HOM= E=20

=20 SUBSCRIBE=20

=20 CONTACT=20 US

=20 EMAIL = BULLETIN=20

=20 FREE SAMPLE

=20 ADVANCED= =20 SEARCH

NEWS=20
FEATURES=20
SUPPLEMENTS=20
> SERVICES=20
- Events=20
- Forward=20 Features=20
- Multime= dia=20
- Awards=20
- Sibos Daily News=20
- Email=20 Bulletin=20
- Forgotton=20 Password
ADVERTISING=20
DIRECTORY=20
> MAGAZINE=20 ARCHIVE=20
- Browse = Archive=20
- Advanced= =20 Search

OpenCube - The Internets #1 CSS Menu, = Drop Down=20 Menu, Flyout Menu, and Pop Up menu Developer

GROUP PRODUCTS

=20 =20

MEDIA PARTNER

=20 =20

RELATED PUBLICATIONS

=20

July = 2006Features

Quantity and quality

The banking world is increasingly relying on = automated,=20 quantative modelling for credit risk assessments. Nicholas Pratt = asks=20 what room is there for quantitative analysis, particularly for SMEs = where=20 reliable, quantitative data is hard to come by.

There is a recurring sketch on one of the UK=E2=80=99s popular comedy = shows where a=20 man goes into a high street bank to enquire about a loan. Opposite him = sits a=20 moribund, monotoned and bespectacled loans advisor behind the desk and = in front=20 of a standard looking computer. The customer presents his credit = details, they=20 are entered into the computer and the result is awaited. After a few = seconds of=20 comedic tension, the advisor simply says: =E2=80=9CThe computer says = no.=E2=80=9D

The punchline is that =E2=80=9Cthe computer says no=E2=80=9D to every = single customer request=20 =E2=80=94 from starting a bank account to asking for a glass of water. = Presumably, the=20 gag is aiming to satirise the automated nature of much of = today=E2=80=99s retail banking=20 and the way that lending decisions are made.

Automation is obviously a major feature of any bank=E2=80=99s = processes and the=20 development of risk modelling has been geared towards automating the = risk=20 management process and basing credit assessments on quantative rather = than=20 qualitative factors, on statistical rather than judgemental = scorecards.

Much of this emphasis on the quantitative has been enabled by an = unwavering=20 confidence in the robustness and capability of the models used =E2=80=94 = an unwavering=20 confidence that is not always justified. =E2=80=9CThe assessment of = credit risk is the=20 key foundation of the banking industry, but the development of robust = quantative=20 credit risk models is still to some extent in its infancy,=E2=80=9D says = Michael Rees,=20 training and consulting director at Palisade, a UK-based provider of = risk=20 analysis tools.

=E2=80=9CThere have been a number of applications of new modelling = techniques,=20 including advanced credit scoring, credit derivatives modelling and = neural=20 networks. The quality of these models depends both on whether the model = captures=20 key behaviours in the real world and on the accuracy of the = model=E2=80=99s=20 parameters.

=E2=80=9CA big issue relating to both of these points is the lack of = sufficient=20 historic data, and that when such data is available, it may not = necessarily=20 provide an accurate indication as to the future,=E2=80=9D says Rees.

The lack of data is a relative issue and particularly acute in the = sometimes=20 vaguely-defined small and medium enterprise (SME) sector. In some = definitions an=20 SME is a sole trader, in others it is a business pulling in up to $50m = in annual=20 turnover. According to Suresh Sankaran, director of strategic consulting = at risk=20 management firm IPS Sendero, the common link is that they are = =E2=80=9Cgenerally=20 privately held companies and, whilst subject to some if not all = financial=20 disclosure requirements, are inherently different from listed = entities=E2=80=9D.

The most important consideration with those SMEs at the smaller end = of the=20 spectrum, says Sankaran, is the net worth of the owner, and this is a = powerful=20 indicator of the exposure that a bank is willing to have with an SME. = =E2=80=9CSince=20 they are often sole ownerships or privately held, there are no exchange=20 listings. They employ auditors but in most instances, the numbers can be = compromised through window dressing. Basel II has tried to encourage = banks to=20 treat SMEs differently, but the inherent drawback is the shallowness of = data and=20 the lack of quality,=E2=80=9D says Sankaran.

To counter the scarcity of data that exists in the unlisted SME = sector, and=20 inspired by the capital adequacy demands of Basel II, there have been = various=20 efforts made by investment banks to pool financial data and default = experience.=20 In the US there is more of a framework for data consortia than in = Europe, in=20 part due to the greater number of trade associations, however various = endeavours=20 to gather more shared data on either side of the Atlantic have not borne = fruit.

=E2=80=9CData consortia is a big challenge,=E2=80=9D says Paul = Waterhouse, global head of=20 analytics at rating agency Standard & Poor=E2=80=99s Risk Solutions = division. =E2=80=9CIt=E2=80=99s=20 a big investment for banks and also for whoever takes on the process of=20 collating the data.=E2=80=9D Investment banks are understandably = reluctant to commit to=20 many of the data consortia, not solely because of the investment needed = but=20 because they feel they are not getting the reward their efforts = warrant.

=E2=80=9CThere are constantly various consortiums aiming to collect=20 loss-given-default data,=E2=80=9D adds Jennifer Courant, managing = director, Risk Rating=20 Analytics Group, Citigroup. =E2=80=9CCitigroup has participated in some = consortiums in=20 the past.=E2=80=9D

The rating agencies have been prominent in adopting a role as the = industry=E2=80=99s=20 collators of default data, but there are drawbacks to this multi-bank = collation.=20 =E2=80=9CWhen we are collecting data, banks may have different processes = and protocols=20 as concerns to what they regard as a default,=E2=80=9D says Waterhouse. = =E2=80=9CThey also=20 estimate recovery in different ways, so what data you get is not fully=20 homogenous.=E2=80=9D

Elsewhere there are elements of scepticism about the suitability of = rating=20 agencies to act as these self-appointed industry collators. = =E2=80=9CThere has been=20 general unhappiness with data pools run by rating agencies because they = have=20 created commercial products based on this data but paid no royalties to = the=20 owners of the intellectual property =E2=80=94 the banks who provided the = data,=E2=80=9D says Don=20 Van Deventer, founder of Kamakura, a Hawaii-based developer of risk = management=20 software.

=E2=80=9CThe rating agencies do not make loans so any data they have = has either been=20 purchased from commercial sources or given to them by banks. They have = no=20 intrinsic source of data unique to the rating agency, unlike a = lender.=E2=80=9D

Which means that many banks are left to rely on their own proprietary = data=20 and models when it comes to SMEs. =E2=80=9CWe have been systematically = capturing data,=20 creating an infrastructure and storing the data in a manner that it can = be=20 cleaned and its integrity maintained,=E2=80=9D says Courant at = Citigroup.

=E2=80=9CWhen modelling SMEs, where there are defaults, we test them = extensively. We=20 look for consistencies across credit-sensitive variables, like debt = service=20 coverage or leverage. It is a case of common-sense credit analysis. We = have a=20 very controlled and regulated processes and it is continually = reviewed.=E2=80=9D

The models are not based solely on the banks=E2=80=99 own default = data. When it comes=20 to SMEs other data from sources, such as regional credit bureaus and=20 macro-economic factors, are involved but also various qualitative = factors.=20 =E2=80=9CGiven the shallowness of financial ratios, it is very important = to incorporate=20 qualitative factors,=E2=80=9D says Sankaran.

=E2=80=9CMore often than not, the qualitative factors have more = weight attached to=20 them than the financial ratios, and I know of several organisations who=20 deliberately assign a lesser weighting to the financial statements and = ratios of=20 an SME and apply a greater weight to the qualitative factors. This is = done to=20 ensure that a grain of cynicism is incorporated into the default = modelling=20 framework.

=E2=80=9CA great example would be the use of a net income to total = assets ratio. This=20 ratio has been proven to be a very accurate predictor of default, but = when we=20 did a project in Taipei, we found this was not the case. The bottom-line = in this=20 project was that since most of the SMEs were cheating on their taxes, = the ratio=20 as it is practiced was meaningless, so we used the inverse as a = qualitative=20 variable.=E2=80=9D

The weighting of qualitative versus quantitative factors is a dilemma = for=20 banks, particularly in the SME sector where qualitative or judgmental = issues=20 carry so much more influence. In the age of mathematically constructed = models=20 and automated processing, how do these judgmental factors fit in?

=E2=80=9CThe qualitative factors can be objectified,=E2=80=9D says = Waterhouse. =E2=80=9CThe banks=20 that we have worked with that employ qualitative analysis have all of = their=20 international analysts working off the same objectified scorecards so = that there=20 is a consistent approach within the bank. But in terms of how much = qualitative=20 evaluation ought to be done, it is a question of the size of the risk = and the=20 cost of analysis on a single counterparty. A purely quantative approach = may be=20 justified on the grounds that you do not want to invest more in = assessing a=20 fully accurate qualitative model if it will exceed the cost of the sum = being=20 underwritten.=E2=80=9D

Citigroup employs judgmental scorecard models for rating segments = with little=20 data history or too few defaults, and their use is governed by a = stringent and=20 centralised policy and approval process. But the trend is unavoidably = one=20 towards quantification and models will be increasingly based on = mathematics not=20 judgments. =E2=80=9CIn the future, it is likely that judgmental = scorecards will become=20 more statistical in nature and the judgmental interventions will be = reduced,=E2=80=9D=20 says Courant. =E2=80=9CJudgmental methodologies continue to be used = where appropriate=20 and when they are used there are policy-driven controls in = place.=E2=80=9D

Rees at Palisade believes the trend is towards quantative modelling, = not=20 least because banks are still unsure how their various judgmental = scorecards=20 will be treated by the Basel II regulators when it comes to seeking = their=20 approval for banks=E2=80=99 internally developed credit models and the = increased need to=20 demonstrate structured and robust approaches to decision making.

=E2=80=9CThe challenges for designers of the associated information = systems will=20 include how to allow judgment to become integrated in these quantitative = processes, how to implement systems that allow for sensitivity tests, = both to=20 the assumptions used in an individual model and as different models are = compared=20 to each other.=E2=80=9D

Bookmark with: (Wh= at=20 is this?)

Delicious=20

Digg=20

reddit=20

StumbleUpon

Social Bookmarking

Social bookmarking allows users to save and = categorise a=20 personal collection of bookmarks and share them with others. This is = different=20 to using your own browser bookmarks which are available using the menus = within=20 your web browser.

Use the links below to share this article on = the social=20 bookmarking site of your choice.

Read more about social bookmarking at Wikipedia - = Social=20 Bookmarking

Clo= se=20