Академический Документы
Профессиональный Документы
Культура Документы
• Tracking error
• Credit Duration
• Average Credit
Rating
This paper does not imply that the value of high quality credit what is the likelihood that another borrower will default.
analysis and stock selection in the management of portfolios Often, due to industry factors, defaults occur in groups, such
to reduce credit losses is in any way diminished. It addresses as the telecom equipment sector in Europe or the airline
the portfolio construction elements of credit portfolio industry in the US. Portfolios where the assets have high
management. default correlation are more likely to experience default
clustering than those with low default correlations.
It must always be remembered that any risk measure is only Concentration refers to holding large positions in specific
as good as the assumptions underlying the modelling. This is names. As defaults can occur randomly, the size of an
as true for the approaches suggested in this paper as it is for investors holding when an asset defaults will directly impact
the simpler ones discussed. A thorough knowledge and the loss effect on the portfolio. A large holding will cause a
understanding of the nature and implications of any large loss and a small holding a small loss.
modelling assumptions is a necessary prerequisite for
effective use of that model in real world applications. The key parameters that impact the credit risk of a portfolio
are:
Credit risk • Underlying credit risks,
• Default correlations and
What is credit risk? • Exposure positions (concentrations).
Risk is the chance that an investment will not generate its The following chart shows the actual historical default rates
expected return. In the case of bonds this may incorporate for triple-B rated credits since 1970.
interest rate risk and credit risk. Credit (or default) risk is the
risk that the specific investment or portfolio will not generate
the returns as expected, due to the failure of the borrower to
Default frequency for Baa-rated credits
repay the principal and interest when due. 1.4%
1.2%
Market risk versus credit risk
1.0%
It is important, when looking at credit risk, to distinguish Average default rate
0.8%
between credit risk and ‘market risk’. Market risk arises when 0.17% per annum
the market price of assets changes due to changing market 0.6%
factors. These may be interest rates or credit spreads. This 0.4%
paper does not focus on the market risk (interest rates or
0.2%
spreads) of bonds.
0.0%
Due to the fact that, unlike equities, bonds mature, market
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
risk only exists to the extent that an investor’s investment or
19
19
19
19
19
19
19
19
19
19
19
19
19
19
19
20
20
20
return horizon differs from the maturity profile of the
underlying investments. Thus an investor with a one year
horizon who buys a 10 year bond will face market risk when
Source: Moody’s Investor Services
they go to sell that bond (with nine years remaining to
maturity) in the market one year later. If the investment
horizon and the investment are the same then the investor The chart shows that triple-B rated credits default ‘on
will earn exactly what they expected to, as long as the bond average’ at a rate of 17 basis points per year (0.17% p.a.). But
does not default. they never default at exactly 17 basis points per year (0.17%
p.a.). Sometimes no triple-B’s default and in some years quite
When an investor’s actual return and expected return are the a few do. It is also important to note that these are
same, there is no risk. Thus, in this case the only risk faced by ‘averages’ across the entire rated universe of bonds, whereas
the investor is the risk of the borrower defaulting. portfolios will always own a much smaller number of bonds.
The impact of this concentration can be significant. While the
‘average’ triple-B default rate in 2002 was 1.21% a portfolio of
What is portfolio credit risk? 10 triple-B’s that owned WorldCom and Enron (defaulted Dec
When making a credit investment there is always the risk that 2001) would have had a triple-B default rate of 20%!
the borrower might default on its debt. For strong
investment grade borrowers this risk is very low but still Credit migration is a term often used when describing credit
exists. Therefore, anyone investing in credit must expect a portfolios. Migration refers to the changing of a credits
certain rate of default. Defaults are expected and the rating through time. A bond issued by a company rated
occurrence of a default does not directly constitute risk per single-A might experience some difficulty or additional
se. Portfolio credit risk is the risk that defaults, or losses due leverage and be downgraded to a triple-B or double-B rating.
to defaults, run at a different (especially higher) level than is Migration might also be an improvement in rating. Migration
expected. Best practice credit portfolio management is does impact the market value of assets and a less
primarily about ensuring that losses due to defaults do not concentrated portfolio (smaller positions) will be less
run at a higher rate than expected. susceptible to the impact of price changes due to specific
names migrating. However it must be remembered that
Portfolio credit risk is derived from the credit risk of the migration is a measure of market risk and not, except in the
underlying bonds, but two other factors are equally extreme of migrating to a defaulted state, credit risk. If an
important; correlation and concentration. Correlation is a investor’s investment maturity matches their investment
term used to identify things that move together, that is the horizon the only risk they face is credit or default risk. As
strength of the relationship between variables. Default migration is a market risk and not a credit risk it will not be
correlation tells us, given a certain borrower has defaulted, explicitly covered in this paper.
Colonial First State Global Asset Management 3
Thus 3.63% of issues rated BBB have defaulted within 10 years However, even if a more realistic scale is used, the ‘average’
of them being rated BBB. It can easily be seen that stepping does not consider the degree of portfolio diversification or
down in credit quality does NOT result in a linear increase in concentration which reduces the effectiveness of this
default probability. The default risk increase is significant and measure. While BBB securities are expected to default with
non-linear as the rating falls. The following chart of one, five an annual probability of 0.17% this does not mean that
and ten year default rates by Rating Category clearly displays exactly 0.17% of BBB’s will default each year. Some years
this. none will default and in other years the default rate might be
above 1%. This is the nature of risk and it is exactly this
volatility of actual default outcomes, around the long run
expected default rates, that generate portfolio risk.
the potential to deliver a significantly different ‘actual This approach has two major flaws
outcome’ between the portfolios is high.
• The first is that it is indifferent between strategies
Clearly, knowing that a portfolio has an average rating of x that lower total risk and those that increase total risk
provides no meaningful information about its risk – it may of the managed portfolio if their variance (tracking
however give you an indication of the returns you should error) to benchmark is the same.
expect to receive.
• The second is that it assumes that volatility is the
Credit duration complete and appropriate measure of risk – thereby
assuming normally distributed portfolios. The fact
Another component of ‘credit risk’ is credit duration. As with that returns from bonds are not normally distributed
interest rate duration, this gives a measure of how much a means that we need to look deeper at the
portfolio might lose if credit spreads were to widen by a basis distribution. The important part from a risk
point. It is mathematically related, among other things, to management perspective is the downside or the tail
the length of the exposure. Thus a portfolio with credit of the distribution. A portfolio could be built that
duration of 4 would expect to lose 4% if credit spreads were had small volatility but a large tail risk and this
to widen by 1% and a duration of 2 would indicate that a tracking error would not capture this risk.
portfolio would expect to lose 2% if credit spreads were to
widen by 1%. It is a measure of market risk and not credit
default or migration risk. A further problem is that most tracking error risk analysis is
done on an ex-post basis so that a risk that does not
Credit duration, as it is typically applied, assumes that all eventuate would not register as a risk. Indeed a low volatile
securities irrespective of their but catastrophic worst case outcome asset or portfolio could
present as an attractive investment alternative. For example
• rating writing insurance contracts or selling an option raises
• term to maturity premium and looks like a good return profile until a claim is
• currency made or option exercised.
• country
• industry Looking at the first flaw. The following table describes the
• level of subordination and broad return expectations for two portfolios A & B
• security type
Portfolio A Portfolio B
have credit spreads that widen by the same amount. Expected excess return 2% 2%
Expected tracking error 4% 2%
To further illustrate these assumptions, the use of credit Expected information ratio 0.5 1.0
duration to ‘measure’ credit risk assumes that if the spread on
a 10 year US single B rated subordinated airline bond Using standard portfolio measures of tracking error and
widened by 10 basis points then the spread on an AUD AAA information ratios portfolio B would be preferred to portfolio
rated, 1 year Super Senior Mortgage Backed Security would A. The following chart demonstrates the two different
also widen by exactly 10 basis points. portfolio possibilities in a different way. In this chart it would
be natural to select portfolio A as a preferred portfolio
While credit duration gives a good ‘back of the envelope’ position. However, as soon as the benchmark portfolio is
measure of what might happen to a portfolio if credit spreads introduced the picture becomes less clear. In the chart below
were to uniformly change, it makes no assessment on the the blue star represents the benchmark portfolio. Which of
likelihood of that happening and does not, in any way alternatives A or B is now preferred?
measure credit migration or default risk.
RETURN
Tracking error A B
Tracking error is the approach taken by many participants 10
when reviewing the performance of a portfolio against its
benchmark. Tracking error looks at excess return volatility as 8
a risk measure.
Tracking error measures the relative volatility of the portfolio portfolios cannot be modelled with a normal distribution and
against the benchmark (and not against zero). The cannot be adequately described with a single measure
information ratio is the ratio of the excess return to this (tracking error or otherwise).
tracking error. The idea is that a higher information ratio
demonstrates that higher returns are being earned for each The diagram below plots the theoretical future value
‘unit of risk’. Thus traditional analysis would select portfolio distribution of a credit portfolio. As can be seen a portfolio
B as the preferred portfolio even though it has higher worth $100 today is never going to be worth $200 in a year
absolute volatility. but may be worth zero. This gives the distribution the shape
we see below. Because of this typical portfolio distribution
for bonds, it is essential that, in addition to measuring the
RETURN
volatility of returns (or unexpected loss (UL)), there is a
A B measure of the downside. We must be able to describe the
length and shape of the tail risk as shown in the diagram
10 below. This will give a measure of how much ‘extreme
downside’ risk is in a portfolio.
8 ‘Alpha’ = 2% ‘Alpha’ = 2%
Probability
the modeled portfolio
value
2 6 8 RISK UL
Probability
no market moves and the initial value of the bonds is
equivalent to the expected value. The table below illustrates
the scenarios; CVaR: is the
average of losses
given that the
Probability A B Portfolio losses exceed the
(A+B) VaR number
1 0.03 70 100 170 chosen
2 0.02 90 100 190
3 0.03 100 70 170
4 0.02 100 90 190
5 0.9 100 100 200
$x Portfolio value
The expected value of bond A and B is 98.9 and the expected
value of the portfolio is 197.8. A proposed consistent measure of tail risk (CVaR)
The table below shows the profit and loss under each Just as tracking error has come to be a well understood and
scenario; uniformly described statistic, it is necessary that a similar
uniform statistic be adopted within the industry for the
Probability A B Portfolio measurement of tail risk. Without an accepted standard
(A+B) measure comparison across portfolios would not be possible.
1 0.03 -28.9 1.1 -27.8 It is important to note that even with a standardised measure
2 0.02 -8.9 1.1 -7.8 that underlying assumptions need to be understood as
3 0.03 1.1 -28.9 -27.8 differences here can lead to different results.
4 0.02 1.1 -8.9 -7.8
5 0.9 1.1 1.1 2.2 It is suggested that a 1 year 20% tail is an appropriate market
standard for this measure for the following reasons:
The 95% VaR for bonds A and B is 8.9. The 95% VaR for the
portfolio is 27.8. That is, the VaR of the portfolio is greater • Given that most fixed interest / credit funds in the
than the sum of the individual VaRs for each bond. market are sold with a ‘suggested’ investment
Diversification within the portfolio is not recognised under timeframe of 3-5 years then a tail that measures
VaR. what might be expected to happen once every five
years is a more appropriate statistic.
Another criticism of the usual adopted VaR measures is that
the confidence intervals used are so far down into the tail as • A measure of 20% does more than just measure the
to be meaningless for analysis. It may well make sense for a extreme event and because it is measuring so much
bank to maintain enough capital to sustain a 1 year in 1,000 of the distribution it inherently picks up some of the
event but for a managed fund, which doesn’t maintain ‘unexpected loss’ characteristics of the distribution.
capital, their clients (and business) would have long
disappeared with losses even getting close to this number. Once we have agreed on the part of the distribution to
So what is a more appropriate measure of tail risk? measure it is essential that the industry is consistent in what
it is measuring. We would propose that a measure of the tail
A better measure of tail risk is the use of Conditional Value at that helps to describe the shape of the distribution rather
Risk (CVaR). CVaR is a measure of the tail that helps to than just putting a value on a single point. This is often
describe the shape of the distribution rather than just putting referred to as CVaR or Conditional Value at Risk and it
a value on a single point. It measures the expected loss given measure the expected loss given (conditional on) the loss
(conditional on) the loss being greater than the VaR. being greater than the VaR.
The diagram below shows a portfolio distribution and A unified measure of tail risk can thus be defined as:
highlights the difference between VaR & CVaR.
CVaRq = VaRq + E(X - VaRq|X > VaRq)
4
Technically, this is violating the sub-additivity property. The sub-
additivity property is VaR(A)+VaR(B) ≥ VaR(A+B).
Colonial First State Global Asset Management 7
A real example It can be seen that the tracking portfolio, despite its overall
characteristics being consistent with the benchmark, and
To demonstrate the usefulness of the tail measure against despite it running an ex-ante tracking error of only 32 basis
more traditional measures we have modelled a ‘low tracking points is much more risky than the benchmark.7
error’ portfolio against its benchmark portfolio.
It is not suggested that many managers might have such a
We have used an internationally recognised global broad concentrated portfolio only that a portfolio with a very low
index and a model low tracking error portfolio constructed to tracking error might be carrying a very large amount of risk
replicate the key parameters of that portfolio. The details of against the benchmark.
the tracking error, as reported by use of the BondEdge5 tool
are also detailed. We have modelled the parameters and tail While CFS GAM uses MKMV’s Portfolio Manager tool to
risk of each of the model and benchmark and present them calculate these tail measures, as was discussed previously, a
below. number of different tools are available in the market which
calculate tail risk estimates. While they use different
assumptions and techniques it is our belief that under
Tracking Benchmark consistent modelling assumptions all these models would
derive similar results.
Issuers 41 517
Exposures 53 1,812 It must be remembered that in looking at this portfolio going
forward (and tracking its performance) it may not experience
Duration 4.31 4.29 any defaults and it may well track the index portfolio very
Average Exposure well. However this does not mean that it didn’t take
189 6
(bps) significant credit risk in doing so.
Maximum
740 32
Exposure (bps) Summary of risk measures
Average Credit
BBB A/BBB The following table presents a brief summary of the risk
Rating
measures discussed in this paper and highlights their relative
advantages.
The one month annualised tracking error was calculated
using BondEdge to be 32bps. Measure Incorporates:
Default Concentration Term to Effect of Extreme
Probability effects Maturity Diversification Event Risk
Now consider the credit portfolio statistics calculated from Average
KMV Portfolio Manager6: Rating 9 8 8 8 8
Credit
Duration 8 8 9 8 8
Tracking
Error 9 8 9 8 8
Tracking Benchmark VAR
9 9 9 8 8
Tail Risk
Expected Loss (bpts) 14 9 (CVaR 9 9 9 9 9
Unexpected Loss (bpts) 175 112
80% VaR (bps) 92 63
80% CVaR (bpts) 1,010 378
7
To model the funds the following assumptions were made:
Conclusion Definitions
Correlation: a measure of how two factors move together. Thus default
There are many tools currently used to ‘measure’ credit risk. correlation measures the impact of one borrower defaulting on another
Many of these, while benefiting from simplicity of calculation, borrower
are significantly inadequate in answering the question “How
CVaR: A unified measure of tail risk defined as:
much credit risk am I taking”. Some such as average credit CVaRq = VaRq + E(X - VaRq|X > VaRq)
ratings are not really risk measures at all. Credit duration and where VaR is Value at Risk at the q% probability level.
tracking error are measures of market risk rather than credit
‘default’ risk. While Credit VaR is a major improvement it Default probability: the likelihood of any borrower defaulting in a given period
(usually one year). Thus, based on historical observations, a triple-B security has
suffers from telling nothing about the fatness or length of the a 0.17% likelihood of defaulting in the next year
tail it is measuring.
Diversification: the spreading of investments across a variety of different
securities or borrowers.
This paper has proposed the use of CVaR as a market standard
for the measurement and, importantly, reporting of credit Duration: a measure of the average life of a fixed income investment
risk. It proposes that CVaR be measured at the 1 year 20% incorporating coupon and principal flows. The duration of a security provides a
level as an appropriate level for the funds management good guide to its expected price change given a change in its yield.
industry. Such a benchmark is necessary to enable Expected loss: the amount an investor would expect to lose, over the longer
consistency across the industry and improve reporting term, by investing in credit. It is calculated as the probability of default
transparency and overall credit risk management practices. multiplied by the amount of loss in the event of a default. Thus if the expected
loss given a default of a Baa company was 50c in the $ the expected loss would
be 0.16% x 50% thus 0.08% or 8 basis points
Information Ratio: The ratio of excess returns (of a portfolio relative to its
benchmark) to tracking error. It is intended to show the return for risk of a
portfolio.
Investment Grade
Aa3 AA-
Strong Aa1 A+
Aa2 A
Aa3 A-
Adequate Baa1 BBB+
Baa2 BBB
Baa3 BBB-
Ba1 BB+
or
Ba2 BB Speculative
High Yield
Ba3 BB- Grade
B1 B+
B2 B
B3 B-
currently highly Caa / D CCC / D
/
vulnerable/
defaulted
Distressed
Default
Disclaimer
The information provided in this document is given in good faith and is derived Tail Risk: The risk of an extreme negative event
from sources believed to be accurate. Neither Colonial First State Asset
Management (Australia) Limited, any associated companies, nor any of their Tracking error: A measure of how closely a portfolio is expected (or historically
employees or directors give any warranty of reliability or accuracy nor accept has) tracked its benchmark. Mathematically it is the standard deviation of the
any responsibility arising in any other way including by reason of negligence excess returns of the portfolio over the index.
for errors or omissions herein. This disclaimer is subject to any contrary
provisions of the Trade Practices Act. Unexpected loss: The volatility of loss around the expected loss.
This document is not financial product advice and is intended to provide VaR: Value at Risk (VaR) measures the maximum potential loss on a group of
general information only. It is not a recommendation of any securities offered securities over some time period, given a specified probability. In other words,
by Colonial First State Asset Management (Australia) Limited or any other once a probability or degree of confidence has been set, VAR is the amount
company or person. It should not be considered as a comprehensive statement which represents the statistical maximum loss for a single security or group of
on any matter and should not be relied upon as such. It does not take into securities.
account any person’s individual objectives, financial situation or needs. Any
investor accepting this information should consider whether the information is
appropriate for them and consider talking to a financial adviser before making
an investment decision.
The Funds referred to in this document are issued by Colonial First State
Investments Limited ABN 98 002 348 352. Product Disclosure Statements (PDS)
for the Funds are available from Colonial First State. Investors should consider
the relevant PDS before making an investment decision. Past performance
should not be taken as an indication of future performance.