A) Supervisory process for model validation
A key question handled is - Has delegation/division of tasks between supervisors been applied in practice? Which are the most relevant tasks to be delegated/allocated to home and host supervisors?
Summary of the observation - This is effective where the supervisors collaborate to perform the task they are best placed to conduct. The key point being that delegation needs to work both ways i.e. Host delegating to Home and Home delegating to Host. Generally speaking, home supervisors should inform hosts of centrally performed model reviews, and hosts are responsible for the supervision of local model implementation issues.
More specifically the home supervisor being responsible for the review of the internal governance of model validation (rating process, control environment, stress testing, worldwide model implementation, internal audit etc.), including the review of a sample of centrally developed models, and the examination of the adequacy of the related IT environment.
In the case of centrally developed models that are applied across the banking group (group-wide models), the home supervisor leads the approval work. Where models applied across the banking group are developed/managed/enhanced at a subsidiary banking entity, the home and host supervisors jointly carry out the approval work. Host supervisors are responsible for the assessment of specific local requirements. Host supervisors are generally responsible for the examination/assessment of the implementation of the rating systems developed by the local subsidiary.
Where a model is applied in several countries, the host supervisors of the subsidiaries where the main developmental work has been performed are responsible in consultation with the home supervisor.
- Local and central models A key question addressed was -
- How are local and central models defined? Are differences driven by specificities and/or organizational arrangements of banking groups?
Summary - Here broadly two main approaches were seen to be prevalent on the basis of the experience gained so far.
- The first one uses geographical specificities as the driver for separation and examines whether a particular rating system requires any local aspects to be taken into account. If this is the case,
the model is defined as local; if local aspects do not have a role, the model is considered to be central.
- The second approach focuses on the division of tasks within the banking group. Models developed, tested and validated by a central unit and used on a group-wide basis are defined as central models while models developed, tested and validated locally, and used in one or more entities, are considered to be local models.
The other issues handled in this section being:
- Portfolio classification - Is it possible for banks to adopt the IRB classification for exposures subject to the Standardised approach on a temporary (roll-out) and/or permanent (permanent partial use) basis?
- Use test for new models - How can the use test requirement be applied practically for new models?
- Supervisory assessment of group-wide models - What is the role of home and host supervisors in the validation of central models?
- Language of IRB/AMA application - In which language do banks have to submit the application to use internal models to the home supervisor? Is the approach consistent across banking groups?
B) Pillar 1 technical issues. In pillar 2 key topics addressed are around default defintion and downturn LGD.
- Definition of default - Which definition of default (DoD) is applied in practice across a cross-border banking group? How is the default of individual entities related to the default of groups?
Summary - Some groups use a different DoD for the consolidated calculation, whereas others use a single DoD across all the group’s entities. However, the first approach is not perceived to be a major problem.
- Downturn LGD - What are the banks’ methodologies to estimate the “downturn LGD”, as requested by the CRD? What is the supervisory approach?
Summary - Different approaches are explianed via different real life examples. Example 1 is around UK FSA. The UK Financial Services Authority has published a paper in which the results of an empirical exercise on downturn LGD estimates are presented1; data were gathered from 12 firms. The main outcome at this early stage is that the degree of variation of the two key downturn parameters (i.e. reduction in property value and probability of possession given default) is quite large. Therefore, some “reference values” for these variables are provided (-40% and 35% respectively), which could be used by each bank for discussion with supervisors. As soon as firms improve their estimation techniques, and the available data increase, FSA expects the thinking on this topic to evolve.
Example 2 is around Spain. The Bank of Spain has published a paper2 in which the requirements for downturn LGD for residential mortgages are presented. The document defines the following concepts: realised, long-run average and downturn LGD; it also requires a minimum segmentation in the estimation process based on risk drivers; and finally, it identifies the estimation procedures accepted.
Other issues covered in this section being-
- Estimation and validation of risk parameters in “low-default-portfolios” - What are the approaches followed by banks to estimate and validate risk parameters for “low default portfolios”? What is the supervisory approach?
- Project finance - What are the banks’ methodologies for estimating risk parameters for Project Finance? What is the supervisory approach?
C) Pillar 2 issues
- Scope of application of ICAAP - What is the scope of application of ICAAP?
- Requirements imposed for ICAAP - What are the requirements that banks have to follow for ICAAP?
Acknowledgment and Source: CEBS Paper on Range of Practices on some Basel II implementation issues.
