (HS), but smoothing choices, lookback period (mostly 1 year), and data weighting influence dispersion.
- Supervisory actions: 71% of banks have a total multiplier above the regulatory minimum of 3, with an average of 3.56, including add-ons for backtesting and other charges.
- Model approval level (full or partial) impacts dispersion.
- Non-harmonized stress period for sVaR increases variability.
Diversification:
- Aggregated portfolios show diversification benefits, with lower dispersion than individual portfolios.
Historical evolution and outlook:
- Progressive improvement in data quality and reduction of dispersion since 2019, with methodological adjustments and regulatory clarifications.
- Postponement of FRTB deployment to 2025 led to maintaining the 2024 format for the 2025 exercise.
- In 2026, focus planned on ASA (alternative standardized approach) data collection in preparation for FRTB.
- Planning of a 2027 exercise integrating new FRTB requirements and extending ASA collection.
Competent authorities’ assessment:
- CAs analyzed results and identified most causes of discrepancies.
- Enhanced monitoring of unexplained cases.
- Corrective actions ongoing or planned to reduce unjustified variability.
- Emphasis on continuous supervision and practice improvement.
Methodological limitations:
- Exercises based on hypothetical portfolios, limiting representativeness for real portfolios.
- Difficulty fully assessing effects of complementary supervisory measures.
- Variability linked to diversity of approaches and modeling practices.
- Need to treat banks with partial approvals separately to avoid bias.
Established facts:
- Participation of 43 European banks with homogeneous data on 105 instruments and 168 portfolios.
- Improvement in IMV data quality, with low IQDs for most asset classes except CO.
- Average dispersion of VaR at 14%, sVaR at 27%, IRC at 43%.
- Average sVaR/VaR ratio around 2.12.
- 71% of banks apply a regulatory multiplier above 3.
- Majority HS method (67%) with varied choices impacting dispersion.
Assumptions:
- Observed variability is due to a combination of allowed regulatory choices, internal practices, and supervisory measures.
- Absence of a common stress period for sVaR contributes to increased dispersion.
- Hypothetical portfolios are representative but do not capture all complexities of real portfolios.
Interpretations:
- Reduction in IMV and VaR dispersion reflects better understanding and application of instructions.
- High IRC variability highlights complexity and sensitivity to methodological choices.
- Supervisory actions (multipliers, add-ons) modify capital levels but do not significantly increase relative dispersion.
- Lower dispersion of HS or MC-based measures suggests methodological homogeneity reduces variability.
Uncertainties:
- Some minor undetected errors may still influence results.
- Exact impact of complementary supervisory measures on overall variability remains difficult to quantify.
- Effect of partial model approvals on dispersion requires ongoing monitoring.
- Regulatory evolution (FRTB, ASA) could substantially change future practices and results.
- The 2025 exercise confirms continuous improvement in data quality and an overall reduction in variability of market risk measures calculated by internal models of European banks.
- Residual discrepancies mainly relate to methodological choices, heterogeneous supervisory practices, and differences in stress periods used.
- Competent authorities identified most causes of discrepancies and initiated corrective actions, including enhanced monitoring of unexplained cases.
- Data quality remains a key issue, requiring constant attention and systematic use of clarification tools (Q&A) to avoid misunderstandings.
- Standardization of stress periods for sVaR calculation is recommended to reduce dispersion.
- The postponement of FRTB deployment to 2025 led to maintaining the current format, but preparation for integration of new ASA and AIMA standards is ongoing.
- For 2026, data collection will focus on the alternative standardized approach (ASA) to reduce burden and prepare for regulatory transition.
- The future 2027 exercise will incorporate FRTB requirements and extend ASA collection to a larger number of institutions.
- The EBA plans to review instrument and portfolio design to balance simplicity and data richness.
- Authorities must maintain active supervision and dialogue with banks to ensure consistency and robustness of internal models.
Synthesis note written from the full document by DataSAI Academy. This note comes from the scientific library of the DataSAI Academy, open to all.