via email
KeyCorp’s Response
to the U.S. Banking Agencies’ Advance Notice of Proposed Rulemaking
Regarding New Risk-Based Bank Capital Rules
KeyCorp appreciates this opportunity to
comment on U.S. Banking Agencies’ Advance Notice of Proposed Rulemaking
(ANPR) which concerns the implementation of the New Basel Accord in the
United States. In this response, we follow the structure of the ANPR and
answer specific questions posed by the regulators.
Expected Losses versus Unexpected
Losses
What are the advantages and disadvantages
of the A-IRB approach relative to alternatives, including those that
would allow greater flexibility to use internal models and those that
would be more cautious in incorporating statistical techniques (such as
greater use of credit ratings by external rating agencies)?
When the Basel Committee issued its first
version of the New Accord in June 1999, it decided not to allow banks to
use the results of internal economic capital models in setting
regulatory capital requirements. The Committee suggested, however, that
it might reconsider the use of internal economic capital models in the
future.1
KeyCorp supports the eventual recognition
of internal models for the direct calculation of capital charges. Using
internal models would help meet the New Accord’s goal of aligning
regulatory capital more closely with economic capital. We expect that in
due course internal models will be accepted for calculation of credit
risk arising from lending and other credit products as well.
Should the A-IRB capital regime be
based on a framework that allocates capital to EL plus UL, or to UL
only? Which approach would more closely align the regulatory framework
to the internal capital allocation techniques currently used by large
institutions? If the framework were recalibrated solely to UL,
modifications to the rest of the A-IRB framework would be required. The
Agencies seek commenters’ views on issues that would arise as a result
of such recalibration.
The A-IRB approach embodies a definition
of regulatory capital that is not consistent with banks’ internal bank
credit risk management practices. That is, capital in the A-IRB approach
covers both expected loss (EL) and unexpected loss (UL), while banks
typically assign economic capital only to UL.
Indeed, common practice is to have
expected margins cover EL plus a return to capital (due to the need to
generate positive Shareholder-Value-Added). Thus, capital is needed only
to cover UL. If the regulators insist on a separate treatment of EL, it
should be done under Pillar 2. The appropriate test would be a
comparison of the A-IRB measurement of EL with the bank’s loss
provisions plus expected FMI.
Wholesale Exposures: Definitions and
Inputs
The Agencies seek comment on the
proposed definition of wholesale exposures and on the proposed inputs to
the wholesale A-IRB capital formulas. What are views on the proposed
definitions of default, PD, LGD, EAD, and M? Are there specific issues
with the standards for the quantification of PD, LGD, EAD, or M on which
the Agencies should focus? (P. 29)
The definition of default outlined in CP3
and the ANPR should be simplified to correspond more closely to what is
commonly used by risk practitioners. That is, loans that fall under the
corporate and specialized lending models should utilize a default
definition that coincides solely with the incidence of non-accrual or
charge-off status (thus excluding the 90 days past due and other
isolated conditions present in the Accord’s current definition).
We are concerned that, in the absence of
moving the default definition for wholesale loans to be based solely on
the occurrence of non-accrual or charge-off status, core banks will be
forced to track two separate measures of default – one for internal risk
assessment and a second for regulatory capital purposes. This would be a
costly exercise, but one without much impact on risk measurement. This
is because the ultimate measurement of risk is the loss distribution,
and shifting the default definition in incremental amounts will only
serve to shift the mix of PD and LGD in an offsetting fashion. The
impact on measured economic capital will be minimal.
Wholesale Exposures: Formulas and
Other Considerations
The Agencies are seeking comment on
the wholesale A-IRB capital formulas and the resulting capital
requirements. Would this approach provide a meaningful and appropriate
increase in risk sensitivity in the sense that the results are
consistent with alternative assessments of the credit risks associated
with such exposures or the capital needed to support them? If not, where
are there material inconsistencies?
The proposed formulas result in a
reasonable representation of risk.
Does the proposed A-IRB maturity
adjustment appropriately address the risk differences between loans with
differing maturities?(P.37)
The proposed maturity adjustment
appropriately addresses the risk differences between loans with
different maturities, provided that these maturities are above one year.
Basel maturity adjustment is a proxy for mark-to-market definition of
capital where losses are defined via change of value at the one-year
horizon. This change of value includes possibilities of both a default
and a downgrade before or at the horizon. However, for exposures with
remaining maturity shorter than one year (short-term maturity),
downgrades will not produce economic loss at the horizon because, if
there is no default, such an exposure simply will not exist at the
horizon. Therefore, the proposed maturity adjustment can only be applied
to loans with maturities above one year.
However, loans with short-term maturity
have less time to default than one year. Therefore, capital requirements
for short-term exposures are unjustifiably overestimated. We suggest
that, for all loans with remaining maturity less than one year, one-year
PD should be adjusted downwards to reflect the remaining maturity. Under
certain assumptions, there is a simple formula for this adjustment. Let
us assume that, when we divide the one-year interval into an arbitrary
number of smaller periods of equal length, conditionally on surviving up
to the beginning of the period, probability of obligor defaulting during
each period is the same. Then, probability of default over time T
(maturity of short-term exposure in years) PD(T), and probability of
default over one year (time horizon) PD(1) are related by this formula:
PD(T) = 1 – exp( ln[1-PD(1)] T ) = 1 –
[1-PD(1)] T
This simple formula is very popular
amongst practitioners and would be a sound choice for the PD term
adjustment.
Retail Exposures: Definitions and
Inputs
For the QRE sub-category of retail
exposures only, the Agencies are seeking comment on whether or not to
allow banking organizations to offset a portion of the AIRB capital
requirement relating to EL by demonstrating that their anticipated FMI
for this sub-category is likely to more than sufficiently cover EL over
the next year.
As indicated above, expected margins must
at least cover expected credit and operating losses for all forms of
credit, not just qualifying revolving retail credits. Therefore, for all
credit exposures, capital should be redefined to cover only UL. If the
regulators redefine capital and introduce a separate treatment of EL (as
indicated in the Attachment to October 11, 2003 Basel Press Release),
the EL treatment (same for all credit exposures, not just QRE) should be
done under Pillar 2. As we mentioned above, the appropriate test would
be a comparison of the A-IRB measurement of one-year EL with the bank’s
loss provisions plus expected FMI.
The Agencies are seeking comment on
the proposed definitions of the retail AIRB exposure category and
sub-categories. Do the proposed categories provide a reasonable balance
between the need for differential treatment to achieve risk-sensitivity
and the desire to avoid excessive complexity in the retail A-IRB
framework? What are views on the proposed approach to inclusion of SMEs
in the other retail category?
We agree generally with proposed
definitions of the retail sub-categories, but wish to note that, in
future iterations of the U.S. regulatory policy, capital for HELOCs and
other home equity loans should not be the same as capital for
residential mortgages. In particular, we believe that the asset
correlations for home equity loans should be lower than the ones for
residential mortgages (see explanation below). Ideally, home equity
exposures should be put into a separate sub-category with its own
correlation function. If this is not feasible, home equity loans and
lines of credit could be treated under “other retail” sub-category.
The Agencies are also seeking views on
the proposed approach to defining the risk inputs for the retail A-IRB
framework. Is the proposed degree of flexibility in their calculation,
including the application of specific floors, appropriate? What are
views on the issues associated with undrawn retail lines of credit
described here and on the proposed incorporation of FMI in the QRE
capital determination process?
The proposed approach to estimating the
inputs to the regulatory retail capital models is generally appropriate.
However, no floors should be placed on any estimated parameter input.
For example, for single-family residential loans (SFRs), high quality
loans with low loan-to-values (LTVs) and/or private mortgage insurance (PMI)
may have estimated LGDs that are close to zero. The proposed 10% floor
on LGDs is not appropriate for such exposures and should be removed.
Retail Exposures: Formulas
The Agencies are interested in views
on whether partial recognition of FMI should be permitted in cases where
the amount of eligible FMI fails to meet the required minimum. The
Agencies also are interested in views on the level of portfolio
segmentation at which it would be appropriate to perform the FMI
calculation. Would a requirement that FMI eligibility calculations be
performed separately for each portfolio segment effectively allow FMI to
offset EL capital requirements for QREs?
As indicated above, we believe that
Pillar 2 should be used to see whether expected margins plus current
reserves cover expected losses. If the EL treatment is at all necessary
(assuming that capital is defined to cover only UL), the FMI test should
be done under the pillar 2 for all credit exposures (and not just QREs).
Moreover, we do not agree with the current definition of the FMI test (FMI
covering EL plus two standard deviations of the annualized loss). We
believe that one-year FMI plus current reserves should cover one-year EL
only. In this definition of the FMI test, portfolio segmentation is
immaterial.
The Agencies are seeking comment on
the retail A-IRB capital formulas and the resulting capital
requirements, including the specific issues mentioned. Are there
particular retail product lines or retail activities for which the
resulting A-IRB capital requirements would not be appropriate, either
because of a misalignment with underlying risks or because of other
potential consequences?
As we mentioned above, A-IRB capital
formulas should be redefined so that the resulting capital would cover
only UL. After such a redefinition, procyclicality of capital will be
reduced, and the regulators might want to flatten asset correlations as
functions of PD. We do believe that asset correlation for retail
exposures should decrease with increasing PD, but Basel asset
correlations for revolving exposures and other retail exposures are too
steep.
In CP3 and ANPR, home equity loans and
lines are treated under residential mortgages category. We believe that
there are at least two conceptual arguments in favor of separate risk
weight curve for home equity products.
One of the reasons why asset correlation
for residential mortgages is set at such a high level is to take into
account long-term nature of mortgage loans. Basel retail model does not
have the maturity adjustment factor, and the effect of longer maturity
on capital is incorporated into the model through higher asset
correlation. Since typical maturity for home equity loans (10-15 years)
is smaller than one for first mortgages (30 years) by at least a factor
of two, the effective asset correlation for home equity loans should be
lower than the one for first mortgages.
The majority of residential mortgages in
the United States are conforming mortgages, i.e. mortgages insured by
the U.S. government and not kept by banks in their books. The mortgages
banks keep in their books are those that do not qualify for the
government insurance (issued to either consumers with poor credit
quality or consumers who buy expensive houses). Home equity loans and
lines of credit are based on all kinds of mortgages and thus have a much
more diverse customer base than non-conforming first mortgages.
Therefore, the asset correlation for home equity products should be
lower than the one for first mortgages.
Credit Risk Mitigation Techniques
The Agencies are seeking comment on
the proposed nonrecognition of double default effects…The Agencies also
are interested in obtaining commenters’ views on alternative methods for
giving recognition to double default effects in a manner that is
operationally feasible and consistent with safety and soundness. With
regard to the latter, commenters are requested to bear in mind the
concerns outlined in the double default white paper, particularly in
connection with concentrations, wrong-way risk (especially in stress
periods), and the potential for regulatory capital arbitrage. In this
regard, information is solicited on how banking organizations consider
double default effects on credit protection arrangements in their
economic capital calculations and for which types of credit protection
arrangements they consider these effects.
Within the banking book, guarantees can
be used to reduce the regulatory capital charge only to the level
associated with the guarantor, giving no benefit to either the
double-default or double-recovery effect of guarantees. That is, in
order for a loss to occur on a guaranteed credit, both the underlying
obligor and the guarantor would have to fail. This probability is likely
to be significantly lower than the probability of either one failing,
therefore the economic capital allocation for the guaranteed credit
should be considerably lower than for either a direct obligation of the
guarantor or the actual underlying credit. Moreover, some credit
guarantees are written in such a manner that the bank, in the unlikely
event of double default, can seek recoveries from both the underlying
obligor and the guarantor. ANPR recognizes neither of these two risk
reduction benefits.
An excellent treatment of this subject
can be found in a recent white paper produced by staff at the Federal
Reserve Board.2 The paper describes an appropriate analytical
approach to the issue (in the context of the asymptotic single risk
factor model currently being used by Basel’s Advanced IRB approach) and
lays out the important supervisory concerns over the use of guaranteed
credits or credit derivatives that function as guarantees. We believe
that these supervisory concerns can be appropriately treated within the
Pillar 2 process, while the analytical framework can be implemented
relatively quickly within Pillar 1.
The only parameter necessary for the
framework implementation that is not already defined in CP3/ANPR is the
measure of the wrong-way risk (see the paper’s Appendix). This
parameter can be set conservatively at the level of 40%-50% until more
research is done.
Securitizations: General
Considerations
Should the Agencies require
originators to hold dollar-for-dollar capital against all retained
securitization exposures, even if this treatment would result in an
aggregate amount of capital required of the originator that exceeded the
pool’s A-IRB capital charge plus any applicable deductions? Please
provide the underlying rationale.
In absolute terms (i.e., in dollars), the
risk of any tranche (or a set of tranches) cannot exceed the risk of the
underlying pool. This statement is very general and holds under any
definition of risk measure. Therefore, under no circumstances, the
amount of capital required of an originator should exceed the pool’s A-IRB
capital charge.
Dollar-for-dollar capital (whether below
or above KIRB) is an arbitrary constraint. This constraint
should not be introduced for exposures above KIRB and should be removed
from the treatment of originators for exposures below KIRB.
Capital for securitization exposures held by originators should be
computed according to the modified SFA discussed below. Under this
treatment, the total capital requirements for originators will always be
below KIRB (it will equal KIRB when the originator
holds all the tranches defined on a pool).
The Agencies seek comment on the
proposed treatment of securitization exposures held by originators. In
particular, the Agencies seek comment on whether originating banking
organizations should be permitted to calculate A-IRB capital charges for
securitizations exposures below the KIRB threshold based on an external
or inferred rating, when available.
Under the proposed rules, both investors
and originators are required to use the RBA whenever external ratings of
a tranche are available. Only when no external rating available,
originators are allowed to use the SFA. The SFA is based on Gordy/Jones
model,3 which provides reasonably accurate description of the
risk underlying a given tranche. Apart from its dependence upon rating,
this risk (represented by capital) depends on the underlying pool’s
granularity, credit quality and asset correlations, as well as tranche
thickness. Therefore, the RBA, which is primarily ratings-based, is
necessarily inferior to the SFA in terms of describing the risk
underlying a securitization tranche. While the RBA is useful for
investors, who typically do not have complete information on the
underlying pool, the superior SFA should always be used by originators,
who do have this information.
The Agencies seek comment on whether
deduction should be required for all nonrated positions above KIRB. What
are the advantages and disadvantages of the SFA approach versus the
deduction approach?
Deduction is not conceptually justifiable
for any tranche – whether it is below KIRB or above. As we argued above,
the SFA should always be used by originators regardless of the
availability of rating. Moreover, as we suggest below, such supervisory
constraints as the capital floor and dollar-for-dollar capital below
KIRB should be removed from the Supervisory Formula.
Securitizations: Capital Calculation
Approaches
The Agencies seek comment on the
proposed treatment of securitization exposures under the RBA. For rated
securitization exposures, is it appropriate to differentiate risk
weights based on tranche thickness and pool granularity?
Apart from its dependence upon rating,
tranche capital depends on underlying pool’s granularity, credit quality
and asset correlations, as well as tranche thickness. Thus, the RBA is
necessarily less accurate than the SFA. However, accuracy of the RBA can
be improved if some of this dependence is taken into account. This is
what was attempted in CP3 and ANPR via introduction of three separate
capital factors for each rating. We believe that the regulators are on
the right track here, but disagree on the calibration.
We have computed capital according to
Gordy/Jones model for underlying pools of different granularity and
considered tranches of different ratings. We used Moody’s table that
relates ratings to expected losses4 and considered only
infinitesimally thin tranches to remove the difference between the
Moody’s and S&P rating systems. Our calculations clearly show that
granularity has much stronger effect on capital than RBA capital factors
suggest, particularly for highly rated tranches. Another result of our
calculations is that overall level of capital factors is way too high
for high ratings (AAA, AA) and too low for low ratings (BBB and below).
The Agencies seek comment on the
proposed SFA. How might it be simplified without sacrificing significant
risk sensitivity? How useful are the alternative simplified computation
methodologies for N and LGD
The SFA is based on the Gordy/Jones model
with two added supervisory overrides: (i) dollar-for-dollar capital up
to KIRB and (ii) the floor which sets minimum capital of
0.56% for any tranche. Neither of the overrides can be justified
conceptually and both of them lead to significant disparity between the
capital charge and the underlying risk. We are particularly concerned
with the floor because model-based capital for most senior and
super-senior tranches is one or two orders of magnitude less than the
floor. On the other hand, dollar-for-dollar capital up to KIRB leads to
overestimation of capital for narrow mezzanine tranches with credit
enhancement levels in the vicinity of KIRB roughly by a factor of two.
Therefore, we believe that both supervisory overrides should be removed
from the SFA. As an additional benefit, this removal would significantly
simplify the Supervisory Formula.5 If not removed completely,
the floor should be reduced to a few basis points.
____________________________________________
“A New Capital Adequacy Framework,” June 1999, p. 41.
See Erik Heitfield and Norah
Barger, Treatment of Double-Default and Double-Recovery Effects for
Hedged Exposures under Pillar 1 of the Proposed New Capital Accord,
Board of Governors, Federal Reserve System, June 2003.
Michael Gordy and David Jones, Random Tranches, Risk,
March 2003, pages 78-83.
See Table 2 in Moody’s Special Report The Lognormal Method Applied to
ABS Analysis, July 27, 2000.
The capital for a tranche with credit enhancement level L and
thickness T would be just K(L+T) – K(L),
where function K is defined in paragraph 590 on page 117.
| 
