This action might not be possible to undo. Are you sure you want to continue?
CREDITRISK
+
CREDITRISK
+
CREDITRISK
+
CR
DITRISK
+
CREDITRISK
+
CREDITRISK
+
CREDIT
SK
+
CREDITRISK
+
CREDITRISK
+
CREDITRISK
CREDITRISK
+
CREDITRISK
+
CREDITRISK
+
CR
DITRISK
+
CREDITRISK
+
CREDITRISK
+
CREDIT
SK
+
CREDITRISK
+
CREDITRISK
+
CREDITRISK
CREDITRISK
+
CREDITRISK
+
CREDITRISK
+
CR
DITRISK
+
CREDITRISK
+
CREDITRISK
+
CREDIT
SK
+
CREDITRISK
+
CREDITRISK
+
CREDITRISK
CREDITRISK
+
CREDITRISK
+
CREDITRISK
+
CR
DITRISK
+
CREDITRISK
+
CREDITRISK
+
CREDIT
SK
+
CREDITRISK
+
CREDITRISK
+
CREDITRISK
CREDITRISK
+
CREDITRISK
+
CREDITRISK
+
CR
DITRISK
+
CREDITRISK
+
CREDITRISK
+
CREDIT
+ + +
Regulated by SFA
CREDIT FIRST
SUI SSE BOSTON
CREDITRISK
+
A CREDIT RISK MANAGEMENT FRAMEWORK
Copyright © 1997 Credit Suisse First Boston International. All rights reserved.
CREDITRISK
+
is a trademark of Credit Suisse First Boston International in countries of use.
CREDITRISK
+
as described in this document (“ CREDITRISK
+
” ) is a method of credit risk management introduced by Credit Suisse Group.
No represent ation or warranty, express or implied, is made by Credit Suisse First Boston International or any other Credit Suisse Group
company as to the accuracy, completeness, or fitness for any particular purpose of CREDITRISK
+
. Under no circumst ances shall
Credit Suisse First Boston International or any other Credit Suisse Group company have any liability to any other person or any entity
for (a) any loss, damage or other injury in whole or in part caused by, resulting from or relating to, any error (negligent or otherwise), of
Credit Suisse First Boston International or any other Credit Suisse Group company in connection with the compilation, analysis,
interpret ation, communication, publication or delivery of CREDITRISK
+
, or (b) any direct, indirect, special, consequential, incident al or
compensatory damages whatsoever (including, without limit ation, lost profits), in either case caused by reliance upon or otherwise resulting
from or relating to the use of (including the inability to use) CREDITRISK
+
.
Issued and approved by Credit Suisse First Boston International for the purpose of Section 57, Financial Services Act 1986. Regulated by
the Securities and Futures Authority. The products and services referred to are not available to private customers.
is a leading global investment banking firm, providing comprehensive
financial advisory, capit al raising, sales and trading, and financial
products for users and suppliers of capit al around the world. It operates in over 60 offices across
more than 30 countries and six continents and has over 15,000 employees.
CREDIT FIRST
SUI SSE BOSTON
CREDI TRI S K
+
1
Contents
1. Introduction to CREDITRISK
+
3
1.1 Developments in Credit Risk Management 3
1.2 Components of CREDITRISK
+
3
1.3 The CREDITRISK
+
Model 4
1.4 Economic Capital 4
1.5 Applications of CREDITRISK
+
5
1.6 Example Spreadsheet Implementation 5
2. Modelling Credit Risk 6
2.1 Risk Modelling Concepts 6
2.2 Types of Credit Risk 7
2.3 Def ault Rate Behaviour 8
2.4 Modelling Approach 9
2.5 Time Horizon f or Credit Risk Modelling 10
2.6 Data Inputs to Credit Risk Modelling 11
2.7 Correlation and Incorporating the Ef fects of Background Factors 14
2.8 Measuring Concentration 16
3. The CREDITRISK
+
Model 17
3.1 Stages in the Modelling Process 17
3.2 Frequency of Def ault Events 17
3.3 Moving f rom Def ault Events to Def ault Losses 18
3.4 Concentration Risk and Sector Analysis 20
3.5 MultiYear Losses f or a HoldtoMaturity Time Horizon 21
3.6 Summary of the CREDITRISK
+
Model 22
4. Economic Capital for Credit Risk 23
4.1 Introduction to Economic Capital 23
4.2 Economic Capital f or Credit Risk 23
4.3 Scenario Analysis 24
5. Applications of CREDITRISK
+
26
5.1 Introduction 26
5.2 Provisioning f or Credit Risk 26
5.3 RiskBased Credit Limits 29
5.4 Portf olio Management 29
I
t
Appendices
A. The CREDITRISK
+
Model 32
A1 Overview of this Appendix 32
A2 Def ault Events with Fixed Def ault Rates 33
A3 Def ault Losses with Fixed Def ault Rates 35
A4 Loss Distribution with Fixed Def ault Rates 38
A5 Application to MultiYear Losses 39
A6 Def ault Rate Uncertainty 41
A7 Sector Analysis 41
A8 Def ault Events with Variable Def ault Rates 44
A9 Def ault Losses with Variable Def ault Rates 46
A10 Loss Distribution with Variable Def ault Rates 47
A11 Convergence of Variable Def ault Rate Case to Fixed Def ault Rate Case 49
A12 General Sector Analysis 50
A13 Risk Contributions and Pairwise Correlation 52
B. Illustrative Example 58
B1 Example SpreadsheetBased Implementation 58
B2 Example Portf olio and Static Data 58
B3 Example Use of the Spreadsheet Implementation 60
C. Contacts 66
D. Selected Bibliography 68
List of Tables
Table 1: Representations of the def ault rate process 9
Table 2: Oneyear def ault rates (%) 12
Table 3: Def ault rate standard deviations (%) 13
Table 4: Recovery rates by seniority and security (%) 14
Table 5: Mechanisms f or controlling the risk of credit def ault losses 25
Table 6: Provisioning f or dif ferent business lines 28
Table 7: Example of credit risk provisioning 28
Table 8: Example portf olio 59
Table 9: Example mapping table of def ault rate inf ormation 59
Table 10: Example 1A  Risk contributions 64
Table 11: Example 1B  Risk analysis of removed obligors 65
Table 12: Example 1B  Portf olio movement analysis 65
List of Figures
Figure 1: Components of CREDITRISK
+
3
Figure 2: Def ault rate as a continuous random variable 8
Figure 3: Def ault rate as a discrete random variable 9
Figure 4: Rated corporate def aults by number of issuers 12
Figure 5: Def aulted bank loan price distribution 13
Figure 6: CREDITRISK
+
Model  Distribution of def ault events 18
Figure 7: CREDITRISK
+
Model  Distribution of def ault losses 19
Figure 8: Impact of sectors on the loss distribution 21
Figure 9: Economic capital f or credit risk 24
Figure 10: Parts of the credit def ault loss distribution 25
Figure 11: Credit risk provisioning 27
Figure 12: Using risk contributions 31
Figure 13: Flowchart description of Appendix A 33
2
CREDIT FIRST
SUI SSE BOSTON
CREDI TRI S K
+
3
Introdu
to CRE
1.1 Developments in Credit Risk Management
Since the beginning of the 1990s, Credit Suisse First Boston (“CSFB”) has been developing and deploying
new risk management methods. In 1993, Credit Suisse Group launched, in parallel, a major project aimed
at modernising its credit risk management and, using CSFB’s expertise, at developing a more f orward
looking management tool. In December 1996, Credit Suisse Group introduced CREDITRISK
+
 a Credit Risk
Management Framework.
Current areas of development in credit risk management include: modelling credit risk on a portf olio basis;
credit risk provisioning; active portf olio management; credit derivatives; and sophisticated approaches to capital
allocation that more closely ref lect economic risk than the existing regulatory capital regime. CREDITRISK
+
addresses all of these areas and the relationships between them.
CREDITRISK
+
can be applied to credit exposures arising f rom all types of products including corporate and retail
loans, derivatives, and traded bonds.
1.2 Components of CREDITRI SK
+
The components of CREDITRISK
+
and the interrelationships between them are shown in the f ollowing diagram.
Figure 1:
Components of CREDITRI SK
+
CREDITRISK
+
comprises three
main components –a CREDITRISK
+
Model that uses a portfolio
approach, a methodology for
calculating economic capit al
for credit risk, and several
applications of the technology.
Introduction to CREDITRISK
+
1
CREDITRI SK
+
Credit Risk Measurement
Credit Default
Loss Distribution
Scenario Analysis
Provisioning
Limits
Portfolio Management
Economic Capital Applications
Exposures Def ault Rates
CREDITRISK
+
Model
Recovery
Rates
Def ault Rate
Volatilities
A modern approach to credit risk management should address all aspects of credit risk, from quantitative
modelling to the development of practical techniques for its management . In addition to wellestablished credit
risk management techniques, such as individual obligor (borrower, counterparty or issuer) limits and concentration
limits, CREDITRISK
+
reflects the requirements of a modern approach to managing credit risk and comprises three
main components:
•
The CREDITRISK
+
Model that uses a portfolio approach and analytical techniques applied widely in the
insurance industry.
•
A methodology for calculating economic capit al for credit risk.
•
Applications of the credit risk modelling methodology including: (i) a methodology for establishing provisions
on an anticipatory basis, and (ii) a means of measuring diversification and concentration to assist in
portfolio management.
1.3 The CREDITRI SK
+
Model
CREDITRISK
+
is based on a portf olio approach to modelling credit def ault risk that takes into account
information relating to size and maturity of an exposure and the credit quality and systematic risk of an obligor.
The CREDITRISK
+
Model is a statistical model of credit def ault risk that makes no assumptions about the
causes of def ault. This approach is similar to that taken in market risk management, where no attempt is made
to model the causes of market price movements. The CREDITRISK
+
Model considers default rates as continuous
random variables and incorporates the volatility of def ault rates in order to capture the uncertainty in the level
of def ault rates. Of ten, background f actors, such as the state of the economy, may cause the incidence of
def aults to be correlated, even though there is no causal link between them. The ef fects of these background
f actors are incorporated into the CREDITRISK
+
Model through the use of def ault rate volatilities and sector
analysis rather than using def ault correlations as explicit inputs into the model.
Mathematical techniques applied widely in the insurance industry are used to model the sudden event of an
obligor def ault. This approach contrasts with the mathematical techniques typically used in f inance. In f inancial
modelling one is usually concerned with modelling continuous price changes rather than sudden events.
Applying insurance modelling techniques, the analytic CREDITRISK
+
Model captures the essential
characteristics of credit def ault events and allows explicit calculation of a f ull loss distribution f or a portf olio of
credit exposures.
1.4 Economic Capital
The output of the CREDITRISK
+
Model can be used to determine the level of economic capital required to cover
the risk of unexpected credit def ault losses.
Measuring the uncertainty or variability of loss and the relative likelihood of the possible levels of unexpected
losses in a portf olio of credit exposures is f undamental to the ef fective management of credit risk. Economic
capital provides a measure of the risk being taken by a f irm and has several benef its: it is a more appropriate
risk measure than that specif ied under the current regulatory regime; it measures economic risk on a portf olio
basis and takes account of diversif ication and concentration; and, since economic capital ref lects the changing
risk of a portf olio, it can be used f or portf olio management.
4
CREDIT FIRST
SUI SSE BOSTON
CREDI TRI S K
+
5
The CREDITRISK
+
Model is supplemented by scenario analysis in order to identif y the f inancial impact of low
probability but nevertheless plausible events that may not be captured by a statistically based model.
1.5 Applications of CREDITRI SK
+
CREDITRISK
+
includes several applications of the credit risk modelling methodology, including a forwardlooking
provisioning methodology and quantitative portf olio management techniques.
1.6 Example Spreadsheet Implementation
In order to assist the reader of this document, a spreadsheetbased implementation that illustrates the range
of possible outputs of the CREDITRISK
+
Model can be downloaded f rom the Internet (http:/ / www.csf b.com).
1
Introduction
Model
Credit
2.1 Risk Modelling Concepts
2.1.1 Types of Uncertainty Arising in the Modelling Process
A statistically based model can describe many business processes. However, any model is only a
representation of the real world. In the modelling process, there are three types of uncertainty that must be
assessed: process risk, parameter uncertainty and model error.
Process Risk
Process risk arises because the actual observed results are subject to random f luctuations even where the
model describing the loss process and the parameters used by the model are appropriate. Process risk is
usually addressed by expressing the model results to an appropriately high level of conf idence.
Parameter Uncertainty
Parameter uncertainty arises f rom the dif f iculties in obtaining estimates of the parameters used in the model.
The only inf ormation that can be obtained about the underlying process is obtained by observing the results
that it has generated in the past. It is possible to assess the impact of parameter uncertainty by perf orming
sensitivity analysis on the parameter inputs.
Model Error
Model error arises because the proposed model does not correctly ref lect the actual process  alternative
models could produce dif ferent results. Model error is usually the least tractable of the three types of
uncertainty.
2.1.2 Addressing Modelling Issues
As all of these types of uncertainty enter into the modelling process, it is important to be aware of them and
to consider how they can be addressed when developing a credit risk model. Indeed, a realistic assessment of
the potential ef fects of these errors should be made bef ore any decisions are made based on the outputs of
the model.
6
CREDIT FIRST
SUI SSE BOSTON
Modelling Credit Risk 2
CREDI TRI S K
+
7
Modelling Credit Risk
The CREDITRISK
+
Model
makes no assumptions
about the causes of default.
This approach is similar to
that t aken in market risk
management , where no
assumpt ions are made
about the causes of market
price movements.
All portfolios of exposures
exhibit credit default risk, as
the default of an obligor
results in a loss.
CREDITRISK
+
addresses these types of uncertainty in several ways:
•
No assumptions are made about the causes of def ault. This approach is similar to that taken in market risk
management, where no assumptions are made about the causes of market price movements. This not only
reduces the potential model error but also leads to the development of an analytically tractable model.
•
The data requirements f or the CREDITRISK
+
Model have been kept as low as possible, which minimises the
error f rom parameter uncertainty. In the credit environment, empirical data is sparse and dif f icult to obtain.
Even then, the data can be subject to large f luctuations year on year.
•
Concerns about parameter uncertainty are addressed using scenario analysis, in which the ef fects of stress
testing each of the input parameters are quantif ied. For example, increasing def ault rates or def ault rate
volatilities can be used to simulate downturns in the economy.
2.2 Types of Credit Risk
There are two main types of credit risk:
•
Credit spread risk: Credit spread risk is exhibited by portf olios f or which the credit spread is traded and
markedtomarket. Changes in observed credit spreads impact the value of these portf olios.
•
Credit default risk: All portf olios of exposures exhibit credit def ault risk, as the def ault of an obligor results
in a loss.
2.2.1 Credit Spread Risk
Credit spread is the excess return demanded by the market f or assuming a certain credit exposure. Credit
spread risk is the risk of f inancial loss owing to changes in the level of credit spreads used in the markto
market of a product.
Credit spread risk f its more naturally within a market risk management f ramework. In order to manage credit
spread risk, a f irm’s valueatrisk model should take account of value changes caused by the volatility of credit
spreads. Since the distribution of credit spreads may not be normal, a standard variancecovariance approach
to measuring credit spread risk may be inappropriate. However, the historical simulation approach, which does
not make any assumptions about the underlying distribution, used in combination with other techniques,
provides a suitable alternative.
Credit spread risk is only exhibited when a marktomarket accounting policy is applied, such as f or portf olios
of bonds and credit derivatives. In practice, some types of products, such as corporate or retail loans, are
typically accounted f or on an accruals basis. A marktomarket accounting policy would have to be applied to
these products in order to recognise the credit spread risk.
2.2.2 Credit Default Risk
Credit def ault risk is the risk that an obligor is unable to meet its f inancial obligations. In the event of a def ault
of an obligor, a f irm generally incurs a loss equal to the amount owed by the obligor less a recovery amount
which the f irm recovers as a result of f oreclosure, liquidation or restructuring of the def aulted obligor.
All portf olios of exposures exhibit credit def ault risk, as the def ault of an obligor results in a loss.
2
Credit def ault risk is typically associated with exposures that are more likely to be held to maturity, such as
corporate and retail loans and exposures arising f rom derivative portf olios. Bond markets are generally more
liquid than loan markets and theref ore bond positions can be adjusted over a shorter time f rame. However,
where the intention is to maintain a bond portf olio over a longer time f rame, even though the individual
constituents of the portf olio may change, it is equally important to measure the def ault risk that is taken by
holding the portf olio.
CREDITRISK
+
f ocuses on modelling and managing credit def ault risk.
2.3 Default Rate Behaviour
Equity and bond prices are f orwardlooking in nature and are f ormed by investors’ views of the f inancial
prospects of a particular obligor. Hence, they incorporate both the credit quality and the potential credit quality
changes of that obligor.
Theref ore, the def ault rate of a particular obligor, inferred f rom market prices, will vary on a continuous scale
and hence can be viewed as a continuous random variable. In modelling credit risk, one is concerned with
determining the possible f uture outcomes over the chosen time horizon.
The process f or the def ault rate can be represented in two dif ferent ways:
•
Continuous variable: When treated as a continuous variable, the possible def ault rate over a given time
horizon is described by a distribution, which can be specif ied by a def ault rate and a volatility of the def ault
rate. The data requirements f or modelling credit def ault risk are analogous to the data requirements f or
pricing stock options  a f orward stock price and the stock price volatility are used to def ine the f orward
stock price distribution. The f ollowing f igure illustrates the path that a def ault rate may take over time and
the distribution that it could have over that time.
•
Discrete variable: By treating the def ault rate as a discrete variable, a simplif ication of the continuous
process described above is made. A convenient way of making default rates discrete is by assigning credit
ratings to obligors and mapping def ault rates to credit ratings. Using this approach, additional inf ormation
is required in order to model the possible f uture outcomes of the def ault rate. This can be achieved via a
rating transition matrix that specif ies the probability of keeping the same credit rating, and hence the same
value f or the def ault rate, and the probabilities of moving to dif ferent credit ratings and hence to dif ferent
values f or the def ault rate. This is illustrated in the f ollowing f igure.
8
CREDIT FIRST
SUI SSE BOSTON
Possible path of
default rate
Frequency of default
rate outcomes
Figure 2:
Default rate as a
continuous random
variable
D
e
f
a
u
l
t
r
a
t
e
Ti me hori zon
The discrete approach with rating migrations and the continuous approach with a def ault rate volatility are
dif ferent representations of the behaviour of def ault rates. Both approaches achieve the desired end result of
producing a distribution f or the def ault rate.
The above two representations of def ault rate behaviour are summarised in the f ollowing table:
Treatment of default rate Dat a requirements
Continuous variable • Def ault rates
• Volatility of def ault rates
Discrete variable • Credit ratings
• Rating transition matrix
The CREDITRISK
+
Model is a statistical model of credit def ault risk that models def ault rates as continuous
random variables and incorporates the volatility of the def ault rate in order to capture the uncertainty in the
level of the def ault rate. A mapping f rom credit ratings to a set of def ault rates provides a convenient approach
to setting the level of the def ault rate.
2.4 Modelling Approach
2.4.1 Risk Measures
When managing credit risk, there are several measures of risk that are of interest, including the f ollowing:
•
Distribution of loss: The risk manager is interested in obtaining distributions of loss that may arise f rom the
current portf olio. The risk manager needs to answer questions such as “What is the size of loss f or a given
conf idence level?”.
•
Identifying extreme outcomes: The risk manager is also concerned with identif ying extreme or catastrophic
outcomes. These outcomes are usually dif f icult to model statistically but can be addressed through the use
of scenario analysis and concentration limits.
Table 1:
Representations of the
default rate process
CREDI TRI S K
+
9
Modelling Credit Risk
Figure 3:
Default rate as a discrete
random variable
2
Possible path of
default rate
B
AAA
AA
BBB
BB
A
Default
Frequency of default
rate outcomes
D
e
f
a
u
l
t
r
a
t
e
Ti me hori zon
The CREDITRISK
+
Model
treats def ault rates as
continuous random variables
and incorporates def ault
rate volatility to capture the
uncertainty in the level of
the def ault rate.
2.4.2 A Portfolio Approach to Managing Credit Risk
Credit risk can be managed through diversif ication because the number of individual risks in a portf olio of
exposures is usually large. Currently, the primary technique f or controlling credit risk is the use of limit systems,
including individual obligor limits to control the size of exposure, tenor limits to control the maximum maturity
of exposures to obligors, rating exposure limits to control the amount of exposure to obligors of certain credit
ratings, and concentration limits to control concentrations within countries and industry sectors.
The portf olio risk of a particular exposure is determined by f our f actors: (i) the size of the exposure, (ii) the
maturity of the exposure, (iii) the probability of def ault of the obligor, and (iv) the systematic or concentration
risk of the obligor. Credit limits aim to control risk arising f rom each of these f actors individually. The general
ef fect of this approach, when applied in a wellstructured and consistent manner, is to create reasonably well
diversif ied portf olios. However, these limits do not provide a measure of the diversif ication and concentration
of a portf olio.
In order to manage ef fectively a portf olio of exposures, a means of measuring diversif ication and concentration
has to be developed. An approach that incorporates size, maturity, credit quality and systematic risk into a single
portf olio measure is required. CREDITRISK
+
takes such an approach.
2.4.3 Modelling Techniques Used in the CREDITRI SK
+
Model
The economic risk of a portf olio of credit exposures is analogous to the economic risk of a portf olio of
insurance exposures. In both cases, losses can be suf fered f rom a portf olio containing a large number of
individual risks, each with a low probability of occurring. The risk manager is concerned with assessing the
f requency of the unexpected events as well as the severity of the losses.
In order to keep model error to a minimum, no assumptions are made about the causes of def ault.
Mathematical techniques applied widely in the insurance industry are used to model the sudden event of an
obligor def ault. In modelling credit def ault losses one is concerned with sudden events rather than continuous
changes. The essential characteristics of credit def ault events are captured by applying these insurance
modelling techniques. This has the additional benef it that it leads to a credit risk model that is analytically
tractable and hence not subject to the problems of precision that can arise when using a simulationbased
approach. The analytic CREDITRISK
+
Model allows rapid and explicit calculation of a f ull loss distribution f or a
portf olio of credit exposures.
2.5 Time Horizon for Credit Risk Modelling
A key decision that has to be made when modelling credit risk is the choice of time horizon. Generally, the time
horizon chosen should not be shorter than the time f rame over which riskmitigating actions can be taken.
CREDITRISK
+
does not prescribe any one particular time horizon but suggests two possible time horizons that
can provide management inf ormation relevant f or credit risk management:
•
A constant time horizon, such as one year.
•
A holdtomaturity or runof f time horizon.
10
CREDIT FIRST
SUI SSE BOSTON
CREDITRISK
+
is based
on a portfolio approach 
summarising information
about size, maturity, credit
quality and systematic risk
into a single measure.
CREDI TRI S K
+
11
2.5.1 Constant Time Horizon
A constant time horizon is relevant, as it allows all exposures to be considered at the same f uture date.
For various reasons, one year is of ten taken as a suitable time horizon: credit risk mitigating actions can
normally be executed within one year, new capital can be raised to replenish capital eroded by actual credit
losses during the period, and, f urthermore, one year matches the normal accounting period. Given these
f actors, CREDITRISK
+
suggests a time horizon of one year f or credit risk economic capital.
2.5.2 HoldtoMaturity Time Horizon
Alternatively, a holdtomaturity time horizon allows the full term structure of default rates over the lifetime of the
exposures to be recognised. This view of the portf olio enables the risk manager to compare exposures of
dif ferent maturity and credit quality and is an appropriate tool, in addition to the constant time horizon, f or
portf olio management. The role that the CREDITRISK
+
Model plays in active portf olio management is discussed
later in this document.
A benchmark time horizon of one year can be used f or portf olios where there is an intention to maintain
exposures f or longer than the term of the booked transactions (e.g. traded bond portf olios).
2.6 Data Inputs to Credit Risk Modelling
2.6.1 Data Inputs
Any modelling of credit risk is dependent on certain data requirements being met. The quality of this data will
directly af fect the accuracy of the measurement of credit risk and theref ore any decision to be made using the
results should be made only having f ully assessed the potential error f rom uncertainties in the data used.
The inputs used by the CREDITRISK
+
Model are:
•
Credit exposures
•
Obligor def ault rates
•
Obligor def ault rate volatilities and
•
Recovery rates.
The CREDITRISK
+
Model presented in this document does not prescribe the use of any one particular data set
over another. One of the key limitations in modelling credit risk is the lack of comprehensive def ault data.
Where a f irm has its own inf ormation that is judged to be relevant to its portf olio, this can be used as the input
into the model. Alternatively, conservative assumptions can be used while def ault data quality is being improved.
2.6.2 Credit Exposures
The exposures arising f rom separate transactions with an obligor should be aggregated according to the legal
corporate structure and taking into account any rights of setof f .
The CREDITRISK
+
Model is capable of handling all types of instruments that give rise to credit exposure,
including bonds, loans, commitments, f inancial letters of credit and derivative exposures. For some of these
transaction types, it is necessary to make an assumption about the level of exposure in the event of a def ault:
f or example, a f inancial letter of credit will usually be drawn down prior to def ault and theref ore the exposure
at risk should be assumed to be the f ull nominal amount.
In addition, if a multiyear time horizon is being used, it is important that the changing exposures over time are
accurately captured.
2
Modelling Credit Risk
Credit Risk Measurement
Exposures Def ault Rates
CREDITRISK
+
Model
Recovery
Rates
Def ault Rate
Volatilities
Figure 4:
Rated corporate defaults
by number of issuers
Oneyear def ault rates
show signif icant f luctuations
f rom year to year.
12
CREDIT FIRST
SUI SSE BOSTON
Credit Risk Measurement
Exposures Def ault Rates
CREDITRISK
+
Model
Recovery
Rates
Def ault Rate
Volatilities
Table 2:
Oneyear default rates (%)
2.6.3 Default Rates
A def ault rate, which represents the likelihood of a def ault event occurring, should be assigned to each obligor.
This can be achieved in a number of ways, including:
•
Observed credit spreads f rom traded instruments can be used to provide marketassessed probabilities of
def ault.
•
Alternatively, obligor credit ratings, together with a mapping of def ault rates to credit ratings, provide a
convenient way of assigning probabilities of def ault to obligors. The rating agencies publish historic def ault
statistics by rating category f or the population of obligors that they have rated.
Credit rating Oneyear def ault rate
Aaa 0.00
Aa 0.03
A 0.01
Baa 0.12
Ba 1.36
B 7.27
Source: Carty & Lieberman, 1997, Moody’s Investors Service Global Credit Research
A credit rating is an opinion of an obligor’s overall f inancial capacity to meet its f inancial obligations (i.e. its
creditworthiness). This opinion f ocuses on the obligor’s capacity and willingness to meet its f inancial
commitments as they f all due. An assessment of the nature of a particular obligation, including its seniority in
bankruptcy or liquidation, should be perf ormed when considering the recovery rate f or an obligor.
It should be noted that oneyear def ault rates show signif icant variation year on year, as can be seen in the
f ollowing f igure. During periods of economic recession, the number of def aults can be many times the level
observed at other times.
Source: Standard & Poor’s Ratings Perf ormance 1996 (February 1997)
•
Another approach is to calculate def ault probabilities on a continuous scale, which can be used as a
substitute f or the combination of credit ratings and assigned def ault rates.
F
r
e
q
u
e
n
c
y
1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996
7 0
6 0
5 0
4 0
3 0
2 0
1 0
0
CREDI TRI S K
+
13
2
Modelling Credit Risk
Credit Risk Measurement
Exposures Def ault Rates
CREDITRISK
+
Model
Recovery
Rates
Def ault Rate
Volatilities
Credit Risk Measurement
Exposures Def ault Rates
CREDITRISK
+
Model
Recovery
Rates
Def ault Rate
Volatilities
Table 3:
Default rate standard
deviations (%)
Figure 5:
Defaulted bank loan
price distribution
F
r
e
q
u
e
n
c
y
2.6.4 Default Rate Volatilities
Published def ault statistics include average def ault rates over many years. As shown previously, actual
observed def ault rates vary f rom these averages. The amount of variation in def ault rates about these averages
can be described by the volatility (standard deviation) of def ault rates. As can be seen in the f ollowing table,
the standard deviation of def ault rates can be signif icant compared to actual def ault rates, ref lecting the high
f luctuations observed during economic cycles.
Oneyear default rate (%)
Credit rating Average St andard deviation
Aaa 0.00 0.0
Aa 0.03 0.1
A 0.01 0.0
Baa 0.12 0.3
Ba 1.36 1.3
B 7.27 5.1
Source: Carty & Lieberman, 1996, Moody’s Investors Service Global Credit Research
The def ault rate standard deviations in the above table were calculated over the period f rom 1970 to 1996
and theref ore include the ef fect of economic cycles.
2.6.5 Recovery Rates
In the event of a def ault of an obligor, a f irm generally incurs a loss equal to the amount owed by the obligor
less a recovery amount, which the f irm recovers as a result of f oreclosure, liquidation or restructuring of the
def aulted obligor or the sale of the claim. Recovery rates should take account of the seniority of the obligation
and any collateral or security held.
Recovery rates are subject to signif icant variation. For example, the f igure below shows the price distribution
of def aulted bank loans and illustrates that there is a large degree of dispersion.
Source: Def aulted Bank Loan Recoveries (November 1996) , Moody’s Investors Service Global Credit Research
$
0

$
1
0
$
1
1

$
2
0
$
2
1

$
3
0
$
3
1

$
4
0
$
4
1

$
5
0
$
5
1

$
6
0
$
6
1

$
7
0
$
7
1

$
8
0
$
8
1

$
9
0
$
9
1

$
1
0
0
1 4
1 2
1 0
8
6
4
2
0
There is also considerable variation f or obligations of dif fering seniority, as can be seen f rom the standard
deviation of the corporate bond and bank loan recovery rates in the table below.
Seniority and security Average St andard deviation
Senior secured bank loans 71.18 21.09
Senior secured public debt 63.45 26.21
Senior unsecured public debt 47.54 26.29
Senior subordinated public debt 38.28 24.74
Subordinated public debt 28.29 20.09
Junior subordinated public debt 14.66 8.67
Source: Historical Def ault Rates of Corporate Bond Issuers, 19201996 (January 1997) Moody’s Investors Service Global Credit Research
Publicly available recovery rate data indicates that there can be signif icant variation in the level of loss, given
the def ault of an obligor. Theref ore, a caref ul assessment of recovery rate assumptions is required. Given this
uncertainty, stress testing should be perf ormed on the recovery rates in order to calculate the potential loss
distributions under dif ferent scenarios.
2.7 Correlation and Incorporating the Effects of Background Factors
Default correlation impacts the variability of default losses from a portfolio of credit exposures. The CREDITRISK
+
Model incorporates the ef fects of def ault correlations by using def ault rate volatilities and sector analysis.
2.7.1 The Random Nature of Defaults and the Appearance of Correlation
Credit def aults occur as a sequence of events in such a way that it is not possible to f orecast the exact time
of occurrence of any one def ault or the exact total number of def aults. Often, there are background factors
that may cause the incidence of default events to be correlated, even though there is no causal link between
them. For example, if there is an unusually large number of def aults in one particular month, this might be due
to the economy being in recession, which has increased the rates of def ault above their average level. In this
economic situation, it is quite likely that the number of def aults in the f ollowing month will also be high.
Conversely, if there are fewer def aults than on average in one month, because the economy is growing, it is
also likely that there will be fewer def aults than on average in the f ollowing month. The def aults are correlated
but there is no causal link between them  the correlation ef fect observed is due to a background f actor, the
state of the economy, which changes the rates of def ault.
2.7.2 Impact of the Economy on Default Rates
There is general agreement that the state of the economy in a country has a direct impact on observed def ault
rates. A recent report by Standard and Poor’s stated that “A healthy economy in 1996 contributed to a
significant decline in the total number of corporate defaults. Compared to 1995, defaults were reduced by
onehalf….”
1
Another report by Moody’s Investors Service stated that “The sources of [default rate volatility]
are many, but macroeconomic trends are certainly the most influential factors”.
2
As the above quotations indicate and as can be seen in Figure 4 above, there is signif icant variation in the
number of def aults f rom year to year. Furthermore, f or each year, dif ferent industry sectors will be af fected to
dif ferent degrees by the state of the economy. The magnitude of the impact will be dependent on how sensitive
an obligor’s earnings are to various economic f actors, such as the growth rate of the economy and the level of
interest rates.
14
CREDIT FIRST
SUI SSE BOSTON
Table 4:
Recovery rates by
seniority and security (%)
Often, there are
background factors that
may cause the incidence
of defaults to be correlated,
even though there is no
causal link between them.
1
St andard and Poor’s
Ratings Performance
1996, February 1997
2
Moody’s Investors
Service, Corporate Bond
Defaults and Default
Rates, January 1996
CREDI TRI S K
+
15
Economic models that attempt to capture the ef fect of changes in the economy on def ault rates can be
developed in order to specif y the def ault rates f or subsequent use in a credit risk model. However, this
approach can have several weaknesses, including the f ollowing:
•
Since there are limited publicly available def ault rate statistics by country or by industry sector, it is dif f icult
to verif y the accuracy of an economic model used to derive def ault rates.
•
Even if a causal relationship could be established relating def ault rates to certain economic variables, it is
questionable whether such relationships would be stable over several years.
Theref ore, alternative approaches that attempt to capture the observed variability of def ault rates have to be
sought.
2.7.3 Incorporating the Effects of Background Factors
It is possible to incorporate the ef fects of background f actors into the specif ication of def ault rates by allowing
the def ault rate itself to have a probability distribution. This is accomplished by incorporating def ault rate
volatilities into the model.
The CREDITRISK
+
Model models the ef fects of background f actors by using def ault rate volatilities that result
in increased def aults rather than by using def ault correlations as a direct input. Both approaches, the use of
def ault rate volatilities and def ault correlations, give rise to loss distributions with f at tails.
Section 3 of this document describes in detail how the CREDITRISK
+
Model uses def ault rate volatilities in the
modelling of credit def ault risk.
The CREDITRISK
+
Model does not attempt to model correlations explicitly but captures the same concentration
ef fects through the use of def ault rate volatilities and sector analysis
3
. There are various reasons why this
approach has been taken, including the f ollowing:
•
Inst ability of default correlations: Generally, correlations calculated f rom f inancial data show a high degree
of instability. In addition, a calculated correlation can be very dependent on the underlying time period of
the data. A similar instability problem may arise with def ault rate volatilities: however, it is much easier to
perf orm scenario analysis on def ault rate volatilities, owing to the analytically tractable nature of a model
that uses volatilities rather than correlations.
•
Lack of empirical dat a: There is little empirical data on def ault correlations. Def aults themselves are
inf requent events and so there is insuf f icient data on multiple def aults with which to calculate explicit
def ault correlations. Since def ault correlations are dif f icult to calculate directly, some approaches use asset
price correlations to derive def ault correlations, but this can only be considered a proxy. This technique
relies upon additional assumptions about the relationship between asset prices and probabilities of def ault.
Furthermore, it is questionable how stable any relationship, that may be inferred or observed during a period
of normal trading, would be in the event of def ault of a particular obligor. In addition, where there is no asset
price f or the obligor, f or example in a retail portf olio, there is no obvious way of deriving def ault correlations.
2
Modelling Credit Risk
The CREDITRISK
+
Model
captures concentration risk
through the use of default
rate volatilities and sector
analysis.
3
Sector analysis is
discussed in Sections
2.8 and 3.4
2.8 Measuring Concentration
The above discussion has highlighted the f act that there are background f actors that af fect the level of def ault
rates. The state of the economy of each dif ferent country will vary over time and, within each country, dif ferent
industry sectors will be af fected to dif fering degrees. A portf olio of exposures can have concentrations in
particular countries or industry sectors. Theref ore, it is important to be able to capture the ef fect of
concentration risk in a credit risk model.
The CREDITRISK
+
Model described in this document allows concentration risk to be captured using sector
analysis. An exposure can be broken down into an obligorspecif ic element, which is independent of all other
exposures, and nonspecif ic or systematic elements that are sensitive to particular driving f actors, such as
countries or industry sectors.
16
CREDIT FIRST
SUI SSE BOSTON
C
M
CREDI TRI S K
+
17
CREDIT
Model
The CREDITRISK
+
Model
3
Credit Risk Measurement
Exposures Def ault Rates
CREDITRISK
+
Model
Recovery
Rates
Def ault Rate
Volatilities
3.1 Stages in the Modelling Process
The modelling of credit risk is a two stage process, as shown in the f ollowing diagram:
By calculating the distribution of def ault events, the risk manager is able to assess whether the overall credit
quality of the portf olio is either improving or deteriorating. The distribution of losses allows the risk manager to
assess the f inancial impact of the potential losses as well as measuring the amount of diversif ication and
concentration within the portf olio.
3.2 Frequency of Default Events
3.2.1 The Default Process
The CREDITRISK
+
Model makes no assumption about the causes of def ault  credit def aults occur as a
sequence of events in such a way that it is neither possible to f orecast the exact time of occurrence of any
one def ault nor the exact total number of def aults. There is exposure to def ault losses f rom a large number of
obligors and the probability of def ault by any particular obligor is small. This situation is well represented by the
Poisson distribution.
What is the
FREQUENCY
of defaults?
What is the
SEVERITY
of the losses ?
Stage 1
Stage 2
Distribution of
default losses
We consider f irst the distribution of the number of def ault events in a time period, such as one year, within a
portf olio of obligors having a range of dif ferent annual probabilities of def ault. The annual probability of def ault
of each obligor can be conveniently determined by its credit rating and a mapping between def ault rates and
credit ratings. If we do not incorporate the volatility of the default rate, the distribution of the number of default
events will be closely approximated by the Poisson distribution. This is regardless of the individual def ault rate
f or a particular obligor.
However, def ault rates are not constant over time and, as we have seen in the previous section, exhibit a high
degree of variation. Hence, def ault rate variability needs to be incorporated into the model.
3.2.2 Distribution of the Number of Default Events
The CREDITRISK
+
Model models the underlying def ault rates by specif ying a def ault rate and a def ault rate
volatility. This aims to take account of the variation in def ault rates in a pragmatic manner, without introducing
signif icant model error.
The ef fect of using def ault rate volatilities can be clearly seen in the f ollowing f igure, which shows the
distribution of the number of def ault events generated by the CREDITRISK
+
Model when def ault rate volatility
is varied. Although the expected number of def ault events is the same, the distribution becomes signif icantly
skewed to the right when def ault rate volatility is increased. This represents a signif icantly increased risk of an
extreme number of def ault events.
3.3 Moving from Default Events to Default Losses
3.3.1 Distribution of Default Losses
Given the number of default events, we now consider the distribution of losses in the portfolio. The distribution
of losses dif fers f rom the distribution of def ault events because the amount lost in a given def ault depends on
the exposure to the individual obligors. Unlike the variation of def ault probability between obligors, which does
not inf luence the distribution of the total number of def aults, the variation in exposure magnitude results in a
loss distribution that is not Poisson in general. Moreover, inf ormation about the distribution of dif ferent
exposures is essential to the overall distribution. However, it is possible to describe the overall distribution of
losses because its probability generating f unction has a simple closed f orm amenable to computation.
18
CREDIT FIRST
SUI SSE BOSTON
Including default rate volatility
Excluding default rate volatility
Figure 6:
CREDITRI SK
+
Model 
Distribution of default
events
P
r
o
b
a
b
i
l
i
t
y
Number of def aul t s
CREDI TRI S K
+
19
In the event of a def ault of an obligor, a f irm generally incurs a loss equal to the amount owed by the obligor
less a recovery amount, which the f irm obtains as a result of f oreclosure, liquidation or restructuring of the
def aulted obligor. A recovery rate is used to quantif y the amount received f rom this process. Recovery rates
should take account of the seniority of the obligation and any collateral or security held.
In order to reduce the amount of data to be processed, two steps are f irst f ollowed:
•
The exposures are adjusted by anticipated recovery rates in order to calculate the loss in a given def ault.
•
The exposures, net of the above recovery adjustment, are divided into bands of exposure with the level of
exposure in each band being approximated by a common average.
The CREDITRISK
+
Model calculates the probability that a loss of a certain multiple of the chosen unit of
exposure will occur. This allows a f ull loss distribution to be generated, as shown in the f igure below.
3.3.2 Impact of Incorporating Default Rate Volatilities
Figure 7 compares the def ault loss distributions calculated without def ault rate volatility and with def ault rate
volatility. The key features and dif ferences are:
•
Same expected loss: Both def ault loss distributions have the same level of expected losses.
•
Fatter t ail: The key change is the level of losses at the higher percentiles; f or example, the 99th percentile
is signif icantly higher when the impact of the variability of def ault rates is modelled. There is now
considerably more chance of experiencing extreme losses.
Since the tail of the distribution has become f atter, while the expected loss has remained unchanged, it can be
concluded that the variance of the def ault loss distribution has increased. This increase in the variance is due
to the pairwise default correlations between obligors. These pairwise default correlations are incorporated
into the CREDITRISK
+
Model through the default rate volatilities and sector analysis. It should be noted that
when the def ault rate volatilities are set to zero, the def ault events are independent and hence the pairwise
def ault correlations are also zero.
In Appendix A, we give an explicit f ormula f or the pairwise def ault correlations implied by the CREDITRISK
+
Model when def ault rate volatilities are incorporated into the model.
3
CREDITRISK
+
Model
Excluding default rate volatility
Including default rate volatility
Figure 7:
CREDITRI SK
+
Model 
Distribution of default
losses
The CREDITRISK
+
Model
allows explicit calculation of
the loss distribution of a
portf olio of credit exposures.
Si ze of l oss
P
r
o
b
a
b
i
l
i
t
y
3.4 Concentration Risk and Sector Analysis
The CREDITRISK
+
Model measures the benef it of portf olio diversif ication and the impact of concentrations
through the use of sector analysis.
3.4.1 Concentration Risk
Diversif ication arises naturally because the number of individual risks in a portf olio of exposures is usually large.
However, even in a portf olio containing a large number of exposures, there may be an opposing ef fect owing
to concentration risk. Concentration risk results f rom having in the portf olio a number of obligors whose
f ortunes are af fected by a common f actor. In order to quantif y concentration risk, the concepts of systematic
f actors and specif ic f actors are necessary.
Systematic factors
Systematic f actors are background f actors that af fect the f ortunes of a proportion of the obligors in the
portf olio, f or example all those obligors having their domicile in a particular country. The f ortunes of any one
obligor can be af fected by a number of systematic f actors.
Specific factors
In general, the f ortunes of an obligor are af fected to some extent by specif ic f actors unique to the obligor.
Systematic f actors impact the risk of extreme losses f rom a portf olio of credit exposures, while diversif ication
largely eliminates the impact of the specif ic f actors.
Concentration risk is dependent on the systematic f actors af fecting the portf olio. The technique f or measuring
concentration risk is sector analysis.
3.4.2 Sector Analysis  Allocating all Obligors to a Single Sector
The most straightf orward application of the CREDITRISK
+
Model is to allocate all obligors to a single sector.
This approach assumes that a single systematic f actor af fects the individual def ault rate volatility of each
obligor. Furthermore, this use of the model captures all of the concentration risk within the portf olio and
excludes the diversif ication benef it of the f ortunes of individual obligors being subject to a number of
independent systematic f actors.
Theref ore, the most straightf orward application of the CREDITRISK
+
Model, in which all obligors are allocated
to a single sector, generates a prudent estimate of extreme losses.
3.4.3 Sector Analysis  Allocating Obligors to one of Several Sectors
In order to recognise some of the diversif ication benef it of obligors whose f ortunes are af fected by a number
of independent systematic f actors, it can be assumed that each obligor is subject to only one systematic f actor,
which is responsible f or all of the uncertainty of the obligor’s def ault rate. For example, obligors could be
allocated to sectors according to their country of domicile. Once allocated to a sector, the obligor’s def ault rate
and def ault rate volatility are set individually. In this case, a sector can be thought of as a collection of obligors
having the common property that they are inf luenced by the same single systematic f actor.
3.4.4 Sector Analysis  Apportioning Obligors across Several Sectors
A more generalised approach is to assume that the f ortunes of an obligor are af fected by a number of
systematic f actors. The CREDITRISK
+
Model handles this situation by apportioning an obligor across several
sectors rather than allocating an obligor to a single sector.
20
CREDIT FIRST
SUI SSE BOSTON
Concentration risk is
dependent on the systematic
factors affecting the portfolio.
The technique for measuring
concentration risk is sector
analysis.
So f ar it has been assumed that all risk in the portf olio is systematic and allocable to one of the systematic
f actors. In addition to the ef fects of systematic f actors, it is likely that the f ortunes of an obligor are af fected
by f actors specif ic to the obligor. Potentially specif ic risk requires an additional sector to model each obligor,
since the f actor driving specif ic risk f or a given obligor af fects that obligor only. However, the CREDITRISK
+
Model handles specif ic risk without recourse to a large number of sectors by apportioning all obligors’ specif ic
risk to a single “Specif ic Risk Sector”.
3.4.5 The Impact of Sectors on the Loss Distribution
As stated above, the CREDITRISK
+
Model allows the portf olio of exposures to be allocated to sectors to ref lect
the degree of diversif ication and concentration present. The most diversif ied portf olio is obtained when
each exposure is in its own sector and the most concentrated is obtained when the portf olio consists of a
single sector.
The f igure below shows the impact of sectors on the loss distribution. As the number of sectors is increased,
the impact of concentration risk is reduced. The graph illustrates this by plotting the ratio of the 99th percentile
of the credit def ault loss distribution f or a given number of sectors to the 99th percentile of the credit def ault
loss distribution when the portf olio is considered to be a single sector.
3.5 MultiYear Losses for a HoldtoMaturity Time Horizon
As discussed in Section 2.5, the CREDITRISK
+
Model allows risk of the portf olio to be viewed on a holdto
maturity time horizon in order to capture any def ault losses that could occur until maturity of the credit
exposure.
Analysing credit exposures on a multiyear basis enables the risk manager to compare exposures of dif ferent
size, credit quality, and maturity. The loss distribution produced provides, f or any chosen level of conf idence, an
indication of the possible cumulative losses that could be suf fered until all the exposures have matured.
The benef its of looking at portf olio credit risk f rom this viewpoint include the f ollowing:
•
The f ull term structure of def ault probabilities is taken into account.
•
The f ull uncertainty of def ault losses over the life of the portf olio is also captured.
For example, because the oneyear average def ault rates f or investment grade obligors are relatively small but
the corresponding exposures may be large, a oneyear time horizon may not be the best measure f or active
portf olio management. However, a multiyear view will ref lect the f act that def aults f ollow a decline in credit
quality over many years.
CREDI TRI S K
+
21
3
CREDITRISK
+
Model
The CREDITRISK
+
Model
allows the portfolio of
exposures to be decomposed
into sectors to reflect the
degree of diversification and
concentration present.
Figure 8:
Impact of sectors on the
loss distribution
The CREDITRISK
+
Model
allows the risk of the portfolio
to be viewed on a holdto
maturity time horizon in order
to capture any default losses
that could occur until maturity
of the credit exposure.
Number of sect ors
9
9
t
h
p
e
r
c
e
n
t
i
l
e
r
a
t
i
o
0. 50
0. 55
0. 60
0. 65
0. 70
0. 75
0. 80
0. 85
0. 90
0. 95
1. 00
0 1 2 3 4 5 6 7 8 9
3.5.1 Using the CREDITRI SK
+
Model to Calculate MultiYear Loss Distributions
The CREDITRISK
+
Model can be used to calculate multiyear loss distributions by decomposing the exposure
prof ile over time into separate elements of discrete time, with the present value of the remaining exposure in
each time period being assigned a marginal def ault probability relevant to the maturity and credit quality. These
decomposed exposure elements can then be used by the CREDITRISK
+
Model to generate a loss distribution
on a holdtomaturity basis.
3.6 Summary of the CREDITRI SK
+
Model
The key features of the CREDITRISK
+
Model are:
•
The CREDIT RISK
+
Model captures the essential characteristics of credit default events. Credit def ault
events are rare and occur in a random manner with observed def ault rates varying signif icantly f rom year
to year. The approach adopted ref lects these characteristics by making no assumptions about the timing
or causes of def ault events and by incorporating the def ault rate volatility. By taking a portf olio approach,
the benef its of diversif ication that arise f rom a large number of individual risks are f ully captured.
Concentration risk, resulting f rom groups of obligors that are af fected by common f actors, is measured
using sector analysis.
•
The CREDIT RISK
+
Model is scaleable and comput ationally efficient. The CREDITRISK
+
Model is highly
scaleable and hence is capable of handling portf olios containing large numbers of exposures. The low data
requirements and minimum of assumptions make the CREDITRISK
+
Model easy to implement f or a wide
range of credit risk portf olios, regardless of the specif ic nature of the obligors. Furthermore, the ef f iciency
of the model allows comprehensive sensitivity analysis to be perf ormed on a continuing basis, which is a
key requirement f or the ability to quantif y the ef fects of parameter uncertainty.
22
CREDIT FIRST
SUI SSE BOSTON
E
C
CREDI TRI S K
+
23
4.1 Introduction to Economic Capital
4.1.1 The Role of Economic Capital
The analysis of uncertainty is the essence of risk management. Theref ore, measuring the uncertainty or
variability of loss and the related likelihood of the possible levels of unexpected losses in a portf olio of
exposures is f undamental to the ef fective management of credit risk. Suf f icient earnings should be generated
through adequate pricing and provisioning to absorb any expected loss. The expected loss is one of the costs
of transacting business which gives rise to credit risk. However, economic capit al is required as a cushion for
a firm’s risk of unexpected credit default losses, because the actual level of credit losses suf fered in any one
period could be signif icantly higher than the expected level.
4.2 Economic Capital for Credit Risk
4.2.1 Credit Default Loss Distribution
Knowledge of the credit def ault loss distribution arising f rom a portf olio of exposures provides a f irm with
management inf ormation on the amount of capital that the f irm is putting at risk by holding the credit portf olio.
Given that economic capital is necessary as a cushion f or a f irm’s risk of unexpected credit def ault losses, a
percentile level provides a means of determining the level of economic capital f or a required level of
conf idence. In order to capture a signif icant proportion of the tail of the credit def ault loss distribution, the 99th
percentile unexpected loss level over a oneyear time horizon is a suitable def inition f or credit risk economic
capital. This can be seen in the f ollowing f igure.
Econo
Capita
Economic Capital for Credit Risk 4
Economic Capital
Credit Default
Loss Distribution
Scenario Analysis
4.2.2 Benefits and Features of Economic Capital
Economic capital as a measure of risk being taken by a f irm has several features and benef its including the
f ollowing:
•
It is a more appropriate measure of the economic risk than that specified under the current regulatory regime.
•
It measures economic risk on a portf olio basis and hence takes account of the benef its of diversif ication.
•
It is a measure that objectively dif ferentiates between portf olios by taking account of credit quality and size
of exposure.
•
It is a dynamic measure, which ref lects the changing risk of a portf olio and hence can be used as a tool
f or portf olio optimisation.
4.3 Scenario Analysis
4.3.1 The Role of Scenario Analysis
The purpose of scenario analysis is to identif y the f inancial impact of low probability but nevertheless plausible
events that may not be captured by a statistically based model. Theref ore, the use of a credit risk model should
be supplemented by a programme of stress testing of the assumptions used.
There are two types of stress tests that should be perf ormed: (i) scenario analysis within the CREDITRISK
+
Model, and (ii) scenario analysis outside the CREDITRISK
+
Model.
4.3.2 Scenario Analysis within the CREDITRI SK
+
Model
The inputs into the CREDITRISK
+
Model can be stressed individually or in combination. For example, it is
possible to simulate downturns in the economy by increasing def ault rates and def ault rate volatilities  sectors
of the portf olio can be stressed to varying degrees ref lecting the f act that each sector could be af fected to a
dif ferent extent. Similarly, the f inancial impact of rating downgrades can be assessed by increasing the def ault
rate assigned to an obligor. For a derivatives portf olio, this can be extended to include the ef fects of movements
in market rates on credit exposures.
Given the ef f icient manner in which the def ault loss distribution can be calculated, it is possible to calculate
the impact of changing parameter inputs used by the model across a wide range of values.
24
CREDIT FIRST
SUI SSE BOSTON
Expected Loss
99th Percentile
Loss Level
Economic Capital
Figure 9:
Economic capital for
credit risk
Loss
P
r
o
b
a
b
i
l
i
t
y
Economic Capital
Credit Default
Loss Distribution
Scenario Analysis
4.3.3 Scenario Analysis outside the CREDITRI SK
+
Model
Certain stress tests can be dif f icult to perf orm within the CREDITRISK
+
Model: f or example, the impact of
political or f inancial uncertainty within a country. For these types of scenarios, analysis that is conducted
without reference to the outputs of the CREDITRISK
+
Model, such as looking at the exposure at risk f or a given
scenario, provides a realistic means of quantif ying the f inancial impact.
A f irm should control the risk of catastrophic losses through the use of obligor and concentration limits,
keeping any one of these limits within the loss f or the percentile level used to determine the economic capital
given by the CREDITRISK
+
Model.
The f igure below illustrates the way in which the distribution of losses can be considered to be divided into
three parts.
It is possible to control the risk of losses that f all within each of the three parts of the loss distribution in the
f ollowing ways:
Part of loss distribution Control mechanism
Up to Expected Loss Adequate pricing and provisioning
Expected Loss  99th Percentile Loss Economic capital and/ or provisioning
Greater than 99th Percentile Loss Quantif ied using scenario analysis and
controlled with concentration limits
Scenario analysis deals with quantif ying and controlling the risk of extreme losses. Losses up to a certain
conf idence level, such as the 99th percentile level, are controlled by the use of adequate pricing, provisioning
and economic capital. Provisioning f or credit risk is discussed in detail in Section 5.2.
CREDI TRI S K
+
25
4
Economic Capital
Scenario analysis
provides a means of
quantifying cat astrophic
losses  potential losses
can be controlled through
concentration limits.
Figure 10:
Parts of the credit default
loss distribution
Table 5:
Mechanisms for
controlling the risk of
credit default losses
Expected Loss
99th Percentile
Loss Level
Economic Capital
Loss
P
r
o
b
a
b
i
l
i
t
y
Covered by
pricing and
provisioning
Covered by capital
and/ or provisions
Quantif ied using scenario
analysis and controlled
with concentration limits
Applic
CREDIT
5.1 Introduction
CREDITRISK
+
includes several applications of the credit risk modelling technology in the areas of provisioning,
setting riskbased credit limits, and portf olio management.
5.2 Provisioning for Credit Risk
One application of CREDITRISK
+
is in def ining an appropriate credit risk provisioning methodology that ref lects
the credit losses of the portf olio over several years and hence that more accurately presents the true earnings
of the business by matching income with losses.
5.2.1 The Need for Credit Provisions
Generally, current accounting and provisioning policies recognise credit income and credit losses at dif ferent
times, even though the two events are related. Usually, credit loss provisions are made only when exposures
have been identif ied as nonperf orming. These provisions are of ten supplemented with other specif ic and
general credit provisions.
In relation to any portf olio of credit exposures, there is a statistical likelihood that credit def ault losses will occur,
even though the obligors are currently perf orming and it is not possible to identif y specif ically which obligors
will def ault. The level of expected loss ref lects the continuing credit risk associated with the f irm’s existing
perf orming portf olio and is one of the costs of doing creditrelated business. This level of expected loss should
be taken account of when executing any business that has a credit risk impact.
When def ault losses are modelled, it can be observed that the most f requent loss amount will be much lower
than the average, because, occasionally, extremely large losses are suffered, which have the effect of increasing
the average loss. Theref ore, a credit provision is required as a means of protecting against distributing excess
prof its earned during the below average loss years.
26
CREDIT FIRST
SUI SSE BOSTON
Applications of CREDITRISK
+
5
The CREDITRISK
+
credit
risk provisioning methodology
more accurately reflects the
true earnings of the business
by matching income with
losses.
Applications
Provisioning
Limits
Portfolio Management
c
T
CREDI TRI S K
+
27
5
Applications
The Annual Credit Provision
reflects the continuing credit
risk associat ed wit h t he
portfolio and is one of the costs
of doing business that creates
credit risk.
Figure 11:
Credit risk provisioning
The Incremental Credit
Reserve protects against
unexpected credit losses
and is used to absorb losses
that are higher than the
expected level.
5.2.2 Annual Credit Provision (ACP)
The starting point f or provisioning is to separate the existing portf olio into a nonperf orming and a perf orming
portf olio. The nonperf orming portf olio should be f ully provisioned to the expected recovery level available
through f oreclosure, administration or liquidation. Once f ully provisioned, the nonperf orming portf olio should
then be separated out and passed to a specialist team f or ongoing management.
As f or the perf orming portf olio, since no def ault has occurred, one needs a f orwardlooking provisioning
methodology. Under CREDITRISK
+
, the Annual Credit Provision (“ACP”) represents the f uture expected credit
loss on the perf orming portf olio, which is calculated as f ollows:
ACP = Exposure x Default Rate x (100%  Recover Rate)
The ACP should be calculated f requently in order to ref lect the changing credit quality of the portf olio. The ACP
is the f irst element of the credit provisioning methodology.
5.2.3 Incremental Credit Reserve (ICR)
The ACP represents only the expected or average level of credit losses. As experience shows, actual losses
that occur in any one year may be higher or lower than this amount, depending on the economic environment,
interest rates, etc. In f act, a better way of viewing the annual credit loss of the portf olio is as a distribution of
possible losses (outcomes), whose average equals the ACP but has a small probability of much larger losses.
In order to absorb these variations in credit losses f rom year to year, a second element of the provisioning
methodology, the Incremental Credit Reserve (“ICR”), can be established.
The CREDITRISK
+
Model provides inf ormation on the distribution of possible losses in the perf orming credit
portf olio. The ICR provides protection against unexpected credit losses (i.e. in excess of the ACP) and is
subject to a cap derived f rom the CREDITRISK
+
Model (the “ICR Cap”). The ICR Cap represents an extreme
case of possible credit losses (e.g. the 99th percentile loss level) on the perf orming portf olio.
ACP=Average level of credit losses
ICR Cap =99th Percentile
Loss Level
Typical
ICR
Loss
P
r
o
b
a
b
i
l
i
t
y
5.2.4 Provisioning for Different Business Lines
The credit risk provisioning methodology described above relates to credit risk arising f rom a loans business
where the income is accounted f or on an accruals basis rather than by markingtomarket.
A credit risk provision can also be established f or other credit business lines, such as traded bond portf olios
and derivatives portf olios. In each case, the CREDITRISK
+
Model provides the inf ormation required in order to
establish the provision that ensures that the accounting principle of matching income with losses is maintained.
For example, f or a portf olio of bonds, part of the expected loss is incorporated within the market price and
hence only the incremental credit reserve is required. This is described in the f ollowing table.
Portfolio type Accounting treatment Provision
Loan Accrual • ACP (1 year) charge to P&L each year
(Counterparty risk) (credit neutral) • ICR (1 year)
Derivatives Marktomarket • ACP (f ull maturity) held as marktomarket adjustment
(Counterparty risk) (credit neutral) • ICR (1 year)
Bond Marktomarket • ICR (1 year) to support business and protect against
(Issuer risk) (credit inclusive) distribution of prof its
5.2.5 Managing the Credit Risk Provision
As credit def aults occur, loans or exposures are moved f rom the perf orming to the nonperf orming portf olio
and hence provisioned to the expected recovery level. This increase in provision is then charged f irst against
the ACP and then, to the extent necessary, against the ICR. To the extent that actual credit losses are less than
the ACP within any given year, the balance is credited to the ICR up to the ICR Cap, beyond which the balance
is taken into P&L. This ensures that the ICR is replenished during low loss years f ollowing a large unexpected
loss, but that the ICR never exceeds the ICR Cap.
A worked example can be seen in the table below:
Year 1 2 3 4 5
Assumptions
Actual loan losses 500 600 300 300 650
ACP 500 525 550 610 625
ICR  Initial level 1,900    
ICR Cap 2,000 2,100 2,200 2,250 2,300
Income St atement
Operating prof it 2,100 2,100 2,205 2,315 2,430
Less: ACP (500) (525) (550) (610) (625)
Add: excess unutilised provision over ICR Cap 0 0 0 135 0
Pretax prof it 1,600 1,575 1,655 1,840 1,805
ICR (pre cap) 1,900 1,825 2,075 2,385 2,225
ICR Cap (as above) 2,000 2,100 2,200 2,250 2,300
Excess unutilised provision over ICR Cap 0 0 0 135 0
ICR (with cap applied) 1,900 1,825 2,075 2,250 2,225
28
CREDIT FIRST
SUI SSE BOSTON
Table 6:
Provisioning for different
business lines
Table 7:
Example of credit risk
provisioning
CREDI TRI S K
+
29
5.3 RiskBased Credit Limits
A system of individual credit limits is a wellestablished means of managing credit risk. Monitoring exposures
against limits provides a trigger mechanism f or identif ying potentially unwanted exposures that require active
management.
5.3.1 Standard Credit Limits
The system of credit limits may be viewed f rom a dif ferent perspective, if applying the methodologies described
within this document.
In particular, in order to equalise a f irm’s risk appetite between obligors as a means of diversif ying its portf olio,
a credit limit system could aim to have a large number of exposures with equal expected losses. The expected
loss f or each obligor can be calculated as the def ault rate multiplied by the exposure amount less the expected
recovery. This means that individual credit limits should be set at levels that are inversely proportional to the
def ault rate corresponding to the obligor rating.
As might be expected, this methodology gives larger limits f or better ratings and shorter maturities, but has the
benef it of allowing a f irm to relate the size and tenor of limits f or dif ferent rating categories to each other.
This approach can be extended to base limits on equalising the portf olio risk contribution f or each obligor.
A discussion on risk contributions and their use in portf olio management is provided later in this section.
5.3.2 Concentration Limits
Any excess country or industry sector concentration can have a negative ef fect on portf olio diversif ication and
increase the riskiness of the portf olio. As a result, a comprehensive set of country and industry sector limits is
required to address concentration issues in the portfolio. Concentration limits have the effect of limiting the loss
from identified scenarios and is a powerful technique for managing “tail” risk and controlling catastrophic losses.
5.4 Portfolio Management
The CREDITRISK
+
Model makes the process of controlling and managing credit risk more objective by
incorporating into a single measure all of the f actors that determine the amount of risk.
5.4.1 Introduction
Currently, the primary technique f or controlling credit risk is through the use of limit systems, including:
•
Individual obligor limits to control the size of exposure
•
Tenor limits to control the maximum maturity of transactions with obligors
•
Rating exposure limits to control the amount of exposure to obligors of certain credit ratings and
•
Concentration limits to control concentrations within countries and industry sectors.
5
Applications
A system of st andard
credit limits can be supple
mented with portfolio level
risk information to manage
credit risk.
Applications
Provisioning
Limits
Portfolio Management
5.4.2 Identifying Risky Exposures
The risk of a particular exposure is determined by f our f actors: (i) the size of exposure, (ii) the maturity of the
exposure, (iii) the probability of def ault, and (iv) the systematic or concentration risk of the obligor. Credit limits
aim to control risk arising f rom each of these f actors individually. However, f or managing risks on a portf olio
basis, with the aim of creating a diversif ied portf olio, a dif ferent measurement, which incorporates size, maturity,
credit quality and systematic risk into a single measure, is required.
5.4.3 Measuring Diversification
The loss distribution and the level of economic capital required to support a portf olio are measures of portf olio
diversif ication that take account of the size, maturity, credit quality and systematic risk of each exposure.
If the portf olio were less diversif ied, the spread of the distribution curve would be wider and a higher level of
economic capital would be required. Conversely, if the portf olio were more diversif ied, a lower level of economic
capital would be required. These measures can be used in managing a portf olio of exposures.
5.4.4 Portfolio Management using Risk Contributions
The risk contribution of an exposure is def ined as the incremental ef fect on a chosen percentile level of the
loss distribution when the exposure is removed f rom the existing portf olio. If the percentile level chosen is the
same as that used f or calculating economic capital, the risk contribution is the incremental ef fect on the
amount of economic capital required to support the portf olio.
Risk contributions have several features including the f ollowing:
•
The total of the risk contributions f or the individual obligors is approximately equal to the risk of the entire
portf olio.
•
Risk contributions allow the ef fect of a potential change in the portf olio (e.g. the removal of an exposure)
to be measured.
•
In general, a portf olio can be ef fectively managed by f ocusing on a relatively few obligors that represent a
signif icant proportion of the risk but constitute a relatively small proportion of the absolute portf olio
exposures.
Theref ore, risk contributions can be used in portf olio management. By ranking obligors in decreasing order of
risk contribution, the obligors that require the most economic capital can easily be identif ied.
This is illustrated in the f ollowing example. A portf olio was created f rom which a small number of exposures
with the highest risk contributions were removed. The ef fect on the loss distribution and the levels f or the
expected loss and the economic capital can be seen in the f igure opposite.
30
CREDIT FIRST
SUI SSE BOSTON
For managing credit
risks on a portfolio basis,
with the aim of creating
a diversified portfolio, a
different measure that
incorporates the magnitude
of the exposure, maturity,
credit quality and systematic
risk into a single measure
is required.
In general, a portfolio
can be effectively managed
by focusing on a relatively
few obligors that represent
a significant proportion of
the risk but constitute a
relatively small proportion
of the absolute portfolio
exposures.
Applications
Provisioning
Limits
Portfolio Management
The reduction in the 99th percentile loss level is larger than the reduction in the expected loss level, which
leads to an overall reduction in the economic capital required to support the portf olio.
5.4.5 Techniques for Distributing Credit Risk
Once obligors representing a signif icant proportion of the risk have been identif ied, there are several
techniques f or distributing credit risk that can be applied. These include the f ollowing:
•
Collateralisation: In the context of the CREDITRISK
+
Model, taking collateral has the ef fect of reducing the
severity of the loss given that the obligor has def aulted.
•
Asset securitisations: Asset securitisations involve the packaging of assets into a bond, which is then sold
to investors.
•
Credit derivatives: Credit derivatives are a means of transferring credit risk f rom one obligor to another,
while allowing client relationships to be maintained.
CREDI TRI S K
+
31
5
Applications
Original 99th Percentile
Loss Level
New 99th Percentile
Loss Level
New Expected Loss
Original Expected Loss
Figure 12:
Using risk contributions
Inf ormation about risk
contributions can be used
to facilitate risk management
and efficient use of economic
capital.
New Economic Capital
Original Economic Capital
Loss
P
r
o
b
a
b
i
l
i
t
y
A1 Overview of this Appendix
This appendix presents an analytic technique f or generating the f ull distribution of losses f rom a portf olio of
credit exposures. The technique is valid f or any portf olio where the def ault rate f or each obligor is small and
generates both oneyear and multiyear loss distributions.
The appendix applies the concepts discussed in Sections 2 and 3 of this document. The key concepts are:
•
Def ault rates are stochastic.
•
The level of def ault rates af fects the incidence of def ault events but there is no causal relationship between
the events.
In order to f acilitate the explanation of the CREDITRISK
+
Model, we f irst consider the case in which the mean
default rate for each obligor in the portfolio is fixed. We then generalise the technique to the case in which the
mean def ault rate is stochastic. The modelling stages of the CREDITRISK
+
Model and the relationships between
the dif ferent sections of this appendix are shown in the f igure opposite.
32
CREDIT FIRST
SUI SSE BOSTON
CREDIT
Model
Appendix A  The CREDITRISK
+
Model A
A2 Default Events with Fixed Default Rates
In Sections A2 to A5 we develop the theory of the distribution of credit def ault losses under the assumption
that the def ault rate is f ixed f or each obligor. Given this assumption and the f act that there is no causal
relationship between def ault events, we interpret def ault events to be independent. In Sections A6 onwards,
the assumption of f ixed def ault rates is relaxed, which introduces dependence between def ault events.
In Section A13 this dependence is quantif ied by calculating the correlation between def ault events implied by
the CREDITRISK
+
Model.
CREDI TRI S K
+
33
A
The CREDITRISK
+
Model
Figure 13:
Flowchart description of
Appendix A
T
Def ault events with
f ixed def ault rates
Def ault losses with
f ixed def ault rates
Calculation procedure
f or loss distribution with
f ixed def ault rates
Convergence of variable
def ault rate case to f ixed
def ault rate case
Application to
multiyear losses
Def ault rate uncertainty Sector analysis
Def ault events with
variable def ault rates
Def ault losses with
variable def ault rates
Calculation procedure f or
loss distribution with
variable def ault rates
General sector analysis
Risk contributions and
pairwise correlations
A2
A3
A4
A6 A7
A8
A9
A10
A12 A13
A11
A5
A2.1 Default Events
Credit def aults occur as a sequence of events in such a way that it is not possible to f orecast the exact time
of occurrence of any one def ault or the exact total number of def aults. In this section we derive the basic
statistical theory of such processes in the context of credit def ault risk.
Consider a portf olio consisting of N obligors. In line with the above assumptions, it is assumed that each
exposure has a def inite known probability of def aulting over a oneyear time horizon. Thus
(1)
To analyse the distribution of losses arising f rom the whole portf olio, we introduce the probability generating
f unction def ined in terms of an auxiliary variable z by
(2)
An individual obligor either defaults or does not default. The probability generating function for a single obligor
is easy to compute explicitly as
(3)
As a consequence of independence between def ault events, the probability generating f unction f or the whole
portf olio is the product of the individual probability generating f unctions. Theref ore
(4)
It is convenient to write this in the f orm
(5)
Suppose next that the individual probabilities of def ault are unif ormly small. This is characteristic of portf olios
of credit exposures. Given that the probabilities of def ault are small, powers of those probabilities can be
ignored. Thus, the logarithms can be replaced using the expression
4
(6)
and, in the limit, equation (5) becomes
(7)
where we write
(8)
f or the expected number of def ault events in one year f rom the whole portf olio.
To identif y the distribution corresponding to this probability generating f unction, we expand F(z) in its
Taylor series:
(9)
34
CREDIT FIRST
SUI SSE BOSTON
A obligor for default of y probabilit Annual ·
A
p
∑
∞
·
·
0
defaults) n ( ) (
n
n
z p z F
) 1 ( 1 1 ) ( − + · + − · z p z p p z F
A A A A
∏ ∏
− + · ·
A
A
A
A
z p z F z F )) 1 ( 1 ( ) ( ) (
∑
− + ·
A
A
z p z F )) 1 ( 1 log( ) ( log
) 1 ( )) 1 ( 1 log( − · − + z p z p
A A
4
This approximat ion
ignores t erms of
degree 2 and higher in
the default probabilities.
The expressions derived
from this approximation
are exact in the limit as
the probabilities of default
tend to zero, and give
good approximations
in practice.
) 1 (
) 1 (
) (
−
−
·
∑
·
z
z p
e e z F
A
A
µ
∑
·
A
A
p µ
n
n
n
z z
z
n
e
e e e z F
∑
∞
·
−
− −
· · ·
0
) 1 (
!
) (
µ
µ
µ µ µ
CREDI TRI S K
+
35
Thus if the probabilities of individual def ault are small, although not necessarily equal, then f rom equation (9)
we deduce that the probability of realising n def ault events in the portf olio in one year is given by
(10)
A2.2 Summary
In equation (10) we have obtained the wellknown Poisson distribution f or the distribution of the number of
def aults under our initial assumptions. The f ollowing should be noted:
•
The distribution has only one parameter, the expected number of def aults µ. The distribution does not
depend on the number of exposures in the portf olio or the individual probabilities of def ault provided that
they are unif ormly small.
•
There is no necessity f or the exposures to have equal probabilities of def ault; indeed, the probability of
def ault can be individually specif ied f or each exposure if suf f icient inf ormation is available.
The Poisson distribution with mean µ can be shown to have standard deviation given by √µ. Historical evidence
of the standard deviation of def ault event f requencies exists in the f orm of yearonyear def ault rate tables.
Such data suggests that the actual standard deviation is invariably much larger than √µ. Thus, our initial
assumption of f ixed def ault rates cannot account f or observed data. Bef ore addressing this in Section A6, we
f irst consider the derivation of the credit loss distribution f rom the results on def ault events above, retaining
our initial assumptions f or now.
A3 Default Losses with Fixed Default Rates
A3.1 Introduction
Under our initial assumptions, the distribution of numbers of def aults in a portf olio of exposures in one year
has been obtained. However, our main objective is to understand the likelihood of suf fering given levels of loss
f rom the portf olio, rather than given numbers of def aults. The distributions are dif ferent because the same level
of def ault loss could arise equally f rom a single large def ault or f rom a number of smaller def aults in the same
year. Unlike the variation of default probability between exposures, which does not influence the distribution of
the total number of def aults, dif fering exposure amounts result in a loss distribution that is not Poisson in
general. Moreover, inf ormation about the distribution of dif ferent exposures is essential to the overall
distribution. However, it is possible to describe the overall distribution because its probability generating
f unction has a simple closed f orm amenable to computation.
A3.2 Using Exposure Bands
The f irst step in obtaining the distribution of losses f rom the portf olio in an amenable f orm is to group the
exposures in the portf olio into bands. This has the ef fect of signif icantly reducing the amount of data that must
be incorporated into the calculation.
Banding introduces an approximation into the calculation. However, provided the number of exposures is large
and the width of the bands is small compared with the average exposure size characteristic of the portf olio, the
approximation is insignif icant. Intuitively, this corresponds to the f act that the precise amounts of exposures in
a portf olio cannot be critical in determining the overall risk.
A
The CREDITRISK
+
Model
!
defaults) (n y Probabilit
n
e
n
µ
µ −
·
Once the appropriate notation has been set up, an estimate of the ef fect of banding on the mean and standard
deviation of the portf olio is given below.
A3.3 Notation
In this section, the notation used f or the exposure banding described above is detailed.
Reference Symbol
Obligor A
Exposure L
A
Probability of def ault P
A
Expected Loss λ
A
In order to perf orm the calculations, a unit amount of exposure L, denominated in a base currency, is chosen.
For each obligor A, def ine numbers ε
A
and ν
A
by writing
and (11)
Thus, ν
A
and ε
A
are the exposure and expected loss, respectively, of the obligor, expressed as multiples of
the unit.
The key step is to round each exposure size ν
A
to the nearest whole number. This step replaces each exposure
amount L
A
by the nearest integer multiple of L. If a suitable size f or the unit L is chosen, then, af ter the rounding
has been perf ormed f or a large portf olio, there will be a relatively small number of possible values f or ν
A
each
shared by several obligors.
The portf olio can then be divided into m exposure bands, indexed by j, where 1 ≤ j ≤ m. With respect to the
exposure bands, we make the f ollowing def initions
Reference Symbol
Common exposure in Exposure Band j in units of L ν
j
Expected loss in Exposure Band j in units of L ε
j
Expected number of def aults in Exposure Band j µ
j
The f ollowing relations hold, expressing the expected loss in terms of the probability of def ault events
;hence (12)
Note that, because we have rounded the ν
j
to make them whole numbers, the expected loss ε
A
will be af fected,
by equation (12) unless a compensating rounding adjustment is made to the expected number of def ault
events µ
j
. If no adjustment is made, the rounding process will result in a small rounding up of the expected loss.
Under the assumption stated above, that the unit size is small relative to the typical exposure size of the
portf olio, these approaches each have an immaterial ef fect on the loss distribution. In the rest of this Appendix
it is assumed that an adjustment to the default probabilities to preserve the expected losses is made. Provided
the exposure sizes are rounded up, then, as will be shown in Section A4.2, the rounding leads to a small
overstatement of the standard deviation.
36
CREDIT FIRST
SUI SSE BOSTON
A A
L L ν × ·
A A
L ε λ × ·
j j j
µ ν ε × ·
∑
·
· ·
j A
A
A
A
j
j
j
ν ν
ν
ε
ν
ε
µ
:
CREDI TRI S K
+
37
As in equation (8), let µ stand f or the total expected number of def ault events in the portf olio in one year. Since
µ is the sum of the expected number of def ault events in each exposure band, we have
(13)
A3.4 The Distribution of Default Losses
We have analysed the distribution of def ault events under our initial assumptions. We now proceed to derive
the distribution of def ault losses.
Intuitively, the def ault loss analysis involves a second element of randomness, because some def aults lead to
larger losses than others through the variation in exposure amounts over the portf olio. As with def ault events,
the second random ef fect is best described mathematically through its probability generating f unction. Thus,
let G(z) be the probability generating f unction f or losses expressed in multiples of the unit L of exposure:
(14)
The exposures in the portf olio are assumed to be independent. Theref ore, the exposure bands are independent,
and the probability generating f unction can be written as a product over the exposure bands
(15)
However, by treating each exposure band as a portf olio and using equation (9), we obtain
(16)
Theref ore
(17)
This is the desired f ormula f or the probability generating f unction f or losses f rom the portf olio as a whole.
In the next section, we show how to use the probability generating f unction to derive the actual distribution
of losses under our initial assumptions.
For later reference, equation (17) can be restated in a slightly dif ferent f orm. First, def ine a polynomial P(z)
as f ollows
(18)
where we have used equations (12) and (13) f or the total number µ of def aults in the portf olio. The probability
generating f unction in equation (17) can now be expressed as
(19)
This functional form for G(z) expresses mathematically the compounding of two sources of uncertainty arising,
respectively, f rom the Poisson randomness of the incidence of def ault events and the variability of exposure
amounts within the portf olio.
A
The CREDITRISK
+
Model
∑ ∑
· ·
· ·
m
j
j
j
m
j
j
1 1
ν
ε
µ µ
∑
∞
·
× · ·
0
L) n losses aggregate ( ) (
n
n
z p z G
) ( ) (
1
z G z G
m
i
i ∏
·
·
j
j j j
j
j
z n
n n
n
j n
j
e z
n
e
z p z G
ν
µ µ ν
µ
ν
µ
+ −
∞
·
∞
·
−
· · ·
∑ ∑
0 0
!
defaults) n ( ) (
∑ ∑
· ·
· ·
+ −
·
+ −
∏
m
j
m
j
j
j j
j
j j
z m
j
z
e e z G
1 1
1
) (
ν
ν
µ µ
µ µ
∑
,
_
¸
¸
∑
,
_
¸
¸
·
∑
·
·
·
·
m
j
j
j
m
j
j
j
m
j
j
j
j
z
z
z P
1
1
1
) (
ν
ε
ν
ε
µ
µ
ν
ν
)) ( ( ) (
) 1 ) ( (
z P F e z G
z P
· ·
− µ
Note that G(z) depends only on the data ν and ε. Theref ore, to obtain the distribution of losses f or a large
portf olio of credit risks, all that is needed is knowledge of the dif ferent sizes of exposures ν within the portf olio,
together with the share ε of expected loss arising f rom each exposure size. This is typically a very small amount
of data, even f or a large portf olio.
A4 Loss Distribution with Fixed Default Rates
A4.1 Calculation Procedure
In this section, a computationally ef f icient means of deriving the actual distribution of credit losses is derived
f rom the probability generating f unction given by equation (17). In Section A10, this approach will be
generalised to compute the distribution f or the CREDITRISK
+
Model.
For n an integer let A
n
be the probability of a loss of nxL, or n units f rom the portf olio. We wish to compute A
n
ef f iciently. Comparing the def inition in equation (14) with the Taylor series expansion f or G(z), we have
(20)
In our case G(z) is given in closed f orm by equation (17). Using Leibnitz’s f ormula we have
(21)
However
(22)
and by def inition
(23)
Theref ore
(24)
Using the relation ε
j
= ν
j
x µ
j
f rom equation (12), the f ollowing recurrence relationship is obtained
(25)
This recurrence relationship allows very quick computation of the distribution. In order to commence the
computation, we have the f ollowing f ormula f or the f irst term, which expresses the probability of no loss arising
f rom the portf olio
(26)
38
CREDIT FIRST
SUI SSE BOSTON
n
z
n
n
A
dz
z G d
n
p · ·
· 0
) (
!
1
) nL of loss (
0
1
1
1
0
). (
!
1 ) (
!
1
·
·
−
−
·
,
_
¸
¸
·
∑
z
m
j
v
j
n
n
z
n
n
j
z
dz
d
z G
dz
d
n
dz
z G d
n
µ
0
1
0 1
1
1
1
1
) (
1
!
1
·
−
· ·
+
+
− −
− −
∑ ∑
,
_
¸
¸
,
_
¸
¸ −
·
z
n
k
m
j
j
k
k
k n
k n
j
z
dz
d
z G
dz
d
k
n
n
ν
µ
¹
'
¹ − · +
·
,
_
¸
¸
·
·
+
+
∑
otherwise 0
some for 1 k if )! 1 (
0
1
1
1
j v k
z
dz
d
j j
z
m
j
v
j
k
k
j
µ
µ
1
0
1
1
)! 1 ( ) (
− −
·
− −
− −
− − ·
k n
z
k n
k n
A k n z G
dz
d
j
j
j
v n
n v j
j j
k n j
j v k
n k
n
A
n
v
A k n k
k
n
n
A
−
≤
− −
− ·
− ≤
∑ ∑
· − − +
,
_
¸
¸
−
·
:
1
some for 1
1
)! 1 ( )! 1 (
1
!
1
µ
µ
j
j
v n
n v j
j
n
A
n
A
−
≤
∑
·
:
ε
∑
· · · ·
·
−
−
m
j j
j
e e P F G A
1
)) 0 ( ( ) 0 (
0
ν
ε
µ
CREDI TRI S K
+
39
Again, it is worthwhile to note that the calculation depends only on knowledge of ε and ν. In practice, these
represent a very small amount of data even f or a large portf olio consisting of many exposures.
A4.2 Precision Using Exposure Bands
The banding process described in Section A3 introduces a small degree of approximation into the data. In this
section, we show that the approximation error is normally not material by considering the ef fect on the portf olio
mean and standard deviation.
In terms of the notation above, the total portf olio expected loss ε and total portf olio standard deviation σ are
; (27)
where the expected loss and standard deviation are expressed in the chosen unit L.
In order to represent the banding, suppose that the above are expressions f or the “true” mean and standard
deviation, but that now the ν are rounded to integer multiples of the unit as explained above. This process
introduces an error; however, write
where (28)
Each τ
j
is at most of absolute size one. It is assumed that the exposures are rounded up, so that each τ
j
is positive.
The expected loss is unaf fected by the method of rounding chosen, because its expression is independent of
the banded exposure amounts. This was noted above.
For the standard deviation, we have
(29)
where ε is the expected loss f or the portf olio. Taking square roots and neglecting higherorder terms in the
Taylor series we obtain
(30)
For a real portf olio, the expected loss ε and the quantity 2σ are of the same order. We conclude that:
•
The expected loss calculated by the model is unaf fected by the banding process.
•
The standard deviation is overstated by an amount comparable with the chosen unit size.
A5 Application to MultiYear Losses
A5.1 Introduction
The recurrence relation above was derived on the basis of a oneyear loss distribution. In this section it is
shown how the initial model can be applied over a multiyear time horizon.
As in the discussion over a oneyear horizon, consider a portf olio of obligors with small probabilities of def ault.
For simplicity it is assumed that the f uture of the portf olio is divided into years. The exposures are permitted to
vary f rom year to year. In particular, each exposure has an individual maturity corresponding to the normal
maturity of bonds, loans or other instruments.
A
The CREDITRISK
+
Model
∑
·
·
m
j
j
1
ε ε
∑
·
× ·
m
j
j j
1
2
ε ν σ
j j j
τ ν ν + ·
ˆ
1 0 ≤ ≤
j
τ
ε σ ε σ ε τ σ ε ν σ σ + · + ≤ × + · × · ≤
∑ ∑ ∑
· · ·
2
1
2
1
2
1
2 2
ˆ
ˆ
m
j
j
m
j
j j
m
j
j j
σ
ε
σ
σ
ε
σ σ σ
2
2
1 ˆ
2
+ ·
,
_
¸
¸
+ ≤ ≤
A5.2 Term Structure of Default
In order to address a multiyear horizon, marginal rates of def ault must be specif ied f or each f uture year f or
each obligor in the portf olio. Collectively, such marginal def ault rates give the term structure of def ault f or
the portf olio.
A5.3 Notation
Fix the f ollowing notation:
Reference Symbol
(t)
Probability of def ault of exposure j in year t p
j
(t) (t)
Amount of exposure j in year t L
j
= Lν
j
(t) (t)
Expected loss in exposure j in year t λ
j
= Lε
j
As for the oneyear discussion, L is the unit of exposure and the ν
j
(t)
are dimensionless whole numbers. Under
the natural assumption that def aults by the same exposure in dif ferent years are mutually exclusive, the
probability generating f unction f or multiyear losses f rom a single exposure is given by
(31)
For the generating f unction of total losses, we have
(32)
In the limit of small probabilities of def ault we argue as f or equation (6) to obtain
(33)
and we obtain
(34)
and
(35)
The probability generating f unction f or the loss distribution is theref ore given by
(36)
This has the same f orm as the one year probability generating f unction (17). Theref ore, the recurrence relation
given by equation (25) f or the distribution of losses over one year is also applicable to the calculation of the
multiyear distribution of losses
(37)
40
CREDIT FIRST
SUI SSE BOSTON
∑ ∑ ∑
· · ·
− + · + − ·
T
t
t
j
T
t
t
j
T
t
t
j j
t
j
t
j
z p z p p z G
0
) (
0
) (
0
) (
) 1 ( 1 1 ) (
) ( ) (
ν ν
∑ ∑
,
_
¸
¸
− + ·
· j
T
t
t
j
t
j
z p z G
1
) (
) 1 ( 1 log ) ( log
) (
ν
∑ ∑
· ·
− ·
,
_
¸
¸
− +
T
t
t
j
T
t
t
j
t
j
t
j
z p z p
1
) (
1
) (
) 1 ( ) 1 ( 1 log
) ( ) (
ν ν
∑ ∑ ∑∑
,
_
¸
¸
· − ·
·
· ν
ν ν
ν
ν
) (
) (
: ,
) (
1
) (
) 1 ( ) ( log
t
j
t
j
t j
t
j
j
T
t
t
j
p z z p z G
∑ ∑ ∑ ∑ ∑
·
·
·
· ·
· ·
T
t
j
t
j
t
j
T
t
j
t
j
t j
t
j
t
j
t
j
t
j
p p
1
: ,
) (
) (
1
: ,
) (
: ,
) (
) ( ) ( ) (
ν ν ν ν ν ν
ν
ε
∑
−
·
t j
t
j
t
j
t
j
z
e z G
,
) (
) (
) (
) 1 (
) (
ν
ν
ε
∑
≤
−
·
n v t j
v n
t
j
n
t
j
t
j
A
n
A
) (
) (
; ,
) (
ε
CREDI TRI S K
+
41
A6 Default Rate Uncertainty
The previous sections developed the theory of the loss distribution f rom a portf olio of obligors, each of which
has a f ixed probability of def ault. In the f ollowing sections, the CREDITRISK
+
Model will be developed f rom this
theory by incorporating def ault rate uncertainty and sector analysis. These concepts are introduced in this
section and Section A7 respectively.
Published statistics on the incidence of def ault events, f or example among rated companies in a given country,
show that the number of def ault events, and theref ore the average probability of def ault f or such entities,
exhibits wide variation f rom year to year
5
.
Such yearonyear statistics may be thought of as samples f rom a random variable whose expected value
represents an average rate of def ault over many years. The appearance of randomness is due to the incidence
of f actors, such as the state of the economy, that inf luence the f ortunes of obligors. The standard deviation of
the variable measures our uncertainty as to the actual def ault rate that will be exhibited over a given year.
Owing to def ault rate uncertainty, there is a chance that def ault rates will turn out to be higher over, f or example,
the next year than their average over many years suggests. This in turn leads to a higher chance of
experiencing extreme losses.
The situation may be summarised by the f ollowing three intuitive f acts about def ault rate uncertainty:
•
Observed def ault probabilities are volatile over time, even f or obligors having comparable credit quality.
•
The variability of def ault probabilities can be related to underlying variability in a relatively small number of
background f actors, such as the state of the economy, which af fect the f ortunes of obligors. For example,
a downward trend in the state of the economy may make most obligors in a portf olio more likely to def ault.
•
However, a change in the economy or another f actor will not cause obligors to def ault with certainty.
Whatever the state of the economy, actual def aults should still be relatively rare events. Theref ore the
analysis above which considered rare events is relevant in a suitably modif ied f orm.
The second point made above is that uncertainty arises f rom f actors that may af fect a large number of obligors
in the same way. In order to measure this ef fect and hence quantif y the impact of individual def ault rate
volatilities at the portf olio level, the concept of sector analysis is necessary. This concept is introduced in the
next section.
A7 Sector Analysis
A7.1 Introduction
It was noted above that the variability of def ault rates can be related to the inf luence of a relatively small
number of background f actors on the obligors within a portf olio. In order to measure the ef fect of these f actors,
it is necessary to quantif y the extent to which each f actor has an inf luence on a given portf olio of obligors.
A f actor such as the economy of a particular country may be considered to have a unif orm inf luence on obligors
whose domicile is within that country, but relatively little inf luence on other obligors in a multinational portf olio.
In this section, the measurement of background f actors is addressed by dividing the obligors among dif ferent
sectors, where each sector is a collection of obligors under the common inf luence of a major f actor af fecting
def ault rates. An initial example might be a division of the portf olio according to the country of domicile of each
obligor. In Section A12, a more general sector analysis, which allows f or the f act that, in reality, obligors may
be under the simultaneous inf luence of a number of major f actors, is presented.
A
The CREDITRISK
+
Model
5
If the default rates of obligors
were fixed, default events
would still have a nonzero
st andard deviation arising
from the randomness of the
default events themselves.
However, as remarked in
Section A2.2, comparison
with historic dat a shows
that observed volatility is
far higher than can be
accounted for in this way.
A7.2 Further Notation
New notation is needed to keep track of the division of the portf olio into sectors and to record the volatility of
the def ault rate f or each sector. Write S
k
: 1 ≤ k ≤ n f or the sectors, each of which should be thought of f or now
as a subset of the collection of obligors.
The CREDITRISK
+
Model regards each sector as driven by a single underlying f actor, which explains the
variability over time in the average total def ault rate measured f or that sector. The underlying f actor inf luences
the sector through the total expected rate of def aults in the sector, which is then modelled as a random variable
with mean µ
k
and standard deviation σ
k
specif ied f or each sector. The standard deviation will ref lect the degree
to which the probabilities of def ault of the obligors in the portf olio are liable to all be more or less than their
average levels. For example, in a sector consisting of a large number of obligors of low credit quality, the mean
def ault rate might be 5% per annum and the standard deviation of the actual def ault rate might be a similar
quantity. Then there will be a substantial chance in any year of the actual average probability of def ault in the
sector being, say, 10% instead of 5%. In turn it is much more likely that, say, 12% of the obligors will actually
def ault in that year. Had the standard deviation been zero, ref lecting that we were certain about the probability
of each obligor def aulting, then a year in which as many as 12% of the obligors def ault would have been a
much more remote possibility.
The table below summarises the new notation to specif y the sector decomposition of the portf olio. In particular,
f or each sector we introduce a random variable x
k
representing the average def ault rate over the sector.
The mean of x
k
is µ
k
and the standard deviation is σ
k
.
Sector S
k
: 1 ≤ k ≤ n
Random variable representing the mean number of def aults x
k
Longterm annual average number of def aults  mean of x
k
µ
k
Standard deviation of x
k
σ
k
For each sector, the data requirements are set out below. Our original notation set up in Section A3.3 is also
repeated f or comparison
Exposure Dat a within Sector Previous New
Not ation Not ation
Base unit of exposure L L
(k) (k)
Exposure sizes in units L
j
= Lν
j
L
j
= Lν
j
1≤ j ≤ m 1≤ k ≤ n ;1≤ j ≤ m(k)
(k) (k)
Expected loss in each exposure band in units λ
j
= Lε
j
λ
j
= Lε
j
1≤ j ≤ m 1≤ k ≤ n ;1≤ j ≤ m(k)
The mean µ
k
is related to the expected loss data by the f ollowing relation which is the analogue of
equation (13):
(38)
42
CREDIT FIRST
SUI SSE BOSTON
∑
·
·
) (
1
) (
) (
k m
j
k
j
k
j
k
ν
ε
µ
CREDI TRI S K
+
43
A7.3 Estimating the Variability of the Default Rate
For each sector, in addition to the expected total rate of def ault µ
k
over the sector given by equation (38),
we must specif y a standard deviation σ
k
of the total expected rate of def ault. We discuss a convenient way
to estimate the standard deviation by reference to equation (38) f or the mean. Although equation (38)
is expressed in terms of exposure bands, it can equivalently be expanded as a sum over all the obligors in
the sector
(39)
where the summation extends over all obligors A belonging to sector k, and the relation
(40)
expresses the average probability of def ault of the obligor over the time period. To obtain an estimate of the
standard deviation of each sector, we assume that, together with a probability of def ault p
A
, a standard deviation
σ
A
has been assigned f or the def ault rate f or each obligor within the sector. A convenient way to do this is to
assume that the standard deviation depends on the credit quality of the obligor. This pragmatic method
assumes that the credit quality of the obligors within a sector is a more signif icant inf luence on the volatility of
the expected def ault f requency than the nature of the sector.
We obtain an estimate of σ
k
f rom the set σ
A
of obligor standard deviations by an averaging process. Recall
that only one random variable, x
k
is held to account f or the unif orm variability of each of the probabilities
of def ault. That is, the actual def ault probability f or each obligor in the sector will be modelled as a random
variable proportional to x
k
, whose mean is equal to the specif ied mean def ault rate f or that obligor. To express
this dependence write x
A
f or the random def ault probability of the single obligor A. Our assumption can then
be written
(41)
Note that the mean of x
A
is correctly specif ied as p
A
by this equation. Assuming equation (41), in particular,
we have
(42)
where we have used equation (39). The sum runs over all obligors in the sector. We estimate the standard
deviation of this sector so as to ensure that this condition holds. Thus the standard deviation of the mean
def ault rate f or a sector is estimated as the sum of the estimated standard deviations f or each obligor in the
sector. An alternative and more intuitive description of the standard deviation σ
k
determined in this way is that
the ratio of σ
k
to the mean µ
k
is an average of the ratio of standard deviation to mean f or each obligor,
weighted by their contribution to the def ault rate. This is easily seen as f ollows. By equation (42)
(43)
A
The CREDITRISK
+
Model
∑
·
A
A
A
k
ν
ε
µ
A
A
A
p ·
ν
ε
k
k
A
A
A
x
x
µ ν
ε
·
k
A
A
A
k
k
A
k
k
A
A
A
A
σ
ν
ε
µ
σ
µ
σ
ν
ε
σ · · ·
∑ ∑ ∑
1
∑
∑
∑
∑
,
_
¸
¸
· ·
A
A
A
A
A
A
A
A
A
A
k
k
p
p
p
p
σ
σ
µ
σ
According to historical experience, the ratio σ
A
/ p
A
is typically of the order of one, so that the standard deviation
of the number of def aults observed year on year among obligors of similar credit quality is typically of the
same order as the average annual number of def aults. Equation (43) shows that the same is true f or
each sector, as one would expect. In the absence of detailed data, the obligor specif ic estimates of the ratio
σ
A
/ p
A
can be replaced by a single f lat ratio. Then, writing ω
k
f or this unif orm ratio, equation (43) reduces to
the simple f orm
(44)
If the nature of the sector made it more appropriate to estimate the standard deviation σ
k
directly, this would
be equivalent to estimating the f lat ratio ω
k
directly.
A8 Default Events with Variable Default Rates
A8.1 Conditional Default Rate
In this section, the distribution of def ault events f or the CREDITRISK
+
Model is obtained. This is achieved by
calculation of the probability generating function. Most of the work has been done already in the calculation of
the probability generating f unction in equation (7) when the def ault rate is f ixed. As in equation (2), the
probability generating f unction f or def ault events is written
(45)
Because the sectors are independent, F(z) can be written as a product over the sectors
(46)
We theref ore f ocus on the determination of F(z) f or a single sector. In the notation of Section A7, the average
def ault rate in sector k is a random variable, written x
k
, with mean µ
k
and standard deviation σ
k
. Conditional
on the value of x
k
, we can write down the probability generating f unction f or the distribution of def ault events
as f ollows
(47)
where equation (7) has been used. Suppose that x
k
has probability density f unction ƒ
k
(x), so that
(48)
Then, the probability generating f unction f or def ault events in one sector is the average of the conditional
probability generating f unction given by equation (47) over all possible values of the mean def ault rate, as the
f ollowing computation shows:
(49)
In order to obtain an explicit f ormula f or the probability generating f unction, an appropriate distribution f or X
k
must be chosen. We make the key assumption that x
k
has the Gamma distribution with mean µ
k
and standard
deviation σ
k
.
The Gamma distribution is chosen as an analytically tractable twoparameter distribution. Bef ore proceeding to
evaluate equation (49) explicitly, the basic properties of the Gamma distribution are stated.
44
CREDIT FIRST
SUI SSE BOSTON
k k k
µ ω σ × ·
∑
∞
·
·
0
defaults) n ( ) (
n
n
z p z F
∏
·
·
n
k
k
z F z F
1
) ( ) (
[ ]
) 1 (
) (
−
· ·
z x
k k
e x x z F
dx x f dx x x x P
k k
) ( ) ( · + ≤ ≤
∫
∑
∫
∑
∞
·
−
∞
·
∞
·
∞
·
· · ·
0
) 1 (
0
0
0
) ( ) ( ) defaults n ( ) defaults n ( ) (
x
z x
n
x
n
n
n
k
dx x f e dx x f x P z z P z F
CREDI TRI S K
+
45
A8.2 Properties of the Gamma Distribution
The Gamma distribution, written Γ(α, β), is a skew distribution, which approximates to the Normal distribution
when its mean is large. The probability density f unction f or a Γ(α, β)  distributed random variable X is
given by
(50)
∞
where Γ(α) = ∫ e
x
x
α1
dx is the Gamma f unction.
x = 0
The Gamma distribution Γ(α, β) is a two parameter distribution, f ully described by its mean and standard
deviation. These are related to the def ining parameters as f ollows
and (51)
Hence, f or sector k, the parameters of the related Gamma distribution are given by
and (52)
A8.3 Distribution of Default Events in a Single Sector
With the choice of Gamma distribution f or the f unction ƒ(x), the expression f or the probability generating
f unction
(53)
given by equation (49), can be directly evaluated. By substitution, change of variable and def inition of the
Gamma integral
(54)
Upon rearrangement this becomes, f or sector k
(55)
This is the probability generating f unction of the distribution of def ault events arising f rom sector k.
It is possible to identif y the distribution of def ault events underlying this probability generating f unction.
By expanding F
k
(z) in its Taylor series
(56)
the f ollowing explicit f ormula is obtained
(57)
This can be identif ied as the probability density of the Negative Binomial distribution.
A
The CREDITRISK
+
Model
dx x e dx x f dx x X x P
x
1
) (
1
) ( ) (
−
−
Γ
· · + ≤ ≤
α β
α
α β
αβ µ ·
2 2
αβ σ ·
2 2
/
k k k
σ µ α ·
k k k
µ σ β /
2
·
∫
∞
·
−
·
0
) 1 (
) ( ) (
x
z x
k
dx x f e z F
α α α α
α
α α
α
β
β β β α β
α
β β α β α β
) 1 (
1
) 1 )( (
) (
1 1 ) (
1
) (
) (
1 1
0
1
1
1
0
1
) 1 (
z z
z
dy
e
z
y
dx
x e
e z F
y
y
x
x
z x
k
− +
·
− + Γ
Γ
·
− +
,
_
¸
¸
− + Γ
·
Γ
·
− −
∞
·
−
−
−
−
∞
·
−
−
−
∫ ∫
k
k
k
k
k
k
p
z p
p
z F
k
β
β
α
+
·
,
_
¸
¸
−
−
·
1
where
1
1
) (
∑
∞
·
,
_
¸
¸ − +
− ·
1
1
) 1 ( ) (
n
n n
k
k
k k
z p
n
n
p z F
k
α
α
n
k
k
k
p
n
n
p P
k
,
_
¸
¸ − +
− ·
1
) 1 ( ) defaults n (
α
α
A8.4 Summary
The portf olio has been divided into n sectors with annual def ault rates distributed according to
(58)
The probability generating f unction f or def ault events f rom the whole portf olio is given by
(59)
where the parameters α
k
, β
k
and p
k
are given by
; and ; (60)
The default event distribution for each sector is Negative Binomial. The default event distribution for the whole
portf olio is not Negative Binomial in general but is an independent sum of the Negative Binomial sector
distributions. The corresponding product decomposition of the probability generating f unction is given by
equation (59).
A9 Default Losses with Variable Default Rates
A9.1 Introduction
The probability generating f unction in equation (59) gives f ull inf ormation about the occurrence of def ault
events in the portf olio. In order to pass f rom def ault events to def ault losses, this distribution must be
compounded with the inf ormation about the distribution of exposures. In Section A3.4, we perf ormed this
process conditional on a f ixed mean def ault rate. We now generalise this process to incorporate the volatility
of def ault rates.
A9.2 The Distribution of Default Losses
By analogy with equation (14), we introduce a second probability generating f unction G(z), the probability
generating f unction f or losses f rom the portf olio. Thus let
(61)
be the probability generating f unction of the distribution of loss amounts. We seek a closed f orm expression
f or G(z) and a means of ef f iciently computing G(z).
As f or the distribution of def ault events, sector independence gives a product decomposition of the probability
generating f unction
(62)
where G
k
(z) is the loss probability generating f unction f or sector k, 1 ≤ k ≤ n.
By analogy with equation (18), we def ine polynomials P
k
(z), 1 ≤ k ≤ n, by
(63)
46
CREDIT FIRST
SUI SSE BOSTON
) , (
k k
β α Γ
∏ ∏
· ·
,
_
¸
¸
−
−
· ·
n
k
k
k
n
k
k
k
z p
p
z F z F
1 1
1
1
) ( ) (
α
2 2
/
k k k
σ µ α ·
k k k
µ σ β /
2
· ) 1 /(
k k k
p β β + ·
∑
∞
·
× · ·
0
L) n losses aggregate ( ) (
n
n
z p z G
∏
·
·
n
k
k
z G z G
1
) ( ) (
∑
,
_
¸
¸
·
∑
,
_
¸
¸
∑
,
_
¸
¸
·
·
·
· ) (
1
) (
) (
) (
1
) (
) (
) (
1
) (
) (
) (
) (
1
) (
k m
j
k
j
k
j
k
k m
j
k
j
k
j
k m
j
k
j
k
j
k
k
j
k
j
z
z
z P
ν
ν
ν
ε
µ
ν
ε
ν
ε
CREDI TRI S K
+
47
where the expression f or µ
k
in equation (38) has been used. The P
k
(z) provide the link between def ault events
and losses, because the f ollowing relation holds
(64)
This is directly analogous to the formula G(z)=F(P(z)) obtained in equation (19) for a fixed mean default rate,
except that there is now one such relation f or each sector. In order to see that the relation continues to hold
in the present case, we expand equation (63) as a sum over individual obligors belonging to sector k. Thus
(65)
By equation (41) we have
(66)
The lef t hand side of equation (66) is the probability generating f unction of the distribution of losses where
each obligor A has def ault rate x
A
. This can be seen by comparing equation (17)  the expressions are the
same, except that in equation (17) terms with the same exposure amount have been collected.
Just as in equation (53), which expresses F
k
(z) as an integral of the Poisson probability generating f unction
over the space of possible values of the random variable x
k
, a conditional probability argument shows that G
k
(z)
is the integral of the lef t hand side of equation (66) over the same space. Thus
(67)
Where the last step f ollows f rom equation (66). By substitution into equation (55) and taking the product over
each sector, we obtain
(68)
This is a closed f orm expression f or the probability generating f unction. In the next section a recurrence relation
f or computing the distribution of losses f rom this expression is derived.
A10 Loss Distribution with Variable Default Rates
In this section, a recurrence relation, suitable f or explicitly calculating the distribution of losses f rom equation
(68), is presented. The relation is a f orm of the recurrence relation in Section A4, derived f or a wider class of
distributions.
A
The CREDITRISK
+
Model
)) ( ( ) ( z P F z G
k k k
·
∑
∑
∑
· ·
A
A
A
k
A
A
A
A
A
A
k
A
A
z
z
z P
ν
ν
ν
ε
µ
ν
ε
ν
ε
1
) (
( )
( ) 1 ) (
) 1 (
1
−
∑ −
∑ − ∑ ∑ + −
· · ·
z P x
z
x
z x z x x
k k
A
A
A
A
k
k
A
A
A
A A
A
A A
e e e e
ν
ν ν
ν
ε
µ
∫
∫
∑
∫
∞
·
−
∞
·
− ∞
·
∞
·
·
∑
· ·
0
) 1 ) ( (
0
) 1 (
0
0
) (
) ( ) ( ) nL of Loss ( ) (
k
k k
k
A
v
A
A
k
x
k k k
z P x
x
k k k
z x
n
x
k k k k
n
k
dx x f e
dx x f e dx x f x P z z G
∏
∑
∏
·
·
·
,
_
¸
¸
−
−
· ·
n
k
k m
j
k
j
k
j
k
k
k
n
k
k
k
k
j
z
p
p
z G z G
1
) (
1
) (
) (
1
) (
1
1
) ( ) (
α
ν
ν
ε
µ
A10.1 General Recurrence Relation
Suppose, in general, a power series expansion
(69)
def ines a f unction G(z) which satisf ies the dif ferential equation
(70)
where A and B are polynomials given respectively by
(71)
In other words, we require that the logarithmic derivative of G(z) be a rational f unction. Then, the terms
of the power series expansion in equation (69) satisf y the f ollowing recurrence relation
(72)
To see this, rearrange the dif ferential equation (70) as f ollows
(73)
By dif ferentiating G term by term, this leads to
(74)
For n ≥ 0 the terms in z
n
on the lef t hand and right hand side respectively are
and (75)
Equating these expressions and rearranging we obtain
(76)
or equivalently
(77)
as required.
A10.2 Application
In equation (68), the probability generating f unction of losses was derived in the f orm
(78)
48
CREDIT FIRST
SUI SSE BOSTON
∑
∞
·
·
0
) (
n
n
n
z A z G
) (
) ( ) (
) (
1
)) ( log (
z B
z A
dz
z dG
z G
z G
dz
d
· ·
s
s
r
r
z b b z B
z a a z A
+ + ·
+ + ·
... ) (
... ) (
0
0
,
_
¸
¸
− −
+
·
∑ ∑
− −
·
− +
·
− +
) 1 , 1 min(
0
1
) , min(
0 0
1
) (
) 1 (
1
n s
j
j n j
n r
i
i n i n
A j n b A a
n b
A
G z A
dz
dG
z B ) ( ) ( ·
,
_
¸
¸
,
_
¸
¸
·
,
_
¸
¸
+
,
_
¸
¸
∑ ∑ ∑ ∑
∞
· ·
∞
·
+
· 0 0 0
1
0
) 1 (
n
n
n
r
i
i
i
n
n
n
s
j
j
j
z A z a z A n z b
∑
·
− +
− +
) , min(
0
1
) 1 (
n s
j
j n j
A j n b
∑
·
−
) , min(
0
n r
i
i n i
A a
∑ ∑
·
− +
·
− +
− + − · +
) , min(
1
1
) , min(
0
1 0
) 1 ( ) 1 (
n s
j
j n j
n r
i
i n i n
A j n b A a A n b
∑ ∑
− −
·
− +
·
− +
− − · +
) 1 , 1 min(
0
1
) , min(
0
1 0
) ( ) 1 (
n s
j
j n j
n r
i
i n i n
A j n b A a A n b
∏
∑
∏
·
·
·
,
_
¸
¸
−
−
· ·
n
k
k m
j
k
j
k
j
k
k
k
n
k
k
k
k
j
z
p
p
z G z G
1
) (
1
) (
) (
1
) (
1
1
) ( ) (
α
ν
ν
ε
µ
CREDI TRI S K
+
49
Taking logarithmic derivatives with respect to z, it f ollows that
(79)
This expresses G’(z)/ G(z) as a rational f unction. Accordingly, af ter calculation of polynomials A(z) and B(z)
such that
(80)
the calculation in Section A10.1 is applicable and leads to a recurrence relation f or the loss amount distribution.
Note that the summation described in equation (80) must be perf ormed directly by adding the rational
summands. Provided the unit size is chosen so that the exposures ν
j
and theref ore the degrees of numerators
and denominators of the rational summands are not too large, this computation can be perf ormed quickly.
A11 Convergence of Variable Default Rate Case to Fixed Default Rate Case
Although the CREDITRISK
+
Model is designed to incorporate the ef fects of variability in the average rates of
def ault, there are two circumstances in which the CREDITRISK
+
Model behaves as if def ault rates were f ixed.
These are where:
•
The standard deviation of the mean def ault rate f or each sector tends to zero.
•
The number of sectors tends to inf inity.
In particular, the ef fect of either a large number of sectors or a low standard deviation of def ault rates on the
portf olio is the same; the behaviour in either case is as if def ault rates were f ixed. In the section on generalised
sector analysis, this f act will be used to f acilitate the analysis of specif ic risk within a portf olio. In this section,
a proof is given of the f irst convergence f act. The proof of the second convergence f act is similar.
The proof proceeds by showing that the probability generating f unction f or the CREDITRISK
+
Model converges
to the f orm
(81)
which is the probability generating f unction in equations (17) f or losses conditional on a f ixed mean def ault
rate. The CREDITRISK
+
Model has the f ollowing probability generating f unction f or def ault losses, given at
equation (68)
(82)
where , , and
A
The CREDITRISK
+
Model
∑
∑
∑
∑
·
·
·
−
·
−
·
′
·
′
n
k
k m
j
k
j
k
j
k
k
k m
j
k
j
k
k k
n
k
k
k
k
j
k
j
z
p
z
p
z G
z G
z G
z G
1
) (
1
) (
) (
) (
1
1
) (
1
) (
) (
1
) (
) (
) (
) (
ν
ν
ν
ε
µ
ε
µ
α
∑
∑
∑
·
·
·
−
−
·
n
k
k m
j
k
j
k
j
k
k
k m
j
k
j
k
k k
k
j
k
j
z
p
z
p
z B
z A
1
) (
1
) (
) (
) (
1
1
) (
) (
) (
1
) (
) (
ν
ν
ν
ε
µ
ε
µ
α
∑
−
,
_
¸
¸
·
·
m
j
j
j
j
z
e z G
1
) 1 (
) (
ν
ν
ε
∏
∑
∏
·
·
·
,
_
¸
¸
−
−
· ·
n
k
k m
j
k
j
k
j
k
k
k
n
k
k
k
k
j
z
p
p
z G z G
1
) (
1
) (
) (
1
) (
1
1
) ( ) (
α
ν
ν
ε
µ
∑
·
·
) (
1
) (
) (
k m
k
k
j
k
j
k
ν
ε
µ ) 1 /(
k k k
p β β + ·
k k k
µ σ β /
2
·
2 2
/
k k k
σ µ α ·
We consider the limit where σ
k
tends to zero. Then
β
k
¡ 0; p
k
= β
k
/ (1+ β
k
) ¡ 0 and α
k
= µ
k
/ β
k
¡ µ
k
/ p
k
Theref ore
(83)
In the limit
(84)
On collecting terms in the exponent having common values n across dif ferent values of k, the summation over
k is eliminated
(85)
as required.
A12 General Sector Analysis
A12.1 Introduction
In the derivation of the CREDITRISK
+
Model probability generating f unction f or the distribution of losses in
Section A9, it was assumed throughout that the portf olio is divided into sectors, each of which is a subset of
the set of obligors. This corresponds to a situation in which obligors f all into classes, each of which is driven
by one f actor but all of which are mutually independent.
We now consider a more generalised situation in which, as bef ore, a relatively small number of f actors explain
the systematic volatility of def ault rates in the portf olio, but it is not necessarily the case that the def ault rate
of an individual obligor depends on only one of the f actors. In these more general circumstances, it is not
possible to describe the portf olio with sectors consisting of groupings of the obligors, but the CREDITRISK
+
Model incorporates this situation in the same way as bef ore, replacing the concept of a sector with that of
a systematic f actor.
To understand how to generalise the sector analysis already presented, we reexamine the derivation of the
probability generating f unction f or the CREDITRISK
+
Model. In equation (68), the probability generating f unction
was derived by expressing it as a product over the sectors and then integrating with respect to the distribution
of def ault rates f or each sector:
(86)
However, this expression can also be viewed as a multiple integral
(87)
We regard the integrand as the probability density f unction of a compound Poisson distribution f or any
given set of values of the means x
k
, 1 ≤ k ≤ n. However, we are simultaneously uncertain about all these values.
Theref ore, the probability density f unction is then integrated over the space of all possible states represented
by the values of the x
k
and weighted by their associated probability density f unctions.
50
CREDIT FIRST
SUI SSE BOSTON
∏
∑
∏
∑
∏
·
·
·
·
·
,
_
¸
¸
−
−
→
,
_
¸
¸
−
−
· ·
n
k
p
k m
j
k
j
k
j
k
k
k
n
k
k m
j
k
j
k
j
k
k
k
n
k
k
k
k
k
j
k
k
j
z
p
p
z
p
p
z G z G
1
) (
1
) (
) (
1
) (
1
) (
) (
1
) ( ) (
1
1
1
1
) ( ) (
µ
ν
α
ν
ν
ε
µ
ν
ε
µ
∑ ∑
· →
+ −
·
∑
−
∏
·
k j
k
j
k
j
k
j
k j
k
j
k
j
k m
j
k
j
k
j
k
j
k
z
n
k
z
e e e z G
,
) (
) (
) (
,
) (
) (
) (
1
) (
) (
) (
1
) (
ν ν
ν
ε
ν
ε
ν
ε
µ
∑
·
−
j
j
j
j
z
e z G
) 1 (
) (
ν
ν
ε
∏
∫
∏
·
∞
·
−
·
· ·
n
k
x
k k k
z P x
n
k
k
dx x f e z G z G
k k
1
0
) 1 ) ( (
1
) ( ) ( ) (
∫
∏
∫
∞
·
·
−
∞
·
∑
·
·
0
1
) 1 ) ( (
0
) ( ... ) (
1
1 n
n
k
k k
x
k
n
k
k k
z P x
x
dx x f e z G
CREDI TRI S K
+
51
Using equation (66), we can examine the exponent in the integrand in its equivalent f orm
(88)
where we have used the delta notation
(89)
To generalise the concept of a sector in these circumstances, allowing each obligor to be inf luenced by more
than one f actor x
k
, we replace the delta f unction with an allocation of the obligors among sectors by choosing,
f or each obligor A, numbers
(90)
The allocation θ
Ak
represents the extent to which the def ault probability of obligor A is af fected by the f actor
underlying sector k. The sector analysis discussed in Section A7 corresponds to the special case
(91)
In the general case the expression in equation (88) is replaced by
(92)
where each obligor contributes a term
where (93)
Equation (65) is replaced by
where (94)
A12.2 Performing the Sector Decomposition
In this section, we show how to assimilate the data f or the CREDITRISK
+
Model f or generalised sector analysis.
We assume that f or each obligor in the portf olio an estimate has been made of the extent to which the volatility
of the obligor’s def ault rate is explained by the f actor k. As explained in Section A12.1, this is expressed by a
choice of number
(95)
f or each sector k and obligor A in the portf olio. The number θ
Ak
represents our judgement of the extent to
which the state of sector k inf luences the f ortunes of obligor A.
As in the special case discussed in Section A7, we must also provide estimates of the mean and standard
deviation f or each sector. We indicate a method of estimating these parameters, assuming again that estimates
have been obtained of both quantities f or each obligor by reference to obligor credit quality.
A
The CREDITRISK
+
Model
∑ ∑ ∑ ∑
· ∈ ·
− · − · −
n
k k A
A
A
k
k
Ak
k A
A
A
k
k
n
k
k k
A A
z
x
z
x
z P x
1 , 1
) 1 ( ) 1 ( ) 1 ) ( (
ν ν
ν
ε
µ
δ
ν
ε
µ
¹
'
¹
∈
∉
·
k A
k A
Ak
1
0
δ
∑
·
·
n
k
Ak Ak
1
1 : θ θ
Ak Ak
δ θ ·
∑ ∑
− · −
· k A
A
A
k
k
Ak
n
k
k k
A
z
x
z P x
, 1
) 1 ( ) 1 ) ( (
ν
ν
ε
µ
θ
) 1 ( −
A
z x
A
ν
∑
·
·
n
k
k
k
Ak
A
A
A
x
x
1
µ
θ
ν
ε
∑
·
A
A
A
Ak
k
k
A
z z P
ν
ν
ε
θ
µ
1
) (
∑
·
A
A
A
Ak k
ν
ε
θ µ
∑
·
·
n
k
Ak Ak
1
1 : θ θ
The mean f or each sector is the sum of contributions f rom each obligor, but now weighted by the allocations
θ
Ak
. Thus
(96)
Then, by analogy with equation (43), we express the ratio σ
k
/ µ
k
as a weighted average of contributions f rom
each obligor
;hence (97)
This estimates the standard deviation f or each f actor. The discussion in Section A7 is recaptured when
θ
Ak
= δ
Ak
as discussed above.
A12.3 Incorporating Specific Factors
So f ar we have assumed all variability in def ault rates in the portf olio to be systematic. Potentially, we require
an additional sector to model f actors specif ic to each obligor.
However, specif ic f actors can be modelled without recourse to a large number of sectors. It was remarked in
Section A11 that assigning a zero variance to a sector is equivalent to assuming that the sector is itself a
portf olio composed of a large number of subsectors. Hence, f or a portf olio containing a large number of
obligors, only one sector is necessary in order to incorporate specif ic f actors. Let the specif ic f actor sector be
sector 1. Then, f or each obligor A, the proportion of the variance of the expected def ault f requency f or that
obligor that is explained by specif ic risk is θ
A1
. Sector 1 would be assigned a total standard deviation given by
equation (97). However, f or the specif ic f actor sector only, this standard deviation can be set to zero.
The specif ic f actor sector then behaves as the limit of a large number of sectors, one f or each obligor in the
portf olio, with independent variability of their def ault rate. The lost standard deviation represented by σ
1
is a
measure of the benef it of the presence of specif ic f actors in the portf olio.
A13 Risk Contributions and Pairwise Correlation
A13.1 Introduction
In this section, we derive f ormulae f or two usef ul measures connected with the def ault loss distribution, as
f ollows:
•
Risk contributions are def ined as the contributions made by each obligor to the unexpected loss of the
portf olio, measured either by a chosen percentile level or the standard deviation.
•
Pairwise correlations between def ault events give a measure of the extent to which concentration risk is
present in the portf olio.
A13.2 Risk Contributions
In this section, we derive a f ormula f or the contribution of an individual obligor to the standard deviation of the
loss distribution in the CREDITRISK
+
Model.
For a portf olio of obligors A having exposure E
A
, the risk contribution f or obligor A can be def ined as the
marginal ef fect of the presence of E
A
on the standard deviation of the distribution of credit losses. Alternatively,
the risk contribution can be def ined as the marginal ef fect of the presence of E
A
on some other measure of
portf olio aggregate risk, such as a given loss percentile.
52
CREDIT FIRST
SUI SSE BOSTON
∑
·
A
A Ak k
µ θ µ
∑
∑
·
A
A Ak
A
A
A
A Ak
k
k
µ θ
µ
σ
µ θ
µ
σ
∑
·
A
A Ak k
σ θ σ
CREDI TRI S K
+
53
In the f irst case, an analytic f ormula f or the risk contribution is possible. The risk contribution can be written
, or equivalently (98)
Moreover, f or most models including the CREDITRISK
+
Model, the risk contributions def ined by equation (98)
add up to the standard deviation. This is because of the variancecovariance f ormula
(99)
where σ
A
and σ
B
are the standard deviations of the def ault event indicator f or each obligor. Provided the model
is such that the correlation coef f icients are independent of the exposures, equation (99) expresses the
variance as a homogeneous quadratic polynomial in the exposures. Hence, by a general property of
homogeneous polynomials we have
(100)
If the marginal ef fect on a given percentile is chosen as the def inition of risk contributions, then an analytic
f ormula will not be possible. Instead, one can use the approximation described next.
Let ε, σ and X be the expected loss, the standard deviation of losses and the loss at a given percentile level
f rom the distribution. Def ine the multiplier to the given percentile as ξ where
(101)
Then, we can def ine risk contributions to the percentile in terms of the contributions to the standard deviation
by writing
(102)
Then, in view of equations (100) and (101), we have
(103)
In the analysis below, we will concentrate on the determination of the contributions to the standard deviation.
In order to evaluate the right hand side of equation (98), we derive analytic f ormulae f or mean and variance of
the distributions of def ault events and def ault losses in the CREDITRISK
+
Model. We use the f ollowing
def initions, referring to a sector k, which are consistent with the notation used previously. Since the mean and
variance of the distribution of losses in the CREDITRISK
+
Model are both additive across sectors, we can work
with one sector f or most of the analysis. For ease of notation, we have theref ore suppressed the reference to
sector k where it is not necessary.
Reference Symbol Mean Variance
Loss severity polynomial (equation 94) P(z)
2
Def ault event probability generating f unction conditional E(z,x) µ
E
σ
E
on mean x
2
Probability density f unction f or mean x f(x) µ
f
σ
f
2
Def ault event probability generating f unction F(z) µ
F
= µ
k
σ
F
2
CREDITRISK
+
Model probability generating f unction G(z) µ
G
= ε
k
σ
G
A
The CREDITRISK
+
Model
A
A A
E
E RC
∂
∂
·
σ
A
A
A
E
E
RC
∂
∂
·
2
2
σ
σ
∑
·
B A
B A B A AB
E E
,
2
σ σ ρ σ
σ
σ
σ σ
σ
· ·
∂
∂
·
∑ ∑
2
2
2
1
2 2
A
A
A
A
A
E
E RC
X · +ξσ ε
A A A
RC C R ξ ε + ·
ˆ
( ) X RC C R
A
A A
A
A
· + · + ·
∑ ∑
ξσ ε ξ ε
ˆ
Here µ
k
and ε
k
are the mean number of def ault events in sector k and the expected loss f rom sector k
respectively. We have
(104)
This is merely a restatement of equation (64). Also, by equation (53)
(105)
For the probability generating f unctions E, F and G, we have, by general properties of probability generating
f unctions
, and (106)
, and
(107)
By def inition of x, we have
(108)
Because E(z, x) is the probability generating f unction of a Poisson distribution, we also have
(109)
By equations (105) and (106), bringing the differentiation by the auxiliary variable z under the integration sign,
we obtain
(110)
Similarly, using equations (105) and (107)
(111)
Hence
(112)
Equations (110) and (112) are the mean and variance of the distribution of def ault events. To provide the link
to the moments of the loss distribution, we use equation (104), which yields, by the chain rule
; (113)
Hence
(114)
54
CREDIT FIRST
SUI SSE BOSTON
)) ( ( ) ( z P F z G ·
∫
·
x
dx x f x z E z F ) ( ) , ( ) (
) 1 ( ) 1 (
2
2
2 2
dz
dG
dz
G d
G G
+ · + µ σ
) 1 (
dz
dE
E
· µ ) 1 (
dz
dF
F
· µ ) 1 (
dz
dG
G
· µ
) 1 ( ) 1 (
2
2
2 2
dz
dE
dz
E d
E E
+ · + µ σ ) 1 ( ) 1 (
2
2
2 2
dz
dF
dz
F d
F F
+ · + µ σ
x x
E
· ) ( µ
E E
µ σ ·
2
f
x x
E F
dx x xf dx x f x µ µ µ · · ·
∫ ∫
) ( ) ( ) (
( ) ( )
2 2 2 2 2 2 2
) ( ) (
f f f
x
E E
x
E E F F
dx x f x d x f µ σ µ µ µ µ σ µ σ + + · + · + · +
∫ ∫
2 2
f f F
σ µ σ + ·
dz
dP
z P
dz
dF
z
dz
dG
)) ( ( ) ( ·
2
2
2
2
2
2
2
) (
)) ( ( )) ( ( ) (
dz
z P d
z P
dz
dF
dz
dP
z P
dz
F d
z
dz
G d
+
,
_
¸
¸
·
2
2
2
2
2
2
2
) 1 ( )). 1 ( ( ) 1 ( )). 1 ( ( ) 1 ( )) 1 ( ( ) 1 ( )) 1 ( (
,
_
¸
¸
− + + ·
dz
dP
P
dz
dF
dz
dP
P
dz
dF
dz
P d
P
dz
dF
dz
dP
P
dz
F d
G
σ
CREDI TRI S K
+
55
Successive dif ferentiation of equation (94) yields
; and (115)
On substituting equations (112) and (115) into equation (114), we obtain
(116)
Substituting f or ε
k
, we obtain
(117)
Finally, summing over sectors gives the standard deviation of the CREDITRISK
+
Model Loss Distribution f or the
whole portf olio
(118)
Note that this is the standard deviation of the actual distribution of losses. As in the earlier sections, the
σ
k
denote the standard deviations of the f actors driving the def ault rates in each sector.
Risk contributions can now be derived directly by dif ferentiating equation (118). Thus, by equation (98)
(119)
Hence
(120)
where we have interchanged E
A
and ν
A
f or notational convenience. Hence
(121)
This is the required f ormula f or risk contributions to the standard deviation. As remarked above, it can be shown
explicitly that the risk contributions add up to the standard deviation of the portf olio loss distribution. Thus, f rom
equation (121),
(122)
Hence, using equation (118)
(123)
as required.
A
The CREDITRISK
+
Model
k
k
A
A Ak
k
dz
dP
µ
ε
ε θ
µ
· ·
∑
1
) 1 (
∑
− ·
A
A A Ak
k
dz
P d
) 1 (
1
) 1 (
2
2
ν ε θ
µ
1 ) 1 ( · P
( )
2 2
2 2 2
1
,
_
¸
¸
− +
,
_
¸
¸
− + ·
∑ ∑ ∑
A
A Ak
A
A A Ak
A
A Ak
k
k k F G
ε θ ν ε θ ε θ
µ
µ µ σ σ
( )
∑ ∑
+
,
_
¸
¸
· − +
,
_
¸
¸
+ ·
A
A A Ak
k
k
k k
A
A A Ak
k
k
k k G
ν ε θ
µ
ε
σ ε ν ε θ
µ
ε
µ σ σ
2
2 2
2
2 2 2
∑ ∑
+
,
_
¸
¸
·
· A
A A
n
k
k
k
k
ν ε
µ
σ
ε σ
1
2
2 2
A
A
A
A A
E
E
E
E RC
∂
∂
·
∂
∂
·
2
2
σ
σ
σ
,
_
¸
¸
,
_
¸
¸
+ ·
,
_
¸
¸
,
_
¸
¸
+
∂
∂
·
∑ ∑ ∑
k
A Ak k
k
k
A A
A
k
k
k
k
B
B B
A
A
A
E
E
E
E
RC µ θ ε
µ
σ
µ
σ
ε
µ
σ
ν ε
σ
2 2
2 2
2
2
2
,
_
¸
¸
,
_
¸
¸
+ ·
∑
k
Ak k
k
k
A
A A
A
E
E
RC θ ε
µ
σ
σ
µ
2
∑∑ ∑ ∑ ∑ ∑
,
_
¸
¸
+ ·
,
_
¸
¸
,
_
¸
¸
+ ·
A k
k Ak
A A
k
k
A
A A
A k
k Ak
k
k
A
A A
A
A
E E
E
E
RC ε θ
σ
µ
µ
σ
σ
µ
ε θ
µ
σ
σ
µ
2
2
2
σ
σ
σ
ε
µ
σ
ν ε
σ σ
µ
θ ε
µ
σ
σ
ν ε
· ·
,
_
¸
¸
,
_
¸
¸
+ ·
,
_
¸
¸
+ ·
∑ ∑ ∑ ∑ ∑ ∑
2
2
2 2
1
k
k
k
k
A
A A
k A
A A
Ak k
k
k
A
A A
A
A
E
RC
A13.3 Pairwise Correlation
In this section, we derive a formula for the pairwise correlation between default events in the CREDITRISK
+
Model.
To def ine caref ully the pairwise correlation over a time period ∆t, we associate to each obligor its indicator
f unction I
A
, which is the random variable having the values
(124)
Then, the correlation ρ between def ault of two obligors A and B in the time period ∆t is def ined by
(125)
That is, the statistical correlation between the indicator functions of A and B in the time period. If the expected
values of I
A
, I
B
and the product I
AB
are µ
A
, µ
B
and µ
AB
, respectively, then µ
A
, µ
B
and µ
AB
are, respectively, the
expected number of defaults of A, B and of both obligors in the time period. Then, because the indicator functions
can only take on the values 0 or 1, the standard expression for correlation reduces to the following form:
(126)
We seek an expression f or the right hand side of equation (126) in the context of the CREDITRISK
+
Model.
We take two distinct obligors A and B and make the following definitions, where the general sector decomposition
is used with n sectors.
Reference Obligor A Obligor B
Time period ∆t
Instantaneous def ault probability PA PB
Expected number of def aults µ
A
= 1  e
p
A
∆t
≈ p
A
∆t µ
B
= 1  e
p
B
∆t
≈ p
B
∆t
Sector decomposition θ
AK
; 1≤ k ≤ n θ
BK
; 1≤ k ≤ n
The unknown term in equation (126) is the expected joint def ault expectation µ
AB
. Since A and B are distinct,
f or any realised values of the sector means x
k
, 1 ≤ k ≤ n the events of def ault are independent, we have, writing
x
A
and x
B
as in equation (93)
(127)
where, as shown in the table, we have approximated the integrand, ignoring higher powers of the def ault
expectations and using the f ollowing approximation
6
(128)
In view of the sector decomposition, we have, by equation (93)
and (129)
For convenience, def ine coef f icients ω
kk’
by writing
(130)
56
CREDIT FIRST
SUI SSE BOSTON
¹
'
¹
·
otherwise 0
period time the in defaults A obligor if 1
A
I
) , (
B A AB
I I ρ ρ ·
2 1 2 2 1 2
) ( ) (
B B A A
B A AB
AB
µ µ µ µ
µ µ µ
ρ
− −
−
·
∫ ∫
∏
·
·
1
1
) ( ...
x x
n
k
k k k B A AB
n
dx x f x x µ
B A
t x t x
x x e e
B A
≈ − −
∆ − ∆ −
) 1 )( 1 (
6
The integration in (127) can
be performed without using
the approximation (128) to
give an exact value for the
correlation. It is interesting
to note that, while the exact
integration, which has a
similar form to equation (54),
depends on knowledge of
the probability generating
function f, only the mean
and st andard deviation of
f are required to estimate
the approximate correlation
given by equat ion (13 8).
In t his sense t he choice
of gamma distribution for
the variability of the mean
default rate is irrelevant to
correlation considerations.
∑
·
·
n
k
A Ak
k
k
A
x
x
1
µ θ
µ
∑
·
·
n
k
B Bk
k
k
B
x
x
1
µ θ
µ
B A
k k
k B Ak
k k
µ µ
µ µ
θ θ
ω
′
′
′
·
CREDI TRI S K
+
57
Then
(131)
and we deduce that
(132)
Hence
(133)
or
(134)
However
(135)
Thus
(136)
Substituting f or ω
kk
, we obtain
(137)
This simplif ies to
(138)
Equation (138) is the f ormula f or def ault event correlation between distinct obligors A and B in the
CREDITRISK
+
Model. Equation (138) is valid wherever the likely probabilities of def ault over the time period in
question are small, taking into account their standard deviation. It is not universally valid. Note, in particular, that
the value of ρ
AB
given by the f ormula can be more than one if too large values of the means and standard
deviations are chosen. This corresponds to the approximation used at equation (128)  the lef thand side is
clearly always less than unity while the approximating f unction is unbounded.
We note two salient features of equation (138):
•
If the obligors A and B have no sector in common then the correlation between them will be zero. This is
because no systematic f actor af fects them both.
•
If it is accepted that, as suggested by historical data, the ratios σ
k
/ µ
k
are of the order of unity, then
depending on the sector decomposition the correlation has the same order as the term √(µ
A
µ
B
) in the
equation. This is the geometric mean of the two def ault probabilities. Theref ore, in general one would
expect def ault correlations to typically be of the same order of magnitude as def ault probabilities
themselves.
A
The CREDITRISK
+
Model
∫ ∫
∏ ∑
· ′
′ ′
·
1
1 ,
) ( ...
x x
n
k
k k k
k k
k k k k AB
n
dx x f x x ω µ
∑
∫
∏
∫
∫
∏ ∑
∫∫
·
≠
≠ ·
′ ≠
′
≠ ·
′
≠
′ ′ ′ ′
+
·
′
n
k
k j x
n
k j j
j j j
x
k k k k kk
k k j x
n
k k j j
j j j
k k
x x
k k k k k k k k k k AB
j k
j k k
dx x f dx x f x
dx x f dx dx x f x f x x
1
;
; 1
2
, :
, ; 1
,
) ( ) (
) ( ) ( ) (
ω
ω µ
( )
∑ ∑
· ′ ≠ ·
′
+ + ·
n
k
k k kk
n
k k k k
k k k k AB
1
2 2
1 ,
'
σ µ ω µ µ ω µ
( )
∑ ∑
· ′ ≠ ·
′
+ + ·
n
k
k k kk
n
k k k k
k k k k AB
1
2 2
1 ,
'
σ µ ω µ µ ω µ
B A
n
k
k
k
B Bk
n
k
k
k
A Ak
n
k k
k k k k
µ µ µ
µ
µ θ
µ
µ
µ θ
µ µ ω ·
,
_
¸
¸
,
_
¸
¸
·
∑ ∑ ∑
· · · ′
′
1 1 1 ,
'
∑
·
+ ·
n
k
k kk B A AB
1
2
σ ω µ µ µ
( )
∑
·
−
,
_
¸
¸
·
− −
−
·
n
k
k
k
B A Bk Ak B A
B B A A
B A AB
AB
1
2
2 1
2 1 2 2 1 2
) ( ) (
µ
σ
µ µ θ θ µ µ
µ µ µ µ
µ µ µ
ρ
( )
∑
·
,
_
¸
¸
·
n
k
k
k
Bk Ak B A AB
1
2
2 1
µ
σ
θ θ µ µ ρ
B1 Example SpreadsheetBased Implementation
The purpose of this appendix is to illustrate the application of the CREDITRISK
+
Model to an example portf olio
of exposures with the use of a spreadsheetbased implementation of the model.
The implementation, consisting of a single spreadsheet together with an addin, can be downloaded f rom the
Internet (http:/ / www.csf b.com) to reproduce the results shown in this appendix. The spreadsheet contains
three examples of the use of the CREDITRISK
+
Model. In addition, the spreadsheet can be used on a user
def ined portf olio.
For illustrative purposes, we have limited the example portf olio size to only 25 obligors. However, the
spreadsheet implementation has been designed to allow analysis of portf olios of realistic size. Up to 4,000
individual obligors and up to 8 sectors can be handled by the spreadsheet implementation. However, there is
no limit, in principle, to the number of obligors that can be handled by the CREDITRISK
+
Model. Increasing the
number of obligors has only a limited impact on the processing time.
B2 Example Portfolio and Static Data
The three examples are based on a portf olio consisting of 25 obligors of varying credit quality and size of
exposure. The exposure amounts are net of recovery. Details of this portf olio are given in Table 8 opposite.
58
CREDIT FIRST
SUI SSE BOSTON
Illustra
Examp
Appendix B  Illustrative Example B
CREDI TRI S K
+
59
B
Illustrative Example
a
p
Table 8:
Example portfolio
Table 9:
Example mapping table of
default rate information
Credit
Name Exposure Rating
1 358,475 H
2 1,089,819 H
3 1,799,710 F
4 1,933,116 G
5 2,317,327 G
6 2,410,929 G
7 2,652,184 H
8 2,957,685 G
9 3,137,989 D
10 3,204,044 D
11 4,727,724 A
12 4,830,517 D
13 4,912,097 D
14 4,928,989 H
15 5,042,312 F
16 5,320,364 E
17 5,435,457 D
18 5,517,586 C
19 5,764,596 E
20 5,847,845 C
21 6,466,533 H
22 6,480,322 H
23 7,727,651 B
24 15,410,906 F
25 20,238,895 E
The example uses a credit rating scale to assign def ault rates and def ault rate volatilities to each obligor.
A table giving an example mapping f rom credit ratings to a set of def ault rates and def ault rate volatilities is
given. The table is shown as Table 9 below. The credit rating scale and other data in the table are designed f or
the purposes of the example only.
Credit Mean St andard
Rating Default rate Deviation
A 1.50% 0.75%
B 1.60% 0.80%
C 3.00% 1.50%
D 5.00% 2.50%
E 7.50% 3.75%
F 10.00% 5.00%
G 15.00% 7.50%
H 30.00% 15.00%
B3 Example use of the Spreadsheet Implementation
Three examples are given, each based on the same portf olio, as f ollows:
•
All obligors are allocated to a single sector.
•
Each obligor is allocated to only one sector. In this example, countries are the sectors. This assumes that
each obligor is subject to only one systematic f actor, which is responsible f or all of the uncertainty of the
obligor’s def ault rate.
•
Each obligor is apportioned to a number of sectors. Again, countries are the sectors. This ref lects the
situation in which the f ortunes of an obligor are af fected by a number of systematic f actors.
The examples are installed on the spreadsheet implementation, together with the results generated by the
model. For each example, the inputs to the model have been set to generate the f ollowing:
•
Percentiles of loss.
•
Full loss distribution.
•
Credit risk provision.
•
Risk contributions.
In this section, the steps to reproducing these results using the model implementation are described.
B3.1 Activating the CREDITRI SK
+
Model
Choose one of the three example worksheets to reproduce the results. Each worksheet is equipped with
a macro button. Press the button to activate the model implementation.
B3.2 Data Input Screen
On activation, the model will show the Data Input Screen. This screen is used to set the worksheet ranges of
data to be read in to the model and to specif y the f orm of output data required. The Data Input Screen has
been preset to the correct ranges corresponding to the layout of each example worksheet.
60
CREDIT FIRST
SUI SSE BOSTON
Activate Model
CREDI TRI S K
+
61
Press the Proceed button on the Data Entry Screen to proceed to the next step.
B3.3 Input Data Check
The model implementation has been preset to identif y errors in the data read in bef ore the calculation
commences. The user is given the option of switching of f this f acility via the Data Input Screen. The model
implementation ensures that the data satisf ies the f ollowing three criteria:
•
The sector allocation table contains only numeric data.
•
The decomposition of each obligor to the various sectors adds up to 100%.
•
A sector must contain at least one allocation entry.
The Input Data Check screen indicates the location of an error in the sector allocation table.
B
Illustrative Example
Data Entry Screen
Press the Proceed button
on the Data Entry Screen
to proceed to the next step.
Credit Suisse First Boston
Example 1A
Data Input Ranges
Obligor Name $ B$ 11:$ B$ 35
Exposure $ C$ 11:$ C$ 35
Mean Default Rate $ E$ 11:$ E$ 35
Standard Deviation $ F$ 11:$ F$ 35
Range of Sectors $ G$ 11:$ G$ 35
Confirm Number of Sectors 1
Optional Settings
Use Sector 1 for Specific Risk
X Check Input Data
Display Options
X Preliminary Statistical Data
X Percentile Losses
Data Output Ranges
Percentiles Output Range $ M$ 11:$ N$ 11
Distribution Output Range $ P$ 11:$ Q$ 11
Risk Contributions Output Range $ I$ 11:$ K$ 11
Percentile for Risk Contributions 99
X Print Percentiles to Worksheet
X Print Distribution to Worksheet
X Print Risk Contributions to Worksheet
Proceed Alter Percentiles
Cancel
Press the Proceed button on the Input Data Check Screen to proceed with the calculation.
B3.4 Portfolio Loss Distribution Summary Statistics
The model implementation has been preset to display summary statistics of the portf olio loss distribution.
The user is given the option to switch of f this f acility via the Data Input Screen.
Press the Proceed button to proceed to the next step. The model implementation now calculates the f ull
distribution of losses.
B3.5 Percentile Losses, Loss Distribution and Credit Risk Provision
The model implementation has been preset to display summary statistics of the portf olio loss distribution.
The user can switch of f this f acility via the Data Input Screen.
62
CREDIT FIRST
SUI SSE BOSTON
Input Data Check Screen
Press the Proceed button
on the Input Data Check
Screen to proceed with
the calculation.
Input Data Check Screen
Press the Proceed button
to proceed to the next step.
The model implementation
now calculates the full
distribution of losses.
Credit Suisse First Boston
No Errors Were Detected In The Input Data
Data Input Error Trapping
Error Type
Sector allocation cells with nonnumeric entries
None
Obligors whose sector decomposition does not sum to 100%
None
Sectors having zero expected loss
None
Proceed
Return to Input Screen
Credit Suisse First Boston
Portfolio Loss Distribution Summary Statistics
Portfolio Aggregate Exposure 130,513,072
Portfolio Expected Loss 14,221,863
Portfolio Standard Deviation 12,668,742
Amounts are st ated in the input units of currency
Proceed
Return to Input Screen
CREDI TRI S K
+
63
Press the Proceed button to output the loss percentiles, the loss distribution, and the risk contributions to
the worksheet.
Loss Distribution
A graph of the loss distribution has been set up on each worksheet using the results generated f rom the
step above.
Credit Risk Provision
From the summary statistic data above, the Annual Credit Provision (ACP) is given by the Expected Loss, i.e.
14,221,863. If the 99th percentile level is chosen as the determining level f or the Incremental Credit Reserve
Cap (ICR Cap), then the ICR Cap is 55,311,503.
B3.6 Risk Contributions
In each example, the model implementation has been preset to output risk contributions f or each obligor in the
example portf olio. The risk contribution calculated by the model is def ined as the marginal impact of the obligor
on a chosen percentile of the loss distribution. The model implementation has been preset to calculate risk
contributions by reference to the 99th percentile loss. This setting can be altered to a dif ferent percentile via
the Data Entry Screen.
B
Illustrative Example
Input Data Check Screen
Press the Proceed button
to output the loss percentiles,
the loss distribution, and
the risk contributions to the
worksheet.
Credit Loss Distribution
0.00%
0.50%
1.00%
1.50%
2.00%
2.50%
0 10, 000, 000 20, 000, 000 30, 000, 000 40, 000, 000 50, 000, 000 60, 000, 000 70, 000, 000
Loss
M
a
r
g
i
n
a
l
P
r
o
b
a
b
i
l
i
t
y
Credit Suisse First Boston
Summary Statistical Data
Aggregate Exposure 130,513,072
Expected Loss 14,221,863
Standard Deviation 12,668,742
Loss Percentiles
50.00 11,089,455
75.00 20,498,062
95.00 38,908,486
97.50 46,152,128
99.00 55,311,503
99.50 62,033,181
99.75 68,612,540
99.90 77,133,478
Proceed
Return to Input Screen
B3.7 Using Risk Contributions For Portfolio Management
Example 1 has been split into two examples, 1A and 1B, to illustrate the use of risk contributions in portf olio
management as f ollows:
•
In example 1A, all 25 obligors are included in the portf olio. Table 10 shows that obligors 24 and 25 have
the largest risk contributions.
•
In example 1B, obligors 24 and 25 have been removed f rom the portf olio. The other portf olio data is
unchanged.
The risk contribution output f rom example 1A is repeated in the table below.
Expected Risk
Name Loss Contribution
1 107,543 228,711
2 326,946 764,758
3 179,971 426,743
4 289,967 716,735
5 347,599 896,874
6 361,639 910,914
7 795,655 2,163,988
8 443,653 1,199,910
9 156,899 434,047
10 160,202 437,350
11 70,916 225,356
12 241,526 756,325
13 245,605 794,754
14 1,478,697 4,773,594
15 504,231 1,602,530
16 399,027 1,330,448
17 271,773 892,720
18 165,528 560,564
19 432,345 1,477,654
20 175,435 593,559
21 1,939,960 6,850,969
22 1,944,097 7,110,748
23 123,642 487,938
24 1,541,091 9,056,197
25 1,517,917 10,618,120
The ef fect on the test portf olio of removing obligors 24 and 25 is shown in table 12 below. Removing these
obligors in example 1B has two ef fects on the portf olio:
•
The expected loss of the portf olio has been reduced by 3,059,008 f rom 14,221,863 to 11,162,856. The
amount of expected loss removed is exactly equal to the expected losses f rom the two removed obligors
because expected loss is additive across the portf olio. Thus the ACP provision in respect of the portf olio
can be reduced by 3,059,008.
•
The 99th percentile loss from the portfolio has declined by 15,364,646 from 55,311,503 to 39,946,857.
This is approximately predicted by the total risk contributions of 19,674,317 from the two obligors removed.
The risk contributions give an estimate of the effect of removing the obligors. Thus, if the 99th percentile
loss is used as the ICR Cap for the portfolio, then the ICR Cap can be reduced by 15,364,646. Furthermore,
if the same percentile is used as the benchmark confidence level for determining economic capital, then the
amount of economic capital required to support the portfolio is reduced by the same amount.
64
CREDIT FIRST
SUI SSE BOSTON
Table 10:
Example 1A Risk
Contributions
CREDI TRI S K
+
65
The tables below summarise the portf olio movement and the risk details of the removed obligors.
Expected Risk
Name Exposure Loss Contribution
24 15,410,906 1,541,091 9,056,197
25 20,238,895 1,517,917 10,618,120
Total 35,649,801 3,059,008 19,674,317
Absolute %
Example 1A Example 1B Movement Movement
Exposure 130,513,072 94,863,271 (35,649,801) 27.3%
Mean 14,221,863 11,162,856 (3,059,007) 21.5%
99th Percentile 55,311,503 39,946,857 (15,364,646) 27.8%
Although the example incorporates unrealistic levels of def ault rates, the percentage movements in Table 12
illustrate a general feature of portf olio risk management. The removal of the obligors with the largest risk
contributions f rom a portf olio has a greater impact on the portf olio risk, as measured by the 99th or percentile
loss, than on the expected loss of the portf olio. Theref ore, a signif icant reduction in the economic capital
required to support a portf olio of credit exposures can be achieved by f ocusing on the management of a small
number of obligors with large risk contributions.
Thus in this case, removal of two obligors, representing 21.5% of the expected loss of the portf olio, has
eliminated 27.8% of the total risk as measured by the 99th percentile loss.
B3.8 Setting the Percentile Levels
The model implementation provides a f acility to change the percentile loss levels calculated and output by the
model. This f acility is accessed f rom the Data Entry Screen.
B
Illustrative Example
Table 11:
Example 1B  Risk
analysis of removed
obligors
Table 12:
Example 1B  Portfolio
movement analysis
Data Entry Screen
Credit Suisse First Boston
Percentiles must be
chosen as numbers
between 0 and 99.9
OK
Cancel
➡
Percentile Levels Setting
50.0
75.0
95.0
97.5
99.0
99.5
99.75
99.9
Number of points required 8
For more inf ormation about CREDITRISK
+
, please contact any of the f ollowing:
Portfolio Management and Credit Derivatives
London John Chrystal 441718883235
john.crystal@csf b.com
Mark Venn 441718884279
mark.venn@csf b.com
New York JeanFrancois Dreyfus 12123255919
jeanf rancois.dreyf us@csf b.com
Stephen Lazarus 12123255911
stephen.lazarus@csf b.com
66
CREDIT FIRST
SUI SSE BOSTON
Conta
Appendix C  Contacts
CREDI TRI S K
+
67
Risk Management
London Mark Holmes 441718882426
mark.holmes@csf b.com
Andrew Cross 441718883839
andrew.cross@csf b.com
Tom Wilde 441718882235
tom.wilde@csf b.com
C
Contacts
cts
C
1. Carty, Lea V., Lieberman, Dana. (January 1997) Historical Default Rates of Corporate Bond Issuers,
19201996, Moody’s Investor Services, Global Credit Research.
2. Carty, Lea V., Lieberman, Dana. (November 1996) Defaulted Bank Loan Recoveries, Moody’s Investor
Services, Global Credit Research.
3. (April 1997) CreditMetrics. J.P.Morgan & Co. Incorporated.
4. Daykin, C.D., Pentikäinen, T., Pesonen, M. (1994) Practical Risk Theory for Actuaries. Chapman & Hall
5. Kealhofer, Stephen. (1995) “Managing Def ault Risk in Portf olios of Derivatives”. Derivative Credit Risk:
Advances in Measurement and Management. Renaissance Risk Publications
6. Marvin, Sally G. (December 1996) Capital Allocation: A Study of Current and Evolving Practices in
Selected Banks. Of f ice of the Comptroller of the Currency
7. Panjer, Harry H., Willmot, Gordon E. (1992) Insurance Risk Models. Society of Actuaries.
8. (February 1997) Ratings Performance 1996 Stability & Transition. Standard & Poor’s.
9. Theodore, Samuel S., Madelain, M. (March 1997) Modern CreditRisk Management And The Use of
Credit Derivatives: European Banks’ Brave New World (And Its Limits). Moody’s Investor Services,
Global Credit Research.
68
CREDIT FIRST
SUI SSE BOSTON
Select
Bibliog
Appendix D  Selected Bibliography D
t
g
C
D
S
C
D
S
C
D
S
C
D
S
C
D
CREDIT FIRST
SUI SSE BOSTON
One Cabot Square, London E14 4QJ
Regulated by SFA
CREDIT SUISSE
FIRST BOSTON
is a leading global investment banking firm, providing comprehensive financial advisory, capital raising, sales and trading, and financial
products for users and suppliers of capital around the world. It operates in over 60 offices across more than 30 countries and six continents and has over 15,000 employees.
Copyright ©1997 Credit Suisse First Boston International. All rights reserved. CREDITR ISK + is a trademark of Credit Suisse First Boston International in countries of use. CREDITR ISK + as described in this document (“CREDITR ISK +”) is a method of credit risk management introduced by Credit Suisse Group. No representation or warranty, express or implied, is made by Credit Suisse First Boston International or any other Credit Suisse Group company as to the accuracy, completeness, or fitness for any particular purpose of CREDITR ISK +. Under no circumstances shall Credit Suisse First Boston International or any other Credit Suisse Group company have any liability to any other person or any entity for (a) any loss, damage or other injury in whole or in part caused by, resulting from or relating to, any error (negligent or otherwise), of Credit Suisse First Boston International or any other Credit Suisse Group company in connection with the compilation, analysis, interpretation, communication, publication or delivery of CREDITR ISK +, or (b) any direct, indirect, special, consequential, incidental or compensatory damages whatsoever (including, without limitation, lost profits), in either case caused by reliance upon or otherwise resulting from or relating to the use of (including the inability to use) CREDITR ISK +. Issued and approved by Credit Suisse First Boston International for the purpose of Section 57, Financial Services Act 1986. Regulated by the Securities and Futures Authority. The products and services referred to are not available to private customers.
Contents
1.
Introduction to CREDITRISK+
1.1 1.2 1.3 1.4 1.5 1.6 Developments in Credit Risk Management Components of CREDITR ISK + The CREDITR ISK + Model Economic Capital Applications of CREDITR ISK + Example Spreadsheet Implementation
3 3 3 4 4 5 5 6 6 7 8 9 10 11 14 16 17 17 17 18 20 21 22 23 23 23 24 26 26 26 29 29
2.
Modelling Credit Risk
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 Risk Modelling Concepts Types of Credit Risk Default Rate Behaviour Modelling Approach Time Horizon for Credit Risk Modelling Data Inputs to Credit Risk Modelling Correlation and Incorporating the Effects of Background Factors Measuring Concentration
3.
The CREDITRISK+ Model
3.1 3.2 3.3 3.4 3.5 3.6 Stages in the Modelling Process Frequency of Default Events Moving from Default Events to Default Losses Concentration Risk and Sector Analysis MultiYear Losses for a HoldtoMaturity Time Horizon Summary of the CREDITR ISK + Model
4.
Economic Capital for Credit Risk
4.1 4.2 4.3 Introduction to Economic Capital Economic Capital for Credit Risk Scenario Analysis
5.
Applications of CREDITRISK+
5.1 5.2 5.3 5.4 Introduction Provisioning for Credit Risk RiskBased Credit Limits Portfolio Management
CREDITR ISK +
1
Contacts Selected Bibliography I t List of Tables Table 1: Table 2: Table 3: Table 4: Table 5: Table 6: Table 7: Table 8: Table 9: Table 10: Table 11: Table 12: Representations of the default rate process Oneyear default rates (%) Default rate standard deviations (%) Recovery rates by seniority and security (%) Mechanisms for controlling the risk of credit default losses Provisioning for different business lines Example of credit risk provisioning Example portfolio Example mapping table of default rate information Example 1A .Distribution of default losses Impact of sectors on the loss distribution Economic capital for credit risk 3 8 9 12 13 18 19 21 24 25 27 31 33 Figure 10: Parts of the credit default loss distribution Figure 11: Credit risk provisioning Figure 12: Using risk contributions 2 CREDIT SUISSE FIRST BOSTON Figure 13: Flowchart description of Appendix A .Risk contributions Example 1B . D.Risk analysis of removed obligors Example 1B .Appendices A. Illustrative Example B1 B2 B3 Example SpreadsheetBased Implementation Example Portfolio and Static Data Example Use of the Spreadsheet Implementation C. The CREDITRISK+ Model A1 A2 A3 A4 A5 A6 A7 A8 A9 Overview of this Appendix Default Events with Fixed Default Rates Default Losses with Fixed Default Rates Loss Distribution with Fixed Default Rates Application to MultiYear Losses Default Rate Uncertainty Sector Analysis Default Events with Variable Default Rates Default Losses with Variable Default Rates 32 32 33 35 38 39 41 41 44 46 47 49 50 52 58 58 58 60 66 68 A10 Loss Distribution with Variable Default Rates A11 Convergence of Variable Default Rate Case to Fixed Default Rate Case A12 General Sector Analysis A13 Risk Contributions and Pairwise Correlation B.Portfolio movement analysis 9 12 13 14 25 28 28 59 59 64 65 65 List of Figures Figure 1: Figure 2: Figure 3: Figure 4: Figure 5: Figure 6: Figure 7: Figure 8: Figure 9: Components of CREDITR ISK + Default rate as a continuous random variable Default rate as a discrete random variable Rated corporate defaults by number of issuers Defaulted bank loan price distribution CREDITR ISK + Model .Distribution of default events CREDITR ISK + Model .
In December 1996. at developing a more forwardlooking management tool. and several applications of the technology. in parallel. Credit Suisse Group launched.2 Components of CREDITR ISK + The components of CREDITR ISK + and the interrelationships between them are shown in the following diagram. a methodology for calculating economic capital for credit risk. a major project aimed at modernising its credit risk management and. derivatives.Introdu to CRE Introduction to CREDITRISK+ 1 1. credit risk provisioning. Credit Suisse Group introduced CREDITR ISK + . CREDITR ISK + can be applied to credit exposures arising from all types of products including corporate and retail loans. credit derivatives. Credit Suisse First Boston (“CSFB”) has been developing and deploying new risk management methods. Current areas of development in credit risk management include: modelling credit risk on a portfolio basis. and traded bonds. active portfolio management. 1. using CSFB’s expertise. In 1993. and sophisticated approaches to capital allocation that more closely reflect economic risk than the existing regulatory capital regime.a Credit Risk Management Framework. CREDITR ISK + Credit Risk Measurement Exposures Recovery Rates Default Rates Default Rate Volatilities Figure 1: Components of CREDITR ISK + Economic Capital Credit Default Loss Distribution Applications Provisioning Limits CREDITR ISK + comprises three main components –a CREDITRISK+ Model that uses a portfolio approach. CREDITR ISK + addresses all of these areas and the relationships between them.1 Developments in Credit Risk Management Since the beginning of the 1990s. CREDITR ISK + Model Scenario Analysis Portfolio Management CREDITR ISK + 3 .
the analytic C REDITR ISK + Model captures the essential characteristics of credit default events and allows explicit calculation of a full loss distribution for a portfolio of credit exposures. and (ii) a means of measuring diversification and concentration to assist in portfolio management. Measuring the uncertainty or variability of loss and the relative likelihood of the possible levels of unexpected losses in a portfolio of credit exposures is fundamental to the effective management of credit risk.4 Economic Capital The output of the CREDITR ISK + Model can be used to determine the level of economic capital required to cover the risk of unexpected credit default losses. 1. it measures economic risk on a portfolio basis and takes account of diversification and concentration. 4 CREDIT SUISSE FIRST BOSTON . from quantitative modelling to the development of practical techniques for its management. background factors. Often. This approach is similar to that taken in market risk management. even though there is no causal link between them. and. where no attempt is made to model the causes of market price movements. The CREDITRISK + Model considers default rates as continuous random variables and incorporates the volatility of default rates in order to capture the uncertainty in the level of default rates. it can be used for portfolio management. such as the state of the economy. Applications of the credit risk modelling methodology including: (i) a methodology for establishing provisions on an anticipatory basis. The effects of these background factors are incorporated into the CREDITR ISK + Model through the use of default rate volatilities and sector analysis rather than using default correlations as explicit inputs into the model. In financial modelling one is usually concerned with modelling continuous price changes rather than sudden events. 1. such as individual obligor (borrower. The CREDITR ISK + Model is a statistical model of credit default risk that makes no assumptions about the causes of default. Economic capital provides a measure of the risk being taken by a firm and has several benefits: it is a more appropriate risk measure than that specified under the current regulatory regime.A modern approach to credit risk management should address all aspects of credit risk. A methodology for calculating economic capital for credit risk.3 The CREDITR ISK + Model CREDITR ISK + is based on a portfolio approach to modelling credit default risk that takes into account information relating to size and maturity of an exposure and the credit quality and systematic risk of an obligor. counterparty or issuer) limits and concentration limits. may cause the incidence of defaults to be correlated. This approach contrasts with the mathematical techniques typically used in finance. Mathematical techniques applied widely in the insurance industry are used to model the sudden event of an obligor default. Applying insurance modelling techniques. CREDITRISK + reflects the requirements of a modern approach to managing credit risk and comprises three main components: • • • The CREDITR ISK + Model that uses a portfolio approach and analytical techniques applied widely in the insurance industry. since economic capital reflects the changing risk of a portfolio. In addition to wellestablished credit risk management techniques.
1 1. CREDITR ISK + 5 .csfb.6 Example Spreadsheet Implementation In order to assist the reader of this document.com). 1.5 Applications of CREDITR ISK + CREDITRISK + includes several applications of the credit risk modelling methodology. a spreadsheetbased implementation that illustrates the range of possible outputs of the CREDITR ISK + Model can be downloaded from the Internet (http://www. including a forwardlooking provisioning methodology and quantitative portfolio management techniques.Introduction The CREDITR ISK+ Model is supplemented by scenario analysis in order to identify the financial impact of low probability but nevertheless plausible events that may not be captured by a statistically based model.
there are three types of uncertainty that must be assessed: process risk.alternative models could produce different results. Indeed. 6 CREDIT SUISSE FIRST BOSTON .2 Addressing Modelling Issues As all of these types of uncertainty enter into the modelling process.1 Types of Uncertainty Arising in the Modelling Process A statistically based model can describe many business processes.1. Process risk is usually addressed by expressing the model results to an appropriately high level of confidence.1.1 Risk Modelling Concepts 2. The only information that can be obtained about the underlying process is obtained by observing the results that it has generated in the past. Model error is usually the least tractable of the three types of uncertainty. any model is only a representation of the real world. Model Error Model error arises because the proposed model does not correctly reflect the actual process . parameter uncertainty and model error. However. it is important to be aware of them and to consider how they can be addressed when developing a credit risk model. 2. In the modelling process. a realistic assessment of the potential effects of these errors should be made before any decisions are made based on the outputs of the model. Process Risk Process risk arises because the actual observed results are subject to random fluctuations even where the model describing the loss process and the parameters used by the model are appropriate. It is possible to assess the impact of parameter uncertainty by performing sensitivity analysis on the parameter inputs. Parameter Uncertainty Parameter uncertainty arises from the difficulties in obtaining estimates of the parameters used in the model.Model Credit Modelling Credit Risk 2 2.
a standard variancecovariance approach to measuring credit spread risk may be inappropriate. • No assumptions are made about the causes of default.Modelling Credit Risk CREDITR ISK+ addresses these types of uncertainty in several ways: 2 The CREDITR ISK+ Model makes no assumptions about the causes of default. where no assumptions are made about the causes of market price movements. However. This approach is similar to that taken in market risk management. All portfolios of exposures exhibit credit default risk. increasing default rates or default rate volatilities can be used to simulate downturns in the economy. the historical simulation approach. in which the effects of stress testing each of the input parameters are quantified. Credit spread risk fits more naturally within a market risk management framework.2 Types of Credit Risk There are two main types of credit risk: • • Credit spread risk: Credit spread risk is exhibited by portfolios for which the credit spread is traded and markedtomarket. This not only reduces the potential model error but also leads to the development of an analytically tractable model. some types of products. such as for portfolios of bonds and credit derivatives. Even then. 2. • Concerns about parameter uncertainty are addressed using scenario analysis. Since the distribution of credit spreads may not be normal. provides a suitable alternative. In practice. which does not make any assumptions about the underlying distribution.2. as the default of an obligor results in a loss. as the default of an obligor results in a loss. Changes in observed credit spreads impact the value of these portfolios. 2. used in combination with other techniques. such as corporate or retail loans. Credit spread risk is only exhibited when a marktomarket accounting policy is applied. a firm generally incurs a loss equal to the amount owed by the obligor less a recovery amount which the firm recovers as a result of foreclosure. Credit default risk: All portfolios of exposures exhibit credit default risk. A marktomarket accounting policy would have to be applied to these products in order to recognise the credit spread risk. are typically accounted for on an accruals basis. All portfolios of exposures exhibit credit default risk. as the default of an obligor results in a loss. • The data requirements for the CREDITR ISK + Model have been kept as low as possible. Credit spread risk is the risk of financial loss owing to changes in the level of credit spreads used in the marktomarket of a product. a firm’s valueatrisk model should take account of value changes caused by the volatility of credit spreads. liquidation or restructuring of the defaulted obligor. In the credit environment. where no assumptions are made about the causes of market price movements.1 Credit Spread Risk Credit spread is the excess return demanded by the market for assuming a certain credit exposure.2 Credit Default Risk Credit default risk is the risk that an obligor is unable to meet its financial obligations. In the event of a default of an obligor. the data can be subject to large fluctuations year on year. In order to manage credit spread risk. 2. empirical data is sparse and difficult to obtain.2. which minimises the error from parameter uncertainty. For example. This approach is similar to that taken in market risk management. CREDITR ISK + 7 .
the default rate of a particular obligor.Credit default risk is typically associated with exposures that are more likely to be held to maturity. The process for the default rate can be represented in two different ways: • Continuous variable: When treated as a continuous variable.a forward stock price and the stock price volatility are used to define the forward stock price distribution. Therefore. Using this approach. and hence the same value for the default rate. This can be achieved via a rating transition matrix that specifies the probability of keeping the same credit rating. where the intention is to maintain a bond portfolio over a longer time frame. CREDITR ISK + focuses on modelling and managing credit default risk. and the probabilities of moving to different credit ratings and hence to different values for the default rate. it is equally important to measure the default risk that is taken by holding the portfolio. 2. Hence. inferred from market prices. In modelling credit risk. Figure 2: Default rate as a continuous random Default rate variable Frequency of default rate outcomes Possible path of default rate Time horizon • Discrete variable: By treating the default rate as a discrete variable. The following figure illustrates the path that a default rate may take over time and the distribution that it could have over that time. they incorporate both the credit quality and the potential credit quality changes of that obligor. will vary on a continuous scale and hence can be viewed as a continuous random variable. 8 CREDIT SUISSE FIRST BOSTON . such as corporate and retail loans and exposures arising from derivative portfolios. The data requirements for modelling credit default risk are analogous to the data requirements for pricing stock options . one is concerned with determining the possible future outcomes over the chosen time horizon. additional information is required in order to model the possible future outcomes of the default rate. Bond markets are generally more liquid than loan markets and therefore bond positions can be adjusted over a shorter time frame. even though the individual constituents of the portfolio may change. the possible default rate over a given time horizon is described by a distribution. This is illustrated in the following figure.3 Default Rate Behaviour Equity and bond prices are forwardlooking in nature and are formed by investors’ views of the financial prospects of a particular obligor. which can be specified by a default rate and a volatility of the default rate. However. A convenient way of making default rates discrete is by assigning credit ratings to obligors and mapping default rates to credit ratings. a simplification of the continuous process described above is made.
The above two representations of default rate behaviour are summarised in the following table: Table 1: Treatment of default rate Continuous variable Data requirements • Default rates • Volatility of default rates Representations of the default rate process Discrete variable • Credit ratings • Rating transition matrix The CREDITR ISK+ Model treats default rates as The CREDITR ISK + Model is a statistical model of credit default risk that models default rates as continuous random variables and incorporates the volatility of the default rate in order to capture the uncertainty in the level of the default rate. continuous random variables and incorporates default rate volatility to capture the uncertainty in the level of the default rate. The risk manager needs to answer questions such as “What is the size of loss for a given confidence level?”.Modelling Credit Risk 2 Figure 3: Default Possible path of default rate Frequency of default rate outcomes Default rate as a discrete random variable B Default rate BB BBB A AA AAA Time horizon The discrete approach with rating migrations and the continuous approach with a default rate volatility are different representations of the behaviour of default rates. A mapping from credit ratings to a set of default rates provides a convenient approach to setting the level of the default rate. CREDITR ISK + 9 . Both approaches achieve the desired end result of producing a distribution for the default rate. 2. including the following: • Distribution of loss: The risk manager is interested in obtaining distributions of loss that may arise from the current portfolio. These outcomes are usually difficult to model statistically but can be addressed through the use of scenario analysis and concentration limits. • Identifying extreme outcomes: The risk manager is also concerned with identifying extreme or catastrophic outcomes. there are several measures of risk that are of interest.4 Modelling Approach 2.4.1 Risk Measures When managing credit risk.
In both cases. losses can be suffered from a portfolio containing a large number of individual risks.2 A Portfolio Approach to Managing Credit Risk Credit risk can be managed through diversification because the number of individual risks in a portfolio of exposures is usually large. and (iv) the systematic or concentration risk of the obligor. the time horizon chosen should not be shorter than the time frame over which riskmitigating actions can be taken. The analytic CREDITR ISK + Model allows rapid and explicit calculation of a full loss distribution for a portfolio of credit exposures. Generally. when applied in a wellstructured and consistent manner. This has the additional benefit that it leads to a credit risk model that is analytically tractable and hence not subject to the problems of precision that can arise when using a simulationbased approach.4.5 Time Horizon for Credit Risk Modelling A key decision that has to be made when modelling credit risk is the choice of time horizon. Credit limits aim to control risk arising from each of these factors individually. In modelling credit default losses one is concerned with sudden events rather than continuous changes. such as one year. However. 2. including individual obligor limits to control the size of exposure. credit quality and systematic risk into a single measure.4. The general effect of this approach. The essential characteristics of credit default events are captured by applying these insurance modelling techniques. The portfolio risk of a particular exposure is determined by four factors: (i) the size of the exposure. maturity.3 Modelling Techniques Used in the CREDITR ISK + Model The economic risk of a portfolio of credit exposures is analogous to the economic risk of a portfolio of insurance exposures.2. 2. CREDITR ISK+ is based on a portfolio approach summarising information about size. CREDITR ISK + does not prescribe any one particular time horizon but suggests two possible time horizons that can provide management information relevant for credit risk management: • • A constant time horizon. Mathematical techniques applied widely in the insurance industry are used to model the sudden event of an obligor default. (ii) the maturity of the exposure. these limits do not provide a measure of the diversification and concentration of a portfolio. In order to keep model error to a minimum. no assumptions are made about the causes of default. 10 CREDIT SUISSE FIRST BOSTON . rating exposure limits to control the amount of exposure to obligors of certain credit ratings. (iii) the probability of default of the obligor. each with a low probability of occurring. and concentration limits to control concentrations within countries and industry sectors. maturity. tenor limits to control the maximum maturity of exposures to obligors. A holdtomaturity or runoff time horizon. a means of measuring diversification and concentration has to be developed. the primary technique for controlling credit risk is the use of limit systems. is to create reasonably welldiversified portfolios. An approach that incorporates size. CREDITR ISK + takes such an approach. In order to manage effectively a portfolio of exposures. Currently. credit quality and systematic risk into a single portfolio measure is required. The risk manager is concerned with assessing the frequency of the unexpected events as well as the severity of the losses.
2. loans. commitments. as it allows all exposures to be considered at the same future date.g.5. it is important that the changing exposures over time are accurately captured. The role that the CREDITR ISK + Model plays in active portfolio management is discussed later in this document. 2. a holdtomaturity time horizon allows the full term structure of default rates over the lifetime of the exposures to be recognised. Credit Risk Measurement Exposures Recovery Rates Default Rates Default Rate Volatilities CREDITR ISK+ Model CREDITR ISK + 11 . The quality of this data will directly affect the accuracy of the measurement of credit risk and therefore any decision to be made using the results should be made only having fully assessed the potential error from uncertainties in the data used. 2 2.6 Data Inputs to Credit Risk Modelling 2. financial letters of credit and derivative exposures.1 Constant Time Horizon A constant time horizon is relevant. and. one year matches the normal accounting period. this can be used as the input into the model. in addition to the constant time horizon.Modelling Credit Risk 2.6. traded bond portfolios).5. This view of the portfolio enables the risk manager to compare exposures of different maturity and credit quality and is an appropriate tool. one year is often taken as a suitable time horizon: credit risk mitigating actions can normally be executed within one year. CREDITR ISK + suggests a time horizon of one year for credit risk economic capital. Alternatively.1 Data Inputs Any modelling of credit risk is dependent on certain data requirements being met. conservative assumptions can be used while default data quality is being improved. The CREDITR ISK + Model is capable of handling all types of instruments that give rise to credit exposure. In addition.6.2 HoldtoMaturity Time Horizon Alternatively. new capital can be raised to replenish capital eroded by actual credit losses during the period. furthermore. For some of these transaction types. including bonds. Where a firm has its own information that is judged to be relevant to its portfolio. A benchmark time horizon of one year can be used for portfolios where there is an intention to maintain exposures for longer than the term of the booked transactions (e. The CREDITR ISK + Model presented in this document does not prescribe the use of any one particular data set over another. it is necessary to make an assumption about the level of exposure in the event of a default: for example. a financial letter of credit will usually be drawn down prior to default and therefore the exposure at risk should be assumed to be the full nominal amount.2 Credit Exposures The exposures arising from separate transactions with an obligor should be aggregated according to the legal corporate structure and taking into account any rights of setoff. for portfolio management. if a multiyear time horizon is being used. Given these factors. For various reasons. One of the key limitations in modelling credit risk is the lack of comprehensive default data. The inputs used by the CREDITR ISK + Model are: • • • • Credit exposures Obligor default rates Obligor default rate volatilities and Recovery rates.
03 0. as can be seen in the following figure. An assessment of the nature of a particular obligation. 1997. Alternatively. obligor credit ratings. It should be noted that oneyear default rates show significant variation year on year. should be performed when considering the recovery rate for an obligor. This can be achieved in a number of ways. provide a convenient way of assigning probabilities of default to obligors.00 0. which can be used as a substitute for the combination of credit ratings and assigned default rates. its creditworthiness).36 7.01 0. together with a mapping of default rates to credit ratings.27 Source: Carty & Lieberman. 30 20 10 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 0 Source: Standard & Poor’s Ratings Performance 1996 (February 1997) • Another approach is to calculate default probabilities on a continuous scale. The rating agencies publish historic default statistics by rating category for the population of obligors that they have rated. Moody’s Investors Service Global Credit Research A credit rating is an opinion of an obligor’s overall financial capacity to meet its financial obligations (i.e. including: CREDITR ISK+ Model • • Observed credit spreads from traded instruments can be used to provide marketassessed probabilities of default.12 1. should be assigned to each obligor. Figure 4: Rated corporate defaults by number of issuers Frequency 70 60 50 40 Oneyear default rates show significant fluctuations from year to year. the number of defaults can be many times the level observed at other times.6. This opinion focuses on the obligor’s capacity and willingness to meet its financial commitments as they fall due. which represents the likelihood of a default event occurring. During periods of economic recession.3 Default Rates Credit Risk Measurement Exposures Recovery Rates Default Rates Default Rate Volatilities A default rate. 12 CREDIT SUISSE FIRST BOSTON . Table 2: Oneyear default rates (%) Credit rating Aaa Aa A Baa Ba B Oneyear default rate 0. including its seniority in bankruptcy or liquidation.2.
Moody’s Investors Service Global Credit Research $91$100 0 CREDITR ISK + 13 . actual observed default rates vary from these averages. As can be seen in the following table.3 1.03 0. the standard deviation of default rates can be significant compared to actual default rates. 14 Credit Risk Measurement Exposures Recovery Rates Default Rates Default Rate Volatilities CREDITR ISK+ Model Figure 5: Defaulted bank loan price distribution 12 10 Frequency 8 6 4 2 $21$30 $31$40 $41$50 $51$60 $61$70 $0$10 $71$80 $81$90 $11$20 Source: Defaulted Bank Loan Recoveries (November 1996) .12 1.00 0. For example.0 0. reflecting the high fluctuations observed during economic cycles.6. Moody’s Investors Service Global Credit Research The default rate standard deviations in the above table were calculated over the period from 1970 to 1996 and therefore include the effect of economic cycles. As shown previously. Recovery rates are subject to significant variation. the figure below shows the price distribution of defaulted bank loans and illustrates that there is a large degree of dispersion. 2.Modelling Credit Risk 2.0 0.1 Default rate standard deviations (%) Source: Carty & Lieberman. Recovery rates should take account of the seniority of the obligation and any collateral or security held. 2 Credit Risk Measurement Exposures Recovery Rates Default Rates Default Rate Volatilities CREDITR ISK+ Model Table 3: Oneyear default rate (%) Credit rating Aaa Aa A Baa Ba B Average 0.1 0.6. 1996.36 7. a firm generally incurs a loss equal to the amount owed by the obligor less a recovery amount.27 Standard deviation 0.5 Recovery Rates In the event of a default of an obligor.3 5.4 Default Rate Volatilities Published default statistics include average default rates over many years. which the firm recovers as a result of foreclosure.01 0. liquidation or restructuring of the defaulted obligor or the sale of the claim. The amount of variation in default rates about these averages can be described by the volatility (standard deviation) of default rates.
the correlation effect observed is due to a background factor. Often. 19201996 (January 1997) Moody’s Investors Service Global Credit Research Publicly available recovery rate data indicates that there can be significant variation in the level of loss. which has increased the rates of default above their average level. different industry sectors will be affected to different degrees by the state of the economy.18 63. but macroeconomic trends are certainly the most influential factors”. Compared to 1995. which changes the rates of default. February 1997 Moody’s Investors Service. the state of the economy. For example. 2 1 1 Standard and Poor’s Ratings Performance 1996.29 24. given the default of an obligor. defaults were reduced by onehalf…. Therefore.7.2 Impact of the Economy on Default Rates There is general agreement that the state of the economy in a country has a direct impact on observed default rates.45 47. 2. there are background factors that may cause the incidence of default events to be correlated. as can be seen from the standard deviation of the corporate bond and bank loan recovery rates in the table below.09 8. even though there is no causal link between them. if there are fewer defaults than on average in one month. 2. this might be due to the economy being in recession. Given this uncertainty. The defaults are correlated but there is no causal link between them . January 1996 As the above quotations indicate and as can be seen in Figure 4 above. Furthermore. if there is an unusually large number of defaults in one particular month. Conversely.74 20. 2 14 CREDIT SUISSE FIRST BOSTON . Credit defaults occur as a sequence of events in such a way that it is not possible to forecast the exact time of occurrence of any one default or the exact total number of defaults.66 Standard deviation 21.21 26.1 The Random Nature of Defaults and the Appearance of Correlation Often. there are background factors that may cause the incidence of defaults to be correlated. The magnitude of the impact will be dependent on how sensitive an obligor’s earnings are to various economic factors. a careful assessment of recovery rate assumptions is required. 2. even though there is no causal link between them.28 28.7.54 38. such as the growth rate of the economy and the level of interest rates.” Another report by Moody’s Investors Service stated that “The sources of [default rate volatility] are many. because the economy is growing.29 14. Table 4: Recovery rates by seniority and security (%) Seniority and security Senior secured bank loans Senior secured public debt Senior unsecured public debt Senior subordinated public debt Subordinated public debt Junior subordinated public debt Average 71. it is also likely that there will be fewer defaults than on average in the following month.7 Correlation and Incorporating the Effects of Background Factors Default correlation impacts the variability of default losses from a portfolio of credit exposures. for each year. stress testing should be performed on the recovery rates in order to calculate the potential loss distributions under different scenarios.There is also considerable variation for obligations of differing seniority. it is quite likely that the number of defaults in the following month will also be high. Corporate Bond Defaults and Default Rates.67 Source: Historical Default Rates of Corporate Bond Issuers. The CREDITRISK + Model incorporates the effects of default correlations by using default rate volatilities and sector analysis. there is significant variation in the number of defaults from year to year. A recent report by Standard and Poor’s stated that “A healthy economy in 1996 contributed to a significant decline in the total number of corporate defaults. In this economic situation.09 26.
There are various reasons why this approach has been taken. give rise to loss distributions with fat tails. This is accomplished by incorporating default rate volatilities into the model. Both approaches. there is no obvious way of deriving default correlations. that may be inferred or observed during a period of normal trading. 2.8 and 3. it is much easier to perform scenario analysis on default rate volatilities. 3 Sector analysis is discussed in Sections 2.Modelling Credit Risk Economic models that attempt to capture the effect of changes in the economy on default rates can be developed in order to specify the default rates for subsequent use in a credit risk model. it is questionable how stable any relationship. This technique relies upon additional assumptions about the relationship between asset prices and probabilities of default. for example in a retail portfolio. Furthermore. • Lack of empirical data: There is little empirical data on default correlations.3 Incorporating the Effects of Background Factors It is possible to incorporate the effects of background factors into the specification of default rates by allowing the default rate itself to have a probability distribution. Therefore. • Instability of default correlations: Generally. the use of default rate volatilities and default correlations. Since default correlations are difficult to calculate directly.7. The CREDITRISK + Model does not attempt to model correlations explicitly but captures the same concentration effects through the use of default rate volatilities and sector analysis . where there is no asset price for the obligor. it is questionable whether such relationships would be stable over several years.4 CREDITR ISK + 15 . some approaches use asset price correlations to derive default correlations. Even if a causal relationship could be established relating default rates to certain economic variables. this approach can have several weaknesses. alternative approaches that attempt to capture the observed variability of default rates have to be sought. Defaults themselves are infrequent events and so there is insufficient data on multiple defaults with which to calculate explicit default correlations. A similar instability problem may arise with default rate volatilities: however. a calculated correlation can be very dependent on the underlying time period of the data. The CREDITRISK + Model models the effects of background factors by using default rate volatilities that result in increased defaults rather than by using default correlations as a direct input. including the following: 2 • • Since there are limited publicly available default rate statistics by country or by industry sector. owing to the analytically tractable nature of a model that uses volatilities rather than correlations. Section 3 of this document describes in detail how the CREDITRISK + Model uses default rate volatilities in the modelling of credit default risk. but this can only be considered a proxy. would be in the event of default of a particular obligor. it is difficult to verify the accuracy of an economic model used to derive default rates. including the following: 3 The CREDITR ISK+ Model captures concentration risk through the use of default rate volatilities and sector analysis. In addition. In addition. correlations calculated from financial data show a high degree of instability. However.
The state of the economy of each different country will vary over time and. A portfolio of exposures can have concentrations in particular countries or industry sectors. An exposure can be broken down into an obligorspecific element. within each country. different industry sectors will be affected to differing degrees. C M 16 CREDIT SUISSE FIRST BOSTON . which is independent of all other exposures. The CREDITRISK + Model described in this document allows concentration risk to be captured using sector analysis.2. and nonspecific or systematic elements that are sensitive to particular driving factors. Therefore. such as countries or industry sectors. it is important to be able to capture the effect of concentration risk in a credit risk model.8 Measuring Concentration The above discussion has highlighted the fact that there are background factors that affect the level of default rates.
This situation is well represented by the Poisson distribution. CREDITR ISK + 17 . as shown in the following diagram: Credit Risk Measurement Default Rates Default Rate Volatilities Stage 1 What is the FREQUENCY of defaults? What is the SEVERITY of the losses ? CREDITR ISK+ Model Stage 2 Distribution of default losses By calculating the distribution of default events. 3.CREDIT Model The CREDITRISK+ Model 3 Exposures Recovery Rates 3. There is exposure to default losses from a large number of obligors and the probability of default by any particular obligor is small.1 Stages in the Modelling Process The modelling of credit risk is a two stage process.2 Frequency of Default Events 3.1 The Default Process The CREDITR ISK + Model makes no assumption about the causes of default .credit defaults occur as a sequence of events in such a way that it is neither possible to forecast the exact time of occurrence of any one default nor the exact total number of defaults. The distribution of losses allows the risk manager to assess the financial impact of the potential losses as well as measuring the amount of diversification and concentration within the portfolio.2. the risk manager is able to assess whether the overall credit quality of the portfolio is either improving or deteriorating.
exhibit a high degree of variation. 3. Hence.3 Moving from Default Events to Default Losses 3.We consider first the distribution of the number of default events in a time period. such as one year. This is regardless of the individual default rate for a particular obligor. within a portfolio of obligors having a range of different annual probabilities of default. as we have seen in the previous section. Moreover. without introducing significant model error. The annual probability of default of each obligor can be conveniently determined by its credit rating and a mapping between default rates and credit ratings. it is possible to describe the overall distribution of losses because its probability generating function has a simple closed form amenable to computation. If we do not incorporate the volatility of the default rate. default rate variability needs to be incorporated into the model. The effect of using default rate volatilities can be clearly seen in the following figure. Although the expected number of default events is the same.3. The distribution of losses differs from the distribution of default events because the amount lost in a given default depends on the exposure to the individual obligors. However. Figure 6: CREDITR ISK+ Model Distribution of default events Probability Including default rate volatility Excluding default rate volatility Number of defaults 3. 18 CREDIT SUISSE FIRST BOSTON .2 Distribution of the Number of Default Events The CREDITR ISK + Model models the underlying default rates by specifying a default rate and a default rate volatility. the variation in exposure magnitude results in a loss distribution that is not Poisson in general. which does not influence the distribution of the total number of defaults. the distribution becomes significantly skewed to the right when default rate volatility is increased. information about the distribution of different exposures is essential to the overall distribution. we now consider the distribution of losses in the portfolio. This aims to take account of the variation in default rates in a pragmatic manner.1 Distribution of Default Losses Given the number of default events. default rates are not constant over time and. the distribution of the number of default events will be closely approximated by the Poisson distribution. which shows the distribution of the number of default events generated by the CREDITR ISK + Model when default rate volatility is varied.2. However. Unlike the variation of default probability between obligors. This represents a significantly increased risk of an extreme number of default events.
This increase in the variance is due to the pairwise default correlations between obligors. as shown in the figure below. liquidation or restructuring of the defaulted obligor. we give an explicit formula for the pairwise default correlations implied by the CREDITR ISK + Model when default rate volatilities are incorporated into the model. Fatter tail: The key change is the level of losses at the higher percentiles. Recovery rates should take account of the seniority of the obligation and any collateral or security held. net of the above recovery adjustment. two steps are first followed: 3 • • The exposures are adjusted by anticipated recovery rates in order to calculate the loss in a given default. There is now considerably more chance of experiencing extreme losses. In order to reduce the amount of data to be processed. for example. It should be noted that when the default rate volatilities are set to zero. which the firm obtains as a result of foreclosure.2 Impact of Incorporating Default Rate Volatilities Figure 7 compares the default loss distributions calculated without default rate volatility and with default rate volatility. the 99th percentile is significantly higher when the impact of the variability of default rates is modelled. This allows a full loss distribution to be generated. These pairwise default correlations are incorporated into the CREDITR ISK + Model through the default rate volatilities and sector analysis. CREDITR ISK + 19 . the default events are independent and hence the pairwise default correlations are also zero. Figure 7: CREDITR ISK+ Model Distribution of default losses Probability Excluding default rate volatility Including default rate volatility The CREDITR ISK+ Model allows explicit calculation of the loss distribution of a portfolio of credit exposures. while the expected loss has remained unchanged. Since the tail of the distribution has become fatter. a firm generally incurs a loss equal to the amount owed by the obligor less a recovery amount. A recovery rate is used to quantify the amount received from this process. Size of loss 3. The exposures. are divided into bands of exposure with the level of exposure in each band being approximated by a common average.CREDITRISK + Model In the event of a default of an obligor. it can be concluded that the variance of the default loss distribution has increased. The key features and differences are: • • Same expected loss: Both default loss distributions have the same level of expected losses. In Appendix A. The CREDITR ISK + Model calculates the probability that a loss of a certain multiple of the chosen unit of exposure will occur.3.
In this case. while diversification largely eliminates the impact of the specific factors. a sector can be thought of as a collection of obligors having the common property that they are influenced by the same single systematic factor. For example. for example all those obligors having their domicile in a particular country. 3. generates a prudent estimate of extreme losses.4 Sector Analysis .Allocating Obligors to one of Several Sectors In order to recognise some of the diversification benefit of obligors whose fortunes are affected by a number of independent systematic factors. Concentration risk results from having in the portfolio a number of obligors whose fortunes are affected by a common factor. the most straightforward application of the CREDITR ISK+ Model.Allocating all Obligors to a Single Sector The most straightforward application of the CREDITR ISK + Model is to allocate all obligors to a single sector.3 Sector Analysis . Systematic factors Systematic factors are background factors that affect the fortunes of a proportion of the obligors in the portfolio.4. In order to quantify concentration risk.2 Sector Analysis . 20 CREDIT SUISSE FIRST BOSTON . Furthermore. even in a portfolio containing a large number of exposures. it can be assumed that each obligor is subject to only one systematic factor. Once allocated to a sector. which is responsible for all of the uncertainty of the obligor’s default rate.4. obligors could be allocated to sectors according to their country of domicile.4. The technique for measuring concentration risk is sector analysis.Apportioning Obligors across Several Sectors A more generalised approach is to assume that the fortunes of an obligor are affected by a number of systematic factors. This approach assumes that a single systematic factor affects the individual default rate volatility of each obligor. this use of the model captures all of the concentration risk within the portfolio and excludes the diversification benefit of the fortunes of individual obligors being subject to a number of independent systematic factors.4 Concentration Risk and Sector Analysis The CREDITR ISK + Model measures the benefit of portfolio diversification and the impact of concentrations through the use of sector analysis. in which all obligors are allocated to a single sector. 3. Systematic factors impact the risk of extreme losses from a portfolio of credit exposures. Specific factors In general. The technique for measuring concentration risk is sector analysis. there may be an opposing effect owing to concentration risk. the concepts of systematic factors and specific factors are necessary. the fortunes of an obligor are affected to some extent by specific factors unique to the obligor.3. Diversification arises naturally because the number of individual risks in a portfolio of exposures is usually large. Therefore. the obligor’s default rate and default rate volatility are set individually. The fortunes of any one obligor can be affected by a number of systematic factors.1 Concentration Risk Concentration risk is dependent on the systematic factors affecting the portfolio. The CREDITR ISK + Model handles this situation by apportioning an obligor across several sectors rather than allocating an obligor to a single sector.4. However. Concentration risk is dependent on the systematic factors affecting the portfolio. 3. 3.
CREDITRISK + Model
So far it has been assumed that all risk in the portfolio is systematic and allocable to one of the systematic factors. In addition to the effects of systematic factors, it is likely that the fortunes of an obligor are affected by factors specific to the obligor. Potentially specific risk requires an additional sector to model each obligor, since the factor driving specific risk for a given obligor affects that obligor only. However, the CREDITR ISK + Model handles specific risk without recourse to a large number of sectors by apportioning all obligors’ specific risk to a single “Specific Risk Sector”. 3.4.5 The Impact of Sectors on the Loss Distribution As stated above, the CREDITR ISK + Model allows the portfolio of exposures to be allocated to sectors to reflect the degree of diversification and concentration present. The most diversified portfolio is obtained when each exposure is in its own sector and the most concentrated is obtained when the portfolio consists of a single sector. The figure below shows the impact of sectors on the loss distribution. As the number of sectors is increased, the impact of concentration risk is reduced. The graph illustrates this by plotting the ratio of the 99th percentile of the credit default loss distribution for a given number of sectors to the 99th percentile of the credit default loss distribution when the portfolio is considered to be a single sector.
1.00 0.95 0.90 0.85
3
The CREDITR ISK + Model allows the portfolio of exposures to be decomposed into sectors to reflect the degree of diversification and concentration present.
Figure 8: Impact of sectors on the loss distribution
99th percentile ratio
0.80 0.75 0.70 0.65 0.60 0.55 0.50 0 1 2 3 4 5 6 7 8 9
Number of sectors
3.5 MultiYear Losses for a HoldtoMaturity Time Horizon
As discussed in Section 2.5, the CREDITR ISK + Model allows risk of the portfolio to be viewed on a holdtomaturity time horizon in order to capture any default losses that could occur until maturity of the credit exposure. Analysing credit exposures on a multiyear basis enables the risk manager to compare exposures of different size, credit quality, and maturity. The loss distribution produced provides, for any chosen level of confidence, an indication of the possible cumulative losses that could be suffered until all the exposures have matured. The benefits of looking at portfolio credit risk from this viewpoint include the following:
The CREDITR ISK + Model allows the risk of the portfolio to be viewed on a holdtomaturity time horizon in order to capture any default losses that could occur until maturity of the credit exposure.
• •
The full term structure of default probabilities is taken into account. The full uncertainty of default losses over the life of the portfolio is also captured.
For example, because the oneyear average default rates for investment grade obligors are relatively small but the corresponding exposures may be large, a oneyear time horizon may not be the best measure for active portfolio management. However, a multiyear view will reflect the fact that defaults follow a decline in credit quality over many years.
CREDITR ISK +
21
3.5.1 Using the CREDITR ISK + Model to Calculate MultiYear Loss Distributions The CREDITR ISK + Model can be used to calculate multiyear loss distributions by decomposing the exposure profile over time into separate elements of discrete time, with the present value of the remaining exposure in each time period being assigned a marginal default probability relevant to the maturity and credit quality. These decomposed exposure elements can then be used by the CREDITR ISK + Model to generate a loss distribution on a holdtomaturity basis.
3.6 Summary of the CREDITR ISK + Model
The key features of the CREDITR ISK + Model are:
•
The CREDIT R ISK + Model captures the essential characteristics of credit default events. Credit default events are rare and occur in a random manner with observed default rates varying significantly from year to year. The approach adopted reflects these characteristics by making no assumptions about the timing or causes of default events and by incorporating the default rate volatility. By taking a portfolio approach, the benefits of diversification that arise from a large number of individual risks are fully captured. Concentration risk, resulting from groups of obligors that are affected by common factors, is measured using sector analysis.
•
The CREDIT R ISK + Model is scaleable and computationally efficient. The CREDITR ISK + Model is highly scaleable and hence is capable of handling portfolios containing large numbers of exposures. The low data requirements and minimum of assumptions make the CREDITR ISK + Model easy to implement for a wide range of credit risk portfolios, regardless of the specific nature of the obligors. Furthermore, the efficiency of the model allows comprehensive sensitivity analysis to be performed on a continuing basis, which is a key requirement for the ability to quantify the effects of parameter uncertainty.
E C
22
CREDIT SUISSE
FIRST BOSTON
Econo Capita
Economic Capital for Credit Risk
4
4.1 Introduction to Economic Capital
4.1.1 The Role of Economic Capital The analysis of uncertainty is the essence of risk management. Therefore, measuring the uncertainty or variability of loss and the related likelihood of the possible levels of unexpected losses in a portfolio of exposures is fundamental to the effective management of credit risk. Sufficient earnings should be generated through adequate pricing and provisioning to absorb any expected loss. The expected loss is one of the costs of transacting business which gives rise to credit risk. However, economic capital is required as a cushion for a firm’s risk of unexpected credit default losses, because the actual level of credit losses suffered in any one period could be significantly higher than the expected level.
4.2 Economic Capital for Credit Risk
4.2.1 Credit Default Loss Distribution Knowledge of the credit default loss distribution arising from a portfolio of exposures provides a firm with management information on the amount of capital that the firm is putting at risk by holding the credit portfolio. Given that economic capital is necessary as a cushion for a firm’s risk of unexpected credit default losses, a percentile level provides a means of determining the level of economic capital for a required level of confidence. In order to capture a significant proportion of the tail of the credit default loss distribution, the 99th percentile unexpected loss level over a oneyear time horizon is a suitable definition for credit risk economic capital. This can be seen in the following figure.
Scenario Analysis
Economic Capital
Credit Default Loss Distribution
CREDITR ISK +
23
the use of a credit risk model should be supplemented by a programme of stress testing of the assumptions used.2. It measures economic risk on a portfolio basis and hence takes account of the benefits of diversification. For example.3. Similarly. Given the efficient manner in which the default loss distribution can be calculated. 4. Scenario Analysis There are two types of stress tests that should be performed: (i) scenario analysis within the CREDITR ISK + Model. Therefore. It is a measure that objectively differentiates between portfolios by taking account of credit quality and size of exposure.2 Benefits and Features of Economic Capital Economic capital as a measure of risk being taken by a firm has several features and benefits including the following: • • • • It is a more appropriate measure of the economic risk than that specified under the current regulatory regime. this can be extended to include the effects of movements in market rates on credit exposures. it is possible to calculate the impact of changing parameter inputs used by the model across a wide range of values.3. the financial impact of rating downgrades can be assessed by increasing the default rate assigned to an obligor. For a derivatives portfolio.3 Scenario Analysis 4.sectors of the portfolio can be stressed to varying degrees reflecting the fact that each sector could be affected to a different extent. which reflects the changing risk of a portfolio and hence can be used as a tool for portfolio optimisation.1 The Role of Scenario Analysis Economic Capital Credit Default Loss Distribution The purpose of scenario analysis is to identify the financial impact of low probability but nevertheless plausible events that may not be captured by a statistically based model.Figure 9: Economic capital for credit risk 99th Percentile Loss Level Economic Capital Probability Expected Loss Loss 4. 24 CREDIT SUISSE FIRST BOSTON . 4. it is possible to simulate downturns in the economy by increasing default rates and default rate volatilities . and (ii) scenario analysis outside the CREDITR ISK + Model.2 Scenario Analysis within the CREDITR ISK + Model The inputs into the CREDITR ISK + Model can be stressed individually or in combination. It is a dynamic measure.
3.potential losses can be controlled through concentration limits. the impact of political or financial uncertainty within a country. For these types of scenarios.3 Scenario Analysis outside the CREDITR ISK + Model Certain stress tests can be difficult to perform within the CREDITR ISK + Model: for example. provides a realistic means of quantifying the financial impact. such as the 99th percentile level. Losses up to a certain confidence level.99th Percentile Loss Greater than 99th Percentile Loss Control mechanism Adequate pricing and provisioning Economic capital and/or provisioning Quantified using scenario analysis and controlled with concentration limits Mechanisms for controlling the risk of credit default losses Scenario analysis deals with quantifying and controlling the risk of extreme losses.Economic Capital 4. A firm should control the risk of catastrophic losses through the use of obligor and concentration limits. Figure 10: Expected Loss Parts of the credit default loss distribution 99th Percentile Loss Level Probability Economic Capital Loss Covered by pricing and provisioning Covered by capital and/or provisions Quantified using scenario analysis and controlled with concentration limits It is possible to control the risk of losses that fall within each of the three parts of the loss distribution in the following ways: Table 5: Part of loss distribution Up to Expected Loss Expected Loss . such as looking at the exposure at risk for a given scenario. analysis that is conducted without reference to the outputs of the CREDITR ISK + Model. Provisioning for credit risk is discussed in detail in Section 5. 4 Scenario analysis provides a means of quantifying catastrophic losses . CREDITR ISK + 25 . provisioning and economic capital. keeping any one of these limits within the loss for the percentile level used to determine the economic capital given by the CREDITR ISK + Model.2. are controlled by the use of adequate pricing. The figure below illustrates the way in which the distribution of losses can be considered to be divided into three parts.
This level of expected loss should be taken account of when executing any business that has a credit risk impact. credit loss provisions are made only when exposures have been identified as nonperforming. even though the two events are related. The CREDITR ISK + credit risk provisioning methodology more accurately reflects the true earnings of the business by matching income with losses. even though the obligors are currently performing and it is not possible to identify specifically which obligors will default. When default losses are modelled. extremely large losses are suffered. there is a statistical likelihood that credit default losses will occur.1 The Need for Credit Provisions Generally. In relation to any portfolio of credit exposures. 5. a credit provision is required as a means of protecting against distributing excess profits earned during the below average loss years. 5 5. it can be observed that the most frequent loss amount will be much lower than the average. 26 CREDIT SUISSE FIRST BOSTON .Applic CREDIT Applications of CREDITRISK+ 5. The level of expected loss reflects the continuing credit risk associated with the firm’s existing performing portfolio and is one of the costs of doing creditrelated business.1 Introduction CREDITR ISK + includes several applications of the credit risk modelling technology in the areas of provisioning. These provisions are often supplemented with other specific and general credit provisions. occasionally. current accounting and provisioning policies recognise credit income and credit losses at different times. because. Therefore. which have the effect of increasing the average loss. and portfolio management. Usually.2 Provisioning for Credit Risk Applications Provisioning Limits Portfolio Management One application of CREDITR ISK + is in defining an appropriate credit risk provisioning methodology that reflects the credit losses of the portfolio over several years and hence that more accurately presents the true earnings of the business by matching income with losses.2. setting riskbased credit limits.
The nonperforming portfolio should be fully provisioned to the expected recovery level available through foreclosure. Under CREDITR ISK +.2. As for the performing portfolio. The ICR provides protection against unexpected credit losses (i. ACP=Average level of credit losses Figure 11: Credit risk provisioning ICR Cap =99th Percentile Loss Level Typical ICR Probability The Incremental Credit Reserve protects against unexpected credit losses and is used to absorb losses that are higher than the expected level.e. can be established.2. whose average equals the ACP but has a small probability of much larger losses. one needs a forwardlooking provisioning methodology. a better way of viewing the annual credit loss of the portfolio is as a distribution of possible losses (outcomes).2 Annual Credit Provision (ACP) The starting point for provisioning is to separate the existing portfolio into a nonperforming and a performing portfolio. Once fully provisioned. The ICR Cap represents an extreme case of possible credit losses (e. The ACP is the first element of the credit provisioning methodology. since no default has occurred. As experience shows. actual losses that occur in any one year may be higher or lower than this amount. In order to absorb these variations in credit losses from year to year. the Incremental Credit Reserve (“ICR”).Recover Rate) 5 The Annual Credit Provision reflects the continuing credit risk associated with the portfolio and is one of the costs of doing business that creates credit risk.g. the Annual Credit Provision (“ACP”) represents the future expected credit loss on the performing portfolio. T The ACP should be calculated frequently in order to reflect the changing credit quality of the portfolio. etc. in excess of the ACP) and is subject to a cap derived from the CREDITR ISK + Model (the “ICR Cap”). Loss The CREDITR ISK + Model provides information on the distribution of possible losses in the performing credit portfolio.3 Incremental Credit Reserve (ICR) The ACP represents only the expected or average level of credit losses. interest rates. the nonperforming portfolio should then be separated out and passed to a specialist team for ongoing management. CREDITR ISK + 27 . which is calculated as follows: ACP = Exposure x Default Rate x (100% .c Applications 5. a second element of the provisioning methodology. In fact. depending on the economic environment. the 99th percentile loss level) on the performing portfolio. administration or liquidation. 5.
such as traded bond portfolios and derivatives portfolios.000 0 1.300 0 2.600 1. but that the ICR never exceeds the ICR Cap. To the extent that actual credit losses are less than the ACP within any given year. to the extent necessary. beyond which the balance is taken into P&L.250 2. for a portfolio of bonds. the CREDITR ISK + Model provides the information required in order to establish the provision that ensures that the accounting principle of matching income with losses is maintained.655 2.200 0 2.2.430 (625) 0 1.225 28 CREDIT SUISSE FIRST BOSTON .000 600 525 2. This increase in provision is then charged first against the ACP and then.100 0 1.315 (610) 135 1.225 2. the balance is credited to the ICR up to the ICR Cap.250 135 2.900 2.825 2.250 650 625 2. This is described in the following table.385 2.5 Managing the Credit Risk Provision As credit defaults occur.100 (500) 0 1.900 2.900 2.575 1.4 Provisioning for Different Business Lines The credit risk provisioning methodology described above relates to credit risk arising from a loans business where the income is accounted for on an accruals basis rather than by markingtomarket. A credit risk provision can also be established for other credit business lines.805 2. against the ICR.840 2. This ensures that the ICR is replenished during low loss years following a large unexpected loss.100 300 550 2. Table 6: Provisioning for different business lines Portfolio type Loan (Counterparty risk) Derivatives (Counterparty risk) Bond (Issuer risk) Accounting treatment Accrual (credit neutral) Marktomarket (credit neutral) Marktomarket (credit inclusive) Provision • ACP (1 year) charge to P&L each year • ICR (1 year) • ACP (full maturity) held as marktomarket adjustment • ICR (1 year) • ICR (1 year) to support business and protect against distribution of profits 5.5. A worked example can be seen in the table below: Table 7: Example of credit risk provisioning Year Assumptions Actual loan losses ACP ICR .200 300 610 2.2.Initial level ICR Cap Income Statement Operating profit Less: ACP Add: excess unutilised provision over ICR Cap Pretax profit ICR (pre cap) ICR Cap (as above) Excess unutilised provision over ICR Cap ICR (with cap applied) 1 2 3 4 5 500 500 1.300 2.100 (525) 0 1. part of the expected loss is incorporated within the market price and hence only the incremental credit reserve is required. In each case. loans or exposures are moved from the performing to the nonperforming portfolio and hence provisioned to the expected recovery level.205 (550) 0 1.075 2. For example.825 2.075 2.
5. 5. 5 Applications Provisioning Limits Portfolio Management A system of standard credit limits can be supplemented with portfolio level risk information to manage credit risk.4.3.1 Introduction Currently.1 Standard Credit Limits The system of credit limits may be viewed from a different perspective. Concentration limits have the effect of limiting the loss from identified scenarios and is a powerful technique for managing “tail” risk and controlling catastrophic losses. The expected loss for each obligor can be calculated as the default rate multiplied by the exposure amount less the expected recovery. As might be expected. a credit limit system could aim to have a large number of exposures with equal expected losses. including: • • • • Individual obligor limits to control the size of exposure Tenor limits to control the maximum maturity of transactions with obligors Rating exposure limits to control the amount of exposure to obligors of certain credit ratings and Concentration limits to control concentrations within countries and industry sectors. 5. but has the benefit of allowing a firm to relate the size and tenor of limits for different rating categories to each other.3 RiskBased Credit Limits A system of individual credit limits is a wellestablished means of managing credit risk.2 Concentration Limits Any excess country or industry sector concentration can have a negative effect on portfolio diversification and increase the riskiness of the portfolio.4 Portfolio Management The CREDITR ISK + Model makes the process of controlling and managing credit risk more objective by incorporating into a single measure all of the factors that determine the amount of risk. This approach can be extended to base limits on equalising the portfolio risk contribution for each obligor. This means that individual credit limits should be set at levels that are inversely proportional to the default rate corresponding to the obligor rating. A discussion on risk contributions and their use in portfolio management is provided later in this section. the primary technique for controlling credit risk is through the use of limit systems.3. Monitoring exposures against limits provides a trigger mechanism for identifying potentially unwanted exposures that require active management. 5.Applications 5. if applying the methodologies described within this document. in order to equalise a firm’s risk appetite between obligors as a means of diversifying its portfolio. In particular. this methodology gives larger limits for better ratings and shorter maturities. CREDITR ISK + 29 . a comprehensive set of country and industry sector limits is required to address concentration issues in the portfolio. As a result.
risk contributions can be used in portfolio management. 30 CREDIT SUISSE FIRST BOSTON . a different measure that incorporates the magnitude of the exposure. maturity. with the aim of creating a diversified portfolio. Credit limits aim to control risk arising from each of these factors individually. maturity. • • • The total of the risk contributions for the individual obligors is approximately equal to the risk of the entire portfolio. with the aim of creating a diversified portfolio. The risk of a particular exposure is determined by four factors: (i) the size of exposure.2 Identifying Risky Exposures For managing credit risks on a portfolio basis. a portfolio can be effectively managed by focusing on a relatively few obligors that represent a significant proportion of the risk but constitute a relatively small proportion of the absolute portfolio exposures. In general. for managing risks on a portfolio basis. If the portfolio were less diversified. if the portfolio were more diversified.g. is required. which incorporates size.3 Measuring Diversification The loss distribution and the level of economic capital required to support a portfolio are measures of portfolio diversification that take account of the size. 5. (iii) the probability of default.4. a lower level of economic capital would be required. credit quality and systematic risk of each exposure. the removal of an exposure) to be measured. the obligors that require the most economic capital can easily be identified. a different measurement. and (iv) the systematic or concentration risk of the obligor. The effect on the loss distribution and the levels for the expected loss and the economic capital can be seen in the figure opposite. Risk contributions have several features including the following: In general.4. However. maturity. the spread of the distribution curve would be wider and a higher level of economic capital would be required. A portfolio was created from which a small number of exposures with the highest risk contributions were removed. If the percentile level chosen is the same as that used for calculating economic capital. These measures can be used in managing a portfolio of exposures. credit quality and systematic risk into a single measure.4 Portfolio Management using Risk Contributions Applications Provisioning Limits Portfolio Management The risk contribution of an exposure is defined as the incremental effect on a chosen percentile level of the loss distribution when the exposure is removed from the existing portfolio. 5. credit quality and systematic risk into a single measure is required. a portfolio can be effectively managed by focusing on a relatively few obligors that represent a significant proportion of the risk but constitute a relatively small proportion of the absolute portfolio exposures.5. This is illustrated in the following example. (ii) the maturity of the exposure. the risk contribution is the incremental effect on the amount of economic capital required to support the portfolio. Therefore.4. Risk contributions allow the effect of a potential change in the portfolio (e. Conversely. By ranking obligors in decreasing order of risk contribution.
taking collateral has the effect of reducing the severity of the loss given that the obligor has defaulted.Applications New Expected Loss Original Expected Loss Figure 12: 5 Using risk contributions Probability Information about risk New 99th Percentile Loss Level Original 99th Percentile Loss Level contributions can be used to facilitate risk management and efficient use of economic capital. New Economic Capital Original Economic Capital Loss The reduction in the 99th percentile loss level is larger than the reduction in the expected loss level. which leads to an overall reduction in the economic capital required to support the portfolio. which is then sold to investors. there are several techniques for distributing credit risk that can be applied. CREDITR ISK + 31 .5 Techniques for Distributing Credit Risk Once obligors representing a significant proportion of the risk have been identified. Credit derivatives: Credit derivatives are a means of transferring credit risk from one obligor to another. while allowing client relationships to be maintained. 5. Asset securitisations: Asset securitisations involve the packaging of assets into a bond. These include the following: • • • Collateralisation: In the context of the CREDITR ISK + Model.4.
The key concepts are: • • Default rates are stochastic. The level of default rates affects the incidence of default events but there is no causal relationship between the events. 32 CREDIT SUISSE FIRST BOSTON .CREDIT Model Appendix A . We then generalise the technique to the case in which the mean default rate is stochastic. The technique is valid for any portfolio where the default rate for each obligor is small and generates both oneyear and multiyear loss distributions. In order to facilitate the explanation of the CREDITR ISK + Model. The modelling stages of the CREDITR ISK + Model and the relationships between the different sections of this appendix are shown in the figure opposite. we first consider the case in which the mean default rate for each obligor in the portfolio is fixed.The CREDITRISK + Model A1 Overview of this Appendix A This appendix presents an analytic technique for generating the full distribution of losses from a portfolio of credit exposures. The appendix applies the concepts discussed in Sections 2 and 3 of this document.
which introduces dependence between default events. In Sections A6 onwards. Given this assumption and the fact that there is no causal relationship between default events. we interpret default events to be independent. In Section A13 this dependence is quantified by calculating the correlation between default events implied by the CREDITR ISK + Model. CREDITR ISK + 33 .T Application to multiyear losses A5 The CREDITRISK + Model A Figure 13: Flowchart description of Appendix A Default events with fixed default rates A2 Default losses with fixed default rates A3 Calculation procedure for loss distribution with fixed default rates A4 Convergence of variable default rate case to fixed default rate case A11 Default rate uncertainty A6 Sector analysis A7 Default events with variable default rates A8 Default losses with variable default rates A9 Calculation procedure for loss distribution with variable default rates A10 General sector analysis A12 Risk contributions and pairwise correlations A13 A2 Default Events with Fixed Default Rates In Sections A2 to A5 we develop the theory of the distribution of credit default losses under the assumption that the default rate is fixed for each obligor. the assumption of fixed default rates is relaxed.
Therefore F (z ) = It is convenient to write this in the form ∏ FA ( z) = ∏ (1+ pA (z − 1)) A A ∑ log(1+ pA (z − 1)) A (4) logF (z ) = (5) Suppose next that the individual probabilities of default are uniformly small. (7) where we write µ= ∑ pA A (8) for the expected number of default events in one year from the whole portfolio. In this section we derive the basic statistical theory of such processes in the context of credit default risk. Given that the probabilities of default are small. Thus p A = Annual probability of default for obligor A (1) To analyse the distribution of losses arising from the whole portfolio. Thus. it is assumed that each exposure has a definite known probability of defaulting over a oneyear time horizon. This is characteristic of portfolios of credit exposures. Consider a portfolio consisting of N obligors. the probability generating function for the whole portfolio is the product of the individual probability generating functions. we introduce the probability generating function defined in terms of an auxiliary variable z by ∞ = F ( z) = ∑ p(ndefaults)z n n 0 (2) An individual obligor either defaults or does not default. To identify the distribution corresponding to this probability generating function. and give good approximations in practice. The expressions derived from this approximation are exact in the limit as the probabilities of default tend to zero. we expand F(z) in its Taylor series: F ( z) = e µ ( z −1) = e − µe µz = e −µ µ n n z n=0 n! ∑ ∞ (9) 34 CREDIT SUISSE FIRST BOSTON . In line with the above assumptions. powers of those probabilities can be ignored. The probability generating function for a single obligor is easy to compute explicitly as F A ( z) = 1 − p A + p A z = 1 + p A ( z − 1) (3) As a consequence of independence between default events. the logarithms can be replaced using the expression 4 log(1+ p A (z − 1)) = p A ( z − 1) and.1 Default Events Credit defaults occur as a sequence of events in such a way that it is not possible to forecast the exact time of occurrence of any one default or the exact total number of defaults. in the limit. equation (5) becomes (6) 4 ∑ pA ( z −1) F (z ) = e A = e µ (z −1) This approximation ignores terms of degree 2 and higher in the default probabilities.A2.
1 Default Losses with Fixed Default Rates Introduction Under our initial assumptions.The CREDITRISK + Model Thus if the probabilities of individual default are small. the expected number of defaults µ. our initial assumption of fixed default rates cannot account for observed data. provided the number of exposures is large and the width of the bands is small compared with the average exposure size characteristic of the portfolio. we first consider the derivation of the credit loss distribution from the results on default events above. A3. our main objective is to understand the likelihood of suffering given levels of loss from the portfolio. it is possible to describe the overall distribution because its probability generating function has a simple closed form amenable to computation. However. The distributions are different because the same level of default loss could arise equally from a single large default or from a number of smaller defaults in the same year. Intuitively.2 Using Exposure Bands The first step in obtaining the distribution of losses from the portfolio in an amenable form is to group the exposures in the portfolio into bands. This has the effect of significantly reducing the amount of data that must be incorporated into the calculation. indeed. Moreover. Before addressing this in Section A6. differing exposure amounts result in a loss distribution that is not Poisson in general. A3 A3. retaining our initial assumptions for now. Historical evidence of the standard deviation of default event frequencies exists in the form of yearonyear default rate tables. Such data suggests that the actual standard deviation is invariably much larger than √µ. the probability of default can be individually specified for each exposure if sufficient information is available. The Poisson distribution with mean µ can be shown to have standard deviation given by √µ. The distribution does not depend on the number of exposures in the portfolio or the individual probabilities of default provided that they are uniformly small. Thus. this corresponds to the fact that the precise amounts of exposures in a portfolio cannot be critical in determining the overall risk. the approximation is insignificant. However. although not necessarily equal. the distribution of numbers of defaults in a portfolio of exposures in one year has been obtained. rather than given numbers of defaults. CREDITR ISK + 35 . Banding introduces an approximation into the calculation. The following should be noted: • The distribution has only one parameter. which does not influence the distribution of the total number of defaults. However.2 Summary e −µ µ n n! (10) In equation (10) we have obtained the wellknown Poisson distribution for the distribution of the number of defaults under our initial assumptions. Unlike the variation of default probability between exposures. • There is no necessity for the exposures to have equal probabilities of default. information about the distribution of different exposures is essential to the overall distribution. then from equation (9) we deduce that the probability of realising n default events in the portfolio in one year is given by A Probability (n defaults)= A2.
indexed by j. that the unit size is small relative to the typical exposure size of the portfolio. 36 CREDIT SUISSE FIRST BOSTON . after the rounding has been performed for a large portfolio. respectively. an estimate of the effect of banding on the mean and standard deviation of the portfolio is given below. The portfolio can then be divided into m exposure bands. the expected loss εA will be affected. a unit amount of exposure L. νA and εA are the exposure and expected loss. In the rest of this Appendix it is assumed that an adjustment to the default probabilities to preserve the expected losses is made. With respect to the exposure bands. If no adjustment is made.Once the appropriate notation has been set up. as will be shown in Section A4. the rounding process will result in a small rounding up of the expected loss. then. If a suitable size for the unit L is chosen.3 Notation In this section. because we have rounded the νj to make them whole numbers. then. define numbers εA and νA by writing L A = L × ν A and λA = L × ε A (11) Thus. by equation (12) unless a compensating rounding adjustment is made to the expected number of default events µj. of the obligor.2. This step replaces each exposure amount LA by the nearest integer multiple of L.hence µ j = εj νj = ∑ A : ν =ν A εA νA j (12) Note that. where 1 ≤ j ≤ m. Provided the exposure sizes are rounded up. the notation used for the exposure banding described above is detailed. expressing the expected loss in terms of the probability of default events ε j = ν j × µ j . these approaches each have an immaterial effect on the loss distribution. is chosen. the rounding leads to a small overstatement of the standard deviation. expressed as multiples of the unit. Reference Obligor Exposure Probability of default Expected Loss Symbol A LA PA λA In order to perform the calculations. we make the following definitions Reference Common exposure in Exposure Band j in units of L Expected loss in Exposure Band j in units of L Expected number of defaults in Exposure Band j Symbol νj εj µj The following relations hold. denominated in a base currency. there will be a relatively small number of possible values for νA each shared by several obligors. For each obligor A. Under the assumption stated above. A3. The key step is to round each exposure size νA to the nearest whole number.
from the Poisson randomness of the incidence of default events and the variability of exposure amounts within the portfolio.The CREDITRISK + Model As in equation (8). define a polynomial P(z) as follows m εj ν ∑ z j j =1 ν j = m εj ∑ j =1 ν j ∑ µj z P (z ) = j =1 m νj µ (18) where we have used equations (12) and (13) for the total number µ of defaults in the portfolio. Since A µ is the sum of the expected number of default events in each exposure band. because some defaults lead to larger losses than others through the variation in exposure amounts over the portfolio.4 The Distribution of Default Losses We have analysed the distribution of default events under our initial assumptions. respectively. the exposure bands are independent. the second random effect is best described mathematically through its probability generating function. and the probability generating function can be written as a product over the exposure bands G(z ) = ∏Gi ( z ) =1 i m (15) However. we show how to use the probability generating function to derive the actual distribution of losses under our initial assumptions. We now proceed to derive the distribution of default losses. As with default events. The probability generating function in equation (17) can now be expressed as G(z ) = e µ (P ( z ) −1) = F (P ( z )) (19) This functional form for G(z) expresses mathematically the compounding of two sources of uncertainty arising. Thus. the default loss analysis involves a second element of randomness. Therefore. Intuitively. First. we obtain G j ( z) = Therefore n =0 ∑ ∞ p (n defaults) z nν j = n= 0 ∑ ∞ e −µ j µn j n! m z nν j =e −µ j + µ j z νj (16) G(z ) = ∏e 1 j= m − µ j +µ j z νj =e −∑ µ j + ∑ µ j z j =1 j =1 m νj (17) This is the desired formula for the probability generating function for losses from the portfolio as a whole. let µ stand for the total expected number of default events in the portfolio in one year. CREDITR ISK + 37 . by treating each exposure band as a portfolio and using equation (9). For later reference. we have µ= ∑µj = ∑ j =1 m m εj (13) j =1ν j A3. let G(z) be the probability generating function for losses expressed in multiples of the unit L of exposure: ∞ = G(z ) = ∑0p ( aggregate losses = n × L) z n n (14) The exposures in the portfolio are assumed to be independent. equation (17) can be restated in a slightly different form. In the next section.
together with the share ε of expected loss arising from each exposure size. the following recurrence relationship is obtained An = j v j ≤n ∑ : εj n An−v j (25) This recurrence relationship allows very quick computation of the distribution. We wish to compute An efficiently. all that is needed is knowledge of the different sizes of exposures ν within the portfolio. This is typically a very small amount of data. For n an integer let An be the probability of a loss of nxL. Comparing the definition in equation (14) with the Taylor series expansion for G(z). Therefore. A4 A4. In order to commence the computation. even for a large portfolio. we have p (lossof nL ) = 1 d nG( z) n! dzn = An z= 0 (20) In our case G(z) is given in closed form by equation (17). this approach will be generalised to compute the distribution for the CREDITR ISK + Model.Note that G(z) depends only on the data ν and ε. we have the following formula for the first term. d n− 1 n! dz dz ∑µj z =1 j m j =1 m vj z =0 (21) 1 n −1 n − 1 d n−k −1 d k +1 k dzn −k −1 G( z) dzk +1 n! k =0 ∑ ∑µ jz νj z =0 µ (k + 1)! if k = v j − 1 for somej v µj z j = j ∑ 0 otherwise j =1 z= 0 m (22) and by definition d n −k −1 G( z ) = (n − k − 1)! An −k −1 dzn− k −1 z =0 Therefore (23) An = k ≤n − k =v j −1for some j ∑ 1 1 n − 1 (k + 1)!(n − k − 1)!µ j An−k −1 = n ! k µ jv j An −v j j v j ≤n n ∑ : (24) Using the relation εj = νj x µj from equation (12).1 Loss Distribution with Fixed Default Rates Calculation Procedure In this section. to obtain the distribution of losses for a large portfolio of credit risks. a computationally efficient means of deriving the actual distribution of credit losses is derived from the probability generating function given by equation (17). which expresses the probability of no loss arising from the portfolio −∑ m εj (26) A0 = G (0 ) = F (P (0 )) = e −µ =e j = 1ν j 38 CREDIT SUISSE FIRST BOSTON . or n units from the portfolio. In Section A10. Using Leibnitz’s formula we have 1 d nG (z ) n ! dzn = However +1 dk +1 dzk = z =0 −1 1 dn G( z ).
The expected loss is unaffected by the method of rounding chosen. In this section. For the standard deviation. suppose that the above are expressions for the “true” mean and standard deviation. In practice.The CREDITRISK + Model Again. We conclude that: (30) • • A5 The expected loss calculated by the model is unaffected by the banding process. each exposure has an individual maturity corresponding to the normal maturity of bonds. In order to represent the banding. it is worthwhile to note that the calculation depends only on knowledge of ε and ν. Taking square roots and neglecting higherorder terms in the Taylor series we obtain ε ε σ ≤ σˆ ≤ σ 1 + =σ + 2σ 2σ 2 For a real portfolio. we have σ 2 ≤ σˆ 2 = ∑νˆ j × ε j = σ 2 + ∑ τ j × ε j ≤ σ 2 + ∑ ε j = σ 2 + ε j =1 j =1 j =1 m m m (29) where ε is the expected loss for the portfolio. As in the discussion over a oneyear horizon. It is assumed that the exposures are rounded up. loans or other instruments. In particular.1 The recurrence relation above was derived on the basis of a oneyear loss distribution. however. so that each τj is positive. For simplicity it is assumed that the future of the portfolio is divided into years. In this section it is shown how the initial model can be applied over a multiyear time horizon.2 Precision Using Exposure Bands A The banding process described in Section A3 introduces a small degree of approximation into the data. but that now the ν are rounded to integer multiples of the unit as explained above. In terms of the notation above. these represent a very small amount of data even for a large portfolio consisting of many exposures. the expected loss ε and the quantity 2σ are of the same order. consider a portfolio of obligors with small probabilities of default. A4. we show that the approximation error is normally not material by considering the effect on the portfolio mean and standard deviation. The exposures are permitted to vary from year to year. the total portfolio expected loss ε and total portfolio standard deviation σ are ε = ∑ ε j . because its expression is independent of the banded exposure amounts. This was noted above. The standard deviation is overstated by an amount comparable with the chosen unit size. Application to MultiYear Losses Introduction A5. CREDITR ISK + 39 . write νˆ j = ν j + τ j where 0 ≤ τ j ≤ 1 (28) Each τj is at most of absolute size one. σ 2 = ∑ν j × ε j j =1 j =1 m m (27) where the expected loss and standard deviation are expressed in the chosen unit L. This process introduces an error.
t . the recurrence relation given by equation (25) for the distribution of losses over one year is also applicable to the calculation of the multiyear distribution of losses An = j .v (jt ) ≤n ∑ ε (jt ) n An−v ( t ) j (37) 40 CREDIT SUISSE FIRST BOSTON . Under the natural assumption that defaults by the same exposure in different years are mutually exclusive. marginal rates of default must be specified for each future year for each obligor in the portfolio.t ν j = ν : ∑ (34) j . the probability generating function for multiyear losses from a single exposure is given by Gj ( z ) = 1 − t ∑ p(jt ) + ∑ pj(t )z =0 =0 t T T ν (jt ) = 1+ t ∑ p(jt ) ( z =0 ν (t ) j T ν (jt ) − 1) (31) For the generating function of total losses. Collectively.t :ν (jt ) =ν ∑ p j(t ) = t =1 j .2 Term Structure of Default In order to address a multiyear horizon. we have logG( z ) = ∑ log 1+ ∑ p(jt ) ( z j t =1 ν (t ) j T − 1) (32) In the limit of small probabilities of default we argue as for equation (6) to obtain log1+ and we obtain ∑ =1 t T p (t ) ( z j − 1) = ∑ p(jt ) (z =1 t T ν (jt ) − 1) (33) logG( z ) = and ∑∑ = j T p (t ) ( z j ν (t ) j − 1) = t 1 ∑ ν zν p (t ) (t ) j j . Therefore.t ε (jt ) ν (jt ) (z (t ) νj (36) This has the same form as the one year probability generating function (17).:ν (jt ) =ν ∑ ∑ T p j(t ) = ∑ ∑ =1 t −1) T ε (jt ) (t ) j .3 Notation Fix the following notation: Reference Probability of default of exposure j in year t Symbol (t) pj L j = Lν j (t) (t) (t) Amount of exposure j in year t Expected loss in exposure j in year t λ j = Lε j (t) As for the oneyear discussion.A5. A5. L is the unit of exposure and the ν j(t) are dimensionless whole numbers.:ν (jt ) =ν ν j (35) The probability generating function for the loss distribution is therefore given by G(z ) = e ∑ j . such marginal default rates give the term structure of default for the portfolio.
which affect the fortunes of obligors. the measurement of background factors is addressed by dividing the obligors among different sectors. as remarked in Section A2. it is necessary to quantify the extent to which each factor has an influence on a given portfolio of obligors. In the following sections.2. in reality. For example. Published statistics on the incidence of default events. In Section A12. is presented. The variability of default probabilities can be related to underlying variability in a relatively small number of background factors. A7 A7. A factor such as the economy of a particular country may be considered to have a uniform influence on obligors whose domicile is within that country. The second point made above is that uncertainty arises from factors that may affect a large number of obligors in the same way. show that the number of default events. This in turn leads to a higher chance of experiencing extreme losses. Such yearonyear statistics may be thought of as samples from a random variable whose expected value represents an average rate of default over many years. Therefore the analysis above which considered rare events is relevant in a suitably modified form. These concepts are introduced in this section and Section A7 respectively. obligors may be under the simultaneous influence of a number of major factors. An initial example might be a division of the portfolio according to the country of domicile of each obligor. In order to measure this effect and hence quantify the impact of individual default rate volatilities at the portfolio level. a downward trend in the state of the economy may make most obligors in a portfolio more likely to default. the next year than their average over many years suggests. but relatively little influence on other obligors in a multinational portfolio. actual defaults should still be relatively rare events. The standard deviation of the variable measures our uncertainty as to the actual default rate that will be exhibited over a given year. a more general sector analysis. CREDITR ISK + 41 . exhibits wide variation from year to year . • However. 5 If the default rates of obligors were fixed. for example. that influence the fortunes of obligors. and therefore the average probability of default for such entities. such as the state of the economy. default events would still have a nonzero standard deviation arising from the randomness of the default events themselves. which allows for the fact that. where each sector is a collection of obligors under the common influence of a major factor affecting default rates. Owing to default rate uncertainty. Whatever the state of the economy. there is a chance that default rates will turn out to be higher over.1 Sector Analysis Introduction It was noted above that the variability of default rates can be related to the influence of a relatively small number of background factors on the obligors within a portfolio.The CREDITRISK + Model A6 Default Rate Uncertainty A The previous sections developed the theory of the loss distribution from a portfolio of obligors. for example among rated companies in a given country. This concept is introduced in the next section. a change in the economy or another factor will not cause obligors to default with certainty. In this section. each of which has a fixed probability of default. the concept of sector analysis is necessary. However. The situation may be summarised by the following three intuitive facts about default rate uncertainty: 5 • • Observed default probabilities are volatile over time. the CREDITR ISK + Model will be developed from this theory by incorporating default rate uncertainty and sector analysis. even for obligors having comparable credit quality. The appearance of randomness is due to the incidence of factors. In order to measure the effect of these factors. comparison with historic data shows that observed volatility is far higher than can be accounted for in this way. such as the state of the economy.
which is then modelled as a random variable with mean µk and standard deviation σk specified for each sector. for each sector we introduce a random variable xk representing the average default rate over the sector. each of which should be thought of for now as a subset of the collection of obligors.A7. The mean of xk is µk and the standard deviation is σk.3 is also repeated for comparison Exposure Data within Sector Previous Notation L L j = Lν j 1≤ j ≤ m Expected loss in each exposure band in units New Notation L (k) Lj = Base unit of exposure Exposure sizes in units Lν j (k) 1≤ k ≤ n . in a sector consisting of a large number of obligors of low credit quality. then a year in which as many as 12% of the obligors default would have been a much more remote possibility. 10% instead of 5%. The standard deviation will reflect the degree to which the probabilities of default of the obligors in the portfolio are liable to all be more or less than their average levels. The underlying factor influences the sector through the total expected rate of defaults in the sector. say. the data requirements are set out below. the mean default rate might be 5% per annum and the standard deviation of the actual default rate might be a similar quantity. 12% of the obligors will actually default in that year. In particular. say. The CREDITR ISK + Model regards each sector as driven by a single underlying factor. which explains the variability over time in the average total default rate measured for that sector.2 Further Notation New notation is needed to keep track of the division of the portfolio into sectors and to record the volatility of the default rate for each sector. Had the standard deviation been zero. Write Sk: 1 ≤ k ≤ n for the sectors.mean of xk Standard deviation of xk µk σk For each sector.1≤ j ≤ m(k) (k) (k) The mean µk is related to the expected loss data by the following relation which is the analogue of equation (13): m(k ) j 1 µk = ∑ ν(k ) = j ε (j k ) (38) 42 CREDIT SUISSE FIRST BOSTON . reflecting that we were certain about the probability of each obligor defaulting. For example.1≤ j ≤ m(k) λ j = Lεj 1≤ j ≤ m λ j = Lεj 1≤ k ≤ n . In turn it is much more likely that. The table below summarises the new notation to specify the sector decomposition of the portfolio. Then there will be a substantial chance in any year of the actual average probability of default in the sector being. Sk : 1 ≤ k ≤ n xk Sector Random variable representing the mean number of defaults Longterm annual average number of defaults . Our original notation set up in Section A3.
a standard deviation σA has been assigned for the default rate for each obligor within the sector. The sum runs over all obligors in the sector.3 Estimating the Variability of the Default Rate A we must specify a standard deviation σk of the total expected rate of default. A convenient way to do this is to assume that the standard deviation depends on the credit quality of the obligor. An alternative and more intuitive description of the standard deviation σk determined in this way is that σk = µk σA ∑ σ A ∑ pA pA A A (43) ∑ pA A = ∑ pA A CREDITR ISK + 43 . By equation (42) sector. in particular. it can equivalently be expanded as a sum over all the obligors in the sector µk = ∑ A εA νA (39) where the summation extends over all obligors A belonging to sector k. to estimate the standard deviation by reference to equation (38) for the mean.The CREDITRISK + Model A7. weighted by their contribution to the default rate. and the relation εA = pA νA (40) expresses the average probability of default of the obligor over the time period. We estimate the standard deviation of this sector so as to ensure that this condition holds. We discuss a convenient way For each sector. whose mean is equal to the specified mean default rate for that obligor. in addition to the expected total rate of default µk over the sector given by equation (38). Assuming equation (41). To express this dependence write xA for the random default probability of the single obligor A. the actual default probability for each obligor in the sector will be modelled as a random variable proportional to xk. To obtain an estimate of the standard deviation of each sector. Although equation (38) is expressed in terms of exposure bands. Thus the standard deviation of the mean default rate for a sector is estimated as the sum of the estimated standard deviations for each obligor in the the ratio of σk to the mean µk is an average of the ratio of standard deviation to mean for each obligor. Our assumption can then be written xA = ε A xk νA µ k (41) Note that the mean of xA is correctly specified as pA by this equation. That is. together with a probability of default pA. xk is held to account for the uniform variability of each of the probabilities of default. This is easily seen as follows. we assume that. we have ∑σ A = ∑ ν A A A A ε σk 1 =σ k µk µk ∑ νA A A ε =σ k (42) where we have used equation (39). We obtain an estimate of σk from the set σA of obligor standard deviations by an averaging process. Recall that only one random variable. This pragmatic method assumes that the credit quality of the obligors within a sector is a more significant influence on the volatility of the expected default frequency than the nature of the sector.
According to historical experience. the probability generating function for default events is written ∞ = F (z ) = ∑ p(ndefaults) zn n 0 ∏ Fk (z ) =1 k n (45) Because the sectors are independent. We make the key assumption that xk has the Gamma distribution with mean µk and standard deviation σk. so that (47) P ( x ≤ x k ≤ x + dx) = f k ( x )dx (48) Then. an appropriate distribution for Xk must be chosen. the probability generating function for default events in one sector is the average of the conditional probability generating function given by equation (47) over all possible values of the mean default rate. the distribution of default events for the CREDITR ISK + Model is obtained. so that the standard deviation of the number of defaults observed year on year among obligors of similar credit quality is typically of the same order as the average annual number of defaults. Equation (43) shows that the same is true for each sector. we can write down the probability generating function for the distribution of default events as follows Fk (z ) [x k = x ] = e x (z −1) where equation (7) has been used. as one would expect. Suppose that xk has probability density function ƒk(x). In the notation of Section A7.1 Default Events with Variable Default Rates Conditional Default Rate In this section. Then. F(z) can be written as a product over the sectors F (z ) = (46) We therefore focus on the determination of F(z) for a single sector. The Gamma distribution is chosen as an analytically tractable twoparameter distribution. writing ωk for this uniform ratio. 44 CREDIT SUISSE FIRST BOSTON . this would be equivalent to estimating the flat ratio ωk directly. written xk. the ratio σA /pA is typically of the order of one. the obligor specific estimates of the ratio σA/pA can be replaced by a single flat ratio. In the absence of detailed data. with mean µk and standard deviation σk. Conditional on the value of xk. Most of the work has been done already in the calculation of the probability generating function in equation (7) when the default rate is fixed. A8 A8. equation (43) reduces to the simple form σ k = ω k × µk (44) If the nature of the sector made it more appropriate to estimate the standard deviation σk directly. As in equation (2). This is achieved by calculation of the probability generating function. as the following computation shows: ∞ ∞ ∞ ∞ Fk (z ) = n=0 ∑ P (n defaults) z n = n=0 ∑ zn x= 0 ∫ P (n defaults x )f ( x )dx = e x (z −1)f ( x )dx x =0 ∫ (49) In order to obtain an explicit formula for the probability generating function. Before proceeding to evaluate equation (49) explicitly. the average default rate in sector k is a random variable. the basic properties of the Gamma distribution are stated.
3 Distribution of Default Events in a Single Sector With the choice of Gamma distribution for the function ƒ(x). These are related to the defining parameters as follows µ = αβ and σ 2 = αβ 2 Hence. By expanding Fk(z) in its Taylor series ∞ Fk (z ) = (1 − pk )α k the following explicit formula is obtained ∑ n 1 = n + α k − 1 n n pk z n (56) n + α k − 1 n P (n defaults) = (1 − pk )α k pk n This can be identified as the probability density of the Negative Binomial distribution. x=0 (50) ∞ The Gamma distribution Γ(α.The CREDITRISK + Model A8. It is possible to identify the distribution of default events underlying this probability generating function. β). (57) CREDITR ISK + 45 . is a skew distribution. The probability density function for a Γ(α. for sector k. for sector k 1 − pk Fk (z ) = 1− p z k wherepk = βk 1+ βk (55) This is the probability generating function of the distribution of default events arising from sector k. β) . By substitution. the expression for the probability generating function ∞ Fk ( z ) = x= 0 ∫e x ( z −1) f (x )dx (53) given by equation (49). fully described by its mean and standard deviation.distributed random variable X is given by − 1 e β x α −1dx α β Γ(α ) x P ( x ≤ X ≤ x + dx) = f ( x )dx = where Γ(α) = ∫ ex x α 1dx is the Gamma function. can be directly evaluated. β) is a two parameter distribution. change of variable and definition of the Gamma integral ∞ ∞ − e xα 1 1 y dx = α α β −1 + 1 − z β Γ(α ) β Γ(α ) y =0 − x Fk (z ) = = x =0 ∫ e x ( z −1) Γ(α ) β ∫ α −1 e−y dy β −1 + 1 − z (54) β Γ(α )(1 + β α −1 − z) α = 1 β (1 + β −1 − z )α αk α Upon rearrangement this becomes.2 Properties of the Gamma Distribution A The Gamma distribution. the parameters of the related Gamma distribution are given by 2 2 2 α k = µk / σ k and β k = σ k / µ k (51) (52) A8. which approximates to the Normal distribution when its mean is large. written Γ(α.
the probability generating function for losses from the portfolio. sector independence gives a product decomposition of the probability generating function G(z ) = ∏1Gk ( z ) = k n (62) where Gk(z) is the loss probability generating function for sector k. 1 ≤ k ≤ n.A8. The default event distribution for the whole portfolio is not Negative Binomial in general but is an independent sum of the Negative Binomial sector distributions. By analogy with equation (18). by ε (j k ) ν (k ) ∑ (k ) z j (k ) j =1 ν j 1 m (k ) ε j ν (j k ) z Pk (z ) = = ∑ (k ) µk j =1 ν (k ) m( k ) ε j j ∑ (k ) j =1 ν j m (k ) (63) 46 CREDIT SUISSE FIRST BOSTON . As for the distribution of default events. A9. β k = σ k / µ k and p k = β k /( 1 + β k ) . We seek a closed form expression for G(z) and a means of efficiently computing G(z). The corresponding product decomposition of the probability generating function is given by equation (59). βk and pk are given by 2 2 2 α k = µk / σ k . β k ) The probability generating function for default events from the whole portfolio is given by (58) F (z ) = ∏1 = k n Fk (z ) = ∏ 1 − pk k =1 1 − p k z n αk (59) where the parameters αk.4 Summary The portfolio has been divided into n sectors with annual default rates distributed according to Γ(α k . we performed this process conditional on a fixed mean default rate. Thus let ∞ = G(z ) = ∑ p( aggregate losses = n 0 n × L) z n (61) be the probability generating function of the distribution of loss amounts. We now generalise this process to incorporate the volatility of default rates. we introduce a second probability generating function G(z). A9 A9. (60) The default event distribution for each sector is Negative Binomial. In order to pass from default events to default losses. we define polynomials Pk(z).1 Default Losses with Variable Default Rates Introduction The probability generating function in equation (59) gives full information about the occurrence of default events in the portfolio. this distribution must be compounded with the information about the distribution of exposures. In Section A3.4. 1 ≤ k ≤ n.2 The Distribution of Default Losses By analogy with equation (14).
the expressions are the same. In the next section a recurrence relation for computing the distribution of losses from this expression is derived. By substitution into equation (55) and taking the product over each sector. because the following relation holds A Gk ( z ) = Fk (Pk ( z )) (64) This is directly analogous to the formula G(z)=F(P(z)) obtained in equation (19) for a fixed mean default rate. which expresses Fk(z) as an integral of the Poisson probability generating function over the space of possible values of the random variable xk. suitable for explicitly calculating the distribution of losses from equation (68).The CREDITRISK + Model where the expression for µk in equation (38) has been used. A10 Loss Distribution with Variable Default Rates In this section. In order to see that the relation continues to hold in the present case. except that in equation (17) terms with the same exposure amount have been collected. is presented. except that there is now one such relation for each sector. Thus Gk (z ) = n=0 ∑ ∞ zn x k =0 ∫ ∞ P (Loss of nL x k )fk ( xk )dxk = = ∞ x k =0 ∫ e ∑ x A ( zv A −1) A fk (x k )dxk (67) ∫e x 0 k ∞ x k ( Pk ( z )−1) f k (x k )dxk = Where the last step follows from equation (66). The relation is a form of the recurrence relation in Section A4. Thus Pk (z ) = ∑ ν A zν A A ε ∑ν A A A ε A = 1 µk ∑ ν A zν A A ε A νA ( z −1) νA ε A (65) By equation (41) we have e −∑ x A +∑ x Az ν A A A =e A ∑ x A zν A −1 ( ) = e µk xk ∑ A = e x k (Pk ( z ) −1) (66) The left hand side of equation (66) is the probability generating function of the distribution of losses where each obligor A has default rate xA. we expand equation (63) as a sum over individual obligors belonging to sector k. This can be seen by comparing equation (17) . a conditional probability argument shows that Gk(z) is the integral of the left hand side of equation (66) over the same space. The Pk(z) provide the link between default events and losses. Just as in equation (53). a recurrence relation. we obtain n n 1 − pk G(z ) = Gk ( z ) = m( k ) ε ( k ) ν(k ) p k =1 k =1 j z j 1 − k (k ) µk j =1 ν j ∏ ∏ ∑ αk (68) This is a closed form expression for the probability generating function. CREDITR ISK + 47 . derived for a wider class of distributions.
n −1) j =0 ∑ bj +1(n − j ) An − j (77) In equation (68). a power series expansion ∞ = G(z ) = ∑ An z n n 0 (69) defines a function G(z) which satisfies the differential equation d 1 dG (z ) A( z) (logG( z )) = = dz G (z ) dz B (z ) where A and B are polynomials given respectively by (70) A( z) = a0 + . this leads to dG = A( z ) G dz (73) s b zj j j =0 ∑ ∞ r (n + 1)An +1zn = ai z i n= 0 i =0 ∑ ∑ ∞ An zn n= 0 ∑ (74) For n ≥ 0 the terms in zn on the left hand and right hand side respectively are min(s . Then.n ) i 0 min(r . we require that the logarithmic derivative of G(z) be a rational function..2 Application min(r .n ) 1 ai An−i − b j +1(n − j )An− j b0 (n + 1) i = 0 j =0 ∑ ∑ (72) To see this.n−1) min(r .n ) j 0 ∑ bj (n + 1− j ) A n+1− j = min(r . the probability generating function of losses was derived in the form n n 1 − pk G(z ) = Gk ( z ) = m( k ) ε ( k ) ν(k ) p k =1 k =1 j z j 1 − k µk j =1 ν (k ) j ∏ ∏ ∑ αk (78) 48 CREDIT SUISSE FIRST BOSTON .A10. rearrange the differential equation (70) as follows B (z ) By differentiating G term by term. + bs zs (71) In other words. the terms of the power series expansion in equation (69) satisfy the following recurrence relation An +1 = min( s −1. + ar z r B ( z) = b0 + .. in general..n ) ∑ i =0 ai A n −i − min(s −1. A10.n ) (76) b 0 (n + 1)An +1 = as required.1 General Recurrence Relation Suppose.n ) and ∑ a i A n− i i=0 (75) Equating these expressions and rearranging we obtain b 0 (n + 1)An +1 = or equivalently ∑ ai A n−i − ∑ bj (n + 1− j )A n +1− j = = j 1 min(s ..
Note that the summation described in equation (80) must be performed directly by adding the rational summands. In particular. The proof proceeds by showing that the probability generating function for the CREDITR ISK + Model converges to the form ε j ∑ j =1 ν j e m G(z ) = νj ( z −1) (81) which is the probability generating function in equations (17) for losses conditional on a fixed mean default rate. These are where: • • The standard deviation of the mean default rate for each sector tends to zero. after calculation of polynomials A(z) and B(z) such that A(z ) = B (z ) k ∑ =1 n pk α k µk 1− m( k ) j =1 ∑ ε (jk ) z ∑ ν (k ) −1 j (80) ( ) (k ) pk m k ε j µk j =1 ν (k ) j z ν (k ) j the calculation in Section A10.1 is applicable and leads to a recurrence relation for the loss amount distribution. A11 Convergence of Variable Default Rate Case to Fixed Default Rate Case Although the CREDITR ISK + Model is designed to incorporate the effects of variability in the average rates of default. this computation can be performed quickly. In this section. there are two circumstances in which the CREDITR ISK + Model behaves as if default rates were fixed. the effect of either a large number of sectors or a low standard deviation of default rates on the portfolio is the same. β k = σ k / µ k . p k = β k /( 1 + β k ) and µk = m(k ) k =1 ∑ ν(k ) j ε (j k ) CREDITR ISK + 49 . The proof of the second convergence fact is similar. In the section on generalised sector analysis. the behaviour in either case is as if default rates were fixed. Provided the unit size is chosen so that the exposures νj and therefore the degrees of numerators and denominators of the rational summands are not too large.The CREDITRISK + Model Taking logarithmic derivatives with respect to z. The number of sectors tends to infinity. it follows that A ′ G′( z ) n Gk (z ) n = = G( z) k =1Gk (z ) k =1 ∑ ∑ pk α k µk 1− m( k ) j =1 ∑ ε (jk )z ∑ ν (jk ) −1 (79) ( ) (k ) pk m k ε j µk j =1 ν (j k ) z ν (jk ) This expresses G’(z)/G(z) as a rational function. Accordingly. a proof is given of the first convergence fact. given at equation (68) n n 1 − pk G(z ) = Gk ( z ) = m( k ) ε ( k ) ν(k ) p k =1 k =1 j z j 1 − k (k ) µk j =1 ν j ∏ ∏ ∑ αk (82) 2 2 2 where α k = µk / σ k . this fact will be used to facilitate the analysis of specific risk within a portfolio. The CREDITR ISK + Model has the following probability generating function for default losses.
To understand how to generalise the sector analysis already presented. each of which is a subset of the set of obligors. replacing the concept of a sector with that of a systematic factor. Then βk ¡ 0. we are simultaneously uncertain about all these values. a relatively small number of factors explain the systematic volatility of default rates in the portfolio. this expression can also be viewed as a multiple integral ∞ G(z ) = x1 =0 x n =0 ∫ ∫ . it is not possible to describe the portfolio with sectors consisting of groupings of the obligors. ∞ e ∑ xk (Pk ( z )−1) k =1 n ∏fk ( xk )dxk = k 1 n (87) We regard the integrand as the probability density function of a compound Poisson distribution for any given set of values of the means xk. Therefore. we reexamine the derivation of the probability generating function for the CREDITR ISK + Model. 50 CREDIT SUISSE FIRST BOSTON . as before. However.. In equation (68). the summation over k is eliminated ε G(z ) = e as required. but the CREDITR ISK + Model incorporates this situation in the same way as before. 1 ≤ k ≤ n. each of which is driven by one factor but all of which are mutually independent. it was assumed throughout that the portfolio is divided into sectors. the probability generating function was derived by expressing it as a product over the sectors and then integrating with respect to the distribution of default rates for each sector: G(z ) = ∏1 = k n Gk ( z) = ∏ ∫ e x ( P (z )−1)fk (x k )dxk k =1 k k n ∞ (86) x=0 However. This corresponds to a situation in which obligors fall into classes. the probability density function is then integrated over the space of all possible states represented by the values of the xk and weighted by their associated probability density functions. ∑ ν j (z j j νj −1) (85) A12 General Sector Analysis A12.We consider the limit where σk tends to zero. but it is not necessarily the case that the default rate of an individual obligor depends on only one of the factors.k ν j +∑ j .. In these more general circumstances. pk = βk /(1+ βk ) ¡ 0 and αk = µk /βk ¡ µk /pk Therefore In the limit n n 1 − pk G(z ) = Gk ( z ) = m( k ) ε ( k ) ν(k ) p k =1 j k =1 z j 1 − k (k ) µk j =1 ν j ∏ ∏ ∑ e αk n 1− p k → m(k ) ε (k ) ν (k ) p k =1 j z j 1− k (k ) µk j =1 ν j ∏ ε (jk ) ∑ pk µk (83) G(z ) → ∏e =1 k n − µk m (k ) ε ( k ) ν ( k ) j j ∑ (k ) z j =1 ν j =e −∑ (k ) j .1 Introduction In the derivation of the CREDITR ISK + Model probability generating function for the distribution of losses in Section A9. We now consider a more generalised situation in which.k ε (k ) j ν (k ) j z ν (jk ) (84) On collecting terms in the exponent having common values n across different values of k.
we replace the delta function with an allocation of the obligors among sectors by choosing. numbers n θ Ak : k =1 ∑ θ Ak = 1 (90) The allocation θAk represents the extent to which the default probability of obligor A is affected by the factor underlying sector k. As in the special case discussed in Section A7. As explained in Section A12. assuming again that estimates have been obtained of both quantities for each obligor by reference to obligor credit quality. CREDITR ISK + 51 . this is expressed by a choice of number θ Ak : k 1 ∑ θ Ak = 1 = n (95) for each sector k and obligor A in the portfolio. We indicate a method of estimating these parameters. we show how to assimilate the data for the CREDITR ISK + Model for generalised sector analysis. for each obligor A.The CREDITRISK + Model Using equation (66). The number θAk represents our judgement of the extent to which the state of sector k influences the fortunes of obligor A. The sector analysis discussed in Section A7 corresponds to the special case θ Ak = δ Ak In the general case the expression in equation (88) is replaced by (91) ∑ xk (Pk ( z ) − 1) = ∑θ Ak µk ν A (zν k A k =1 A . we can examine the exponent in the integrand in its equivalent form A k ∑ =1 n xk (Pk ( z ) − 1) = k ∑ ∑ µkk ν A (zν A =1 ∈ A k n x ε A − 1) = ∑ δ Ak µk ν A ( zν k A . we must also provide estimates of the mean and standard deviation for each sector. allowing each obligor to be influenced by more than one factor xk.2 Performing the Sector Decomposition In this section.1. We assume that for each obligor in the portfolio an estimate has been made of the extent to which the volatility of the obligor’s default rate is explained by the factor k. Ak x ε A − 1) (88) where we have used the delta notation 0 A ∉ k δ Ak = 1 A ∈ k (89) To generalise the concept of a sector in these circumstances.k where each obligor contributes a term n x ε A − 1) (92) x A (zν A − 1) where x A = Equation (65) is replaced by εA νA ∑1θ Ak µkk = k n x (93) Pk ( z) = 1 µk ∑θ Ak ν A zν A A ε A where µk = ∑ θ Ak ν A A A ε (94) A12.
It was remarked in Section A11 that assigning a zero variance to a sector is equivalent to assuming that the sector is itself a portfolio composed of a large number of subsectors.1 Introduction In this section. we derive a formula for the contribution of an individual obligor to the standard deviation of the loss distribution in the CREDITR ISK + Model. measured either by a chosen percentile level or the standard deviation. Thus µk = ∑ θ Ak µA A (96) Then. the risk contribution can be defined as the marginal effect of the presence of EA on some other measure of portfolio aggregate risk. the proportion of the variance of the expected default frequency for that obligor that is explained by specific risk is θA1.3 Incorporating Specific Factors So far we have assumed all variability in default rates in the portfolio to be systematic. for each obligor A. as follows: • • Risk contributions are defined as the contributions made by each obligor to the unexpected loss of the portfolio. but now weighted by the allocations θAk. we express the ratio σk/µk as a weighted average of contributions from each obligor σk = µk ∑θ Ak µA µA A A σ ∑θ Ak µA A . A12. However. only one sector is necessary in order to incorporate specific factors. the risk contribution for obligor A can be defined as the marginal effect of the presence of EA on the standard deviation of the distribution of credit losses. by analogy with equation (43). one for each obligor in the portfolio. for a portfolio containing a large number of obligors. we derive formulae for two useful measures connected with the default loss distribution. we require an additional sector to model factors specific to each obligor. Sector 1 would be assigned a total standard deviation given by equation (97). Pairwise correlations between default events give a measure of the extent to which concentration risk is present in the portfolio. 52 CREDIT SUISSE FIRST BOSTON . Hence. Let the specific factor sector be sector 1. this standard deviation can be set to zero. The discussion in Section A7 is recaptured when θAk = δAk as discussed above. The specific factor sector then behaves as the limit of a large number of sectors.The mean for each sector is the sum of contributions from each obligor. for the specific factor sector only. A13. Potentially. However. specific factors can be modelled without recourse to a large number of sectors. Then. Alternatively. such as a given loss percentile.2 Risk Contributions In this section. A13 Risk Contributions and Pairwise Correlation A13. For a portfolio of obligors A having exposure EA.hence σ k = ∑θ Akσ A A (97) This estimates the standard deviation for each factor. with independent variability of their default rate. The lost standard deviation represented by σ1 is a measure of the benefit of the presence of specific factors in the portfolio.
by a general property of homogeneous polynomials we have ∑ RCA = 2σ ∑ E A ∂EA A A 1 ∂σ 2 = 2σ 2 =σ 2σ (100) If the marginal effect on a given percentile is chosen as the definition of risk contributions. the risk contributions defined by equation (98) add up to the standard deviation. Since the mean and variance of the distribution of losses in the CREDITR ISK + Model are both additive across sectors. then an analytic formula will not be possible. for most models including the CREDITR ISK + Model. This is because of the variancecovariance formula σ2 = ∑ ρ ABE AEB σ Aσ B A. an analytic formula for the risk contribution is possible. we have therefore suppressed the reference to sector k where it is not necessary. we can define risk contributions to the percentile in terms of the contributions to the standard deviation by writing ˆ RC A = ε A + ξRCA Then. referring to a sector k. or equivalently RCA = A 2σ ∂E A ∂E A (98) Moreover. one can use the approximation described next. equation (99) expresses the variance as a homogeneous quadratic polynomial in the exposures.The CREDITRISK + Model In the first case. Instead. Define the multiplier to the given percentile as ξ where ε + ξσ = X (101) Then. the standard deviation of losses and the loss at a given percentile level from the distribution. Let ε. we can work with one sector for most of the analysis. We use the following definitions. Hence. we derive analytic formulae for mean and variance of the distributions of default events and default losses in the CREDITR ISK + Model. σ and X be the expected loss.x) Mean Variance µE µf µF = µk µG = εk σE σf 2 f(x) F(z) G(z) 2 2 2 σF σG CREDITR ISK + 53 . In order to evaluate the right hand side of equation (98). which are consistent with the notation used previously. The risk contribution can be written A RCA = E A E ∂σ 2 ∂σ . Reference Loss severity polynomial (equation 94) Default event probability generating function conditional on mean x Probability density function for mean x Default event probability generating function CREDITR ISK + Model probability generating function Symbol P(z) E(z. in view of equations (100) and (101).B (99) where σA and σB are the standard deviations of the default event indicator for each obligor. For ease of notation. we have (102) ∑ RCˆ A = ∑ (ε A + ξRC A ) = ε + ξσ A A =X (103) In the analysis below. we will concentrate on the determination of the contributions to the standard deviation. Provided the model is such that the correlation coefficients are independent of the exposures.
we also have 2 σE = µE (108) (109) By equations (105) and (106). We have G(z ) = F (P (z )) This is merely a restatement of equation (64). we obtain µF = ∫ µ E ( x )f ( x )dx = ∫ xf (x )dx = µf x x (110) Similarly. using equations (105) and (107) 2 2 2 2 2 σ F + µF = ∫ σ E + µE f (x )d x = ∫ µ E + µE f ( x )dx = µf + σ f2 + µf2 x x ( ) ( ) (111) Hence 2 σ F = µf + σ f2 (112) Equations (110) and (112) are the mean and variance of the distribution of default events. (1) 2 dz dz dz dz dz dz dz dz (114) 54 CREDIT SUISSE FIRST BOSTON . we use equation (104). by the chain rule dG dF dP (z ) = (P (z )) dz dz dz Hence 2 σG = . σ F + µF = (1) + (1) and dz dz dz2 dz2 2 2 σ G + µG = 2 2 (106) 2 2 σ E + µE = d G dG (1) + (1) 2 dz dz 2 (107) By definition of x. Also. µF = (1) and µG = (1) dz dz dz d E dE d F dF 2 2 (1) + (1) . by equation (53) (104) F ( z ) = E ( z.x )f ( x )dx x ∫ (105) For the probability generating functions E. bringing the differentiation by the auxiliary variable z under the integration sign. we have µE (x ) = x Because E(z. we have. which yields.Here µk and εk are the mean number of default events in sector k and the expected loss from sector k respectively. x) is the probability generating function of a Poisson distribution. To provide the link to the moments of the loss distribution. (1) − (P (1)). d 2G d 2F dP dF d 2P ( z) (z ) = (P ( z )) (P (z )) + dz dz2 dz2 dz dz2 2 (113) d 2F dP 2 dF d 2P dF dP dF dP 2 (P (1)) (1) + (P (1)) 2 (1) + (P (1)). F and G. by general properties of probability generating functions µE = dE dF dG (1) .
we obtain 1 2 2 2 σ G = σ F + µk − µk µk Substituting for εk . As remarked above. by equation (98) RCA = E A Hence ∂σ E ∂σ 2 = A ∂E A 2σ ∂E A 2 2ε k θ Ak µ A (119) RCA = EA ∂ 2σ ∂E A σ ∑ ε B ν B + ∑ µk B k k 2 2 EA εk = 2σ 2E A µ A + ∑ k σk µ k (120) where we have interchanged EA and νA for notational convenience. summing over sectors gives the standard deviation of the CREDITR ISK + Model Loss Distribution for the whole portfolio n 2σ σ 2 = ∑ εk k µ k =1 k + 2 ∑ ε Aν A A (118) Note that this is the standard deviation of the actual distribution of losses. dP 1 (1) = dz µk ∑θ Akε A = A εk µk and d P 1 (1) = 2 µk dz 2 ∑ θ Akε A(ν A − 1) A 2 (115) On substituting equations (112) and (115) into equation (114). CREDITR ISK + 55 . Thus. the σk denote the standard deviations of the factors driving the default rates in each sector. Hence RC A = EAµA σ E + A ∑ k σk µ k 2 ε k θ Ak (121) This is the required formula for risk contributions to the standard deviation. As in the earlier sections. from equation (121). we obtain 2 σG ( ) ∑ θ Akε A + ∑ θ Akε Aν A − ∑ θ Akε A A A A 2 = σk 2 (116) = ( 2 σk + 2 µk ) εk µ k + 2 ∑ A 2 θ Ak ε Aν A − ε k εk µk + 2 ∑ θ Akε Aν A A (117) Finally. Thus.The CREDITRISK + Model Successive differentiation of equation (94) yields A P (1) = 1 . ∑ RCA = ∑ A A E A µA σ E + A ∑ k 2 σk µ k θ Ak ε k 2 = ∑ A 2 EA µ A + σ ∑∑ A k σk µ k E Aµ A θ Ak ε k σ 2 (122) Hence. using equation (118) ∑ A RCA = σ ε ν ∑ A A + ∑ µk σ A k k εk ∑ θ Ak A E A µA 1 = σ σ σ ∑ ε Aν A + ∑ µ k A k k 2 2 σ 2 εk = σ =σ (123) as required. it can be shown explicitly that the risk contributions add up to the standard deviation of the portfolio loss distribution. Risk contributions can now be derived directly by differentiating equation (118).
where the general sector decomposition is used with n sectors.1≤ k ≤ n µB = 1. we have. We take two distinct obligors A and B and make the following definitions. Then. B and of both obligors in the time period. we derive a formula for the pairwise correlation between default events in the CREDITRISK + Model. ignoring higher powers of the default expectations and using the following approximation 6 (1− e −x A∆t )(1 − e −x B ∆ t ) ≈ x A xB In view of the sector decomposition.1≤ k ≤ n 6 The integration in (127) can be performed without using the approximation (128) to give an exact value for the correlation. we associate to each obligor its indicator function IA. the expected number of defaults of A. To define carefully the pairwise correlation over a time period ∆t. by equation (93) (128) xA = k ∑ µk θ Ak µ A =1 k ωk k ′ = n x and x B = k ∑ µkk θ Bk µ B =1 n x (129) For convenience. for any realised values of the sector means xk. The unknown term in equation (126) is the expected joint default expectation µAB . define coefficients ωkk’ by writing θ Akθ Bk ′ µ µ µ k µk ′ A B (130) 56 CREDIT SUISSE FIRST BOSTON .epA ∆t ≈ pA ∆ t θAK . we have. It is interesting to note that. the correlation ρ between default of two obligors A and B in the time period ∆t is defined by (124) ρ AB = ρ (I A . then µA. IB and the product IAB are µA. Reference Time period Instantaneous default probability Expected number of defaults Sector decomposition Obligor A Obligor B ∆t PA PB µA = 1. respectively.A13. as shown in the table. which is the random variable having the values 1 if obligor A defaults in the time period IA = 0 otherwise Then. which has a similar form to equation (54). the statistical correlation between the indicator functions of A and B in the time period.IB ) (125) That is.. writing xA and xB as in equation (93) n µ AB = ∫ . depends on knowledge of the probability generating function f. In this sense the choice of gamma distribution for the variability of the mean default rate is irrelevant to correlation considerations. µB and µAB are. while the exact integration. only the mean and standard deviation of f are required to estimate the approximate correlation given by equation (138).epB ∆t ≈ pB ∆ t θBK . If the expected values of IA. 1 ≤ k ≤ n the events of default are independent.3 Pairwise Correlation In this section. we have approximated the integrand. respectively.. µB and µAB . the standard expression for correlation reduces to the following form: ρ AB = µ AB − µ A µB 2 12 2 12 (µ A − µ A ) (µB − µB ) (126) We seek an expression for the right hand side of equation (126) in the context of the CREDITR ISK + Model. ∫ x A x B ∏ fk ( xk )dxk x1 xn k =1 (127) where. because the indicator functions can only take on the values 0 or 1. Since A and B are distinct.
. j ≠k . then depending on the sector decomposition the correlation has the same order as the term √(µAµB) in the equation. x :j ≠k .The CREDITRISK + Model Then A µ AB = ∫ . taking into account their standard deviation. Note.k xk xk′ n (132) + j ∑ω kk ∫ x k2f k ( xk )dxk ∫ k =1 xk n x j . CREDITR ISK + 57 . If it is accepted that. Equation (138) is valid wherever the likely probabilities of default over the time period in question are small.k ′ k =1 n (131) and we deduce that µ AB = ∑ ωkk ′ ∫∫ x k x k ′f k (x k )fk ( xk ′ )dxk dxk ′ ∫ ∏f j ′(x j )dx j k ≠k ′ . We note two salient features of equation (138): • • If the obligors A and B have no sector in common then the correlation between them will be zero. It is not universally valid.k ′ j =1. as suggested by historical data. j ≠k j =1.k =1k ≠k ′ ∑ ωk k ′ µk µk ' + ∑ ωkk µk2 + σ k2 k =1 n n n n ( ) (133) µ AB = However k . in particular. Therefore. in general one would expect default correlations to typically be of the same order of magnitude as default probabilities themselves.k 1k k ω µ ∑ ≠ k′ k ′ µk µk ' + ∑ ωkk ( k2 + σ k2 ) = = k 1 (134) n θ µ n θ µ ωk k ′ µk µk ' = ∑ Ak A µ k ∑ Bk B µk = µ A µB ∑=1 k =1 µ k k =1 µ k k . the ratios σk/µk are of the order of unity.k ′ n (135) Thus 2 µ AB = µ A µB + ∑ ω kk σ k k =1 n (136) Substituting for ωkk . This is the geometric mean of the two default probabilities. we obtain ρ AB This simplifies to µ AB − µ A µB = = (µ A µB 2 2 (µ A − µ A )1 2 (µB − µB )1 2 n ) −1 2 σ θ Akθ Bk µ A µB k µ k k =1 ∑ n 2 (137) ρ AB = (µ A µB ) 12 σ θ Akθ Bk k µ k k =1 ∑ 2 (138) Equation (138) is the formula for default event correlation between distinct obligors A and B in the CREDITR ISK + Model. j ≠k ∏f j ( x j )dx j n Hence µ AB = or k . This is because no systematic factor affects them both. that the value of ρAB given by the formula can be more than one if too large values of the means and standard deviations are chosen.. This corresponds to the approximation used at equation (128) .the lefthand side is clearly always less than unity while the approximating function is unbounded. ∫ ∑ ωkk ′x k x k ′ ∏ fk (x k )dxk x1 x n k .
Illustrative Example B1 B Example SpreadsheetBased Implementation The purpose of this appendix is to illustrate the application of the CREDITR ISK + Model to an example portfolio of exposures with the use of a spreadsheetbased implementation of the model. can be downloaded from the Internet (http://www. the spreadsheet implementation has been designed to allow analysis of portfolios of realistic size. B2 Example Portfolio and Static Data The three examples are based on a portfolio consisting of 25 obligors of varying credit quality and size of exposure. However. However. there is no limit. consisting of a single spreadsheet together with an addin. For illustrative purposes. 58 CREDIT SUISSE FIRST BOSTON .Illustra Examp Appendix B . Increasing the number of obligors has only a limited impact on the processing time. Details of this portfolio are given in Table 8 opposite.csfb. the spreadsheet can be used on a userdefined portfolio. in principle. we have limited the example portfolio size to only 25 obligors. The exposure amounts are net of recovery.com) to reproduce the results shown in this appendix.000 individual obligors and up to 8 sectors can be handled by the spreadsheet implementation. Up to 4. In addition. to the number of obligors that can be handled by the CREDITR ISK + Model. The spreadsheet contains three examples of the use of the CREDITR ISK + Model. The implementation.
00% 7.184 2. A table giving an example mapping from credit ratings to a set of default rates and default rate volatilities is given.50% 2.480.652.50% 15.00% Standard Deviation 0.097 4.a p Name 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Credit Rating A B C D E F G H Illustrative Example B Table 8: Example portfolio Exposure 358.00% 7.116 2.912.457 5.895 Credit Rating H H F G G G H G D D A D D H F E D C E C H H B F E The example uses a credit rating scale to assign default rates and default rate volatilities to each obligor.989 3.906 20.685 3.933.50% 10.75% 0.364 5.317.586 5.957.929 2.517.042. Table 9: Mean Default rate 1.928.60% 3.044 4.410.466.724 4. The credit rating scale and other data in the table are designed for the purposes of the example only.50% 1.475 1.312 5.727.710 1.847.80% 1.845 6.00% Example mapping table of default rate information CREDITR ISK + 59 .651 15.238.320.137.410.764.830.819 1.50% 3.322 7.596 5.75% 5.435.327 2.089.00% 15.989 5.204.00% 5.533 6. The table is shown as Table 9 below.517 4.727.00% 30.799.
1 Activating the CREDITR ISK + Model Choose one of the three example worksheets to reproduce the results. Credit risk provision. Press the button to activate the model implementation. together with the results generated by the model. countries are the sectors. Each worksheet is equipped with a macro button. the steps to reproducing these results using the model implementation are described.2 Data Input Screen On activation. This reflects the situation in which the fortunes of an obligor are affected by a number of systematic factors. countries are the sectors. For each example. In this section. which is responsible for all of the uncertainty of the obligor’s default rate. B3. Again. Full loss distribution. Activate Model B3. the model will show the Data Input Screen. 60 CREDIT SUISSE FIRST BOSTON . The examples are installed on the spreadsheet implementation. Risk contributions. as follows: • • All obligors are allocated to a single sector. This screen is used to set the worksheet ranges of data to be read in to the model and to specify the form of output data required. This assumes that each obligor is subject to only one systematic factor.B3 Example use of the Spreadsheet Implementation Three examples are given. the inputs to the model have been set to generate the following: • • • • Percentiles of loss. Each obligor is allocated to only one sector. • Each obligor is apportioned to a number of sectors. The Data Input Screen has been preset to the correct ranges corresponding to the layout of each example worksheet. In this example. each based on the same portfolio.
CREDITR ISK + 61 .3 Input Data Check The model implementation has been preset to identify errors in the data read in before the calculation commences. B3. The model implementation ensures that the data satisfies the following three criteria: • • • The sector allocation table contains only numeric data.Illustrative Example B Credit Suisse First Boston Example 1A Data Input Ranges Obligor Name Exposure Mean Default Rate Standard Deviation Range of Sectors Confirm Number of Sectors Optional Settings Use Sector 1 for Specific Risk X $B$11:$B$35 $C$11:$C$35 $E$11:$E$35 $F$11:$F$35 $G$11:$G$35 1 Check Input Data Display Options X X Preliminary Statistical Data Percentile Losses Data Output Ranges Percentiles Output Range Distribution Output Range Risk Contributions Output Range Percentile for Risk Contributions X X X $M$11:$N$11 $P$11:$Q$11 $I$11:$K$11 99 Print Percentiles to Worksheet Print Distribution to Worksheet Print Risk Contributions to Worksheet Proceed Alter Percentiles Data Entry Screen Cancel Press the Proceed button on the Data Entry Screen to proceed to the next step. Press the Proceed button on the Data Entry Screen to proceed to the next step. The decomposition of each obligor to the various sectors adds up to 100%. The user is given the option of switching off this facility via the Data Input Screen. The Input Data Check screen indicates the location of an error in the sector allocation table. A sector must contain at least one allocation entry.
Return to Input Screen Proceed Press the Proceed button on the Input Data Check Screen to proceed with the calculation.742 Press the Proceed button to proceed to the next step. Credit Suisse First Boston Portfolio Loss Distribution Summary Statistics Portfolio Aggregate Exposure Portfolio Expected Loss Portfolio Standard Deviation Amounts are stated in the input units of currency Input Data Check Screen Press the Proceed button to proceed to the next step. The model implementation now calculates the full distribution of losses.4 Portfolio Loss Distribution Summary Statistics The model implementation has been preset to display summary statistics of the portfolio loss distribution.513.5 Percentile Losses. The user can switch off this facility via the Data Input Screen. The user is given the option to switch off this facility via the Data Input Screen.Credit Suisse First Boston No Errors Were Detected In The Input Data Data Input Error Trapping Error Type Sector allocation cells with nonnumeric entries None Obligors whose sector decomposition does not sum to 100% None Sectors having zero expected loss None Input Data Check Screen Press the Proceed button on the Input Data Check Screen to proceed with the calculation.221. The model implementation now calculates the full distribution of losses. Return to Input Screen Proceed 130.863 12. 62 CREDIT SUISSE FIRST BOSTON .668. B3. B3.072 14. Loss Distribution and Credit Risk Provision The model implementation has been preset to display summary statistics of the portfolio loss distribution.
00 99.6 Risk Contributions In each example.00% 0.486 46.000 40.75 99.498.50% 1.152. If the 99th percentile level is chosen as the determining level for the Incremental Credit Reserve Cap (ICR Cap). the Annual Credit Provision (ACP) is given by the Expected Loss.742 Press the Proceed button to output the loss percentiles.503 62.000.00 97.503.089. The model implementation has been preset to calculate risk contributions by reference to the 99th percentile loss.000 60.668.000 30.478 Input Data Check Screen Press the Proceed button Proceed Return to Input Screen to output the loss percentiles.133. Loss Distribution A graph of the loss distribution has been set up on each worksheet using the results generated from the step above. the loss distribution.221.863 12.000 Loss Credit Risk Provision From the summary statistic data above.311.072 14.00 75. and the risk contributions to the worksheet.033.221. and the risk contributions to the worksheet.50% 0.90 11.50 99.513.000 50.e.062 38.50 99. CREDITR ISK + 63 .863.000.000. the loss distribution.908.181 68.612.00 95.000. B3.000 20. The risk contribution calculated by the model is defined as the marginal impact of the obligor on a chosen percentile of the loss distribution. the model implementation has been preset to output risk contributions for each obligor in the example portfolio. 2. i.540 77.00% 0 10.311.000.000 70.455 20. 130.Illustrative Example B Credit Suisse First Boston Summary Statistical Data Aggregate Exposure Expected Loss Standard Deviation Loss Percentiles 50.000.50% Credit Loss Distribution 2.00% Marginal Probability 1. then the ICR Cap is 55.128 55. 14. This setting can be altered to a different percentile via the Data Entry Screen.000.
221.939.317 from the two obligors removed.971 289. The amount of expected loss removed is exactly equal to the expected losses from the two removed obligors because expected loss is additive across the portfolio.008.602.543 326.946 179.618. The other portfolio data is unchanged.874 910. to illustrate the use of risk contributions in portfolio management as follows: • • In example 1A.530 1. all 25 obligors are included in the portfolio.969 7.655 443.478.653 156. 1A and 1B.364.946.857.850.646.944. • The 99th percentile loss from the portfolio has declined by 15.988 1.027 271.594 1.642 1.748 487.711 764.477.917 Risk Contribution 228. Thus the ACP provision in respect of the portfolio can be reduced by 3.097 123.916 241.856.231 399.960 1.899 160. The risk contribution output from example 1A is repeated in the table below. Table 10: Example 1A Risk Contributions Name 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Expected Loss 107.120 The effect on the test portfolio of removing obligors 24 and 25 is shown in table 12 below.564 1.910 434.008 from 14.202 70.325 794. obligors 24 and 25 have been removed from the portfolio.311. Table 10 shows that obligors 24 and 25 have the largest risk contributions.435 1.350 225.091 1.863 to 11.639 795.914 2.162. 64 CREDIT SUISSE FIRST BOSTON .646 from 55.197 10. Thus.528 432.047 437.720 560.754 4.773 165.654 593.967 347. In example 1B.199.B3.599 361.517.059.559 6.364.7 Using Risk Contributions For Portfolio Management Example 1 has been split into two examples. then the ICR Cap can be reduced by 15.541.758 426.526 245. This is approximately predicted by the total risk contributions of 19. Removing these obligors in example 1B has two effects on the portfolio: • The expected loss of the portfolio has been reduced by 3. The risk contributions give an estimate of the effect of removing the obligors. if the same percentile is used as the benchmark confidence level for determining economic capital.345 175.735 896. Furthermore.448 892.605 1.056.743 716.773.674.938 9.330.110. then the amount of economic capital required to support the portfolio is reduced by the same amount. if the 99th percentile loss is used as the ICR Cap for the portfolio.697 504.503 to 39.059.356 756.163.
75 99.059.056.646) % Movement 27.863 55. the percentage movements in Table 12 illustrate a general feature of portfolio risk management.0 95.541.059.856 39.517.895 35. removal of two obligors. representing 21.801) (3.618.Portfolio movement analysis Name 24 25 Total Exposure 15.8% Although the example incorporates unrealistic levels of default rates.0 97.8 Setting the Percentile Levels The model implementation provides a facility to change the percentile loss levels calculated and output by the model. B3.Illustrative Example The tables below summarise the portfolio movement and the risk details of the removed obligors.3% 21.0 99.9 Number of points required 8 CREDITR ISK + 65 . Therefore.317 Example 1A Exposure Mean 99th Percentile 130.008 Risk Contribution 9. has eliminated 27.120 19. than on the expected loss of the portfolio.801 Expected Loss 1.946.503 Example 1B 94.5 99.410. a significant reduction in the economic capital required to support a portfolio of credit exposures can be achieved by focusing on the management of a small number of obligors with large risk contributions.863.0 75.162.238.917 3.311. B Table 11: Example 1B .649.513.649.072 14. The removal of the obligors with the largest risk contributions from a portfolio has a greater impact on the portfolio risk. Credit Suisse First Boston Percentile Levels Setting OK 50. as measured by the 99th or percentile loss.5 99.5% of the expected loss of the portfolio.906 20.221.197 10.5% 27.271 11.857 Absolute Movement (35.007) (15.674.364.9 ➡ Data Entry Screen Cancel Percentiles must be chosen as numbers between 0 and 99. Thus in this case.091 1.Risk analysis of removed obligors Table 12: Example 1B .8% of the total risk as measured by the 99th percentile loss. This facility is accessed from the Data Entry Screen.
com 12123255911 stephen.dreyfus@csfb.venn@csfb.Conta Appendix C .Contacts For more information about CREDITR ISK +. please contact any of the following: Portfolio Management and Credit Derivatives London John Chrystal 441718883235 john.crystal@csfb.com Stephen Lazarus 66 CREDIT SUISSE FIRST BOSTON .com Mark Venn New York JeanFrancois Dreyfus 12123255919 jeanfrancois.com 441718884279 mark.lazarus@csfb.
com Andrew Cross Tom Wilde Contacts C C CREDITR ISK + 67 .cross@csfb.holmes@csfb.com 441718882235 tom.wilde@csfb.cts Risk Management London Mark Holmes 441718882426 mark.com 441718883839 andrew.
P.. (April 1997) CreditMetrics.. Willmot. Global Credit Research.D. 4. Office of the Comptroller of the Currency 7. Lieberman. 68 CREDIT SUISSE FIRST BOSTON .. Renaissance Risk Publications 6.Morgan & Co.. J. Lieberman.. M. 2. Carty. Moody’s Investor Services. (1994) Practical Risk Theory for Actuaries. Sally G. Madelain. C. Dana. Panjer. (March 1997) Modern CreditRisk Management And The Use of Credit Derivatives: European Banks’ Brave New World (And Its Limits). Dana. (1992) Insurance Risk Models. Kealhofer. (January 1997) Historical Default Rates of Corporate Bond Issuers. Samuel S. 19201996. Moody’s Investor Services. (November 1996) Defaulted Bank Loan Recoveries. Theodore. Moody’s Investor Services. M. (1995) “Managing Default Risk in Portfolios of Derivatives”. Marvin. Stephen. Chapman & Hall 5. Gordon E.. 8. Derivative Credit Risk: Advances in Measurement and Management. Incorporated. Harry H. Pentikäinen. 3. Society of Actuaries.Select Bibliog Appendix D .Selected Bibliography D 1. Global Credit Research. Lea V. Pesonen. 9. Standard & Poor’s. Lea V. Global Credit Research. T. (December 1996) Capital Allocation: A Study of Current and Evolving Practices in Selected Banks. Carty. (February 1997) Ratings Performance 1996 Stability & Transition. Daykin.
t g .
London E14 4QJ Regulated by SFA C D .C D S C D S C D S C CREDIT SUISSE FIRST BOSTON D S One Cabot Square.