EBA欧洲银行-B2-M.-Moscadelli-Discussion_7页_430kb

3rd EBA Policy Research Workshop Summary: "How to Measure the Riskiness of Banks"

Core Content

The 3rd EBA Policy Research Workshop, held on 25 November 2014, focused on the measurement of operational risk in banks, particularly through the analysis of a comprehensive loss database containing over 42,000 observations. The discussion centered around the paper "Evidence, Estimates and Extreme Values from Austria" by Stefan Kerbl (OENB), presented by Marco Moscadelli from the Bank of Italy. The workshop aimed to explore the current methodologies and suggest improvements for better operational risk assessment.

Main Points and Findings

Cross-Section Analysis

• Mean Loss and Bank Size: There is a weak relationship between the mean loss and the size of the bank, especially when the size remains relatively stable.
• Net Interest Income as a Proxy: Net Interest Income is identified as a strong proxy for the total loss amount, ranking second after Own funds requirements for operational risk. This is supported by the Kendall's tau rank correlation coefficient.

Severity Distribution

• Variation in Business Line Riskiness: There is significant variation in the riskiness of Business Lines (BLs) at the 99.9% regulatory percentile. This variation suggests that the current TSA (Top-down Approach) coefficients structure is inadequate.
• Event Types vs. Business Lines: Banks are more capable of categorizing events into Event Types (ETs) rather than Business Lines, indicating that ETs may provide a more accurate representation of operational risk severity.

Suggested Improvements and Areas for Further Work

Cross Section Analysis

1. Regression Analyses: Further regression analyses are recommended to validate or challenge the results in Table 4, especially regarding the relationship between Total Loss Amount and financial indicators.
2. Nonlinear Relationships: The analysis should consider nonlinear relationships and use accuracy measures such as R², Adjusted R², AIC, and BIC to evaluate model performance.
3. Business Indicator: A new indicator called the Business Indicator has been shown to perform well, suggesting that it could be a better measure than traditional financial indicators.

Parametric Distribution

1. Inclusion of "Other" Business Line: The "Other" Business Line (Corporate item) should be included in the analysis, as it often encompasses the most severe losses not clearly attributable to standard Basel BLs.

Generalised Pareto Distribution (GPD)

1. Threshold Selection: The method for selecting the threshold in the GPD should be improved. Current methods often result in thresholds lower than the mean, which can hinder extrapolation beyond high confidence levels.
2. Alternative Estimation Methods: Instead of Maximum Likelihood Estimation (MLE), methods such as PWM (Probability Weighted Moments) or penalised/constrained/weighted MLE could be considered to improve stability.
3. Tail Behavior Assessment: The use of the "ln(p-sigma)" plot proposed by Dutta & Perry is suggested to assess tail behavior in each BL and ET. A flattening of the plot at higher or lower percentile levels indicates potential issues with tail smoothing or data behavior.

Cross Validation

1. Log-Likelihood Clarification: It should be clarified how the log-likelihood of the g-and-h distribution was calculated, given that the density function is complex and difficult to handle.
2. Champernowne Approach: The density function used in the Champernowne approach for log-likelihood calculation needs clarification. If based on the first step, the training and validation sets may not align conceptually.
3. Alternative Validation Methods: If cross-validation proves infeasible, alternative methods such as graphical (e.g., q-q plots) or numerical (e.g., bootstrapping) approaches should be considered to evaluate the fit of proposed distributions.

References and Context

The findings align with the Basel Committee on Banking Supervision's consultative document on revisions to the simpler approaches for operational risk, published in October 2014 and issued for comment by 6 January 2015. This document highlights the need for better calibration of current frameworks and the importance of using more accurate indicators to estimate operational risk capital requirements.

Conclusion

The workshop emphasized the importance of refining operational risk measurement methods, particularly through more detailed cross-sectional and severity distribution analyses. It also called for improved parametric modeling techniques and validation strategies to ensure more accurate and reliable risk assessments across different business lines and event types.