EBA欧洲银行-C2-Duellman-Discussion_7页_144kb

Summary of "Estimating the distribution of total default losses on the Spanish financial system"

Core Content

This document presents a study on estimating the distribution of total default losses in the Spanish financial system, with a focus on systemic risk measurement and its implications for policy. The authors, R. García-Céspedes and M. Moreno, propose an extension of the common default mode model by incorporating random recoveries and a market valuation model. The discussant, Klaus Dullmann, provides critical insights and raises several technical and policy-related issues.

Main Views and Key Information

Economic Perspective

• Systemic Risk Measurement: The paper measures systemic risk by analyzing the total credit risk loss distribution of the Spanish financial system and allocating this risk to individual banks.
• Model Comparisons: Results vary significantly depending on the model used, such as the difference between a constant recovery rate assumption and a random recovery rate assumption, as well as between the simple default mode model and the market valuation model.
• Business Cycle Variability: The loss distribution is analyzed over the business cycle, highlighting the importance of understanding how risk evolves with economic conditions.
• Key Drivers of Uncertainty: Loss estimates are more sensitive to the loss given default (LGD) than to the probability of default (PD), making LGD a crucial factor in risk assessment.

Technical Perspective

• Methodology: The study employs the Important Sampling (IS) method, as developed by Glasserman and Li, to improve the efficiency of risk estimation.
• Model Extensions: The IS method is extended to account for correlated random recoveries and market valuation, enhancing the model's realism.
• Computational Effort: Introducing random LGDs significantly increases the computational burden, which is a key limitation.
• Confidence Intervals: The IS model produces thinner confidence intervals than the standard Monte Carlo (MC) method when the same computational effort is applied.

General Remarks

• Risk Allocation: The marginal Expected Shortfall (ES) approach is appreciated for its ability to allocate systemic risk to individual banks without creating unintended incentives.
• Model Validity: The results are based on a state-of-the-art IS method, but their general validity is affected by several debatable assumptions.
• Assumptions and Limitations:
  • The correlation between macroeconomic factors is assumed to equal GDP correlations, which may not hold in stress periods.
  • The average maturity of all assets in the portfolio is assumed to be 3 years, which could be too simplistic.
  • The market model's loss estimates are highly sensitive to non-linear dependencies.
  • Gaussian assumptions are too restrictive and fail to capture tail dependence.
  • The constant correlation over time assumption is not realistic, as correlations change in stress periods.
  • Correlation may not be sufficient to capture dependencies across banks.

Policy Issues

Policy Issues (1)

• Government Support: The role of government support in the financial system is not fully explored, despite being mentioned.
• Sovereign Risk Impact: Sovereign rating downgrades negatively affect equity prices. The difference between Moody's standalone and with-government-support ratings could provide valuable insights.
• Cyclicality of Risk: The cyclicality of risk contributions is an area for further exploration, as it could inform more nuanced regulatory responses.
• Stochastic LGD: The model could offer new insights by integrating a stochastic LGD effect.
• Government-Banking Link: The interplay between the government and the banking sector is an interesting area for further research.

Policy Issues (2)

• Identification of SIFIs: The paper claims to provide a "basic tool to identify SIFIs" but does not follow through with this objective.
• Size vs. Interconnectedness: Some results suggest that size alone is not a sufficient indicator of systemic importance (SIFIness), contradicting literature that supports this view.
• Interconnectedness and Macro Variables: The question of whether interconnectedness is adequately captured by macroeconomic variables remains open.
• Tail Dependence and Non-Linearity: The Gaussian model setup excludes tail dependence, which is important for risk modeling.
• Time-Variant Correlations: The assumption of constant correlations over time is not realistic, and the paper could benefit from comparing its approach to the Basel Committee's indicator-based methodology for measuring SIFIness.

Technical Remarks

• Data Sources: The correlation between LGD and macroeconomic factors is derived from FDIC data, which may not be the most appropriate source for the Spanish context.
• Country-Specific Considerations: The legal framework of a country, particularly the power balance between borrowers and lenders, heavily influences the recovery rate distribution. Using Spanish data would be more appropriate.
• Model Simplifications: The paper starts with a multi-factor portfolio model but later relies primarily on the Basel single factor setup, which may limit the model's generality.
• Presentation and Structure: The focus of the paper is unclear, with the economic model and IS method both receiving attention. The more technical aspects of IS could be moved to an appendix or a separate paper. Additionally, figures in the appendix are often small and difficult to read, affecting clarity.

Conclusion

The study contributes to the understanding of systemic risk in the Spanish financial system by introducing a more realistic model that accounts for random LGDs and market valuation. However, the paper faces challenges in terms of model assumptions, data relevance, and the clarity of its policy implications. Further research is needed to refine the model and better capture the complexities of systemic risk and interconnectedness.