Upload
truongnguyet
View
222
Download
1
Embed Size (px)
Citation preview
CREDIT RISK RATING ATAUSTRALIAN BANKS
Andrew McDonald and Guy Eastwood
Working Paper 7
ABSTRACT
This working paper summarises a survey of the internal credit risk rating practices at 10
Australian banks. The survey is a response to international proposals to develop more risk
sensitive capital adequacy requirements utilising banks’ internal credit risk ratings. It
examines various elements of the Australian banks’ internal credit risk rating systems,
including the systems’ basic architecture, operating design and applications. Comparisons of
the survey findings with those from similar international surveys are also provided.
JEL Classification Numbers: G21
Keywords: Credit Risk Rating,APRA,Australia, Banks
Table of Contents
1 INTRODUCTION 3
2 OVERVIEW OF INTERNAL CREDIT RISK RATING SYSTEMS 5
2.1 SYSTEM ARCHITECTURE 5
2.2 OPERATING DESIGN 7
2.3 APPLICATIONS 8
3 CREDIT RISK RATING AT AUSTRALIAN BANKS 9
3.1 ARCHITECTURE OF AUSTRALIAN BANKS’ CREDIT RATING SYSTEMS 9
3.1.1 LOSS CONCEPT 9
3.1.2 NUMBER OF RISK GRADES 11
3.1.3 DISTRIBUTION OF EXPOSURES AMONG RISK GRADES 14
3.1.4 POINT-IN-TIME VS THROUGH-THE-CYCLE GRADING 15
3.2 OPERATING DESIGN FEATURES 16
3.2.1 EXPOSURES RATED 16
3.2.2 CUSTOMER PD RATINGS 16
(i) TYPES OF RATING PROCESSES 16
(ii) KEY RATING RISK FACTORS 18
(iii) LINK TO QUANTITATIVE DEFAULT STATISTICS 18
3.2.3 FACILITY LGD RATINGS 20
3.2.4 SYSTEM OVERSIGHT AND CONTROL PROCESSES 21
3.2.5 VALIDATION PROCEDURES 22
3.3 RATING APPLICATIONS 23
4 CONCLUDING REMARKS 25
APPENDIX 1:TYPES OF CREDIT RISK RATING MODELS 27
APPENDIX 2: EXAMPLE TWO-DIMENSIONAL CREDIT RISK RATING MATRIX 29
REFERENCES 30
APRA DECEMBER 20002
1 INTRODUCTION
Credit risk rating has become an important feature of most Australian banks’ credit risk management
systems over the past decade. This reflects the efforts of institutions to strengthen credit management
practices following the asset quality problems of the late 1980s/early 1990s, wider availability and growing
familiarity with rating techniques of increasing sophistication within the industry, and a growing array of
uses to which ratings may be applied. The Australian Prudential Regulation Authority (APRA) – as well
as its predecessor as banking supervisor, the Reserve Bank of Australia – has also sought to encourage the
use of credit rating systems, including through prudential guidelines originally issued to banks in 1995
and recently updated to apply to authorised deposit-taking institutions more generally.1
This paper surveys the internal credit risk rating systems currently utilised by Australian banks and
compares local rating practices with those found internationally. The main motivation for the paper stems
from proposals that have been put forward by the Basel Committee on Banking Supervision for the
reform of international bank capital adequacy guidelines, aimed at increasing the sensitivity of regulatory
capital requirements to differences in institutions’ individual risk profiles. The Committee has
recommended that, in conjunction with a new standardised approach to capital adequacy regulation, an
internal ratings based approach could also form the basis for determining regulatory capital charges. The
approach would link banks’ credit risk capital requirements to internal credit risk ratings for those
institutions with suitably robust and well-developed risk rating systems that are able to meet minimum
supervisory standards.2
The Committee’s recommendation means that prudential regulators, such as APRA, have need to
conduct further research into banks’ internal risk rating systems preparatory to commenting on, and
implementing, the next round of more detailed reform proposals. The stocktake of current industry
practice summarised in this paper constitutes part of that process.
The following section of the paper provides a general overview of banks’ internal credit risk rating
systems. It sets out some basic concepts and a broad structure for the main part of the report, which
summarises the main survey findings. This middle section focuses on the basic structure, operating design
and applications of the local banks’ rating systems and provides some international comparisons. The final
section offers some concluding remarks.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 20003
1 Refer Prudential Standards APS 220 Credit Quality available on the APRA website (www.apra.gov.au).
2 Capital adequacy regulation of banks and other authorised deposit-taking institutions in Australia, as in most countries, is based on international guidelines issued by the Basel Committee on Banking Supervision. The Basel Committee, which comprises central banks and bank supervisory agencies from G-10 countries, operates under the auspices of the Bank for International Settlements (BIS) and consults widely with supervisory agencies in other countries and with industry on prudential matters. The committee’s capital adequacy reform proposals are contained in a consultative paper, A New Capital Adequacy Framework, issued in June 1999.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 20004
Fac
tors
con
side
red
in r
atin
g,eg
Fina
ncia
l ana
lysis
Indu
stry
ana
lysis
Qua
lity
of fi
nanc
ial d
ata
Ext
erna
l rat
ings
Ana
lytic
rat
ings
Tool
s/m
odel
s
Firm
siz
e/va
lue
Man
agem
ent
Term
s of
faci
lity/
LGD
Oth
er c
onsid
erat
ions
Rat
ing
crite
ria
Wri
tten
/for
mal
elem
ents
Subj
ectiv
e/in
form
alel
emen
ts(c
ultu
ral)
Rat
er’s
own
expe
rien
ce a
ndju
dgem
ent
Pre
lim
inar
yra
ting
prop
osed
for
loan
appr
oval
proc
ess
Rel
atio
nshi
pm
anag
eran
d/or
cred
it of
ficer
App
rova
lpr
oces
s(p
er p
olic
y)as
sign
sfin
alra
ting
Ris
k
rati
ng
Port
folio
mon
itori
ng
Loan
loss
res
erve
anal
ysis
Loan
/bus
ines
s lin
epr
icin
g an
dpr
ofita
bilit
y an
alys
is
Inte
rnal
cap
ital
allo
catio
n
Ret
urn
on c
apita
lan
alys
is
Qua
ntita
tive
loss
char
acte
rist
ics
Ass
essin
g at
trac
tiven
ess
of c
usto
mer
rela
tions
hip
Eva
luat
ion
of r
ater
effe
ctiv
enes
s
Adm
inist
rativ
e an
dm
onito
ring
requ
irem
ents
Freq
uenc
y of
loan
revi
ew
Gen
eral
cr
edit
qual
itych
arac
teri
stic
s
Loss
con
cept
Ord
inal
or
card
inal
risk
mea
sure
Num
ber
of r
isk g
rade
s
Poin
t-in
-tim
e vs
thr
ough
-the
-cyc
le g
radi
ng
Arc
hitec
tura
l fe
ature
s
Use
s of
ratings
Ass
ignm
ent
of
ratings
On-
goin
g re
view
by
initi
al r
ater
Peri
odic
rev
iew
of
each
cust
omer
rel
atio
nshi
p
Aim
ed a
t re
view
ing
prof
itabi
lity/
desir
abili
ty a
s w
ell
as c
ondi
tion
Gen
eral
ly c
ondu
cted
by
sam
eau
thor
ities
tha
t ap
prov
e lo
ans
Lin
e/cr
edit
rev
iew
Wat
ch p
roce
sses
Cre
dit
man
agem
ent
revi
ew
Mon
itori
ng p
roce
ss (
ofte
n qu
arte
rly)
focu
sed
on lo
ans
that
exh
ibit
curr
ent
or p
rosp
ectiv
e pr
oble
ms
only
Aim
ed a
t id
entif
ying
bes
t pa
th t
oim
prov
e or
exi
t cr
edit
at lo
wes
t co
st
Con
duct
ed b
y sa
me
auth
oriti
es t
hat
appr
ove
loan
s,al
thou
gh o
ther
s m
aypa
rtic
ipat
e as
wel
l (ie
wor
kout
gro
up)
Rev
iew
of
adeq
uacy
of
unde
rwri
ting
and
mon
itori
ng f
rom
ran
dom
sam
ple
Sam
ple
wei
ghte
d to
war
ds h
ighe
r-r
isk lo
ans
Loan
rev
iew
judg
emen
t is
“fin
al s
ay”
Neg
ativ
e co
nseq
uenc
es f
or i
nitia
l rat
erif
cons
isten
t di
sagr
eem
ents
Rev
iew
pro
cess
es
Ada
pted
fro
m T
reac
y &
Car
ey 1
998
Fig
ure
1:
Ris
k ra
ting
pro
cess
2 OVERVIEW OF INTERNAL CREDIT RISK RATING SYSTEMS
Internal credit risk ratings are summary indicators of the degree of risk inherent in institutions’ individual
credit exposures. In combination, credit ratings also provide a useful snapshot of the overall quality of an
institution’s credit portfolio. A credit rating represents an assessment of the risk of loss from the failure of a
given counterparty to meet debt servicing and other payment obligations on a timely basis. Other, more
traditional, credit assessments tend to be binary in nature (covering decisions such as whether or not to accept
a new credit proposal or continue lending to a particular borrower). Credit risk ratings seek to be more
informative by grouping credit exposures with similar risk characteristics into a larger number of risk grades
or buckets.
There is no standard approach to risk rating credit exposures. Credit risk rating systems vary widely
among institutions, including in their basic structure, operating design and uses; the main elements of
which are illustrated in Figure 1.
2.1 System architecture
The basic structure (or architecture) of a credit risk rating system includes such features as:
• the risk concepts underlying the ratings;
• whether the ratings represent relative (ordinal) or absolute (cardinal) measures of risk; and
• the degree of risk differentiation sought (eg number of risk grades).
Abstracting from portfolio composition effects, the stand-alone credit risk associated with an individual
loan or other credit exposure depends on: (i) the amount at risk; (ii) the term of the exposure; (iii) the
likelihood of the customer defaulting during that period; and (iv) the likely severity of loss if default occurs.
Although an exposure’s term and probability of default are inextricably linked, these risk factors are
handled separately in most credit risk rating systems. Most systems assess the likelihood of customer
default over some common (often implicit) time horizon for all exposures irrespective of actual tenors.3
Consequently, credit risk ratings typically focus on the last two of the above-mentioned risk factors,
which when combined provide a measure of overall expected loss over the relevant time horizon.4
Expected loss in this instance has a statistical meaning, ie ratings do not represent predictions of actual
credit losses for particular loans, instead ratings are indicators of the credit losses likely to arise on average
over similar time periods from large numbers of loans that display similar characteristics.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 20005
3 Differences in the maturities of exposures are usually dealt with at a later stage, eg when ratings are linked to numerical loss statistics for use in quantitative analyses, such as risk-based pricing,provisioning and capital modelling.
4 Expected loss = Probability of default x Loss given default
Table 1 shows that credit risk rating systems can be broadly classified according to whether the ratings
generated are ‘one-’ or ‘two-dimensional’ in form and which of the three loss concepts mentioned above
underlie the ratings, namely probability of default (PD), loss given default (LGD) or expected loss (EL).
One-dimensional ratings usually focus either solely on the probability of customer default (ie ignore the
impact of loss given default) or intermingle an analysis of default risk and loss given default into a single
overall assessment of expected loss. Two-dimensional systems, on the other hand, assign separate ratings
to each credit exposure relating to each of these risk factors.
Table 1: Types of credit risk ratings
Worldwide, a relatively small but increasing number of banks determine for each of their rated exposures:
• an obligor (or customer) risk rating – that reflects the probability of customer default based on
an assessment of quantitative and/or qualitative information relating to the customer’s
prospective financial stability;5 and
• a facility rating – that reflects loss given default. This rating is usually based on facility-level
characteristics such as product type, security coverage and/or seniority ranking.
Some institutions combine these two ratings into an overall composite risk rating, although most
subsequent internal analysis and reporting tends to be carried out at the sub-component level. At other
institutions the composite indicator is achieved simply by bringing the two ratings together in the form
of an alphanumeric label.
While usually classified as a two-dimensional system, a sort of hybrid approach is used at some institutions
around the world. A separate customer-level default risk rating is first determined, which is then adjusted
up or down according to the nature of each of the customer’s credit facilities. Typically, these facility-
based adjustments are restricted to one, or perhaps two, risk grades.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 20006
5 Banks’ policy rules generally specify that all facilities to the same customer be assigned the same PD rating.While there may be instances where a customer defaults on one transaction and not another, as a practical matter, estimating a single PD for each customer is usually sufficient and avoids a more likely problem of having multiple ratings for a customer with little justification for the differences.
One-dimensional ratings
- Single obligor-level rating
- Single combined rating
Two-dimensional ratings
- Hybrid rating
- Composite rating
Loss conceptPD
(obligor-level rating)
LGD
(facility-level rating)
EL
(facility-level rating)
✔
✔
✔ ✔
✔
✔
✔
2.2 Operating design
A system’s operating design denotes the processes by which risk is evaluated and ratings assigned. It also
includes the surrounding controls and other measures designed to promote rating accuracy, integrity and
consistency throughout the rating organisation.
A wide range of risk evaluation processes exists. An important distinguishing feature of these processes
relates to the degree of subjectivity that is involved both in the selection of inputs and the way in which
those inputs are combined to derive ratings. At one extreme, ratings may be based on high-level
definitions judged by experienced lenders using traditional credit analysis – this approach is also followed
by many of the major external ratings agencies, including Moody’s and Standard & Poor’s. At other
institutions, such broad, subjective criteria are supplemented by more explicit guidance. This can involve
tools designed to assist raters in assigning ratings, including the development of quantitative and/or
qualitative rating benchmarks and/or formal debt rating models of varying levels of sophistication.6
However, even where more specific criteria have been introduced, the degree of subjectivity can vary
greatly, eg:
• individual raters may be able to exercise a good deal of discretion at institutions where simple
benchmarks and/or qualitative (rather than quantitative) model inputs are relied on;
• debt rating models differ widely in the extent to which they have been developed from
objective statistical analysis or expert judgement;
• institutions often use a mix of different techniques or models from among which raters must
choose; while
• at most institutions, raters are permitted to use professional judgement to override
model-based assessments.
The degree of subjectivity involved in the rating process can have important implications for the accuracy,
integrity and consistency of ratings within an organisation (Figure 2). Other things being equal,
appropriate controls and other measures designed to promote these desirable rating features will need to
be more extensive (costly) the greater the degree of subjectivity involved in the ratings process and the
greater the potential for rating outcomes to affect the performance assessments (and directly or indirectly
the remuneration) of raters. On the other hand, focusing on purely objective data, and/or reducing raters’
discretion to exercise judgement through model overrides, will ignore potentially relevant information
not taken into account by the rating model. It can also reduce incentives for lending and credit officers
to undertake adequate due diligence when assessing loan applications.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 20007
6 Appendix 1 provides an outline of the main types of debt rating models used by banks.
Figure 2: Impact of subjective rating criteria1
How institutions attempt to work out an appropriate balance between these (and other) competing
influences can have a major impact on the overall design of their rating systems. This includes the choice
of methodologies used to assign ratings and other quality control measures, such as the design of policies
and organisational structures relating to the independent review of ratings, frequency of rating updates,
and responsibilities for the overall oversight and maintenance of the system.
Some other key aspects of a system’s operating design include:
• what credit exposures are covered by the system and what, if any, exposures are excluded;
• where different rating models/techniques are used, especially in different market segments, how
are ratings brought together into a single rating scale; and
• what processes exist to validate the (on-going) effectiveness of the ratings in differentiating risk.
2.3 Applications
As institutions’ familiarity with credit risk rating has increased, so too has the range and sophistication of
the uses to which credit ratings have been applied.
Credit ratings can be used to improve many aspects of credit risk management at the transaction,
customer and portfolio levels within a lending institution. In a recent survey of rating practices at large
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 20008
Moody’s CustomerRating
A1
A2
A3
Baa1
Baa2
Baa3
Ba1
Ba2
Ba3
B1
B2
B3
1 2 3 4 5 6
1 The diagram above shows the range of outcomes from a trial in which one bank asked several relationship and credit review managers to rate the same set of customers using identical information about those customers. Ratings were based on subjective high-level ratings definitions; the bank, at that time, had not introduced debt rating models or other rating tools to assist in the assignment of ratings.
US banks,Treacy & Carey (1998) grouped these uses broadly into analytic and reporting, as well as credit
administration applications. Analytic and reporting uses can include: monitoring and reporting of risk
positions; communication of differentiated credit risk acquisition and portfolio management strategies;
policy limits (eg large exposure limits); as inputs in loss provisioning, economic capital allocation and
risk-adjusted performance measures; and in risk-based pricing and employee compensation
arrangements. Administrative uses include such things as setting delegated credit approval thresholds and
trigger points for more intensive and/or specialist management of impaired and other problematic (eg
watchlist) loans.
Generally speaking, different uses will have different implications for the overall structure of credit rating
systems and the internal controls needed to maintain system integrity. For example, a system that is primarily
intended for tracking broad trends in portfolio credit quality generally would not need to be as sophisticated
as one intended for use in more complex capital allocation and risk-adjusted performance measures.
3 CREDIT RISK RATING AT AUSTRALIAN BANKS
The following sections of this paper describe the internal credit risk rating practices of 10 Australian
banks. Foreign banking groups operating in Australia were excluded from the survey, which also focused
on the banks’ rating of corporate and commercial exposures. Although credit scoring of banks’ consumer
portfolios has become widespread, only one bank in Australia has sought to implement a single rating
scale across the whole of its credit portfolio. A couple of the other banks rate consumer exposures on a
portfolio basis for APRA reporting purposes, however, this does not reflect broad internal reporting
practice at those banks as it tends to obscure higher-level monitoring of credit quality trends. This
situation is consistent with overseas experience; it reflects the fact that banks typically manage their
corporate/commercial and consumer portfolios differently using distinctly separate processes.
3.1 Architecture of Australian Banks’ Credit Rating Systems
3.1.1 Loss concept
As earlier noted, an important aspect of any credit risk rating system is the loss concept used to
differentiate the riskiness of different credit exposures, ie whether the ratings are one- or two-dimensional
in form and whether they focus primarily on PD, LGD, EL or all three credit risk measures.
All 10 of the surveyed Australian banks utilise two-dimensional rating systems. In rating their credit
exposures, each of these banks determines a separate customer-level PD rating, a facility-level LGD rating
and a composite EL rating.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 20009
Other surveys indicate that only a small (though growing) proportion of overseas banks have similarly
structured ratings. Last year, for example, the Basel Committee on Banking Supervision surveyed nearly
30 banks across the G-10 that were identified by their national supervisors as having well developed
internal rating systems.7 In terms of the categories of rating systems identified in Table 1:
• about a third of the surveyed banks utilise two-dimensional ratings (of those, most use hybrid
ratings while only “a small number” assign separate PD and LGD ratings);
• 20 per cent use single facility-level ratings that explicitly take into account both obligor and
transaction specific characteristics; while
• the remainder (about half) assign single obligor-level ratings meant primarily to reflect the risk
of the borrower defaulting.
Two US studies,Treacy & Carey (1998) and English & Nelson (1998), made similar findings.Treacy &
Carey surveyed the top 50 US banks while English & Nelson surveyed over 100 US banks across different
size categories. Not unexpectedly, the latter found a higher proportion of smaller banks using one-
dimensional systems. Among those institutions using a two-dimensional approach, neither study cited
any examples of banks using composite ratings, though Treacy & Carey noted that “a few banks” planned
to shift in that direction.
The differences in the survey results partly reflect differences in the timing of the surveys (combined with
the rapid pace of development in this area) and the smaller number of banks in Australia. Nonetheless,
Australian banks appear overall to have moved more quickly in adopting two-dimensional composite
approaches to credit risk rating compared with their overseas counterparts.
Generally speaking, as internal credit risk ratings have been applied to a wider range of more sophisticated
uses, so too has the need for more accurate, more differentiated and more quantitative measures of risk.
By separately assessing default risk and loss given default, two-dimensional systems (and composite rating
systems in particular) can:
• improve communication about risk (including by conveying more detailed information about
risk profiles and by reducing potential ambiguity over the meaning of risk grades);
• lessen the tendency at some banks to rate primarily on the strength of available security;
• facilitate the development of rating tools to assist in the risk rating process;
• facilitate tracking of the accuracy of risk ratings;
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200010
7 Basel Committee on Banking Supervision (2000a).
• are conceptually better aligned with the more sophisticated provisioning, capital allocation and
risk-based pricing techniques being developed at some institutions; and
• facilitate the alignment of internal and external credit ratings. As few banks have long enough
data sets to assign quantitative values to their internal risk grades based on their own loss
histories; many have sought to align their ratings with the experience of longer-established
external rating schemes (such as Moody’s and/or Standard & Poor’s). Alignment of internal and
external ratings is also likely to facilitate the development of secondary and derivative trading in
individual and pooled loans of customers that normally would not be rated by external
rating agencies.
In recent years, the above considerations have contributed to a greater or lesser extent to all of the
Australian banks either introducing composite systems where no rating system previously existed or
converting from simpler, more subjective, one-dimensional systems. In the case of the larger banks, the
move to composite ratings has been closely linked to the development of portfolio credit risk modelling
capabilities and more statistically based provisioning and capital allocation processes. Among the smaller
banks, however, it has had more to do with the banks’ wanting to improve their transaction-based credit
management processes as they have sought to enter into, or expand, their commercial lending operations,
including in new geographic areas. As part of this process, some banks have introduced externally
developed rating tools. As these latter products tend to focus on probability of default rather than overall
expected loss, an LGD dimension has been added.
The Australian banks combine their PD and LGD ratings to form an overall indicator of risk in a number of ways:
• the majority of banks simply bring the two ratings together in the form of an alphanumeric
label. For example, a rating of 1G might represent a low-risk borrower/high loss-given-default
(eg unsecured) facility while a rating of 7A might represent a relatively high-risk borrower/low
loss-given-default (eg well secured) facility;
• a couple of banks have developed look-up matrices. These matrices assign the same composite
rating to different PD and LGD rating combinations considered by each bank to imply similar
levels of expected loss. A hypothetical example of such a matrix is shown in Appendix 2.
3.1.2 Number of risk grades
Rating systems typically comprise one or more ‘pass’ grades plus several lower-quality ‘non-pass’ grades
reserved for exposures displaying varying degrees of credit impairment. Pass grades usually include those
rating categories that a bank considers to be of acceptable quality for new loans under its usual
underwriting standards and may also include one or two watchlist (or special mention) grades.
Internationally, the number of customer risk grades varies considerably among banks (Graphs 1 & 2).
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200011
Graph 1: Number of pass grades at international banks – Basel survey1
Graph 2: Number of non-pass grades at international banks – Basel survey1
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200012
0
10
20
30
40
0-1 5Number of grades
Perc
ent
of
ban
ks
%
2 3 4 6
1 The distinction between pass and non-pass grades, though commonly used, is not standard nomenclature around the world. Consequently,banks responded inconsistently to the Basel survey when asked to split their risk grades into these two categories. Some banks, for example,appear to have included watchlist grades within the pass category while others have included these grades in the non-pass category; one bank reported that it had no non-pass grades.
0
10
20
30
40
< 5 5-9 10-14 15-19 20-24Number of grades
Perc
ent
of
ban
ks%
A general trend has been for banks to increase the number of pass grades as the range and sophistication
of uses to which their ratings are applied has grown. As long as raters can achieve the finer distinctions
required, rating systems with more risk grades – greater granularity – convey more information than
systems with fewer grades and can enhance a bank’s ability to analyse and model its portfolio of
credit risks.8
Larger banks also tend to have more customer risk grades than smaller banks. The cost/benefit analysis
of maintaining a larger number of risk grades tends to be more favourable for larger institutions. The
latter generally have more complex credit portfolios (comprising many more customers and a wider
spectrum of risk) and are more likely to have introduced other sophisticated techniques of portfolio
analysis that require ratings as inputs. Also, larger banks are usually better positioned, and have more
resources, to develop and support more granular rating systems (though with industry pooling of default
data becoming more prevalent, and the increasing availability of externally developed rating tools, this
may be becoming less of an issue).
With regard to the number of customer risk grades, Australian banks are mostly clustered around the
middle of the international spectrum (Graph 3). Most of the local banks have either 9 or 10 customer
PD grades; one large bank has 22 main grades. Another large bank, which has a general 9-grade
customer rating scale, utilises an expanded 19-grade scale for its large corporate customers. All of the
banks have from 3 to 5 non-pass/watchlist grades, though in some cases these administration grades may
be further sub-divided into a larger number of sub-grades.
Graph 3: Total number of customer risk grades at Australian banks
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200013
0
1
2
3
4
5
6
9 1 2 2Number of grades
Nu
mb
er o
f b
anks
1 10 1 2
8 Banks, however, need to exercise caution so as to avoid going beyond the point where they can no longer makemeaningful distinctions concerning the riskiness of different exposures. Banks that have linked portfolio risk modelling with risk-based pricing and/or profitability measures can face strong pressures in this regard, including from business lines looking for rating scale refinements to assist in meeting pricing and other performance targets.
Tighter clustering in the number of risk grades among Australian banks reflects the fact that several of
the smaller and mid-sized banks have recently (within the past two years) expanded the number of risk
grades as part of wider upgrades of their rating systems. Further change is occurring, with some of the
largest banks in the process of shifting to more granular rating scales. Like a growing number of their
overseas counterparts, the number of risk grades at quite a few Australian banks has been chosen to
facilitate comparisons with the rating scales of the major international rating agencies, Moody’s and
Standard & Poor’s.9
There is less clustering in the number of LGD grades, which are often based on security coverage ratios,
among banks in Australia (Graph 4).
Graph 4: Number of LGD grades at Australian banks
3.1.3 Distribution of exposures among risk grades
Regardless of the overall number of risk grades, the granularity, and therefore usefulness, of a bank’s rating
system will be reduced if credit exposures tend to be concentrated in only one or two risk grades.
Again, Australian banks seem to be broadly in line with current international practice (Graph 5). In
the case of most of the local banks, a maximum of about a third of rated exposures falls within a single
grade. All three of the overseas comparison studies, two of which are able to be depicted in the graph,
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200014
0
1
2
3
4
3 4 5 6 7 8 9 10 16Number of grades
Nu
mb
er o
f b
anks
9 The rating scales for both of these agencies comprise 10 main ‘alpha’ grades (including a default category) or 22 grades including ‘+/-’ modifiers. As some of the lower agency grades tend to fall outside the banks’ stated normal underwriting standards, some of these grades may either be dropped or grouped together when compared with the banks’ internal rating scales.
found similar results. In the case of three of the regional Australian banks, which rely heavily on the
judgement of raters, between 50 and 70 per cent of rated exposures fall within a single grade.
Graph 5: Concentration of exposures in largest risk grade
3.1.4 Point-in-time vs through-the-cycle grading
Institutions tend to be characterised as taking either a through-the-cycle or point-in-time rating approach.
External rating agencies usually state that they take a through-the-cycle approach, ie a borrower’s rating
is meant to stay the same over the course of an economic cycle unless the firm experiences a major
unexpected shock (good or bad) to its perceived long-term condition or the original downside scenario
used to rate the borrower proves to have been too optimistic. Under this approach, observed default
frequencies for particular ratings will tend to vary with the economic cycle.
Banks, on the other hand, including those in Australia, usually say they take more of a point-in-time
approach, ie they base their ratings more on the borrower’s current condition so that a borrower is more
likely to migrate up and down through the ratings as economic conditions change. Under this approach,
observed default rates for particular ratings should tend to be more stable over the economic cycle.
These differences can have important implications when interpreting ratings-based information,
particularly when ratings from different sources are used in combination, eg when mapping a bank’s
internal grades to external agency grades. Failure to take such differences into account can potentially
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200015
0
10
20
30
40
50
60
70
Less than 20 20-29 30-39 40-49 50-59 60-69 70 or morePercent of exposure in largest category
Perc
ent
of
ban
ks
Basel Committee
Treacy & Carey
Australian Banks
%
produce substantially biased estimates of default probability, distorting internal analyses of loan loss
provisioning requirements, appropriate allocation of credit capital, risk-adjusted performance
and pricing.10
3.2 Operating design features
3.2.1 Exposures rated
In principle, the banks apply ratings to all of their corporate, commercial and other transaction-managed
exposures. In practice, exposures below certain thresholds (typically $100,000 or $250,000) tend not to
be rated on an individual basis. At some banks these exposures are assigned a representative rating on a
portfolio basis, at others the exposures remain unrated.
Portfolio ratings may also be assigned to certain other segments of some banks’ loan books. Usually these
segments comprise relatively homogeneous groups of small loans that tend not to be actively managed at
an individual customer level. Such loans include small leasing exposures and exposures that are primarily
assessed on the basis of available security, such as certain stand-alone property exposures or small business
loans secured over the owner’s residence.
3.2.2 Customer PD ratings
(i) Types of rating processes
As outlined in section 2, banks incorporate a wide range of risk evaluation techniques into their rating
processes. In its survey of internal rating systems, the Basel Committee identified three main categories
of risk evaluation processes depending on the degree of judgement involved:
• statistically-based processes – at a small number of the surveyed banks, a statistical default model
or some other quantitative tool was essentially the sole basis for determining ratings – at least
within certain portfolios;
• constrained expert judgement-based processes – were more common. In these systems, ratings
were primarily model-based but relied more heavily on qualitative inputs, non-statistical (ie
expert judgement-based) models and/or permitted to a limited degree (usually one or two
ratings notches at most) judgemental overrides of model-determined ratings; and
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200016
10 Treacy & Carey (1998) provides a good discussion of the potential biases involved. In practice, though, the differences in approach between the external rating agencies and banks might be somewhat less sharp than the above discussion implies. For example, where a bank’s rating system involves a good deal of expert judgement and/or permits judgemental overrides of objective ratings criteria, this could contribute to some ‘stickiness’ in the bank’s ratings. On the other hand, various studies indicate that there is a systematic relationship between agency rating transitions and current economic conditions, particularly among lower rated entities, eg refer Basel Committee on Banking Supervision (2000b), pp 141-4.
• (unconstrained) expert judgement-based processes – more than half of the banks surveyed fell
into this category. Some banks had no formal debt rating models while others had models but
permitted raters considerable discretion to override model-determined ratings based on their
professional judgement.
The Australian scenario is similar to that portrayed for other countries by the Basel Committee study. In
terms of the Basel study, most of the local banks would be classified as having (unconstrained) expert
judgement-based processes. Two of the smallest banks have highly subjective rating systems. The other
banks, over recent years, have introduced a variety of rating assessment tools to assist staff in their rating
determinations, including:
• use of external ratings (where these are available);
• more tailored rating definitions (incorporating more detailed quantitative and qualitative rating
benchmarks, often differentiated by customer size and/or industry); and
• debt rating models.
Several banks (mostly the larger institutions with more complex portfolios) utilise multiple models and
other rating tools; these are either tailored to different parts of the banks’ portfolios or provide different
approaches to rating the same exposures.
The banks’ existing debt rating models include a mix of expert judgement-based and statistical models,
either developed in-house for various segments of the banks’ credit portfolios or purchased from external
vendors. The latter tend to be founded mainly on overseas (mostly US-based) experience, though some
ratings systems developed from locally pooled data are now becoming available.
In more recent times, there has also been a trend towards greater reliance on statistical methods, though
only the largest banks have credit portfolios of sufficient size (and the resources) to develop statistical
models based solely on their own loss histories. Most of the statistical models in use are scorecard-type
models, ie the models utilise historical data on defaulted and non-defaulted borrowers to group (rate)
customers with similar characteristics. A couple of the larger banks have developed such models based
on their own loss experience; others use models purchased from external sources. Several of the larger
banks also use externally developed equity-based models to assist in rating both listed and (large) unlisted
corporate borrowers. These models statistically infer default probabilities from so-called distance-to-
default measures based on borrowers’ liability/asset structures and stock-price volatility data.
Most of the local banks permit judgemental overrides of their rating models; in fact, most emphasise to
their raters that rating models are to be used as tools and should not supplant comprehensive analysis of
exposures for rating purposes. Four banks do not permit overrides (at least by front-line lending staff).
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200017
In terms of the Basel study, these banks’ rating systems fall largely within the middle category (ie
constrained expert judgement-based processes) due to the importance of a range of qualitative elements
in their main rating models.
(ii) Key rating risk factors
For those banks that have highly subjective rating systems and/or allow judgemental overrides of model-
based ratings, traditional credit analysis plays the sole/potentially primary role in assigning final ratings to
credit exposures. In these cases, individual raters have wide discretion over what information is
considered and how it impacts on final rating assessments. The pervading credit culture within the
organisation will be an important factor influencing rating outcomes that will be communicated to raters
through such channels as training programs and feedback via mentoring, performance assessment against
key indicators and other oversight mechanisms.11 As mentioned, a number of banks have also sought to
provide more explicit guidance to raters by including various benchmarks in their formal rating criteria,
eg in the form of key indicator comparison tables.
The various model-based assessments cover a wide spectrum of different variables. Leaving aside equity-
based models, most of these variables can be grouped into a few broad categories relating to the obligor’s
financial condition and capacity/cashflow, management quality and industry risk characteristics. Most
models currently in use involve several variables from each of these categories; many also input
information about account conduct (particularly those models dealing with the banks’ smaller customers).
These categories represent key areas of traditional credit analysis. Virtually all models in use also involve
both quantitative and qualitative inputs. The latter often generate a good deal of debate during model
development. Although there is evidence that such data are predictive, subjective inputs are also more
open to interpretation/manipulation and can be less responsive to changed business circumstances.
(iii) Link to quantitative default statistics
Banks attempt to quantify the loss characteristics of their risk grades for a variety of reasons. As discussed
in section 3.2.5 below, by tracking the ex post performance of ratings, institutions seek to test whether
their rating systems are effective in differentiating among exposures with different degrees of risk. By
linking internal rating scales to numerical estimates of PD, LGD and/or EL, credit ratings can also be
turned to a variety of quantitative applications, including in more sophisticated pricing, profitability and
performance analyses, and as inputs in loss provisioning and economic capital allocation modelling.
Seven of the surveyed Australian banks assign quantitative default characteristics to their internal ratings.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200018
11 For example, inconsistent incentives built into an organisation’s oversight structures and culture can weaken the rating process, eg where business volume key performance indicators take precedence over quality indicators or risk-based performance indicators are not supported by strong quality assurance processes.
The three banks that do not assign quantitative default characteristics to their internal ratings, possess
rating systems which are not well suited to establishing quantitative default estimates. Given the nature
and small size of their commercial lending portfolios, these banks currently see limited benefits from
introducing more sophisticated analysis techniques.
In line with overseas experience, Australian banks currently lack long-term data on the performance of
their internal rating systems. Like their overseas counterparts, the local banks have sought other ways of
assigning quantitative default (and other credit migration) probabilities to their internal risk grades. In
some cases, average expected default frequencies for each risk grade are based on the default rates
predicted by the banks’ statistical debt rating models. In other cases, internal risk grades are mapped to
the rating scales of longer-established external rating schemes for which long-term performance data are
available (ie Moody’s and/or Standard & Poor’s). Some banks use a combination of both methods.
Mapping to external benchmarks is not straightforward and requires considerable judgement on the part
of analysts in comparing banks’ internal rating processes with those of the external agencies. This can
involve comparing the definitions and criteria underpinning their internal grades with those of the rating
agencies; analysing the financial characteristics of internally rated borrowers with standard ratios that
characterise the agency grades; and analysing the grades of internally rated borrowers that also have
external ratings.12
In order to facilitate the mapping process, many banks have developed internal rating scales and associated
rating criteria that attempt to mirror those of the external agencies. Overseas, some banks statistically
model the external agencies’ rating processes. The models are then used to rate the banks’ customers. This
latter approach has not been adopted by any of the Australian banks, though the technique has been applied
by some of the banks to examine how closely their internal processes align with those of the external
agencies. Overall, differences in the approaches adopted by different banks have led to a good deal of
variation among their nominal PD scales, even where ratings categories are meant to be broadly equivalent.
Benchmarks based on external experience (whether that of external rating agencies or an externally
developed debt rating model) also give rise to questions about their applicability to a bank’s particular
client base and lending practices. Inconsistencies in such things as rating approach (eg point-in-time vs
through-the-cycle rating), portfolio composition (eg in the size, nature of operations and domicile of
borrowers) or type of borrowing instrument (eg bonds vs bank loans) can lead to substantial bias when
mapping internal grades to externally sourced PD values.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200019
12 In the latter case, however, there can be a good deal of circularity involved when external ratings are also an input in the internal rating process.
3.2.3 Facility LGD ratings
A more mechanical approach is typically taken when assigning LGD ratings than for PD ratings. LGD
ratings are generally assigned in one of three ways:
• on the basis of security coverage ratios – this is the main method used by all banks for most
types of exposures. Security coverage is usually based on a discounted or ‘shaved’ loan to value
ratio, whereby estimated loan security values are discounted by standardised ‘safe lending’ or
‘extension’ margins.The discount factors seek to cover such factors as uncertainties inherent in
the valuation process (including liquidity considerations) and potential subsequent downturn in
security values.The banks’ policy documents set out standard valuation procedures and discount
factors based on types of security. The ‘discount factors tend to be based largely on management
judgementand traditional industry benchmarks/‘rules of thumb’. The smallest banks tend to
extend value only on real estate or high quality assets, such as cash and government securities,
whereas larger banks recognise a wider range of security types in their rating procedures;
• by directly estimating an expected recovery percentage – eg in the case of impaired assets; or
• by applying a generic classification based on the type of exposure – some banks use this
approach for certain types of exposures, eg in the case of subordinated debt, small leasing
exposures, exposures secured over residential or non-specialised commercial property and
exposures to other banks.
Currently, five Australian banks seek to assign quantitative LGD values to their credit exposures. Usually
this is based on LGD ratings but in some cases also depends on the product type or line of business.
As in other areas of credit risk rating, this aspect of the banks’ systems is heavily constrained by data
limitations. Although most of the local banks are working to reduce some of these limitations, key LGD
parameters are currently a mix of the banks’ own loss histories (mostly restricted to data on Australian-
based customers and covering a limited time span), published studies of bond and commercial loan
experience (mostly data on US-based customers) and a large degree of management judgement.
Where banks have undertaken LGD studies, methodological differences have arisen. For example, while
some banks base LGD estimates on discounted cash flows and the inclusion of workout costs, others
exclude these factors (usually because the required historical data are unavailable or not easily accessible).
The choice of an appropriate basis for discounting delayed cash flows can also affect the results of these
studies; while some banks favour discounting on the basis of contractual rates of interest, others favour a
discount rate that is related to the cost of equity.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200020
3.2.4 System oversight and control processes
Apart from debt rating models, banks in Australia and overseas incorporate a range of other features into
their rating processes designed to enhance the accuracy, integrity and consistency of ratings throughout
their operations. Generally speaking, these measures seek to reduce the degree of subjectivity and/or
inject greater independence into their rating procedures. Some of the main control features include:
• at all banks surveyed, broad ownership of the rating system resides with an independent central
credit management unit that reports to the bank’s Board/Board Risk Management Committee.
Responsibility for overall oversight of the system’s operation falls to this unit. While business
lines might suggest system modifications, this unit is responsible for approving changes to the
system, with major changes also requiring approval from the Board/Board Risk
Management Committee;
• the assignment of credit ratings is integrated into the banks’ normal credit approval/review
processes and is subject to the checks and balances built into those systems. Typically:
– while relationship managers retain the main responsibility for accurately assigning ratings,
credit approval/review policies and processes channel larger and other riskier13 credit
exposures into the bank’s independent credit line and/or to higher or specialist approval levels
within the credit line;
– at banks where overriding is permitted, a higher approval level is required where
lending/credit officers propose modifying the rating recommendations of the bank’s rating models;
– credit assessments and related ratings are subject to formal periodic review, at least annually
in most cases. Most banks seek to enhance the timeliness of credit reviews, and any associated
rating adjustments, by also specifying early review events and/or more frequent periodic
reviews for lower-rated exposures. At some banks, early review policies are supported by
centralised and/or automated monitoring systems;
• among other aspects of credit risk management, ratings at most of the surveyed banks are subject
to regular and ad hoc review by independent credit review teams, with reporting of main trends
to top management levels within the banks and to board risk management and internal audit
committees. Particular exposures are reviewed on a sample basis, with higher proportions of
loans for review typically being drawn from riskier grades and areas of growing concentration.
Poor credit process ratings received from these review teams can lead to a reduction or
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200021
13 For example, at some banks sole lending delegations are substantially reduced in the case of new customers (particularly those seeking to refinance existing loans from other institutions) or existing customers that are branching into new fields of activity.
cancellation of delegated credit authority and, at several banks, can directly affect
staff remuneration;
• several banks also test the performance of their rating systems, including through regular
monitoring of credit migration data against expected outcomes.
In addition, some banks have established, or are considering establishing, automated data transfer linkages
to minimise/eliminate rekeying of ratings input data (eg financial ratios from financial spreading packages)
and of completed ratings. The aim is to improve system efficiency, including by reducing inconsistencies
in different data management systems caused by transcription error, failure to update databases or
potential manipulation of ratings information. Some systems also seek to track potential instances of
‘gaming’ rating models whereby loan officers might alter customer information and re-enter it several
times in order to obtain a better rating recommendation. As a further means of enhancing rating
consistency, efficiency and overall accuracy, at some banks where industry characteristics form an
important input into rating models, an economics or other specialised unit, rather than individual
lending/credit officers, is responsible for inputting relevant industry assessments.
A couple of banks regularly undertake centralised monitoring of model override trends. Such
monitoring can help indicate potential problems in the way rating models are being used within a bank
and/or deterioration in model performance. That said, overrides are not ignored at other banks, as the
review of override decisions is an important aspect of the responsibilities of the banks’ credit line and
credit review teams.
3.2.5 Validation procedures
On-going testing of the system’s ability to distinguish accurately the riskiness of different exposures is an
important element of any credit risk rating system. It also requires a good deal of subjective judgement
(given the long-term nature of credit cycles and short-term record of existing bank systems). Validation
procedures can involve comparing evolving credit migration statistics against expectations and/or
comparing internal ratings with other available rating alternatives, eg external agency ratings and/or
externally developed rating models.
These procedures aim to:
• confirm the continuing ability of the system to rank risk effectively;
• test the accuracy of quantitative PD estimates (where these have been assigned to risk grades);
and/or
• compare the discriminant power of alternative rating schemes.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200022
The larger Australian banks regularly review credit migration data and from time to time undertake
comparison studies of alternative rating systems. While the various analyses of credit migration data
suggest that the banks’ rating systems can rank risk in a relative (or ordinal) sense, the accuracy of their
associated quantitative probability scales remains more problematic given that available data have only
been drawn from the good part of the current credit cycle. Most of the smaller banks are currently
building up their internal default histories but to date have gathered insufficient data to form valid
conclusions as to the efficacy of their rating systems.
3.3 Rating applications
As outlined earlier, the applications for which an internal rating system is utilised provides an important
indication of the degree of sophistication that should be built into the system’s design and operating
features. Banks that utilise internal ratings for relatively sophisticated purposes (eg as inputs in statistically-
based provisioning, capital allocation and risk-based pricing models) require more sophisticated rating
systems that feature adequate risk differentiation, strong control frameworks and ratings that are linked to
quantitative default characteristics. However, where used for less demanding analytical tasks, a less
sophisticated system may be adequate.
Australian banks’ internal risk ratings are used in varying degrees in a wide range of applications which,
on an industry basis, cover the full scope of administrative, reporting and analytic uses outlined in
section 2.3.
Table 2: Administrative uses
For example, the larger banks utilise their internal risk grading systems to assign delegated credit approval
authorities to lending personnel. At these banks, the maximum amount that each lending/credit officer
may approve for any particular obligor varies by risk grade; ie delegated lending authorities are larger for
less risky grades and vice versa. Use of this technique affords the banks greater flexibility in tailoring
lending delegations to the skills and circumstances of particular lending officers. Among the smaller
banks, ratings are used more simply in the delegation process. Typically, lending staff are prohibited from
authorising new lending below a certain threshold rating; above that threshold, lending authority is
typically restricted to a fixed dollar amount irrespective of the obligor’s rating.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200023
Delegated credit approval authorities 5
Problem loan management 10
Number of Banks
All of the Australian banks surveyed also use their rating systems to facilitate problem loan management.
Typically, when an exposure is assigned certain (low) grades, it becomes subject to requirements for:
• more frequent monitoring and reporting on the condition of the obligor and prospects
for repayment;
• the development of a formal rehabilitation or exit strategy; and/or
• transfer to a specialist asset management unit.
Table 3: Analytic uses
All of the Australian banks utilise ratings for portfolio monitoring and management purposes. Among
other things, the banks’ internal rating systems are used to report to high management levels total asset
balances, large exposures and relative change in distributions for each risk grade. This information
provides management with analyses of the mix of loans within the banks’ portfolios and various sub-
portfolios (including by business line, industry and product type), data on problem assets and the risk
profile of assets within pass grades. Ratings are also used to communicate risk-differentiated business
acquisition strategies, eg in developing customer target profiles for particular products.
Four banks have developed statistical provisioning and credit risk capital modelling capabilities. Output
from these models, which incorporate credit ratings as major inputs, is used for a variety of purposes,
usually including:
• benchmarking total capital and general provisioning requirements;
• internal capital allocation;
• internal risk reporting;
• portfolio management;
• the setting of credit risk concentration limits;
• risk-adjusted profitability measurement (using risk-adjusted return on capital and shareholder
value added concepts); and
• developing risk-based pricing benchmarks.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200024
Portfolio management/monitoring 10
Provisioning/capital allocation modelling 4
Risk-based pricing 10
Number of Banks
Some of these banks are also beginning to use the information to affect performance-based remuneration
in some parts of their businesses.
4 CONCLUDING REMARKS
The preceding sections of this report have outlined the current state-of-play regarding credit risk rating
among Australian banks and provided some international comparisons. Generally speaking, Australian
banks’ credit risk rating practices appear to be in line with those of their international peers.
Internal bank rating practices, both locally and overseas, continue to evolve as the experience of rating
institutions builds. As a group, the local banks have moved relatively quickly to adopt two-dimensional
credit risk rating approaches whereby customer default probabilities and expectations of loss in the event
of default are separately rated. In other respects, as elsewhere in the world, considerable differences exist
among the banks’ rating systems, particularly in relation to the detail of how ratings are determined, the
associated quality control processes that have been established in each institution, and the way in which
quantitative values have been assigned to risk grades.
These differences reflect many influences deriving from the particular circumstances of each institution,
including differences in the size and nature of banks’ rated portfolios, the intended applications of ratings,
the capabilities of banks’ available resources and systems, the legacy of past decisions, and costs of change.
As noted elsewhere in the paper, after weighing the various considerations involved, no single ratings
approach is necessarily right for all institutions or for any particular institution over time.
A key challenge faced worldwide by virtually all developers and users of internal credit risk rating
systems, including prudential supervisors looking to utilise banks’ internal ratings for regulatory capital
and other purposes, is the widespread lack of good long-run data on the performance of banks’ loans.
The lack of such data can impact on the ability of an institution to develop effective rating tools. It can
also impede efforts to verify the accuracy and robustness of institutions’ rating systems, assign reliable
quantitative loss estimates to risk grades, and make reliable comparisons of ratings from different
institutions: all important tasks not only from the perspective of banks, themselves, but also their
prudential supervisors (particularly in the context of proposals to utilise banks’ internal ratings for
regulatory capital purposes).
While many banks have implemented data warehousing processes aimed at improving this situation, given
the length of a typical credit cycle, it will be some years yet before these banks have data covering one
full credit cycle let alone several different cycles. Moreover, irrespective of the stage of the credit cycle,
institutions with small rated portfolios may not experience sufficient defaults for many years to come, if
at all, before any reliable statistical analysis of their default experience or the performance of their rating
systems can be undertaken.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200025
Other means by which banks have sought to overcome data deficiencies include industry pooling of
default data, use of other external benchmarks and externally developed rating models. Data pooling and
externally-developed models can be of particular potential benefit to smaller banks, given their generally
more limited data and other resources, and their narrower exposure to the risk spectrum. As discussed,
however, use of external resources also exposes institutions to risks relating to the applicability or
otherwise of the external measures to their individual customer bases and lending practices, and whether
the external benchmarks, models etc will continue to exist and be appropriately maintained.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200026
APPENDIX 1: TYPES OF CREDIT RISK RATING MODELS
Credit risk rating models can be broadly classified as either being based mainly on expert judgement or
statistical analysis.
Expert judgement- or rules-based models are non-statistical models that have been developed on
the basis of professional judgement. The models can take the form of decision trees or, more frequently,
scorecard-type models where both the input variables and associated coefficients are judgementally
derived based on the knowledge and experience of an organisation’s expert lending/credit officers.
Where sufficient data are available, statistical models may be developed utilising a variety of techniques.
For example, some overseas banks seek to model the decision processes of the major rating agencies by
regressing various financial ratios and other variables against the agencies’ ratings. More frequently,
statistical techniques are used to analyse differences in the characteristics of large groups of banks’
defaulted and non-defaulted customers. Usually, one of three basic approaches is adopted:
(i) linear probability models – these models are developed by regressing a selection of quantitative
and/or qualitative variables describing borrower characteristics against a (dichotomous) dependant
variable that takes a value of 1 if a loan is in default or 0 otherwise. One problem with such models
is that the estimated default probabilities are not guaranteed to lie within a range of 0-1. That said,
linear probability models remain the most common means of credit scoring/rating consumer and
other retaillending products;
(ii) logit/probit models – utilise more sophisticated regression techniques that constrain estimated
default probabilities within a 0-1 range. This is achieved by assuming that default probabilities are
distributed in particular ways within that range (eg probit models assume that default probabilities
are normally distributed while logit models assume that a logistic distribution is more appropriate).
Logit and probit models have become more common with increased computing power and the
development of statistical debt rating models for commercial and corporate borrowers;
(iii) linear discriminant models – tend to be more commonly found in academic literature. These
models seek to establish a linear classification rule (or formula) that best distinguishes between
particular groups of borrowers (specifically defaulters and non-defaulters). This is achieved by
analysing a range of variables (borrower characteristics) in order to determine a set of coefficients
that maximises the ‘between group’ variance, while minimising the ‘within group’ variance, of a
sample set of defaulting and non-defaulting borrowers. Linear discriminant models assume that
borrowers belong to discrete categories, unlike the above-mentioned probability models, which
envisage a graduated spectrum of borrowers from low to high risk.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200027
Some other statistical techniques, such as artificial neural networks and genetic algorithms, have an ability
to ‘learn’, ie these techniques can automatically adjust an institution’s rating models as new default data
come to hand that indicate underlying relationships between the characteristics of borrowers and
propensities of borrowers to default are changing or could be better specified. Although several Australian
banks use the techniques for various other purposes, eg potential fraud detection, none of the banks
currently use neural networks or similar techniques for credit risk rating purposes. Similarly, we are not
aware of these techniques being used for ratings purposes overseas, except possibly at a very few banks or
on an experimental basis. One Australian bank has indicated that it is contemplating using neural
networks as a diagnostic check of its existing rating models rather than as its main rating tool.
Some banks also use equity-based or option-theoretic models (usually purchased from external vendors)
for relevant parts of their portfolios. These models use information contained within a borrowing firm’s
share price as an indication of credit risk. US-based KMV Corporation is the main supplier of these
models globally. Moody’s Risk Management Services has also recently developed a public firm rating
model, which takes a hybrid approach that combines a standard probit model with
option-theoretic concepts.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200028
APPENDIX 2: EXAMPLE TWO-DIMENSIONAL CREDIT RISK RATING MATRIX
This hypothetical matrix (or look-up table) illustrates how a 9-grade composite credit risk rating scale
might be derived from an 8-grade PD rating scale and 7-grade LGD rating scale.
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200029
Customer Facility LGD ratingPD Rating
1
2
3
4
5
6
7
8
A B C D E F G
1 1 1 2 2 3 3
1 1 2 2 2 3 3
2 2 3 3 3 4 4
3 3 4 4 4 5 5
4 4 5 5 6 6 6
5 5 6 7 7 7 7
7 7 8 8 8 8 8
8 8 9 9 9 9 9
References
Basel Committee on Banking Supervision (1999), A New Capital Adequacy Framework, June,
(www.bis.org).
Basel Committee on Banking Supervision (2000a), Range of Practice in Banks’ Internal Rating Systems,
January, (www.bis.org).
Basel Committee on Banking Supervision (2000b), ‘Credit Ratings and Complementary Sources
of Credit Quality Information’, Basel Committee on Banking Supervision Working Paper 3, August,
(www.bis.org).
Eales, R (1997), ‘Credit Risk in Corporate Banking – Theory and Practice’, Credit Risk in Banking:
Proceedings of a Conference, Reserve Bank of Australia, May.
English,WB & Nelson,WR (1998), ‘Bank Risk Rating of Business Loans’, Working Papers: Finance &
Economics Discussion Series, Board of Governors of the Federal Reserve System, November,
(www.federalreserve.gov).
Treacy, WF & Carey, MS (1998), ‘Credit Risk Rating at Large US Banks’, Federal Reserve Bulletin,
November, (www.federalreserve.gov).
CREDIT RISK RATING AT AUSTRALIAN BANKS
APRA DECEMBER 200030