CREDIT RATINGS oracles, miracles and much more Idzard van Eeghen ABNAMRO Bank ISDA/BBA/RMA Conference London June 23, 2003

1
CREDIT RATINGSoracles, miracles and much
moreIdzard van EeghenABNAMRO
BankISDA/BBA/RMA Conference London June 23, 2003
2
WHAT THIS PRESENTATION WILL BE ABOUT

• 1. Stakeholders and choices to be made
• 2. Sovereign ratings
• 3. Financial Institutions ratings
• 4. Structured-product ratings

3
THE RATING SYSTEM AS ORACLE
4
THE RATING SYSTEM AS MIRACLE

5
RATINGS AND RATING VALIDATION FOR WHOM AND TO
WHAT PURPOSE?

• RATING SYSTEMS STAKEHOLDERS
• Supervisors, Portfolio Managers
• Risk Management
• Business
• Stakeholders have different and often conflicting
  needs, but have in common that they want credit
  ratings to be transparent.

6
WHAT IS THE OPTIMAL NUMBER OF RATING SYSTEMS?

• Differences in Accounting. E.g.
• IAS, country specific GAAP
• Accountancy formats vary sometimes per Industry
• Treatment of pension costs, salary costs,
  valuation issues
• Differences in risk drivers. E.g.
• country specific
• industry specific (e.g. licences in Pharma,
  regulations in Public sector)
• obligor specific
• ?? There is a natural tendency to increase the
  number of rating systems to infinity

7
NUMBER OF EXPLANATORY VARIABLES SCYLLA OR
CHARYBDIS?

• Scylla the less data available the lower the
  number of
• model inputs that can be proven to
  be statistically
• relevant.
• Charybdis the specialised (complex) nature of
  the
• obligor requires that many aspects be
  included
• if the rating model is to be credible to the
  users.

8
CHOICES MADE

• Criteria
• Transparency for all users
• Intuitive/ logic
• Quantification of risk
• ? Range of rating systems, but optimal number is
  elusive
• ? In each rating system, we group the many
  elements in meaningful Pillars which are
  perceived to be important drivers of risk
• ? Rating models used as much for transparency
  reasons as for quantification of risk

9
SOVEREIGN FOREIGN-CURRENCY RATINGS
Using external ratings for benchmarking and
deriving default rates is problematic See for
example 1-year default rates for
sovereigns Moodys 1985-2002 SP 1975-2002
10
SOVEREIGN LOCAL-CURRENCY RATINGS

• Sovereign local-currency ratings difference
  between foreign currency and local-currency
  rating on average is
• Moodys 0.4 notch
• SP 1.0 notch
• However historic default rates of local-currency
  sovereign debt are only 1/10th of those of
  foreign-currency debt
• SP expects Sovereign default rates to converge
  to those of Corporates. But arent Sovereign
  ratings in fact issue ratings? And is public debt
  comparable to bank debt?

11
FINANCIAL-INSTITUTIONS RATINGS

• Many obligors are investment-grade ? Too few
  defaults to base default rates upon
• Benchmark not always available e.g. hedge funds
• Information required for proper analysis is often
  sensitive for privacy or competitive reasons and
  therefore not disclosed

12
FINANCIAL-INSTITUTIONS RATINGS
Two-sided Gauss-test, 95 confidence
interval. Data Moodys Corporate Bond Default
Database 1983-2000
13
RATINGS ARE MORE THAN 1-YEAR DEFAULT RATES
? Behaviour of ratings in economic cycle. ? Also,
ratings have a memory. Look beyond 1-year
default rates ? full migration matrix ? longer
periods
14
ANALYSING OUTLIERS EXPERT JUDGEMENT REQUIRED

• In depth analysis is done when a model-generated
  rating shows a significant and/or systematic bias
  from external rating or internally approved
  rating. Determining the cause often requires
  expert judgement.
• E.g.

15
RATING STRUCTURED PRODUCTS

• Projects are often structured in such a way as to
  derive the desired external rating.
• ? Band of observed rating categories becomes very
  limited
• ? External rating determines the structure of the
  obligor/credit (the tail wags the dog)
• Benchmarking internal ratings with external
  ratings could lead to de facto copying of
  external rating model. Is that the purpose of
  your internal rating system?

16
FINALLY

• Ratings still have something of an oracle and of
  a miracle in them they are a combination of man
  and machine.