Basel III Risk-based capital G-SIB/ D-SIB
REGULATORY RATIO
H) Accounting definitions and policies
7.4. Rating assignment and parameter estimation
Measuring the credit risk of a transaction involves calculat-ing both the expected and the unexpected loss on the transaction. The unexpected loss is the basis for the calcula-tion of both regulatory and economic capital and refers to a very high, albeit improbable, level of loss that is not considered a recurring cost but must be absorbed by capital.
Measuring risk involves two separate steps: estimating the risk, and then assigning the credit risk parameters: PD, LGD and EAD.
Risk must be estimated on the basis of the entity’s internal experience, i.e., observed defaults for each rating grade, or credit score and recovery experience in defaulted transac-tions.
Low default portfolios: global corporates, banks, non-bank financial institutions and central governments In some portfolios (so-called low default portfolios) there is so little default experience that alternative approaches to parameter estimation must be adopted.
PD and LGD estimation in low default portfolios relies basi-cally on studies performed by external rating agencies, which reflect the pooled experience of the large numbers of entities and countries rated by the agencies.
The parameters estimated for global portfolios vary by type of portfolio but are the same for all the Group’s units. Thus, a financial institution with a rating of 8.5 will have the same PD, regardless of the unit in which the exposure is booked.
Corporates (including SMEs, specialised lending and receivables)
For portfolios of customers that have an account manager assigned to them, the estimation is based on the entity’s own internal experience. The PD is calculated by observing new defaults in the portfolio and relating these defaults to the ratings assigned to the customers concerned.
In SME portfolios, LGD is calculated based on observed re-coveries of defaulted transactions. This calculation takes into account not only the cash inflows and outflows associated with the recovery process but also the timing of these flows, so as to calculate their present value, as well as the direct and indirect costs of recovery. The LGD estimates used for regulatory purposes must be downturn LGD estimates.
Lastly, EAD, or exposure at default, is estimated by com-paring the percent utilisation of committed facilities at the time of default and prior to default in order to estimate the extent to which customers make more use of their credit facilities as they approach default.
In contrast to low default portfolios, SME portfolios have specific rating systems in each Group unit, requiring specific PD calibrations in each case.
Retail portfolios
In portfolios where customers do not have an account man-ager assigned to them but are treated on a pooled or stand-ardised basis, PDs are also estimated based on the entity’s
internal experience, although the data unit for assigning PDs is the transaction, not the customer.
PDs are calculated by observing new defaults and relating each new default to the score assigned to the transaction at the time of approval or, for transactions beyond a certain age, to the customer rating.
As for SME portfolios, LGD is calculated based on observed recovery rates, adjusted to downturn conditions. The EAD estimation is also similar to that of SMEs.
The risk parameters for retail portfolios must be estimated separately for each entity, country and segment and need to be updated at least once a year.
The parameters are then assigned to the transactions re-corded on each unit’s balance sheet, so as to calculate the expected losses and capital requirements associated with the unit’s exposure.
In the case of equity positions, the same ratings and pa-rameters are used as described above, depending on the segment.
Parameter estimation
The following tables present a summary of the parameter models used in the different geographies that have authori-sation to use internal models for the calculation of capital requirements for credit risk:
TABLA 41.
PARAMETERS OF IRB MODELS BY GEOGRAPHICAL AREA Global models
Compo-
nent Portfolio No.
significant models
R’Portfolio RWA (in €’000)
Description of model and methodology
PD Corporates 1 40,861,679 Model which uses the
equivalent agency rating and relates the internal rating to the ODF (S&P) through a regression model
20 Corporates PD > 0.03%
IFIs 2 7,652,721 Model which uses the
equivalent agency rating and relates the internal rating to the ODF (S&P) through a regression model
15 Corporates,
Financial Institutions
PD > 0.03%
Sovereigns 1 405,347 Model which uses the
equivalent agency rating and relates the internal rating to the ODF (S&P) through a regression model
21 Sovereign THERE IS NONE
LGD Non-Bank IFIs 1 2,028,018 External agency data
with conservative
Corporates 46,486,382 Value determined by
the regulator Corporates,
Financial Institutions
45%
Sovereigns 1 405,347 The model uses the
external agency recovery rating and relates it to country macroeconomic variables. Conservative adjustments are applied to the recovery ratings.
- Sovereign
EAD Corporates,
Non-Bank IFIs 1 42,889,696 The drawndown of
committed facilities is modelled based on rating downgrades (credit quality indicators) and this behaviour is extrapolated to default (taking the worst rating in the scale as a proxy at least equal to the current utilisation of the balance at account level
Project Finance and Sovereigns
1 18,507,066 The Corporates -
Non-Bank IFIs model is used Specialised Lending, Sovereign
EAD must be at least equal to the current utilisation of the balance at account level Nota: Excludes RWAs from Santander Factoring y Confirming S.A. and Santander Lease, S.A. E.F.C.
Spain portfolios
Compo-
nent Portfolio No.
significant models
R’Portfolio RWA (in €’000)
Description of model and methodology
3 28,725,260 Statistical models, based on internal default experience. Adjusted to the economic cycle
5 - 12 Corporates PD > 0.03%
Autonomous Communities
1 3,609,635 Model based on an
adjustment to the external rating
Institutions PD > 0.03%
Standardised
corporates 9 2,723,310 Statistical models, based on internal default experience. Adjusted to the economic cycle
10 Retail
portfolios - Other
PD > 0.03%
Retail mortgage 1 14,631,760 Statistical model, based on internal default experience. Adjusted to the economic cycle
non-mortgage 8 3,236,110 Statistical models, based on internal default experience. Adjusted to the economic cycle
3 28,725,260 Model based on internal recovery data. Downturn by selection of worst years in cycle
>10 Corporates
Standardised
corporates 1 2,723,310 Model based on internal recovery data. Downturn by selection of worst years in cycle
>10 Retail portfolios - Other
Retail mortgage 1 14,631,760 Model based on internal recovery data. Downturn by selection of worst years in cycle
non-mortgage 7 3,236,110 Model based on internal recovery data. Downturn by selection of worst years in cycle
3 28,725,260 Statistical model, in which internal data on observed exposure at default is used to obtain a CCF
>10 Corporates EAD must be at least equal to the current utilisation of the balance at account level Standardised
corporates 1 2,723,310 Statistical model, in which internal data on observed exposure at default is used to obtain a CCF
>10 Retail portfolios - Other
EAD must be at least equal to the current utilisation of the balance at account level Retail 2 17,867,870 Statistical model, in which
internal data on observed exposure at default is used to obtain a CCF
>10 Retail portfolios - Revolving and Other
EAD must be at least equal to the current utilisation of the balance at account level Nota: excludes RWAs from Santander Factoring y Confirming S.A,. Santander Lease, S.A. E.F.C. and Santander Consumer España
United Kingdom portfolios
Compo-
nent Portfolio No.
significant models
R’Portfolio RWA (in €’000)
Description of model and methodology
PD Mortgages 1 27,520,816 Statistical model which
produces a PD that is scaled to a cycle average
>10 Retail portfolios - Mortgages
PD > 0.03%
Consumer 1 2,727,757 Statistical model which produces a PD that is scaled to a cycle average
6 - 10 Retail portfolios - Other
PD > 0.03%
Overdrafts 1 2,367,278 Observed default rates segmented in statistical score bands, scaled to a long-term average
Housing 1 1,256,201 Expert judgment rating
model N/A Low
default portfolio
Corporates PD > 0.03%
A&L models
(FIRB) 5 2,663,074 Statistical rating model for Corporates and slotting model for Specialised Lending
LGD Mortgages 1 27,520,816 Loss estimates and write-off probability based on internal data, stressed to a downturn situation
Consumer 1 2,727,757 Loss estimates and write-off probability based on a regression, with expert judgment where appropriate
4 - 5 Retail portfolios - Other
N/A
Overdrafts 1 2,367,278 Loss estimates and write-off probability based on internal data, using a long-term average
Housing 1 1,256,201 Estimate based on data on the realisable value of the collateral
(FIRB) 5 2,663,074 Foundation IRB >10 years
(only
EAD Mortgages 1 27,520,816 Long-term CCD estimates
applied to on- and off-balance-sheet totals
6 - 10 Retail portfolios - Mortgages
EAD must be at least equal to the current utilisation of the balance at account level
Consumer 1 2,727,757 Regression model 7 - 10 Retail
portfolios - Other
EAD must be at least equal to the current utilisation of the balance at account level Overdrafts 1 2,367,278 Long-term CCD estimates
applied to on- and off-balance-sheet totals
8 - 10 Retail portfolios - Revolving
EAD must be at least equal to the current utilisation of the balance at account level Social
Housing 1 1,256,201 Estimate based on data N/A Low
default portfolio
Corporates EAD must be at least equal to the current utilisation of the balance at account level A&L models
(FIRB) 5 2,663,074 Foundation IRB >10 years
(only least equal to the current utilisation of the balance at account level
Portugal portfolios
Compo-
nent Portfolio No.
significant models
R’Portfolio RWA (in €’000)
Description of model and methodology
3 4,117,982 Statistical model, based on internal data which calibrates the scoring model, performing a cyclical adjustment
>10 Corporates PD > 0.03%
Standardised
corporates 2 474,793 Statistical model, based on internal data which calibrates the scoring model, performing a cyclical adjustment
>10 Retail portfolios - Other
PD > 0.03%
Retail mortgage 1 3,027,685 Statistical model, based on internal data which calibrates the scoring model, performing a cyclical adjustment
non-mortgage 4 816,717 Statistical model, based
on internal data which calibrates the scoring model, performing a cyclical adjustment
3 4,117,982 Statistical model, based on internal recovery data.
Downturn by selection of worst years in cycle
6 - 10 Corporates
Standardised
corporates 2 474,793 Statistical model, based on internal recovery data.
Downturn by selection of worst years in cycle
6 - 10 Retail portfolios - Other
Retail mortgage 1 3,027,685 Statistical model, based on internal recovery data.
Downturn by selection of worst years in cycle
6 - 10 Retail
non-mortgage 4 816,717 Statistical model, based on internal recovery data. Downturn by selection of worst years in cycle
3 4,117,982 Statistical model, in which internal data on observed exposure at default is used to obtain a CCF
6 - 10 Corporates CCF > 0
Standardised
corporates 2 474,793 Statistical model, in which internal data on observed exposure at default is used to obtain a CCF
6 - 10 Retail portfolios - Other
CCF > 0
Retail 3 3,844,402 Statistical model, in which internal data on observed exposure at default is used to obtain a CCF
nent Portfolio No.
significant models
R’Portfolio RWA (in €’000)
Description of model and methodology
México 1 3,289,753 Statistical model, based on internal default experience.
Adjusted to the economic cycle
6-10 Corporate/
SMEs PD > 0.03%