Barclays 2008 Annual Report Download - page 87

1
Business review
Barclays PLC Annual Report 2008 85
It is Barclays philosophy to embed Basel II models as extensively as
possible in the portfolio management process. This is an ongoing initiative
and we expect greater convergence over time. However, in some cases
there are sound business reasons for having different models for capital
allocations and internal processes.
EAD models within retail portfolios are split into two main
methodological categories. The less complex models derive product level
credit conversion factors (CCFs) from historical balance migrations; these
are frequently further segmented at a delinquency bucket level. The most
sophisticated EAD models are behavioural based, determining customer
level CCFs from characteristics of the individual facility.
Retail LGD models are built using bespoke methods chosen to best
model the observed recovery process. In a number of secured portfolios,
structural models are often used which parameterise the LGD drivers
giving models which can easily be updated to reflect current market
trends. Models based on historical cash collected curves are often utilised
in portfolios where recoveries are not based on the recovery of a single
source of collateral. Finally, in some instances regression techniques are
used to generate predicted LGDs based on account characteristics. In all
instances bespoke country level factors are derived to discount recovery
flows to the point of default. For capital calculations, customised
economic downturn adjustments are made to adjust losses to stressed
conditions.
Most retail models within Barclays are built in-house, although
occasionally external consultants will be contracted to build models on
behalf of the businesses. Whilst most models are statistically or
empirically derived, some expert lender models (similar to those described
above in the wholesale context) are used, particularly where data
limitations preclude a more sophisticated approach.
Where models are used in the calculation of regulatory capital, the
definition of default is in line with the regulatory definition of default
requirements i.e. for UK portfolios the default definition is 180 days past
due whilst international regulators may have different rules. In some cases,
for models not used in regulatory capital calculations, in order to maximise
model suitability, different default definitions are used. However, in all
cases EAD and LGD models are appropriately aligned.
The control mechanisms for the rating system
Each of the business risk teams is responsible for the design, oversight and
performance of the individual credit rating models – PD, LGD and EAD –
that comprise the credit rating system for a particular customer within each
asset class. Group-wide standards in each of these areas are set by Group
Risk and are governed through a series of committees with responsibility
for oversight, modelling and credit measurement methodologies.
Through their day-to-day activities, key senior management in Group
Credit Risk, the businesses and the business risk teams have a good
understanding of the operation and design of the rating systems used.
For example:
– The respective Business Risk Heads or equivalents are responsible for
supplying a robust rating system.
– The Group Risk Director, Credit Risk Director and Wholesale and Retail
Credit Risk Directors are required to understand the operation and
design of the rating system used to assess and manage credit risk in
order to carry out their responsibilities effectively. This extends to the
Business CEOs, Business Risk Directors and the Commercial/ Managing
Directors or equivalent.
In addition, Group Model Risk Policy requires that all models be validated
as part of the model build (see page 79). This is an iterative process that is
carried out by the model owner. Additionally, a formal independent review
is carried out after each model is built to check that it is robust, meets all
internal and external standards and is documented appropriately. These
reviews must be documented and conducted by personnel who are
independent of those involved in the model-building process. The results
of the review are required to be signed off by an appropriate authority.
In addition to the independent review, post implementation and
annual reviews take place for each model. These reviews are designed to
ensure compliance with policy requirements such as:
– integration of models into the business process
– compliance with the model risk policy
– continuation of a robust governance process around model data inputs
and use of outputs
Model performance is monitored regularly; frequency of monitoring is
monthly for those models that are applicable to higher volume or volatile
portfolios, and quarterly for lower volume or less volatile portfolios. Model
monitoring includes coverage of the following characteristics: utility,
stability, efficiency, accuracy, portfolio and data.
Model owners set performance ranges and define appropriate actions
for their models. As part of the regular monitoring, the performance of the
models is compared with these operational ranges. If breaches occur, the
model owner reports these to the approval body appropriate for the
materiality of the model. The model approver is responsible for ensuring
completion of the defined action, which may ultimately be a complete
rebuild of the model.