Barclays 2007 Annual Report Download - page 95

1
Business review
Barclays PLC Annual Report 2007 93
Expert lender models are used for parts of the portfolio where the
risk drivers are specific to a particular counterparty, but where there
is insufficient data to support the construction of a statistical model.
These models utilise the knowledge of credit experts that have in depth
experience of the specific customer type being modelled. Where possible,
the characteristics identified by the expert lenders for use in these models
are linked during the modelling process to the Merton framework. This
linkage ensures that the model is intuitive and that there is some economic
rationale for the default process that is being captured by the model.
For any of the portfolios where we have a low number of default
observations we adopt specific rules to ensure that the calibration of
the model meets the Basel II and FSA criteria for conservatism. We have
developed our own internal policy which describes specific criteria for
the use of parametric (e.g. Pluto Tasche) and non-parametric low default
portfolio calibration techniques.
Statistical models such as behavioural and application scorecards are
used for our high volume portfolios such as SME. The model builds
typically incorporate the use of large amounts of internal data, combined
with supplemental data from external data suppliers. Where external data
is sourced to validate or enhance internally-held data as part of the risk
assessment process or to support model development and BAU operation,
a similar approach is adopted towards ensuring data quality to that
applied to the management of internal data. This entails adherence to the
Group’s procurement and supplier management process, including the
agreement of specifications and service level agreements.
Typically, modellers do not manipulate external data before using it as
input to the model estimation or validation procedure. Changes required
in the estimation and validation process are documented in the model
build papers.
For all the above asset classes we use the Basel II definition of default,
utilising the 90 day past due criteria as the final trigger of default.
Our retail banking operations have long and extensive experience of using
credit models in assessing and managing risk in their businesses and as a
result models play an integral role in retail approval and customer
management (e.g. limit setting, cross-sell etc.) processes.
Models used include application and behavioural scorecards and/or
PD/LGD and EAD models. These may be used in isolation, in combination
to produce measures of forecast loss or as part of a suite of models that
underpin risk/reward based decisions. The score cut-offs will be set at the
appropriate level depending on the specific objective, such as ensuring all
the accepted accounts meet the minimum required return on EC. It is
Barclays philosophy to embed the Basel models as extensively as possible
in the portfolio management process. This is an ongoing initiative and we
expect greater convergence over time.
In line with Basel II requirements, Barclays will use all available relevant
data, including data relating to other Barclays accounts and external
agency data. Barclays does not use pooled data.
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 derived,
some expert lender models are used, particularly where data limitations
preclude a more sophisticated approach. For mortgage originations
Barclays use a third party scorecard (Omniscore), supported by a series
of policy rules, to arrive at a lending decision.
All new models, including third party models, are measured against the
required Group minimum standards as detailed in the Barclays Model
Risk Policy.
For retail asset classes, Basel II specifies that the definition of default must
include a trigger based on days past due, with the number of days being
between 90 and 180. All Barclays advanced internal ratings-based models
are compliant with this, with the majority using 180 days as the trigger. In
all cases LGD models are specified so that they have a definition of default
aligned to that used in the corresponding PD model.
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 89). 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 can include 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.