Barclays 2008 Annual Report Download - page 85

1
Business review
Barclays PLC Annual Report 2008 83
1
Business review
Barclays probability of default grades (wholesale)
DG/TTC Default Probability
Band >=Min Mid <Max
1 0.00% 0.010% 0.02%
2 0.02% 0.025% 0.03%
3 0.03% 0.040% 0.05%
4 0.05% 0.075% 0.10%
5 0.10% 0.125% 0.15%
6 0.15% 0.175% 0.20%
7 0.20% 0.225% 0.25%
8 0.25% 0.275% 0.30%
9 0.30% 0.350% 0.40%
10 0.40% 0.450% 0.50%
11 0.50% 0.550% 0.60%
12 0.60% 0.900% 1.20%
13 1.20% 1.375% 1.55%
14 1.55% 1.850% 2.15%
15 2.15% 2.600% 3.05%
16 3.05% 3.750% 4.45%
17 4.45% 5.400% 6.35%
18 6.35% 7.500% 8.65%
19 8.65% 10.000% 11.35%
20 11.35% 15.000% 18.65%
21 18.65% 30.000% 100.00%
two different expressions of PD depending on the purpose for which it is
used. For the purposes of calculating regulatory and economic capital,
long-run average through-the-cycle PDs are required. However, for the
purposes of pricing, PDs should represent the best estimate of probability
of default, typically in the next 12 months, dependent on the current
position in the credit cycle. Hence, point-in-time PDs are also required.
Each PD model outputs a point-in-time (PIT), through-the-cycle
(TTC) or a hybrid, e.g. a 50:50 blend, default estimate. Conversion
techniques appropriate to the portfolio are then applied to calculate both
PIT and TTC estimates. Industry and location of the counterparty and an
understanding of the current and long-term credit conditions are
considered in deriving the appropriate conversion. Two ratings are
therefore recorded for each client, the PIT and the TTC estimates.
Barclays internal rating system also differentiates between wholesale
and retail customers. For wholesale portfolios, the rating system is
constructed to ensure that each client receives the same rating
independent of the part of the business with which they are dealing. To
achieve this, a model hierarchy is adopted which requires users to adopt a
specific approach to rating each counterparty depending upon the nature
of the business and its location.
A range of methods is approved for estimating wholesale
counterparty PDs. These include bespoke grading models developed
within the Barclays Group (Internal Models), vendor models such as
MKMV Credit Edge and RiskCalc, and a conversion of external alphabet
ratings from either S&P, Moody’s or Fitch. Retail models, especially those
used for capital purposes, are almost exclusively built internally using
Barclays data, although in some cases bureau models may be used in
conjunction with these models. In addition, in some low data/low default
environments external developments may be utilised for decision-making
purposes.
A key element of the Barclays Wholesale framework is the probability
of default distribution, which maps PDs into internal grades both for PIT
(default grades) and TTC (TTC band) purposes. This has been developed
to record differences in the probability of default risk at meaningful levels
throughout the risk range. In contrast to wholesale businesses, retail areas
do not bucket exposures into generic grades for account management
purposes (although they may be used for reporting purposes). Instead,
accounts are managed either at a granular level or based on bespoke
segmentations.
Exposure at default (EAD) represents the expected level of usage of
the credit facility when default occurs. At default, the customer may not
have drawn the loan fully or may already have repaid some of the principal,
so that exposure is typically less than the approved loan limit. When the
Group evaluates loans, it takes exposure at default into consideration,
using its extensive historical experience. It recognises that customers may
make heavier than average usage of their facilities as they approach
default. The lower bound of EAD is the actual outstanding balance at
calculation of EAD. For derivative instruments, exposure in the event of
default is the estimated cost of replacing contracts with a positive value
should counterparties fail to perform their obligations.
When a customer defaults, some part of the amount outstanding on
the loan is usually recovered. The part that is not recovered, the actual loss,
together with the economic costs associated with the recovery process,
comprise the loss given default (LGD), which is expressed as a percentage
of EAD. Using historical information, the Group estimates how much is
likely to be lost, on average, for various types of loans in the event of
default.
The level of LGD depends principally on: the type of collateral (if any);
the seniority or subordination of the exposure; the industry in which the
customer operates (if a business); the length of time taken for the recovery
process and the timing of all associated cash flows; and the jurisdiction
applicable and work-out expenses. The outcome is also dependent on
economic conditions that may determine, for example, the prices that can
be realised for assets, whether a business can readily be refinanced or the
availability of a repayment source for personal customers.
The ratings process
The term ‘internal ratings’ usually refers to internally calculated estimates
of PD. These ratings are combined with EAD and LGD in the range of
applications described previously. The ‘ratings process’ refers to the use
of PD, EAD and LGD across the Group. In Barclays, the rating process is
defined by each business. For central government and banks, institutions
and corporate customers many of the models used in the rating process
are shared across businesses as the models are customer specific. For
retail exposures, the ratings models are usually unique to the business and
product type e.g. mortgages, credit cards, and consumer loans.
Wholesale Approaches
A bespoke model has been built for PD and LGD for Sovereign ratings.
For Sovereigns where there is no externally available rating, we use an
internally developed PD scorecard. The scorecard has been developed
using historic data on Sovereigns from an external data provider covering
a wide range of qualitative and quantitative information. Our LGD model
is based on resolved recoveries in the public domain, with a significant
element of conservatism added to compensate for the small sample size.