2014年-EBA欧洲银行管理局_20131217_Report_on_variability_of_Market_RWA_103页_2mb

Summary of the Report on Variability of Risk Weighted Assets for Market Risk Portfolios

Core Content

This report presents the findings from the 2013 Hypothetical Portfolio Exercise (HPE), conducted by the European Banking Authority (EBA) in parallel with the Basel Committee's SIGTB exercise. The goal was to assess variability in market Risk Weighted Assets (RWA) generated by banks' internal models, focusing on Value at Risk (VaR), Stressed Value at Risk (SVaR), and Incremental Risk Charge (IRC) models. The report also includes an analysis of correlation trading portfolios under the All Price Risk (APR) model.

Main Features of the 2013 HPE

• Number of Portfolios: The exercise included 35 general portfolios (28 individual and 7 aggregated) and 7 correlation trading portfolios.
• Asset Classes: The portfolios covered Equity, Interest Rates, Foreign Exchange (FX), Commodities, and Credit.
• Data Collection: Participating banks provided an Initial Market Valuation (IMV) as of 10 May 2013 and were requested to deliver a one-year Profit and Loss (P&L) vector for VaR calculations.
• Data Scope: The analysis focused on banks using Historical Simulation (HS) for VaR, with data collected from 13 EU jurisdictions and 17 banks.

Key Findings

Variability in VaR

• The overall average variability for VaR is 33%.
• Variability decreases significantly when using a homogenised VaR metric:
  • Individual portfolios: ~30% reduction.
  • Aggregated portfolios: ~50% reduction.
• However, variability increases in some portfolios:
  • 4 portfolios (11.4%) show increased variability in 'VaR Alt'.
  • 10 portfolios (28.6%) show increased variability in 'VaR Comp'.
• P&L Analysis: For HS banks, the main driver of variability in VaR capital outcomes is the volatility in the P&L, not the correlation between banks' P&L.

Variability in SVaR

• The average variability for SVaR is 44%, which is higher than VaR.
• The stressed period used for SVaR is not normalised, making variability analysis less reliable.
• Variability is not consistently observed across all portfolios.

Variability in IRC

• The average variability for IRC is 77%, significantly higher than VaR.
• IRC variability is particularly high for bespoke portfolios (25–28), with some showing up to 81%.
• For more 'plain vanilla' portfolios, the variability decreases to 55%.
• For corporate risk portfolios, the variability further reduces to 38%, which is comparable to regulatory VaR.

Variability in APR

• The average variability for APR is 81%, the highest among all metrics.
• The standardised APR calculation shows even higher variability at 133%.
• Most banks are reducing exposure to Correlation Trading Portfolios (CTP), which are typically in run-down mode.

Diversification Benefit (DB)

• Larger aggregated portfolios exhibit greater DB than smaller ones.
• Alt VaR shows lower dispersion in DB than regulatory VaR.
• Stressed VaR exhibits higher dispersion in DB than regulatory VaR.
• IRC DB has a median level of 41–42%, significantly lower than VaR and SVaR (70–80%), but with higher dispersion than regulatory VaR (5% vs 30%).

Dispersion in Capital Outcome

• The most 'inclusive' portfolios (29 and 30) show lower variability (28–26%) compared to individual and aggregated portfolios.
• Variability is not driven by complexity on an aggregated basis.
• Regulatory add-ons did not influence variability in this exercise.
• Applying 3 multipliers instead of regulatory ones resulted in the same max-min range but increased the variation coefficient due to lower average capital.

Main Drivers of Variability

• Modelling Choices: Banks have different approaches to modelling risk factors, such as basis risk in CDS vs bonds, index vs components, and forward equity volatility.
• Supervisory Actions: Regulatory differences in the treatment of sovereign exposures (e.g., exclusion from IRC) contribute to variability.
• Data Periods: The non-homogenised stressed period for SVaR introduces variability that may not be due to model differences alone.

Limitations and Notes

• The sample of banks is limited (maximum of 10), which affects the reliability of conclusions.
• Some portfolios were not provided by banks due to model limitations or internal authorisation issues.
• Outlier values were excluded from the analysis, especially for VaR and SVaR, to ensure data consistency.
• For the APR portfolios, no data points were eliminated, indicating a more consistent reporting process.

Conclusion

The report highlights that variability in market RWA is influenced by both regulatory treatment and banks' internal models. While VaR shows the lowest variability, IRC and APR exhibit higher dispersion. The analysis of P&L data suggests that volatility in actual performance is a more significant driver than model-based correlation. The study also demonstrates that aggregated portfolios show less variability due to diversification effects, and that regulatory add-ons do not significantly affect the observed variability.