Barclays 2008 Annual Report Download - page 86

84 Barclays PLC Annual Report 2008 |Find out more at www.barclays.com/annualreport08
To construct ratings for institutions, corporates, specialised lending
and purchased corporate receivables and equity exposures, we use
external models, rating agencies and internally constructed models. The
applicability of each of these approaches to our customers has been
validated by us to internal rating standards. The data used in validating
these primary indicators are representative of the population of the bank’s
actual obligors and exposures and its long-term experience.
Internally built PD models are also widely used. We employ a range of
methods in the construction of these models. The basic types of PD
modelling approaches used are:
– Structural
– Expert lender
– Statistical
Structural models incorporate in their specification the elements of the
industry-accepted Merton framework to identify the distance to default
for a counterparty. This relies upon the modeller having access to specific
time series data or data proxies for the portfolio. Data samples used to
build and validate these models are typically constructed by adding
together data sets from internal default observations with comparable
externally obtained data sets from commercial providers such as rating
agencies and industry gathering consortia.
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.
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 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 Small/Medium Enterprises
(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.
In wholesale portfolios the main approaches to calculate LGD aim to
establish the affects of drivers (including industry, collateral coverage,
recovery periods, seniority and costs) by looking at Barclays historical
experience, supplemented with other external information where
necessary. Estimates built using historical information are reviewed to
establish whether they can be expected to be representative of future loss
rates, and adjusted if necessary.
In a similar fashion, wholesale EAD models estimate the potential
utilisation of headroom based on historical information also considering
the future outlook of client behaviour.
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.
Derivative counterparty credit risk measurement
The magnitude of trading exposure is determined by considering the
current mark to market of the contract, the historic volatility of the
underlying asset and the time to maturity. This allows calculation of a
credit equivalent exposure (CEE) for such exposures using a stochastic
method and a 98% confidence level.
Retail Approaches
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 customer approval and management
processes.
Models used include PD models, mostly in the form of application and
behavioural scorecards, as well as LGD and EAD models.
Application scorecards are derived from the historically observed
performance of new clients. They are built using customer demographic
and financial information, supplemented by credit bureau information
where available. Through statistical techniques, the relationship between
these candidate variables and the default marker is quantified to produce
output scores reflecting a PD. These scores are used primarily for new
customer decisioning but are, in some cases, also used to allocate PDs to
new customers for the purposes of capital calculation.
Behavioural scorecards are derived from the historically observed
performance of existing clients as well as being supplemented by the
same data as is used for application scoring, including the use of bureau
data. The techniques used to derive the output are the same as for
application scoring. The output scores are used for existing customer
management activities as well as for allocating PDs to existing customers
for the purposes of capital calculation.
Risk management
Credit risk management
Measurement, reporting and internal ratings